UvA-DARE is a service provided by the library of the University of Amsterdam (http://dare.uva.nl)

UvA-DARE (Digital Academic Repository)

Common wisdom versus facts: How entrepreneurs differ in their behavioral traits from otheroccupational groups

Koudstaal, M.

Link to publication

Citation for published version (APA):Koudstaal, M. (2016). Common wisdom versus facts: How entrepreneurs differ in their behavioral traits fromother occupational groups.

General rightsIt is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s),other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).

Disclaimer/Complaints regulationsIf you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, statingyour reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Askthe Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam,The Netherlands. You will be contacted as soon as possible.

Download date: 19 Mar 2021

2Ever since the influential work of Knight (1921), much research has been devoted to entrepreneurs and what makes them such a distinct breed. As a result, a large body of research has now suggested that entrepreneurs might be fundamentally different from others in their risk appetites, (over)optimism, and their more intuitive decision-making styles. However, even though this may sound appealing, the existing empirical evidence in each of these separate areas is far from clear-cut, in contrast to the common wisdom that can be found in many blogs on the internet. The aim of this dissertation is therefore twofold: (1) To gain further insight into the unique behavioral traits of entrepreneurs in comparison to other occupational groups - more specifically managers and employees - and (2) to test the validity of the common wisdom within this area. While many interesting topics can be further explored within this framework, this dissertation focuses on the following four traits on which the academic literature has mostly focused and for which common wisdom seems to be rather prevalent: risk & uncertainty (Chapter 2), optimism & overconfidence (Chapter 3), intuitive versus contemplative decision-making (Chapter 4), and preferences for individual versus team incentives (Chapter 5). What do you think the conclusion will be? Martin Koudstaal (1982) has graduated cum laude in Economics at the University of Amsterdam and also holds an MPhil from the renowned Tinbergen Institute. Besides this, he has gained extensive work experience as a management consultant in the financial services sector at Double Effect and now works at Rabobank New York as of May 2016.

657Universiteit van Amsterdam

Comm

on Wisdom

versus Facts: How

Entrepreneurs differ in their Behavioral Traits from other O

ccupational Groups M

artin Koudstaal

Common Wisdom versus Facts:How Entrepreneurs differ in their Behavioral Traits from other Occupational Groups

Martin Koudstaal

COMMON WISDOM VERSUS FACTS:

HOW ENTREPRENEURS DIFFER

IN THEIR BEHAVIORAL

TRAITS FROM OTHER

OCCUPATIONAL GROUPS

ISBN: 978 90 5170 809 7

Cover design: Crasborn Graphic Designers bno, Valkenburg a/d Geul

Graphic design: Pieter Vonk Design, www.pietervonk.com

This book is no. 657 of the Tinbergen Institute Research Series, established through

cooperation between Thela Thesis and the Tinbergen Institute. A list of books which

already appeared in the series can be found in the back.

COMMON WISDOM VERSUS FACTS: HOWENTREPRENEURS DIFFER IN THEIRBEHAVIORAL TRAITS FROM OTHER

OCCUPATIONAL GROUPS

ACADEMISCH PROEFSCHRIFT

ter verkrijging van de graad van doctor

aan de Universiteit van Amsterdam

op gezag van de Rector Magnificus

prof. dr. ir. K.I.J. Maex

ten overstaan van een door het college voor promoties

ingestelde commissie,

in het openbaar te verdedigen in de Agnietenkapel

op donderdag 15 september 2016, te 10:00 uur

door

Martin Koudstaal

geboren te Hellevoetsluis

PROMOTIECOMMISSIE

Promotores:

Prof. dr. C.M. van Praag, Copenhagen Business School en Universiteit van Amsterdam

Prof. dr. R. Sloof, Universiteit van Amsterdam

Overige leden:

Prof. dr. R.M.W.J. Beetsma, Universiteit van Amsterdam

Dr. G. Dushnitsky, London Business School

Prof. dr. S. Estrin, London School of Economics and Political Science

Prof. dr. A.H.G. Rinnooy Kan, Universiteit van Amsterdam

Dr. J. Sol, Universiteit van Amsterdam

Faculteit Economie en Bedrijfskunde

Acknowledgements

[November 23, 2012, 12:10h]: I’m stuck in traffic on my way to the University to meet

Mirjam van Praag and Randolph Sloof. The goal of our talk will be to assess

PhD possibilities for me, on my request. While I’m trying to rush past the other

cars as quickly as possible, all of a sudden the phone rings. It’s Mirjam. A massive

“Oh no!” crosses my mind. Totally forgot that our appointment was re-scheduled to

12:10h instead of 13:00h, how stupid! I decide to hit the gas even more than I already

did, and arrive about thirty-five minutes later at the Uni, miraculously without any

scratches. I park my car literally as horrible as possible and run to the 10th floor of

the E-building on the Roetersstraat. Fortunately, I know my way here after six years

of college. I arrive about 12:50h at Mirjam’s office, where Mirjam and Randolph

have been waiting for me for forty minutes. What a start of my potential career change.

[February 7, 2013 ]: I’m hired. Yes, I agree that this sounds very strange after such

a terrible first impression, but it’s actually the truth. In retrospect, two possible ex-

planations are in my view most likely to explain this. First, Mirjam and Randolph

must have had an amazing confidence in my abilities, and I’m very grateful to them

for this. Second, I think that the match between the gaps in the literature on the one

hand and my research interests on the other hand was absolutely spot on. What a

luck. Everybody is very excited to make this research a success.

[October 1, 2013 ]: Our first survey on risk-taking is live. With the fantastic help of

Remko Wouters and Janneke van Nispen (both Synpact / Week van de Onderne-

mer) and Arko van Brakel, Jelle Ganzeveld and Fabrizio Tomasi (all VNO-NCW

De Baak), it’s been sent to almost 15,000 entrepreneurs and 6,000 managers. For em-

ployees, we work together with Marcel Paquay, Tom Grooteman, Paula van

Assema and Peter Schouten of CG Selecties. We hope to obtain about 300 com-

pletes in all categories, but actually, we don’t know what to expect. We have never

done such an experiment on such a large scale. Two weeks later, the final samples turn

out to be 910 entrepreneurs, 397 managers and 981 employees. BOOM! Everyone is

amazed about the head start of our research program. After carefully outlining the

quite complicated research design in a call of no less than 56 minutes, notary Olivier

Spier is willing to do the drawing of the prize winners of the survey. Faiza Shafi

from Pakistan builds a great and user-friendly website for us so that we can dispatch

personalized reports to all participants.

[May & December 2014, March 2015 ]: Our second, third and fourth surveys are live

(topics: optimism & overconfidence, decision-making style, and willingness to pay for

team incentives). This time with the great help of Remko Wouters and Olav van

Gogh (both Synpact / Week van de Ondernemer) and Arko van Brakel, Rolien

van Dijk, Marisja Verweij, Annemieke van Herk, Luuk Willems and Fabrizio

Tomasi (all VNO-NCW De Baak). We now know what kind of firework to expect, and

we successfully manage to keep up the good work. In Wave 2, 875 entrepreneurs, 516

managers and 667 employees complete our survey, and in Wave 3, 697 entrepreneurs,

265 managers and 969 employees participate. In Wave 4, we have 400 entrepreneurs,

155 managers and 609 employees. Olivier Spier is fortunately still willing to do the

drawings, even though this will probably never be his ‘flagship’ project.

[July 26, 2014 ]: Thanks to the experienced PR consultant Thomas Cordes, the re-

sults of our first research wave on risk taking are published on the front page of the

Dutch newspaper ‘het Financieele Dagblad’ ! Editor Rob de Lange has written a very

catchy and interesting story about our general research program and the results of the

first round. The article is well received and several people indicate by email that would

like to participate as well. Erik Boer of the Amsterdam Center for Entrepreneurship

(ACE) does a great job in promoting the article across the ACE network.

[March 16 - April 12, 2015 ]: Thanks to Mirjam, I can spend a month at the renowned

London School of Economics (LSE). Saul Estrin is an incredible host who does all

he can to make my stay as valuable as possible. I can attend interesting seminars and

meet great faculty people.

[April 19, 2015 ]: The first serious academic success is a fact; our first paper on risk

taking is accepted for publication in Management Science! A great venue for our pa-

per and considered one of the top journals among business schools, which is cool. Two

weeks earlier, our paper also received the “Best Paper Award” on the 2015 Annual

Meeting of the Academy of Management, which was great news as well. I feel very

much indebted to Mirjam and Randolph, who have once again showed the amazing

power of their combination. Also the numerous chats with my colleagues Eszter Czi-

bor, Laura Rosendahl Huber, Silvia Dominguez Martinez, Sabina Albrecht,

Thomas Buser and Joeri Sol have definitely contributed a lot, so thanks guys!

[July, 2015 ]: Glad to be a participant in the 4th Entrepreneurship Residence Week in

beautiful Oxford, which is a yearly event organized by Saul, Mirjam and Dorothy

Cooke. Was great to meet up with some of the best entrepreneurship scholars.

[August, 2015 ]: The first paper without my academic parents is a fact. Together with

Laura and Eszter, the two outstanding co-authors, we have come up with a novel

way of testing the willingness to be in a team. The results seem promising so far.

[September, 2015 ]: Participated in the 2015 SEI conference at the London Business

School, which is the most productive conference I believe I have attended. Gary

Dushnitsky and Keyvan Vakili are great hosts who not only create a fruitful atmo-

sphere, but have also read all papers and provide useful feedback.

[March, 2016 ]: It’s only a few more months before I will hopefully successfully defend

my dissertation. I’m now realizing that all of the above would not have been possible

without my family and friends. First of all, my dearest darling (and at the time of

writing: my wife) Charlotte Koudstaal, who has been a great support and believer

all along. It’s just great to be in love with you and to have such an incredible soul mate

at the same time. In a slight different vein, the same can be said about my mum and

dad Adrie & Peter Koudstaal, who always keep on supporting and contributing

to my new crazy ideas. My brothers Stefan Koudstaal and Thomas Koudstaal

and their fantastic wife and girlfriend, Jolein Overdijk and Kirsten Plucker, also

deserve a big ‘box’ / hug. I think we are a great family, and it’s a great pleasure to see

everyone being happy and very successful in his/her own authentic way. I also feel very

welcome and at home when with Paul, Jacqueline and Angelique Valkenet, and

I would like to thank them for always being so interested in my PhD and for providing

great support. Finally, I would like to thank Arthur Meilink, Lotte Wanschers,

Frank van Aartrijk, Leonie Westerman, Freek Copper, Roza van der Heide,

Marieke Groenendijk, Bert Honing, Esther Kennes, Merel Mulder, Pieter

Vonk, Patrick Stastra, Frank Smithuis, Jotham Sietsma, David Duindam,

Sanne Pasman, Maja Oomen, Leonie van der Heyden, Ron & Denise van

der Marel, Joost & Marjolein de Wit and all my former TI classmates for

all the fun and support in the past years. A successful defense is therefore definitely

attributable to you as well.

Martin Koudstaal

Contents

1 Introduction 1

1.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Sampling and Methodology . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Results and overall conclusion . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Risk & Uncertainty 11

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Measurement and sampling . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2.1 Measurement of risk, loss and ambiguity aversion . . . . . . . . 16

2.2.2 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.3 Incentives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.4.1 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.4.2 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.4.3 Second experiment with alternative elicitations of loss aversion . 33

2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Appendices Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3 Optimism & Overconfidence 71

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.2 Design, measurement and sampling . . . . . . . . . . . . . . . . . . . . 75

3.2.1 Measurement of optimism . . . . . . . . . . . . . . . . . . . . . 75

3.2.2 Measurement of overconfidence . . . . . . . . . . . . . . . . . . 76

3.2.3 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.2.4 Incentives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.4.1 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.4.2 Heterogeneity checks . . . . . . . . . . . . . . . . . . . . . . . . 88

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Appendices Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4 Intuitive versus Contemplative Decision-Making 101

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

4.2 Measurement and sampling . . . . . . . . . . . . . . . . . . . . . . . . 104

4.2.1 Objective measures of decision-making . . . . . . . . . . . . . . 104

4.2.2 Subjective measures of decision-making . . . . . . . . . . . . . . 106

4.2.3 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.2.4 Incentives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

4.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.4.1 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.4.2 Heterogeneous effects of response times on contemplativeness . . 113

4.4.3 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . 117

4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

Appendices Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

5 Preference for Invididual versus Team Incentives 125

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.2 Related literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.3 Context and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

5.3.1 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

5.3.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

5.4 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

5.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

5.5.1 Determinants of team choice in the Baseline treatment . . . . . 142

5.5.2 Comparison of team choice between the treatments . . . . . . . 149

5.6 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Appendices Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

6 Summary 161

Bibliography 164

Samenvatting (Summary in Dutch) 183

Chapter 1

Introduction

Ever since the influential work of Knight (1921), much research has been devoted to

entrepreneurs and what makes them such a distinct breed. Interestingly, while Knight

(1921) was mostly pondering why entrepreneurs could earn above normal returns, on

average the opposite holds true in reality (see e.g. Hamilton (2000), although according

to Levine and Rubinstein (2013) the picture is different for incorporated entrepreneurs).

In fact, Hall and Woodward (2010) show that the expected value of an entrepreneurial

venture is actually negative when taking into account the low probability of success,

the high probability of zero exit value and one’s outside options - meaning that peo-

ple should shy away from entrepreneurship instead of embarking on it. Much of the

subsequent research has therefore tried to understand what behavioral traits might

drive individuals to become an entrepreneur despite this poor ex-ante prospect. As a

result, a large body of research has now suggested that behavioral traits like risk ap-

petite, (over)optimism, and a more intuitive decision-making style might explain this

phenomenon, and that entrepreneurs are fundamentally different from others in these

areas. However, even though this may sound appealing, the existing empirical evidence

in each of these separate areas is far from clear-cut, thus leading Astebro et al. (2014)

to conclude that: “Indeed, reviewing the evidence on the roots of entrepreneurship,

what surprises us most is how little we really know.” (p. 65).

While the academic literature has therefore been more conservative in its language,

the opposite seems to apply to a number of blogs and internet websites. These still con-

fidently stick to the common wisdom and spread out statements like: “Entrepreneurs

thrive on risk and uncertainty. To entrepreneurs, risk and uncertainty are part of the

game of entrepreneurship; risk is what makes the game exciting. Managers on the other

hand are conservative and detest risk; they simply avoid it” or: “I am an optimist. I

think you have to be, to be an entrepreneur”.1 Or how about this: “Entrepreneurs, like

1These two statements were taken from: http://www.mytopbusinessideas.com/entrepreneurs-vs-managers/, and http://yourstory.com/2013/09/mark-zuckerberg-techcrunch-disrupt-sf/.

1

other managers, have different management and leadership styles. Many entrepreneurs

make their decisions based primarily upon their “gut”, their intuitive feeling about what

the best choice should be”. And finally: “They [entrepreneurs] are the lone wolves of

the business world, relying only on themselves for inspiration and successful at inno-

vating in isolation; until a business idea takes off and then requires the input of skills,

expertise and manpower”.2

However, as mentioned before, this common wisdom is not (yet) fully supported by

the existing empirical ‘stylized’ facts. The aim of this dissertation is therefore twofold:

(1) to gain further insight into the unique behavioral traits of entrepreneurs in compar-

ison to other occupational groups - more specifically managers and employees - and (2)

to test the validity of the common wisdom as briefly illustrated in the previous para-

graph. While many interesting topics can be further explored within this framework,

this dissertation focuses on the following four traits on which the academic literature

has mostly focused and for which common wisdom seems to be rather prevalent: risk

& uncertainty (Chapter 2), optimism & overconfidence (Chapter 3), intuitive versus

contemplative decision-making (Chapter 4), and preferences for individual versus team

incentives (Chapter 5).3 Before elaborating on these four behavioral traits, the remain-

der of this introduction will first provide the contributions, sampling and methodology,

overall findings and limitations of our research program.

1.1 Contributions

We believe that our research program contributes to the literature in a number of

ways. First, while entrepreneurs have generally been compared to the population at

large, we have used a double control group of managers and employees instead. In

general, we think it makes sense to compare entrepreneurs to other labor market par-

ticipants, since both try to make a living from their work but differ in the way they

do so. The reason for having a double control group is that entrepreneurs and man-

agers are arguably more similar to each other, while employees are included to compare

our study to previous studies. We consider the comparison between entrepreneurs and

managers of particular interest, since both are responsible for taking strategic decisions

and for directing their employees / subordinates, whereas employees in contrast typi-

cally have more operational tasks (see also Table 1.1 for an admittedly simplified and

2The link for the latter quote is: http://www.entrepreneurcountryglobal.com/zoo/item/entre-preneurs-from-lone-wolf-to-pack-leader, and for the former: http://www.forbes.com/sites/ericbasu/2014/03/02/metrics-vs-intuition-which-is-most-important-in-a-startup/#2c0a0b0649ea.

3See e.g. Astebro et al. (2014) for risk preferences (Chapter 2), overconfidence (Chapter 3), andnon-pecuniary benefits such as autonomy (Chapter 5). Chapter 4 is related to e.g. Busenitz andBarney (1997), Levesque and Schade (2005), Burmeister and Schade (2007), and Kahneman (2011).A more detailed overview of the relevant literature will be provided in each of the chapters.

2

rough comparison). Also in terms of background characteristics such as age, gender

and education, entrepreneurs and managers are likely to be similar (and our findings

indeed show this), in contrast to when comparing entrepreneurs and employees. If such

observable differences would extend to unobserved differences such as perseverance or

wealth, no fair comparison can be made between entrepreneurs and non-entrepreneurs

if the latter are proxied by a sample of employees.

Table 1.1 Simplified Differences Between Entrepreneurs, Managers & Employees

Entrepreneurs Managers Employees

Bearing responsiblility for taking strategic decisions Yes Yes No

Bearing responsibility for taking risks Yes Yes No

Bearing responsiblility for hiring and managing personnel Yes Yes No

Bearing full financial consequences of undertaken risks Yes No No

Second, we obtain our results among large samples of entrepreneurs, managers and

employees, thus alleviating concerns that the results are difficult to generalize. Third,

and foremost, our surveys include measures that are not only novel in the entrepreneur-

ship literature, but which also prove to be key differentiating factors that might help

explain some of the previous mixed results (like for instance loss aversion as discussed

in Chapter 2). In summary, we believe that we contribute to the literature through an

enhanced and more encompassing assessment of the differences between entrepreneurs

and comparable others.

1.2 Sampling and Methodology

To gain substantial reach among entrepreneurs, managers and employees, we collabo-

rated with three business partners in each of the four research waves that were executed

between October 2013 and May 2015. For entrepreneurs, we worked together with

“Synpact”, a company that has the digital Rolodex of a selection of small and medium-

sized enterprises, including 15,000 entrepreneurs in the Netherlands. The Rolodex is

supported by frequent contacts through a wide variety of training programs and confer-

ences. For managers, we collaborated with a large and highly reputed training center

called “De Baak”, which is part of the largest and influential employers’ organization

in the Netherlands (“VNO-NCW, MKB-Nederland”). The training center was willing

to send our invitations to all 5,888 managers they have in file. Finally, the employees

3

were recruited via a Dutch market research agency who approached representative

samples of their database of over 70,000 Dutch employees.

Invitations to participate were sent out by each of the three business partners after

approval by us. As will be further described in each of the chapters, all groups were

given exactly 14 days to respond and a reminder was sent out after 7 days. Response

rates were generally in the range of 5 - 15%, which is comparable to e.g. Graham et al.

(2013) among these types of subjects. Moreover, the subsamples of entrepreneurs, man-

agers and employees were generally quite consistent in their background characteristics

across the four research waves, despite the fact that at least 53% of the participants

had not participated in any wave before.

As for the methodology, all waves relied on the same research setup, which is a com-

bination of an online survey and a ‘lab-in-the-field’ experiment (see below a screenshot

of our loss aversion measure in Chapter 2).

Such a combination of an online survey and a lab-in-the-field experiment is not uncom-

mon in the literature (see e.g., DeMartino and Barbato, 2003; Block and Koellinger,

4

2009; and Graham et al., 2013), and combines the benefits of a controlled incentivized

experiment with representative samples of established entrepreneurs, managers and

employees (instead of students as is often the case in laboratory experiments). Hence

we can directly observe what the actual decision-makers of interest, i.e. entrepreneurs

and managers (and also employees), decide instead of trying to extrapolate this from

student behavior in the lab. A second and related benefit is that we have a much more

diverse sample in terms of age, education, experience and income as compared to an

average student sample, thus allowing us to assess the impact of such background char-

acteristics. Furthermore, while a large field dataset with actual behavior would have

been better in the sense that it records actual instead of arguably hypothetical behavior

in a stylized setting, we believe it is very hard - if not, impossible - to find comparable

situations where entrepreneurs and managers have to take the exact same kinds of

decisions under the exact same circumstances. Yet to partly cope with this potential

critique, we did try to elicit actual behavior by linking the decisions in the surveys

to substantial incentives. For instance, when participants preferred a risky bet over a

fixed amount, it meant that they could actually win or lose a substantial amount of

money, just as in real life. Arguably our setup thus compares ‘actual’ decision-making

of entrepreneurs, managers and employees whilst using the benefits of a ‘controlled’

environment.

The incentives in each of our individual research waves were sizable. However, due

to budget limitations and a large number of participants in each wave, we opted to

pay out only a fraction of the participants and only a random selection of the decisions

made. Again, such a practice is not uncommon in the experimental economics literature

and has been shown to lead to similar findings as when paying out all participants or

all decisions (see e.g., Bolle, 1990; Starmer and Sugden, 1991; Cubitt et al., 1998; and

Laury, 2006). The actual amounts won by each of the 20 to 25 winning participants

per wave were in the range of €100 - €785 and were on average €340. The incentive

schemes were not always the same, though, so each of the individual chapters will

further elaborate on the exact specifications used. More generally, we will do the same

for the exact measurement, sampling, and descriptive statistics, such that every chapter

can be read independently.

1.3 Results and overall conclusion

Chapter 2 discusses our study on risk attitudes of entrepreneurs, managers and employ-

ees. Theory predicts that entrepreneurs are more inclined to accept risk and uncertainty

than others, but the empirical evidence is mixed. To better understand the unique be-

havioral characteristics of entrepreneurs and the root causes of these mixed results, we

5

compare entrepreneurs to managers and employees in terms of risk, loss, and ambiguity

aversion (n = 2,288). The results indicate that entrepreneurs perceive themselves as

less risk averse than managers and employees, in line with common wisdom. However,

when we examine our experimental incentivized measures that are based on real deci-

sions, the differences are subtler. Entrepreneurs are only found to be unique in their

lower degree of loss aversion, and not in their risk or ambiguity aversion. We find that

this combination of results might be explained by the fact that perceived risk attitude

is not only correlated to risk aversion, but also to loss aversion. Apparently economists

use a more narrow definition of risk than the forces that drive the behavior of (risk

taking) entrepreneurs.

In Chapter 3 we examine differences in optimism and overconfidence between the

three occupational groups (n = 2,058). We use two survey-based optimism measures

adopted from the psychology literature (dispositional optimism and attributional style),

and two incentivized overconfidence measures adopted from the psychology and eco-

nomics literature (overestimation of one’s own ability and overestimation of a future

stock market closing price). We find that entrepreneurs are more optimistic than others

in their dispositional optimism and in their attributional style when bad events occur.

For the incentivized measures of overconfidence we find no differences between en-

trepreneurs and managers, although both are more prone to it than employees. Finally,

exploration of the within-group heterogeneities yields that high levels of optimism need

not necessarily be unique to entrepreneurs; we find that successful entrepreneurs and

successful managers are both more optimistic but not more overconfident than their less

successful peers. Taken together, it therefore seems that optimism and overconfidence

are traits of (successful) decision-makers in general rather than just entrepreneurs.

Chapter 4 examines how contemplative entrepreneurs, managers and employees are

(n = 1,931). To measure this, we use two well-known subjective measures taken from

the psychology literature and two objective measures based on response times and

the number of contemplative choices made in a set of games taken from Rubinstein

(2016). The latter two measures are directly based on Kahneman’s (2011) System 1

and System 2 type of thinking. The results of this study indicate that entrepreneurs

make less contemplative choices than managers, but not than employees. Surprisingly,

however, we do not find any significant differences in response times between the three

occupational groups. A potential explanation for this inconsistency might be that the

effect of response time on making contemplative choices is not homogeneous, and we

indeed find that the (positive) link between response times and contemplativeness is

significantly stronger for entrepreneurs and managers than for employees. Taking this

into account, entrepreneurs are not only less contemplative than managers, but also

less contemplative than employees whenever their response times are low. The two

6

subjective measures show a similar pattern; entrepreneurs score highest on Faith in

Intuition, while managers score highest on Need for Cognition. We therefore conclude

that entrepreneurs mostly differ in their decision-making styles from managers, and

somewhat less so from employees.

Finally, Chapter 5 deals with preferences for team incentives (n = 1,164). To

measure these, we include two between-subjects treatments: one “baseline” treatment

where only payoff autonomy is at stake and another “joint investment” treatment where

both payoff autonomy and riskiness are at stake. The results of the baseline treatment

first indicate that the general willingness to pay for team incentives is positively affected

by education and optimistic beliefs about the teammate’s performance, while it is

negatively affected by risk aversion and overconfidence. Occupational category, ability

and gender do not have a significant impact. Hence, we do not find any evidence

that entrepreneurs in general are willing to pay more or less for team incentives than

others. Furthermore, when including the data of the “joint investment” treatment,

it surprisingly shows that the loss in decision rights does not automatically lead to

a discount in the willingness to pay for team incentives. Further in-depth analysis

however reveals that there is one exception to this rule; we find that entrepreneurs

who perceive their potential teammate’s risk appetite as too different from their own,

do have a significantly lower willingness to pay. Overall, the latter might thus help

understand: (i) in which specific cases entrepreneurs prefer to work alone and (ii) why

venture teams often have homophilic risk preferences.

Taking all results of the individual chapters together, the conclusion seems war-

ranted that the differences between entrepreneurs, managers and employees are subtler

than expected by common wisdom. In part this is due to the methodology used. We

find that the differences between the three occupational groups are much smaller on the

objectively verifiable measures than on the subjective measures. In other words, while

the self-assessed measures generally confirm the common wisdom, the results on the

objective measures often only show marginally significant differences and sometimes

no significant differences at all. Especially when comparing entrepreneurs to man-

agers, it is found that the differences on the objective measures are smaller than when

comparing entrepreneurs to employees. Following the heated debate in the literature

on the definition of an entrepreneur, one might suspect that our results are different

when using stricter definitions of an entrepreneur and/or of a manager. However, we

generally find that this does not have a substantial impact (see in particular Chapters

2 and 4). The only exception is found in our optimism study in Chapter 3, where

successful managers (e.g. CEOs) cannot be distinguished from entrepreneurs. But in

all other cases, the general differences appear to be robust to using stricter definitions.

As such, our findings keep indicating that entrepreneurs differ only marginally in their

7

behavioral traits from managers (and employees). But the behavioral traits in which

they do differ - being mostly specific elements of decision-making such as loss aversion

and optimism when dealing with bad events - have not previously been documented

systematically in the literature.

1.4 Limitations

One of the main limitations of our study is the descriptive nature of our data, thus

only allowing some correlational observations without making causal inferences. As

such we cannot distinguish whether the differences in behavioral traits between en-

trepreneurs, managers and employees have been developed during entrepreneurship

and employment spells, or already existed before that. For instance, we do not know

if it are the most optimistic individuals who self-select into entrepreneurship, or rather

that entrepreneurs become more optimistic once they turn entrepreneurs. Similarly so,

little can be learned from our study about the potential causal relationship between

these traits and (personal) success. Nevertheless, we believe that our results are useful

given that the entrepreneurship literature is still very much in its explorative stage,

especially when considering empirical validations of theoretically predicted differences

between entrepreneurs and other occupational groups (see e.g., Kihlstrom and Laffont,

1979; Landier and Thesmar, 2009; and Lee, 2015). Additionally, as already noted in

the opening paragraph of this thesis, there is still much to be learned about the unique

behavioral traits of entrepreneurs and about the causes of the mixed evidence so far

(Astebro et al., 2014).

Another limitation of our online surveys is that the experiments are not as con-

trolled as in the laboratory. We thus cannot verify whether entrepreneurs or managers

themselves have filled out the surveys, or whether we have for instance dealt with their

sons or daughters all the time. However, given that most participants took approxi-

mately 10 to 30 minutes to complete the survey, we are not particularly worried about

this. Participants also extensively used the feedback box to personally thank us for the

interesting questions, thus further alleviating the aforementioned concern. Moreover,

even though the exact ‘loss of control’ can still not be fully determined, we consider it

a significant gain that we have more diversity in background characteristics compared

to an average student sample. On top of that, we also value the large numbers of

established entrepreneurs, managers and employees, hence those people who actively

participate in the labor market.

A third limitation is the issue of sample selection. This arises due to the fact that our

response rates are only 5 - 15%, which makes it difficult to assess the impact of possible

non-representativeness. To alleviate this issue, we have therefore compared responders

8

and non-responders to each other on their age and gender (which unfortunately are

the only available items for non-responders) and find no significant differences between

the two groups. Moreover, as mentioned before, most of the participants in the survey

were new and in total we have welcomed 2,408 entrepreneurs, 1,086 managers, and

2,486 employees in one or more of our surveys. We therefore find it promising that the

overall response rates are more in the order of 15 - 30%. Furthermore, we also find it

supportive that there are no major significant differences in background characteristics

between the subsamples of the four waves, arguably suggesting that they might be

representative - of course provided there are no systematic and consistent self-selection

mechanisms across waves.

A fourth and final limitation is that our study has only focused on Dutch en-

trepreneurs, managers and employees. Hence, we do not know whether the differences

encountered in our studies are applicable to the Dutch population only, or whether

there might also be cultural differences that we have not accounted for. This would

therefore be an interesting avenue for follow-up research.

These limitations notwithstanding, we believe that our research program not only

contributes to the literature by further clarifying the unique behavioral traits of en-

trepreneurs in comparison with managers and employees, but also by offering explana-

tions for some of the previous mixed results in this area. Taken all of this into account,

entrepreneurs do indeed appear to be a distinct breed, but much less so than common

wisdom sometimes portrays them to be.

9

10

Chapter 2

Risk & Uncertainty

This chapter is based on Koudstaal, M., R. Sloof, and C.M. van Praag (2015), “Risk,

Uncertainty and Entrepreneurship: Evidence From a Lab-in-the-Field Experiment”,

which is forthcoming in Management Science.

2.1 Introduction

One of the most salient dimensions of entrepreneurship is risk and uncertainty. Eco-

nomic theory predicts that entrepreneurs, as business owning residual claimants, are

less averse towards risk and uncertainty than others. Entrepreneurs assume business

risks in uncertain environments. Their income, wealth, satisfaction and social status

are dependent on the outcomes of their decisions in uncertain situations (Cantillon,

1755; Knight, 1921; Kirzner, 1973; and Kihlstrom and Laffont, 1979). On top of that,

most of the entrepreneurs’ investment portfolios are totally undiversified (Moskowitz

and Vissing-Jorgensen, 2002), also due to capital constraints in the market for en-

trepreneurial finance (e.g., Evans and Jovanovic, 1989; Holtz-Eakin et al., 1994a,b;

Hvide and Møen, 2010; Fairlie and Krashinsky, 2012; and Schmalz et al., 2013).

Notwithstanding this theoretical prediction, the body of empirical evidence on risk,

uncertainty and entrepreneurship is rather mixed (see Holm et al., 2013, Appendix

Table). To reconcile these earlier findings, we conduct a lab-in-the-field experiment

among 910 entrepreneurs, 397 managers and 981 employees in the Netherlands. To

obtain additional measures of loss aversion, we run an additional experiment among

different samples of 697 entrepreneurs, 265 managers and 969 employees.

The recent study by Holm et al. (2013) is most related to ours. They also perform

a large-scale lab-in-the-field experiment with incentives to determine how attitudes

towards risk and uncertainty distinguish entrepreneurs from others. They distinguish

between strategic and non-strategic risk. Strategic risk covers measures of trust and

11

competition. Non-strategic risk is measured in terms of risk aversion and ambiguity

aversion. They find that entrepreneurs are more willing to assume strategic risk but

are not more willing to assume risks lacking a strategic, interactive character. Our

study is distinct from theirs in three ways. First, we compare entrepreneurs (in a

Western country as opposed to China) to managers as well as employees and not to

the local population at large. Second, we use both a survey-based and an experimental,

incentivized measure of risk aversion. Third, we also measure loss aversion and find

that this is the most important difference between entrepreneurs and managers in the

domain of risk and uncertainty.

More generally, our study can be characterized by the following four distinguishing

features. First, we elicit peoples’ risk attitudes using two different measures: one is

an ‘objective’ measure which is incentivized and experimental, based on a multiple

pricelist (MPL) elicitation method (in the style of Holt and Laury, 2002). The other

is ‘subjective’, i.e., survey-based and self-assessed (Dohmen et al., 2011). Both are

well-established within their categories and have been extensively validated and used.1

So far, studies testing differences in risk attitudes between entrepreneurs and others

have either used an incentivized experimental measure in the Holt and Laury style, or

a non-incentivized self-assessed survey-based measure in the spirit of Dohmen et al.

(2011). Interestingly, all studies using experimental measures of risk aversion find no

differences between entrepreneurs and the control group, whereas most of the other

studies do find differences supporting the common wisdom that entrepreneurs are less

risk averse. By using both measures we can contribute to the explanation of the mixed

findings so far.2

Second, besides comparing entrepreneurs and others with respect to risk, we also try

to understand in what related aspects entrepreneurs and managers are different. We

consider both loss aversion, allowing an asymmetric effect of losses and gains on peoples’

utility, and ambiguity aversion, i.e. probabilities are unknown and there is genuine

uncertainty in the Knightian sense.3 By relating the three incentivized experimental

1See Filippin and Crosetto (2014) for a meta-analysis of studies using the Holt and Laury measureto relate risk to gender. For a validity test of the Dohmen et al. (2011) measure, see Dohmen et al.(2007), Bonin et al. (2007), Caliendo et al. (2009), Lonnqvist et al. (2014) and Beauchamp et al.(2012). Overall, the Dohmen question scores highly on re-test reliability within-person and has beenshown to be virtually stable over re-test intervals ranging from three weeks up to almost two years (seeDohmen et al., 2007; Lonnqvist et al., 2014). However, a recent study by Brachert and Hyll (2014)shows that occupational choices may affect the Dohmen test outcomes.

2Examples of studies using Holt and Laury style elicitation of risk attitude are Elston et al. (2006),Macko and Tyszka (2009), Sandri et al. (2010), Burmeister-Lamp et al. (2012) and Holm et al. (2013).Examples of studies on risk and entrepreneurship using non-experimental measures of risk attitudeare Brockhaus (1980), Hull et al. (1980), Caird (1991), Begley (1995), Koh (1996), Sarasvathy et al.(1998), Stewart Jr. et al. (1999), Van Praag and Cramer (2001), Uusitalo (2001), Cramer et al. (2002),Djankov et al. (2006, 2007), Caliendo et al. (2010), Hvide and Panos (2014) and Skriabikova et al.(2014). See also Appendix A (Appendices Chapter 2) or Astebro et al. (2014) for further reference.

3Gachter et al. (2010) is the only study we are aware of that also compares the degree of loss aversion

12

measures of uncertainty (risk, loss, and ambiguity aversion) to the survey-based self-

assessed measure of risk aversion, we can extend our understanding of the relationship

between objective and subjective measures of risk. For instance, we find that subjects’

views of what is risk aversion is in fact a mixture of what economists call risk, loss,

and ambiguity aversion.

Third, we use a double control group. Instead of comparing entrepreneurs with

the general population, we use two tightly defined control groups, viz. managers and

employees.4 We are especially interested in the first control group. Behavioral char-

acteristics of managers and entrepreneurs have been compared in various studies (e.g.

Brockhaus, 1980; Schere, 1982; Begley, 1995; and Stewart Jr. et al., 1999), because the

two groups are arguably very similar. Both are responsible for strategic and complex

decisions and are managing the employees in their companies (if any). Therefore they

are likely to be similar in terms of many observable aspects, such as education, age, and

labor market participation. We indeed observe that the managers and entrepreneurs

in our sample are very similar, whereas the differences in background characteristics

with employees are sizeable. If these differences extend to unobserved characteristics,

such as perseverance or wealth, no fair comparison can be made between entrepreneurs

and non-entrepreneurs. Therefore, employing two relatively similar control groups al-

lows for a potentially cleaner test of behavorial differences between entrepreneurs and

others. Admittedly, managers might also be more similar to entrepreneurs in terms of

their attitudes towards risk and uncertainty. This can be inferred, for instance, from

the fact that they are likely to self-select into positions with strong(er) incentive pay.

In that sense, a comparison between entrepreneurs and managers might lead to un-

derestimating the true differences between entrepreneurs and comparable others. It is

thus important to use a more general control group, too. Using different control groups

may then show to what extent differences are related to the control group used.5

across occupational groups. They find that entrepreneurs are less loss averse on average than othersin the risky choice category. Moreover, managers appear less loss averse than blue-collar workers butnot than white-collar workers. The degree of ambiguity aversion of entrepreneurs has been comparedto students and non-entrepreneurs by Koh (1996), Macko and Tyszka (2009) and Holm et al. (2013)and with managers by Schere (1982). With the exception of Holm et al. (2013), who do not report asignificant difference, the general finding seems to be that entrepreneurs are better able to cope withambiguous situations than both managers and non-entrepreneurs are.

4Many studies have used rather unspecified control groups, such as Van Praag and Cramer (2001),Uusitalo (2001), Cramer et al. (2002), Elston et al. (2006), Djankov et al. (2006, 2007), Macko andTyszka (2009), Caliendo et al. (2010), Sandri et al. (2010), Burmeister-Lamp et al. (2012) and Holmet al. (2013).

5Many studies that have compared entrepreneurs and managers are relatively old and rely on smallsamples and self-assessed measures of risk attitude. The overall findings are mixed, too. Brockhaus(1980) found no differences between the two groups, whereas both Begley (1995) and Stewart Jr.et al. (1999) report lower levels of risk aversion among entrepreneurs than managers. Furthermore, ameta-analytical review by Stewart Jr. and Roth (2001) concludes that managers are more risk aversethan entrepreneurs, although this conclusion is challenged by Miner and Raju (2004) who concludethat the role of risk propensity in entrepreneurship remains unresolved. In a comparison of managers

13

Fourth, following the debate in the literature about who can be considered an

entrepreneur (see for instance Parker, 2009; Hurst and Pugsley, 2011; Levine and

Rubinstein, 2013; or Henrekson and Sanandaji, 2014), we verify our main findings

by using various alternative definitions of the entrepreneur. In our basic sample, an

‘entrepreneur’ is someone who founded, inherited or has taken over a company that

s/he is currently (co-)managing and of which s/he has at least 5% of the shares.6 We

use alternative subsamples that are based on ‘stricter’ definitions of entrepreneurship

(see Lindquist et al., 2015), i.e. those that are arguably more successful and thus more

similar to the ‘Schumpeterian’ entrepreneur. Subsamples used incude for instance: (i)

incorporated entrepreneurs (Levine and Rubinstein, 2013), making up almost half of

the sample, (ii) entrepreneurs with an above median number of employees and (iii)

entrepreneurs with above median income. In comparable ways, we also use various

more selective definitions of our control groups. Managers in the basic sample are

defined as employees in firms not started up by the respondent, and having at least

two direct reports under their responsibility. The stricter definitions limit the sample

to: (i) CEOs (17%), (ii) managers with an above median number of direct reports and

(iii) managers with above median income. Finally, employees are the people who work

in organizations and do not belong to the groups of entrepreneurs and managers (when

using the baseline, i.e., least ‘strict’ definitions for entrepreneurs and managers).

Our findings tell the following story. Entrepreneurs perceive themselves as more

risk tolerant than managers who see themselves, in turn, as being more risk tolerant

than employees. This ranking is consistent with most of the previous studies using sub-

jective measures of risk. However, based on the objective MPL risk aversion measure,

entrepreneurs and managers have similar risk attitudes, but are both less risk averse

than employees. When analyzing the differences in loss - and ambiguity aversion across

the three groups, we show that loss aversion is the missing piece. Whereas all three

groups have similar degrees of ambiguity aversion, entrepreneurs have a significant

lower level of loss aversion than the two other groups. We reconcile these different

findings by relating the subjective risk measure to all three experimental measures.

All three appear to be strongly related to what people self-assess to be their risk atti-

tude. Respondents thus have a notion of ‘risk’ that is different from economists, and

more a mixture of risk and uncertainty. Hence, not only could a distinct degree of risk

aversion of managers and entrepreneurs explain the differences in their self-assessed

risk attitude, these differences may also relate to differences in loss aversion or ambi-

and employees, Graham et al. (2013) show that managers have a lower risk aversion than employees.6Five percent is the cutoff ownership that the tax authority calls ‘a substantial interest’. In our

sample, 88% (65%) of the entrepreneurs in our sample holds at least 30% (51%) of the companyshares.

14

guity aversion.7 All these results are independent of the various definitions we use of

entrepreneurs and managers. In some cases, limiting the sample to more successful

entrepreneurs even strengthens the results.

The loss aversion measure we use in our first experiment records subjects’ willing-

ness to accept a small stakes mixed prospect. This measure can be reasonably criticized

on various grounds. We therefore test the robustness of our results in another large ex-

periment (n = 1, 931). Again it turns out that entrepreneurs are significantly less loss

averse than both managers and employees are. The cleanest evidence comes from com-

paring WTA-WTP gaps (for a fancy bread tray) among the three groups of interest.

Here entrepreneurs also have the lowest loss aversion in riskless choices.

Our two main conclusions are basically two sides of the same coin. First, en-

trepreneurs do differ from managers and employees in their attitude towards risk and

uncertainty, but in a rather subtle way. Second, subjective self-assessed measures of

risk attitude measure more than the economists’ strict notion of risk aversion alone.

The distinguishing trait of entrepreneurs thus becomes apparent only after realizing

that there is more to risk and uncertainty than risk aversion per se.

We think it is rather intuitive that entrepreneurs are indeed different from managers

and employees in the way they deal with risk and uncertainty and that the difference

is related to how losses ‘loom larger than corresponding gains’ (Kahneman and Tver-

sky, 1979; Kahneman and Tversky, 1984; and Tversky and Kahneman, 1992). The

entrepreneur’s position is one in which much more is at stake to be lost than in the role

of a manager. However, our study cannot reveal why entrepreneurs are found to be

less loss averse than managers. Although the general consensus tends to be that pref-

erences in the domain of risk and uncertainty are stable (see e.g. Borghans et al., 2008;

Sahm, 2012; and Fouarge et al., 2014), a recent study by Brachert and Hyll (2014)

casts doubt on the stability of these preferences. Therefore, the descriptive nature of

the study prevents us from drawing causal conclusions. Entrepreneurs might either be

less loss averse types or might become less loss averse when becoming an entrepreneur.

In a similar vein, a managerial context might also affect managerial loss aversion con-

sidering the asymmetry in blame and credit within organizations (Swalm, 1966; and

Kahneman and Lovallo, 1993). We acknowledge that this is a limitation of our study.

This limitation notwithstanding, we believe that our study not only contributes to

the literature by further clarifying the unique behavioral features of entrepreneurs in

comparison with managers and employees, but also by offering an explanation for the

previous mixed results in this area.

7An alternative explanation of the differential difference between subjective and objective riskmeasures across entrepreneurs and managers might be demand effects based on stereotypes, despiteour careful wording in the surveys.

15

In what follows, Section 2.2 first discusses design and measurement issues. Section

2.3 provides the descriptive statistics of our sample, while Section 2.4 reports the em-

pirical findings, including those of our second experiment with alternative elicitations

of loss aversion. Section 2.5 concludes.

2.2 Measurement and sampling

2.2.1 Measurement of risk, loss and ambiguity aversion

Entrepreneurship is associated with risk bearing, uncertainty, gains and losses. The

classic economists and philosophers who laid the foundation of thinking about en-

trepreneurship all but Schumpeter defined the entrepreneur either as a risk bearer

(Cantillon, 1755; Say, 1803; and Marshall, 1930), an uncertainty bearer (Knight, 1921),

or as agents who are less inclined to avoid losses (Knight, 1921; Marshall, 1930).8 In-

tuitively these three different concepts can be understood as follows.

Risk aversion is a concept with a very specific meaning in economics. It is the

willingness of people to sacrifice expected payoffs to circumvent taking risks. In other

words, it measures the extent to which the utility of a guaranteed payoff (for instance

50) is higher than the utility derived from a proposition with the same expected reward

obtained with risk (for instance 100 with 50% probability and 0 with 50% probability).

Risk aversion is involved in decision-making situations where a probability can be

assigned to each possible outcome of the situation.

Loss aversion refers to the notion that decision-makers prefer to avoid losses over

acquiring gains. Loss aversion was first demonstrated by Kahneman and Tversky in

their prospect theory (Kahneman and Tversky, 1979, 1984). Loss aversion implies

that losing 50 will decrease utility or satisfaction by more than the increase in utility

or satisfaction that is associated with a (windfall) gain of 50. Loss aversion explains the

well-known endowment effect that people value the goods and assets they own higher

than identical goods and assets they do not own (Kahneman et al., 1990).

8The earliest philosophic thinker about entrepreneurship Cantillon (1755) defined the entrepreneuras a risk bearer as a consequence of buying and selling at uncertain prices. Say’s entrepreneur (1803)is a risk bearer because of the risk of losing capital and reputation due to the likelihood of failure.Hence, Say defines entrepreneurship in terms of the risk of losses rather than of gains. Marshall’sview on entrepreneurship (1930) is the most common one: entrepreneurs are responsible for assumingthe business risks associated with their enterprise. Marshall also acknowledges that a few extremelyhigh prices will have a disproportionately great attractive force (Marshall, 1930, p. 554) “becauserisk lovers are more attracted by the prospects of a great success than they are deterred by the fear offailure” . Thus, also Marshall pays particular attention to loss aversion. Knight (1921) was the firstto explicitly distinguish between risk and true uncertainty (ambiguity). He defines the entrepreneuras the particular kind of individual who bears uncertainty because business decisions practically neverconcern calculable probabilities (Van Praag, 1999, p. 322).

16

Ambiguity aversion is also known as uncertainty aversion and refers to a preference

for risks with known probabilities over and above risks with unknown probabilities

(true Knightian uncertainty), e.g., Ellsberg (1961) and Holm et al. (2013). Ambiguous

events have a greater degree of uncertainty than risky events not only because the

outcome is uncertain, but also the probability of the realization of that outcome and,

as a consequence, the expected payoff.

Risk aversion

To measure risk aversion empirically, we rely on two measures. The first experimental,

choice-based measure is obtained by using the multiple price list (MPL) format of

Dohmen et al. (2010), which originates from Holt and Laury (2002). Participants are

confronted with a list of ten decisions between two options: a risky one with known

probabilities (Option A) and a safe one (Option B). In each of the ten cases Option A

corresponds to gaining €300 with a 50% chance or gaining €0 with a 50% chance. The

safe Option B on the other hand gradually increases from €25 to €250 (see Figure

G1, Appendix G in Appendices Chapter 2). Instead of asking each participant to

reveal their preferences for every decision, we asked each participant to indicate their

switching point, see e.g. Dohmen et al. (2011) and Gneezy and Pietrasz (2014). For

example, a possible answer was “I prefer Option A in decision 1 and Option B in 2-10”.

The second, survey-based measure of risk aversion is copied from Dohmen et al.

(2011). Participants indicate their self-perceived willingness to take risks in general, as

well as in the two sub-domains of career and financial matters. We employed a 0 - 10

scale, where 0 stood for “Not at all willing to take risks” and 10 for “Very willing to

take risks”. In the design of the questionnaire, this question was widely spaced from

the incentivized risk measure, which came first. The question about willingness to take

risk in general is of main interest, the ones about career and financial matters are used

for robustness checks.

Loss aversion

Loss aversion is measured by means of the MPL applied by Fehr and Goette (2007)

and Gachter et al. (2010), which in essence is like the Holt and Laury price list but

also includes negative payoffs. In this case, Option A consists of a 50% probability of

receiving €6 and a 50% probability of losing an amount between €1 and €10. When

selecting the safe Option B, participants receive €0 (see Figure G2, Appendix G in

Appendices Chapter 2). Again, we are interested in the switching points.

Overall, the small stakes in these lotteries ensure that risk aversion cannot con-

vincingly explain the choice behavior in these decisions, as risk aversion in such small

stake lotteries would imply extreme degrees of risk aversion in high stake gambles (e.g.

17

Rabin, 2000; Wakker, 2005; Fehr and Goette, 2007; and Wakker, 2010). Rabin (2000)

therefore argues that under expected utility theory, people should be risk neutral in

such small stakes gambles. We emphasize in our survey that selecting Option A entails

a real loss of money.

The benefit of the small stakes mixed gamble is that it provides a simple proxy for

loss aversion. Gachter et al. (2010) measure individual subjects’ WTA/WTP ratios

for a toy car and find that these are significantly and highly correlated with the small

stakes lottery choice we use in our experiment (ρ = 0.635). Loss aversion inferred from

risky choices thus correlates strongly with loss aversion inferred from riskless choices,

alleviating to some extent the concern that our loss aversion proxy might be confounded

with risk aversion. Nevertheless, we acknowledge that there is still scope for criticism.

In Subsection 2.4.3 we will therefore discuss two alternative elicitations of loss aversion

that we used in a second experiment.

Ambiguity aversion

Our measure of ambiguity aversion is taken from Fox and Tversky (1995) and Gneezy

and Pietrasz (2014) and uses an MPL structure again. In each of the ten decisions,

we present participants with an Urn A with 50 red balls and 50 black balls, and an

Urn B which has an unknown distribution of red and black balls. The selection of Urn

A pays off €300 if a red ball is drawn and €0 if it is black. If participants choose to

select Urn B, payments vary between €250 and €475 if a red ball is drawn and is €0

in the case of a black ball (see Figure G3, Appendix G in Appendices Chapter 2).

2.2.2 Sampling

According to Holm et al. (2013, p. 1676), obtaining a large-scale experiment involving

hundreds of entrepreneurs and managers “ ... would be a demanding undertaking any-

where in the world. Owners and CEOs of established firms are rarely willing to devote

their scarce time to time-consuming academic studies”. They observe that some ear-

lier studies solved this problem by studying the self-employed, others by using small

(convenience) samples, whereas they themselves have gone to China to perform an

incentivized experiment with affordable monetary awards. Their sample includes 700

private enterprises, excluding start-ups and small-scale household firms, and a ran-

dom sample of 200 individuals as control group. They note that their control group

is not ideal and that “ ... the ideal control group would be one that is identical to the

entrepreneurs except that they are not entrepreneurs” (p. 1677).

We took a different route to obtain a large-scale sample in a Western country

(the Netherlands), including a control group that is rather similar to the group of

18

entrepreneurs. We decided to bring the lab to the field and obtain responses from

participants online. This practice is not uncommon when aiming for a substantial

response from the field (see e.g., DeMartino and Barbato, 2003; Block and Koellinger,

2009; and Graham et al., 2013). As described in Chapter 1, we were able to reach

qualifying participants through our collaborations with “Synpact”, a company with

access to 15,000 entrepreneurs, and “De Baak”, a highly reputed training center with

5,888 managers in file. The same invitation was also sent to a sample of 7,850 employees

which were recruited via a Dutch market research agency with access to more than

70,000 Dutch employees.

Invitations to participate were sent out to the groups of entrepreneurs and managers

on October 1, 2013 (Round 1) and to the employees on November 4, 2013 (Round 2).

All groups had exactly 14 days to respond and non-respondents at that stage received

a reminder after 7 days. Out of all people who received the mailing, 910 entrepreneurs,

397 managers and 981 employees completed the survey. Response rates were thus in

the range of 6 - 12%. These are comparable to the European response rates in e.g.

Graham et al. (2013), and were even high compared to earlier experiences of Syn-

pact and De Baak with non-incentivized surveys. Finally, a comparison of respondents

with non-respondents based on the available observables (age and gender) yields no

significant differences for entrepreneurs and managers. For the responding employees,

however, females were slightly oversampled (53% versus 47%).

Default definitions of entrepreneurs, managers and employees

The qualifying characteristics for inclusion in the entrepreneur sample were: all peo-

ple who have founded, inherited or taken over a company that they are currently

(co-)managing. We also classified participants as ‘entrepreneurs’ who obtained firm

ownership over a company within 5 years after start-up and who are currently its (co-)

manager. Individuals qualify for inclusion in the sample of ‘managers’ if they are em-

ployed by an organization that they did not start up themselves and have at least two

subordinates for whom they are directly responsible. We also classify project managers

as ‘managers’ in case they have overall responsibility for their projects and at least two

direct reporting lines. People belong to the group of ‘employees’ if they are employed

by an organization and do not belong to the first two groups.9

9Participants who are both entrepreneurs and managers or employees, and therefore eligible formultiple subsamples, were instructed to select the one generating most of their income. With theexception of 12 participants, these instructions were followed up adequately.

19

2.2.3 Incentives

Respondents were requested to first complete two parts of incentivized games and

then fill out the rest of the survey, including the subjective measure of risk aversion

and several background questions. This paper reports the results of the second part

of incentivized games only. All participants first received instructions about what to

expect in general and about the reward structure. Instructions also included examples

to get familiar with the experimental setup. The total questionnaire took 14 minutes on

average, including possible breaks that people took while being online. Except for the

general risk question, all decisions in our experiment were made incentive compatible

and thus had real financial consequences if one was a selected as prize winner. This

was clearly communicated.

Incentives are such that participants can earn a maximum of €675 (€200 in Part

1 and €475 in Part 2) and a minimum of €90 (€100 in Part 1 and -€10 in Part

2), depending on their choices and luck. The luck component consist of three ele-

ments. First, decisions involve a random draw whenever a participant selects a risky

or ambiguous option. Second, in each of the two parts, only one decision is randomly

selected for payment. Such a procedure is quite common in the literature (see e.g.

Laury, 2006; Dohmen et al., 2011) and is according to Azrieli et al. (2012) the only

incentive-compatible way to utilize the MPL method. Third, only a random selection

of participants are selected as winners and actually paid out. Given a limited budget

and the income levels of the participants we chose to pay out substantial (instead of

very small) amounts to a few (instead of all) randomly selected participants. Bolle

(1990), Starmer and Sugden (1991), Cubitt et al. (1998), and Laury (2006) all show

that this payment procedure does not lead to different results compared to either the

case where all participants are paid (Bolle, 1990; Starmer and Sugden, 1991; Cubitt

et al., 1998) or the case where all decisions are paid (Laury 2006). In Round 1 we

randomly selected two winners from each day’s completed participants’ files in the first

week and one winner per day in the second week. This resulted in 21 prize winners in

Round 1 in total. In the second round, we paid out five participants. Overall, chances

of getting paid out were 1/62 in the Round 1 and 1/196 in Round 2. This was unknown

to the participants (and ourselves) beforehand.

Our incentive structure has two potential drawbacks. First, although the prize

structure was very transparant throughout, the probability of winning was unknown

ex ante. Second, the realized probability of winning turned out to be low, due to un-

expectedly high response rates, thereby diluting incentives. The unknown probability

of winning might be problematic if entrepreneurs have systematically different beliefs

about these probabilities. For instance, entrepreneurs might be more optimistic and

therefore face stronger perceived incentives than non-entrepreneurs. We find an indi-

20

cator that these differences in beliefs do not play an important role in a robustness test

using data from a second experiment where the estimated probability of winning was

communicated ex ante (see Subsection 2.4.3).

Because of the low probability of winning (ex post), some people might even believe

that the probability of winning is so low, that they consider the decision hypothetical.

This might weaken the validity of our approach, although hypothetical risk elicitations

seem to be correlated with incentivized ones (e.g., Dohmen et al., 2011).

Draw and payment schedule

To foster trust, all prize winners as well as all other random draws were performed by a

civil-law notary who also monitored a legitimate course of the payouts. The procedure

at the notary was as follows. Before the start of the experiment, it was agreed that we

would pay out the 15th and 30th participant of each day in the first week and the 15th

participant of each day in the second week. The daily rankings were established based

on the registered end time of each survey. Furthermore, we also determined a payment

schedule prior to the experiment which outlined the two winning choices in Part 1 and

Part 2 for each prize winner and whether s/he was lucky when taking the risky option.

The most involved part was to settle the ambiguity in our ambiguity aversion measure.

Here we took two draws from two urns with 101 numbers (0 until 100). The first draw

rendered a benchmark number that corresponded to the percentage of winning (e.g. 88

leads to an 88% chance of winning). The second draw from the other urn determined

if a participant was lucky, which occurred whenever the second number was lower than

or equal to the first number. Overall, these series of draws yielded a payment schedule

that was accustomed to every choice a prize winner could make. Participants were

unaware of this procedure.

2.3 Descriptive statistics

Panel A in Table 2.1 shows the sample descriptive statistics (n = 2, 288) of the measures

of risk, loss, and ambiguity aversion. Panel B reports the correlations between these

variables. For ease of presentation we have reversely coded the survey measure of risk:

a higher value implies a stronger aversion to risk. Note, however, that the levels of

the different measures in Panel A are not directly comparable. Furthermore, for the

experimental measures of risk, loss, and ambiguity aversion we worked with the number

of safe options that a participant chose (in the case of ambiguity aversion; the number

of risky (as opposed to uncertain) options). The more safe (in the case of ambiguity

aversion; risky) options a participant preferred, the more averse he/she was.

21

Table 2.1 Descriptive Statistics

Panel A: Means Observations Mean St. dev. Minimum Maximum

Risk Aversion

- Survey measure1 2,288 3.67 1.79 0 10

- Experimental measure 2,288 5.38 2.76 0 10

Loss Aversion 2,288 5.05 2.65 0 10

Ambiguity Aversion 2,288 5.74 3.64 0 10

Panel B: Correlations Risk Risk Loss Ambiguity

Survey (S) or Aversion Aversion Aversion Aversion

Experimental (E) (S) (E) (E) (E)

Risk Aversion

- Survey measure1 -

- Experimental measure 0.17 *** -

Loss Aversion 0.12 *** 0.05 *** -

Ambiguity Aversion 0.05 *** -0.05 *** -0.01 -

1 Reverse coded measure of “Willingness to take risks”.

* Denotes statistical significance at the 10% level; ** at the 5% level; *** at the 1% level.

Panel B shows that most of the correlations between the measures are rather low. The

correlation with the highest absolute value is the one between the two measures of risk

attitude. The survey-based measure of risk attitude is also correlated significantly with

both loss aversion and ambiguity aversion, but to a lower degree. The low correlations

between the three experimental measures support the idea that these measures capture

distinct behavorial aspects of risk and uncertainty.

Table 2.2 shows the descriptive statistics of some characteristics that are used to

define stricter subsamples of (more successful) entrepreneurs and managers. Panel A

shows the income distribution of each of the three samples according to the answer

categories used in the questionnaire.10 Entrepreneurs are over-represented in both

10We allowed participants to keep their income private, so Panel A reports the distribution of theavailable data points. Overall, 656 entrepreneurs, 329 managers and 820 employees were willing to

22

tails of the income distribution relative to managers, which is a common observation

(Hamilton, 2000). We do not observe substantial differences between the average level

of the entrepreneurial and managerial incomes, though. Both are higher than the

income level of employees. For entrepreneurs and managers, the median income is

in the category of €50,001 - €75,000. For employees the median value falls in the

€25,001 - €50,000 category, in line with the modal income in the Netherlands in 2013

(€33,500).

Panel B shows that 82% of the entrepreneurs in our sample are the founder of their

firms, a commonly used stricter definition of entrepreneurs. 14% of the firms have

been acquired through a takeover and in 4% of the cases, the entrepreneurs have bought

themselves into the business they currently (co-)manage within five years after its start-

up. For managers we are interested in subsamples of CEOs (17%) and all managers

except those who are responsible for projects rather than people (82%). Panel C shows

that 20% (38%) of the entrepreneurs are currently managing and leading young firms

in their start-up (survival) phase. Some studies define entrepreneurs exclusively as the

owner/managers of start-ups (e.g., Brockhaus, 1980), whereas other studies explicitly

take them out (Holm et al., 2013). We shall use the same distinctions to test the

robustness of our results against using various definitions of the entrepreneur. Panel C

also shows that almost half of the entrepreneur sample consists of incorporated business

owners. This enables us to limit the sample of entrepreneurs to incorporated business

owners consistent with, for instance, Levine and Rubinstein (2013). The right hand

side of Panel C shows the age and size distributions of the firms that managers and

employees work for. As expected these distributions are similar, but different from

the ones of entrepreneurial firms. The latter are younger (see panel C) and smaller

(see Panel D). As a robustness check we shall split the sample of entrepreneurs and

managers according to the age and size distribution of their firms. Managers in smaller

and younger, i.e., more entrepreneurial firms, may be more similar to entrepreneurs.

Panel D of Table 2.2 shows the distribution of the number of employees employed

by entrepreneurs and supervised by managers. 17% of the entrepreneurs have zero

employees and 43% have at most one. We also consider a stricter definition of en-

trepreneurship based on the number of employees they employ (cf. e.g. Tag et al.,

2013) and perform a similar exercise for managers.

Table 2.3 compares background characteristics of the three subsamples. Entrepre-

neurs and managers are similar in terms of the most commonly used background charac-

teristics: their age, the percentage of females as well as their experience and educational

background. Employees are different in terms of their background characteristics com-

share their income level (which equals 72%, 83%, and 84%, respectively). Comparing responders andnon-responders on the income question with each other shows no differences in terms of average age,gender, education and experience (in two-sample t-tests).

23

Table 2.2 Descriptives of Variables to Define Sample Splits

Entrepreneurs Managers Employees

(n = 910) (n = 397) (n = 981)

Panel A: Income Panel A: Income

< €25,000 23% < €25,000 2% 26%

€25,001 - €50,000 20% €25,001 - €50,000 17% 58%

€50,001 - €75,000 19% €50,001 - €75,000 34% 12%

€75,001 - €125,000 20% €75,001 - €125,000 36% 3%

€125,001 - €200,000 11% €125,001 - €200,000 8% 1%

€200,001 - €300,000 4% €200,001 - €300,000 2% 0%

€300,001 - €400,000 1% €300,001 - €400,000 0% 0%

> €400,000 2% > €400,000 1% 0%

Panel B: Entrepreneur characteristics Panel B: Manager characteristics

Founder 82% CEO 17% -

Business taken over 14% General Manager 65% -

Joined firm within 5 yrs 4% Project Manager 18% -

Panel C: Firm age and legal structure Panel C: Firm age and size

Start-up phase (0 - 3 yrs) 20% Firm age ≤ 5 yrs 5% 6%

Survival phase (0 - 5 yrs) 38% Firm age 6 - 50 yrs 50% 55%

Firm age > 50 yrs 45% 39%

Incorporated 49% Firm size ≤ 25 FTE 13% 19%

Sole propriotership 38% Firm size 26 - 1000 FTE 53% 50%

Other 13% Firm size > 1000 FTE 34% 31%

Panel D: Firm size (no. of FTE) Panel D: Management level (direct reports)

Less than 2 43% 2 - 5 45% -

2 - 5 25% 6 - 10 30% -

6 - 10 10% 11 - 25 19% -

11 - 25 11% 26 - 50 4% -

26 - 50 5% More than 50 2% -

More than 50 7%

24

pared to the other two groups: they are somewhat younger (mean age is 41), more

likely to be female and they have lower educational degrees on average.

Table 2.3 Background Characteristics

Entrepreneurs Managers Employees

(n = 910) (n = 397) (n = 981)

Age 47.36 a 46.45 c 41.24 a,c

Female (dummy) 0.25 a 0.28 c 0.53 a,c

Education (highest degree): d f d,f

- High School 4% 2% 3%

- Lower intermediate vocational degree 12% 11% 34%

- College education 46% 42% 42%

- University education 38% 45% 21%

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

d) Significant difference between entrepreneurs and employees at the 5% level (Kolmogorov-Smirnov test)

e) Significant difference between entrepreneurs and managers at the 5% level (Kolmogorov-Smirnov test)

f) Significant difference between managers and employees at the 5% level (Kolmogorov-Smirnov test)

2.4 Results

2.4.1 Main results

To get a first impression of our main findings, Table 2.4 shows the means of the four

measures of risk and uncertainty for each of the three groups of interest. The first

column in Table 2.4 shows that entrepreneurs subjectively assess themselves as less

risk averse than managers. Managers, in turn, rate themselves as less risk averse than

employees. Two sample t-tests reveal that the differences between entrepreneurs and

employees, entrepreneurs and managers, and managers and employees are all highly

significant (p < 0.001 in all cases). Ranksum and Kolmogorov-Smirnov tests confirm

these results.

The second column shows that the experimental measure of risk aversion is not

significantly lower for entrepreneurs when compared to managers, although both en-

trepreneurs and managers are significantly less risk averse than employees.11 The rest

11In terms of the certainty equivalents per group, we find that the average CE category is equal to€125 - €150 for entrepreneurs and managers and equal to €100 - €125 for employees. As expected,all average values are below the expected value of €150. The associated CRRA coefficients would be

25

of the table reveals that the raw differences in terms of loss aversion show a similar pat-

tern: entrepreneurs are least loss averse, followed by managers and employees. Here the

difference between entrepreneurs and managers is significant at the 5% level, whereas

the difference between managers and employees is not (p = 0.08). The last column of

Table 2.4 reveals an unexpected pattern: entrepreneurs and managers, who are equally

ambiguity averse, are more ambiguity averse than employees. Finally, a closer exami-

nation of the CDFs of the four measures for the subsamples of entrepreneurs,managers

and employees (see Figure G4, Appendix G in Appendices Chapter 2) shows that the

differences in mean values are not driven by extreme values.12

Table 2.4 Raw Differences in Risk, Loss, and Ambiguity Aversion

Risk Risk Loss Ambiguity

Aversion Aversion Aversion Aversion

Survey (S) or

Experimental (E) (S) (E) (E) (E)

Entrepreneurs (n = 910) 3.10 a,b 5.03 a 4.77 a,b 5.88 a

Managers (n = 397) 3.69 b,c 5.17 c 5.08 b 5.90 c

Employees (n = 981) 4.20 a,c 5.78 a,c 5.29 a 5.54 a,c

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

In Table 2.5 the output of ordered probit regressions for each of the four behavorial

variables is depicted. Control variables such as age, gender, education, experience and

income are included. Columns ‘a’ show the results excluding some arguably endogenous

variables, i.e., education, experience and income, whereas columns ‘b’ include those as

explanatory variables (analogous to e.g. Dohmen et al., 2010). Note that the number

approximately 0 - 0.21 for entrepreneurs and managers and 0.21 - 0.37 for employees, the latter beingin line with the lower bound of the study of Holt and Laury (2002) among students (but see Rabin,2000, for a criticism of this approach).

12The CDFs for ambiguity aversion reveal that a large fraction of participants (± 30%) always re-frained from the ambiguous option and preferred the risky one. These findings are roughly comparableto those in Gneezy and Pietrasz (2014), who find for their overall sample a percentage of 24% thatnever chose the ambiguous option (and of 30% for men). With hindsight a potential drawback of ourmeasure is that subjects could not choose the winning color that would apply in case they would optfor the uncertain urn themselves. Choosing the risky urn may then be guided by pessimistic beliefsabout the success probability of the assigned winning color (red) in the uncertain urn, for instancesparkled by a fear that we as experimenters would want to economize on our budget and stack the deckagainst them. This may explain the slightly higher percentage we find as compared to Gneezy andPietrasz (2014), who properly do have subjects self selecting their success color. See Trautmann andVan der Kuilen (forthcoming) for a general review of measuring ambiguity attitudes experimentally,including a brief discussion of the mixed evidence on subjects having such strategic perceptions.

26

of observations in column ‘b’ drops because not every participant was willing to share

his/her income level.

Table 2.5 paints a similar picture as Table 2.4. The first estimation equation shows

that entrepreneurs view themselves as less risk averse than managers (see the Wald test

in the last row of the table), whereas both entrepreneurs and managers are less risk

averse than employees. These findings are largely consistent with previous studies using

survey-based measures of risk aversion. The second set of estimates supports the view

arisen from Table 2.4 that entrepreneurs are similar to managers when taking risky

decisions in an experimental and incentivized environment, even though they perceive

themselves as more risk-taking than managers. Again we find that both entrepreneurs

and managers are less risk averse than employees with similar background character-

istics. The third set of results shows that one behavioral characteristic is unique for

entrepreneurs: a lower level of loss aversion. The fourth and final set of results indicates

that the differences between employees on the one hand and entrepreneurs and man-

agers on the other hand in terms of ambiguity aversion disappear when including more

controls in the equation. Apparently, entrepreneurs, managers and employees that are

comparable in terms of their age, gender, education, income and experience do not

show differences in their attitudes towards ambiguity. This result was also obtained by

Holm et al. (2013).

The control variables also have different associations with the survey-based measure

of risk than with all three experimental measures. Older people claim to be less willing

to take risks in general (consistent with Dohmen et al., 2011) but none of the three

experimental measures is significantly associated with age. Females are less risk taking

according to the survey-based measure (also consistent with Dohmen et al., 2011) but

the choice based measures are no different for females than for males. The latter result

is largely consistent with the conclusions from a recent meta-analysis about gender

differences in risk attitudes elicited by this type of games (Filippin and Crosetto, 2014).

Surprisingly, for education we find a slightly positive effect using the survey-based

Dohmen measure of risk appetite (as opposed to e.g. Harrison et al., 2007), but no

significant effect when using any of the experimental measures. People with higher

incomes view themselves as less risk-averse (by comparison; in Dohmen et al. (2010)

the effect of household income is the same but just insignificant). Interestingly, higher

income people are less loss averse according to the experimental measure but not less

risk averse or ambiguity averse.

27

Tab

le2.5

Ris

k,

Loss

,an

dA

mb

igu

ity

Avers

ion

of

Entr

ep

ren

eu

rs,

Man

agers

an

dE

mp

loyees

(1a)

(1b

)(2

a)

(2b

)(3

a)

(3b

)(4

a)

(4b

)

Dep

.va

riab

le:

Ris

kR

isk

Ris

kR

isk

Loss

Loss

Am

big

uit

yA

mb

igu

ity

Ave

rsio

nA

vers

ion

Aver

sion

Ave

rsio

nA

vers

ion

Aver

sion

Aver

sion

Aver

sion

Su

rvey

(S)

or

Exp

.(E

)(S

)(S

)(E

)(E

)(E

)(E

)(E

)(E

)

Entr

epre

neu

r-0

.655

∗∗∗

-0.6

50∗∗

∗-0

.261∗∗

∗-0

.282∗

∗∗-0

.200∗∗

∗-0

.207∗∗

∗0.1

04∗∗

0.0

20

[-12

.43]

[-9.

23]

[-5.0

5]

[-4.0

3]

[-3.8

9]

[-3.0

6]

[1.9

9]

[0.2

9]

Man

ager

-0.2

56∗∗

∗-0

.203

∗∗∗

-0.2

03∗

∗∗-0

.254∗

∗∗-0

.093

-0.0

18

0.1

20∗

0.0

48

[4.2

3][-

2.63

][-

3.3

2]

[-3.2

9]

[-1.5

2]

[-0.2

3]

[1.8

8]

[0.5

9]

Age

0.03

0∗∗

0.03

7∗∗

-0.0

21

-0.0

27∗

-0.0

13

-0.0

08

0.0

06

0.0

10

[2.2

0][2

.28]

[-1.5

1]

[-1.6

8]

[-0.9

8]

[-0.5

4]

[0.4

7]

[0.6

9]

Age

2/

100

-0.0

02-0

.003

0.0

02

0.0

03∗

0.0

02

0.0

01

-0.0

01

-0.0

02

[-1.

57]

[-1.

59]

[1.6

0]

[1.8

1]

[1.0

3]

[0.6

9]

[-0.7

6]

[-0.9

5]

Fem

ale

0.23

4∗∗∗

0.13

4∗∗

0.0

31

-0.0

01

0.0

61

0.0

37

-0.0

02

-0.0

11

[4.9

8][2

.49]

[0.6

6]

[-0.0

2]

[1.3

1]

[0.6

6]

[-0.0

3]

[-0.2

1]

Ed

uca

tion

0.08

7∗∗∗

0.0

35

0.0

30

0.0

33

[2.5

9][1

.10]

[0.9

6]

[1.0

1]

Exp

erie

nce

0.00

4-0

.001

-0.0

01

-0.0

03

[1.2

1][-

0.1

1]

[-0.2

6]

[-0.9

4]

Ln

(in

com

e)-0

.134

∗∗0.0

08

-0.0

67∗

∗0.0

47

[3.7

4][0

.24]

[-1.9

8]

[1.3

1]

con

stan

t1.

598∗

∗∗2.

669∗

∗∗2.3

70∗

∗∗2.3

93∗∗

∗1.5

35∗∗

∗2.0

93∗∗

∗0.9

95

0.4

12∗∗

∗

[5.3

4][5

.45]

[8.0

7]

[5.2

7]

[5.6

3]

[4.7

4]

[0.9

0]

[3.5

1]

Ob

s.2,

288

1,80

52,2

88

1,8

05

2,2

88

1,8

05

2,2

88

1,8

05

Log

lik.

-4,2

35.3

-3,3

33.9

-5,0

55.4

-3,9

70.4

-4,9

20.8

-3,8

65.8

-4,9

19.7

-3,8

41.1

EN

T=

MA

N1

<0.

01**

*<

0.01

***

0.3

10.6

00.0

6*

<0.0

1***

0.8

00.3

61

Th

isre

port

sth

ep

-valu

eof

the

Wald

test

‘Entr

epre

neu

r’=

‘Man

ager

’.

*D

enote

sst

ati

stic

al

sign

ifica

nce

at

the

10%

level

;**

at

the

5%

level

;***

at

the

1%

level

.

28

Table 2.6 Relationship between Subjective and Objective Measures

(1) (2)

Dep. variable: Risk Risk

Aversion Aversion

Survey (S) or Experimental (E) (S) (S)

Risk Aversion (E) 0.107∗∗∗ 0.088∗∗∗

[7.36] [6.20]

Loss Aversion (E) 0.072∗∗∗ 0.058∗∗

[4.71] [3.90]

Ambiguity Aversion (E) 0.032∗∗∗ 0.036∗∗∗

[3.03] [3.56]

Entrepreneur -1.005∗∗∗

[-12.64]

Manager -0.448∗∗∗

[-4.51]

constant 2.547∗∗∗ 3.175∗∗∗

[20.62] [23.76]

Obs. 2,288 2,288

Log lik. -4,522.4 -4,443.2

* Denotes statistical significance at the 10% level; ** at the 5% level; *** at the 1% level.

In an effort to reconcile the abovementioned findings, we have also directly compared

the subjective and objective measures of risk and ambiguity (see also Ding et al. (2010)

and Willebrands et al. (2012) for comparisons of subjective and objective risk mea-

sures). Table 2.6 shows that the subjective assessments of respondents’ risk attitudes

are not only correlated with the experimental risk measure, but also with loss aversion

and ambiguity aversion. All three coefficients in the ordered probit regression on risk

attitude are highly significant and have the expected positive sign. The measure of

risk has the highest association with the self-assessed risk attitude, but both loss and

ambiguity aversion play a significant part too in the explanation of the self-assessed

value. The result is the same both without (column 1) and with (column 2) controls

for entrepreneurs and managers.

Overall, we conclude from Table 2.6 that subjective assessments of risk attitude

proxy for more than just risk aversion. This could explain why entrepreneurs perceive

themselves as being less risk averse than managers while the objective measure of risk

aversion has a similar value for entrepreneurs and managers. Entrepreneurs might

29

perceive themselves less risk averse based on their lower level of loss aversion that

they (rightly or wrongly) mix up with the economist’s definition of risk aversion. An

additional explanation for the larger differences between managers and entrepreneurs in

the survey risk measure compared to the experimental measure might be experimenter

demand effects based on stereotypes; according to common wisdom entrepreneurs are

expected to be more risk taking and their own subjective assessment may partly reflect

these general expectations. In our second experiment (see Subsection 2.4.3) we tried

to reduce experimenter demand effects by being less upfront about the purpose of our

study. We obtain the same results, however, which alleviates the concerns that demand

effects might be the main driver.

2.4.2 Robustness checks

In this section we will first test to what extent the results remain the same when using

stricter definitions of entrepreneurs and managers. Table 2.2 shows that the samples of

entrepreneurs, managers and, to a lesser extent, employees are suitable for the creation

of subsamples based on alternative and common definitions of entrepreneurs, managers

and employees. Table 2.7 displays the main result of Table 2.5, using various alternative

definitions. Thus, each coefficient is obtained in a separate regression (see Table 2.5

for the specifics of these regressions).

For entrepreneurs we use a set of stricter definitions in congruence with the lit-

erature mentioned earlier in Section 2.3. We use the subsets of: (i) entrepreneurs

with an incorporated firm, thereby mainly excluding the own-account self-employed,

(ii) entrepreneurs with an above median number of fulltime equivalent employees in

their company, (iii) entrepreneurs with above median incomes, (iv) entrepreneurs that

have founded their business, instead of obtaining it through takeover or buy-in, (v)

entrepreneurs in the survival phase (firm age ≤ 5 years) and (vi) entrepreneurs past

their survival phase (firm age > 5 years). Panel A of Table 2.7 shows the results of

confronting the data with these alternative definitions of the entrepreneur. For man-

agers and employees we employ the original samples. The last line in Panel A shows

the result of Table 2.5 again.

The panel shows a clear pattern consistent with the findings in Table 2.5. Whatever

definition of the entrepreneur is used, they assess themselves as more risk taking than

both managers and employees. Using objective measures of risk and uncertainty, the

data show again that entrepreneurs and managers are equally risk averse, but less

so than employees. The only notable and significant difference with the benchmark

appears when limiting the sample to incorporated entrepreneurs. They are significantly

less risk averse than both managers and employees. We found similar results when only

30

Table 2.7 Differences in Risk Attitude using Stricter Definitions

(1) (2) (3) (4)

Dep. variable: Risk Risk Loss Ambiguity

Aversion Aversion Aversion Aversion

Survey (S) or Experimental (E) (S) (E) (E) (E)

Panel A: Subsets of Entrepreneurs

i) Incorporated (n = 446) -0.702 a,b -0.442 a,b -0.400 a,b 0.077

[-7.84] [-3.78] [-3.43] [0.65]

ii) Above median no. of employees (n = 401) -0.730 a,b -0.282 a -0.270 a,b 0.083

[-7.41] [-2.59] [-2.46] [0.75]

iii) Above median ent. income (n = 377) -0.613 a,b -0.404 a,b -0.109 -0.119

[-5.63] [-3.86] [-1.60] [-0.91]

iv) Founder (n = 757) -0.598 a,b -0.218 a -0.280 a,b 0.011

[-7.71] [-2.81] [-3.68] [0.75]

v) In survival phase (firm ≤ 5 yrs., n = 347) -0.640 a,b -0.257 a -0.258 a,b -0.042

[-6.01] [-2.54] [-2.65] [-0.39]

vi) Not in survival phase (firm > 5 yrs., n = 563) -0.611 a,b -0.239 a -0.249 a,b -0.001

[-7.44] [-2.88] [-2.98] [-0.02]

β(Entrepreneur) in Table 2.5: -0.650 a,b -0.282 a -0.207 a,b 0.020

Panel B: Subsets of Managers

vii) CEO or general manager (n = 324) -0.218 b,c -0.274 c -0.044 b -0.006

[-2.53] [-3.27] [0.52] [-0.08]

viii) CEO (n = 66) -0.319 b,c -0.367 c -0.020 b -0.087

[-2.40] [-2.57] [-0.40] [-0.55]

ix) Above median no. of direct reports (n = 219) -0.197 b,c -0.259 c -0.048 0.083

[-2.31] [-3.04] [-0.49] [0.79]

x) Above median man. income (n = 155) -0.202 b,c -0.370 c -0.010 b 0.009

[-1.97] [-3.22] [-0.06] [0.07]

xi) Manager in a firm > 15 yrs (n = 316) -0.195 b,c -0.247 c -0.033 b 0.125

[-2.22] [-2.86] [-0.38] [1.35]

β(Manager) in Table 2.5: -0.256 b,c -0.254 c -0.018 b 0.048

Panel C: Combinations of A&B

i) vs. viii); p-values Wald tests < 0.01 0.57 0.01 0.67

ii) vs. ix); p-values Wald tests < 0.01 0.59 0.17 0.73

iii) vs. x); p-values Wald tests < 0.01 0.25 0.04 0.46

a) Significant difference entrepreneurs and employees (5% level), b) Significant difference entrepreneurs and managers

(5% level), and c) Significant difference managers and employees (5% level).

31

considering the sample of entrepreneurs (and adding dummies for the various groups),

see Appendix C (Appendices Chapter 2). All in all, loss aversion is the one behavorial

feature that distinguishes entrepreneurs from managers; the results in Table 2.5 turn

out to be robust against using various stricter definitions of (successful) entrepreneur-

ship.

Panel B of Table 2.7 shows the results when varying the definition of a manager

while keeping the baseline samples of entrepreneurs and employees. Again we find that

the main results remain, irrespective of the definition used. We restrict the sample

to: (vii) CEOs or general managers (as opposed to project managers), (viii) CEOs

exclusively, (ix) managers with more than the medium number of direct reports, (x)

managers with above median managerial income, and (xi) managers in firms that are

older than 15 years. The stricter definitions used do not only restrict the sample to

more successful managers but also in some cases to managers that can reasonably

be expected to be more different from entrepreneurs than average, such as the ones

employed in older firms. Again, the last line of the panel shows the benchmark result

for managers copied from Table 2.5.

Panel C finally tests some of the alternative definitions against each other. Whether

we compare entrepreneurs of incorporated firms (i) with CEOs (viii), or whether we

compare entrepreneurs (ii) and managers (ix) with larger spans of control, or higher

than median incomes ((iii) versus (x)), the results remain very similar to the main

findings according to the Wald statistics in each of these cases.13

In Appendix D (see Appendices Chapter 2) we have run another set of regressions

to further examine potential heteregeneous effects. First of all, we pooled the responses

of managers and employees to see if and how entrepreneurs differed from all others.

As might be expected, we find that entrepreneurs stand out from this pooled group in

both risk aversion and loss aversion, but not in ambiguity aversion.14 Second, to better

understand the separate role of the organization they work for, we have restricted the

sample to entrepreneurs and managers in young and small firms (i.e. at most 15 years

old and at most 25 employees, respectively). Obviously, organizational size and age

are lower for entrepreneurs than for managers and just adding controls for firm size

and age to the general regression specification might be insufficient. The results are

qualitatively similar to the main results, although some of the significance levels have

dropped a bit (possibly due to the smaller samples).

13We also employed a stricter definition of employees by limiting that subsample to above medianincome earners. Again the results were the same.

14Here we have just pooled all managers and employees together, but obviously in practice thedistribution of managers and employees is much different. This would lead to even more distinctresults.

32

Another check is based on the idea that many people are belonging to one of the

groups at the time of measurement, but may have been part of another group in the

past. In other words, the distinction between the three groups in terms of typology is

not black-white. Possibly, the differences between the ‘pure’ groups are larger when

taking into account that some individuals belong to ‘gray’ areas. Appendix E (see

Appendices Chapter 2) reports the results from analyses that take this into account.

We first find that 71% of the entrepreneurs in the sample have been managers in

the past and 9% is currently wage employee or manager besides being a business

owner. Moreover, 17 (10)% of the managers (employees) is also an entrepreneur on

the side, whereas 12 (9)% of the managers (employees) have been so in the past.

Apparently, people move out of and especially into entrepreneurship over the course of

their professional lives. Re-running the same regressions, but now controlling for the

gray areas, shows that the effects found in Table 2.5 (and 2.7) do not change when

accounting for past and current positions in the other groups. The coefficients of the

controls that distinguish the gray groups from the ‘pure’ groups have the expected

signs (diminishing the main effect), but they are not significant.

2.4.3 Second experiment with alternative elicitations of loss

aversion

The loss aversion measure we employ in our first experiment records subjects’ will-

ingness to accept a mixed prospect with small gains and losses. According to Rabin

(2000), traditional risk aversion deriving from utility curvature cannot play a role with

small stakes (see also Wakker and Deneffe, 1996). Identifying loss aversion using small

stakes mixed prospects has some drawbacks, though. First, one may reasonably wonder

whether losing 10 euros at most is something entrepreneurs and managers truly worry

about. Second, choices in mixed prospects may be affected by probability weighting

and utility curvature as well, potentially confounding loss aversion with other drivers

of decisions under risk. In December 2014 we therefore ran a second experiment in

which we elicitated loss aversion in two alternative ways. This second survey was sent

out to the same databases of entrepreneurs, managers and employees as described in

Subsection 2.2.2.

The first alternative measure is based on riskless choices and compares subjects’

willingness to accept (WTA) with their willingness to pay (WTP) for a given good.

We employ a between-subjects design similar to Kahneman et al. (1990). Gachter

et al. (2010) measure individual subjects’ WTA/WTP ratios for a toy car using a

within-subjects design and compare the results to those obtained from a between-

33

subjects control treatment. They find no systematic differences in the WTA and WTP

valuations. Reliable estimates of individual WTA/WTP ratios require sufficient time

between the WTA and WTP elicitations, as well as the use of the strategy method (cf.

Gachter et al., 2010). Both of these features are unattractive for our purposes, because

our subject pool requires that we keep the experiment as short and simple as possible.

This motivates our choice for a between-subjects design.

Each of the three occupational groups are randomly cut in halves: either we elicit

their willingness to pay or their willingness to accept for a fancy bread tray. Half of

the sample is offered to buy the tray from us (for a price between €0 and €20, in steps

of €2) using their prize money, whereas the other half is offered the opportunity to

sell the tray (that they obtained as part of their prize money) back to us. Following

Gachter et al. (2010) we employed an incentive compatible elicitation procedure where

subjects indicate their willingness to trade at a randomly drawn price (see Figures 5

and 6, Appendix G in Appendices Chapter 2). We take the midpoint of the resulting

reservation price interval as the inferred reservation price (either WTA or WTP). For

instance, when a participant indicates that he/she buys the bread tray at a price of

€4 but not at a price of €6, we take the midpoint of €4 and €6, i.e. €5, as his/her

reservation value. Conversely, when a participant in the WTA treatment is willing to

let go of the bread tray for €12 but not for €10, we work with a reservation price of

€11.

The second alternative measure of loss aversion we obtain is (again) a measure of

loss aversion in risky choices, but now with much higher stakes. More specifically, we

confront subjects with three lottery choice tasks: one in the gain domain, one in the

mixed domain and one in the loss domain. The one in the gain domain exactly matches

our risk aversion measure in experiment one. Subjects choose between Option A that

gives a 50% chance of winning €300 (and a 50% chance of winning €0) and Option

B that yields a given fixed amount for sure, where the fixed amount ranges from €25

to €250 (in steps of €25). The lottery used in the mixed domain corresponds to our

original loss aversion measure scaled up by a factor of 50. Thus, Option A now gives

a 50% chance of winning €300 and a 50% chance of losing a given amount, where the

loss ranges from €0 to €350 (in steps of €50).15 Third, the loss domain is the mirror

image of the gain domain. Subjects then choose between a 50% chance of losing €300

(Option A) and a sure loss of a given amount (Option B, with the loss ranging from €25

to €250). By adding this third lottery we can investigate whether entrepreneurs are

15Overall, the setup was almost equal to the first experiment and very similar to e.g. Eckel andGrossman (2008). That is, all participants earned a base fee of €375 by completing the other partsof the survey (which are not in scope of this paper), and all gains and losses made were added to /substracted from this base payment. We made it very explicit that making losses entailed really losingmoney. The maximum loss was capped at €350 (instead of €500) such that a participant’s earningscould not go negative.

34

especially different in how they cope with (unavoidable) losses, or rather in how they

tradeoff potential gains against potential losses. Finally, we also include the subjective

unincentivized risk attitude measure again.

We also improved on our first experiment in other ways. To avoid that differences

in beliefs about winning probabilities would potentially confound our results, we now

explicitly informed subjects of the expected likelihood of being selected as a prize

winner. Subjects were told beforehand that, on the basis of our previous experience,

their chances of becoming a prize winner were about 1 in 100. Moreover, this time

we were also less upfront about the purpose of our study. In the first experiment

we mentioned in our cover letter that our study aims to explore “ ... differences in

decision-making between entrepreneurs, managers and employees” (see Appendix B in

Appendices Chapter 2). We did so in the hope that it would increase response rates.

Yet a potential worry might be that this leads to unwanted experimenter demand

effects based on stereotypes. In our second experiment we therefore just mentioned

that our research “ ... aims to study decision-making processes”. Finally, based on

our desire to minimize response times (and the measured response times in the first

experiment for this specific part), we used a bisection procedure that led subjects to

their switching point via 3 to 4 binary choices (see Abdellaoui, 2000; Abdellaoui et al.,

2007; and Abdellaoui et al., 2008 for illustrations).

The final sample is again large. Overall 697 entrepreneurs, 265 managers and 969

employees participated (n = 1,931). Only 18% of them participated in both research

waves. Furthermore, in terms of individual and company characteristics the three sam-

ples proved rather similar to the first experiment. For instance, for each occupational

group we find no differences in gender and education and small differences in age.

The results of the second experiment are shown in Tables 2.8 and 2.9. They confirm

that entrepreneurs are indeed less loss averse than both managers and employees. The

cleanest piece of evidence comes from examining loss aversion in riskless choices, i.e., our

first alternative measure of loss aversion, because there by definition risk motivations

cannot play a role. The final column in Table 2.8 shows that the willingness to pay for

the bread tray does not differ between the three groups. Entrepreneurs, however, have a

significantly lower willingness to accept than both managers and employees. Using the

WTA/WTP ratio based on group averages as a proxy for loss aversion, entrepreneurs

score lowest at 1.65, followed by managers (1.86) and employees (2.00). These values

are in line with those obtained in previous studies that elicited WTAs and WTPs in

a between-subjects design (see e.g. Kahneman et al. (1990) and Gachter et al. (2010),

and see Horowitz and McConnell (2002) for a review of WTA/WTP studies).

35

Table 2.8 Raw Differences in Risk and Loss Aversion (Second Experiment)

Risk Risk Risk Loss Willingness Willingness

Aversion Aversion Aversion Aversion to Accept to Pay

(gain) (loss) (WTA) (WTP)

Survey (S)

or Experimental (E) (S) (E) (E) (E) (E) (E)

Correspondence

with first experiment exact exact - times 50 - -

Entrepreneurs (n = 697) 2.88 a,b 5.16 a 3.86 2.68 a,b € 9.46 a,b € 5.73

Managers (n = 265) 3.33 b,c 5.09 c 3.97 3.51 b € 11.56 b € 6.21

Employees (n = 969) 3.79 a,c 6.07 a,c 3.74 3.70 a € 10.53 a € 5.27

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)c) Significant difference between managers and employees at the 5% level (two-sample t-test)

A more rigourous comparison of the first alternative measure of loss aversion follows

from the Tobit regressions in Table 2.9 that aim to explain the respondents’ inferred

reservation prices. Here employees’ WTP serves as the benchmark category. The

dummy WTA - equal to one if the observed reservation price belongs to a subject in

the seller role rather than in the buyer role - captures the WTA-WTP gap. The main

variables of interest are the interaction terms ‘Entrepreneur x WTA’ and ‘Manager x

WTA’. The former appears highly significant in all three specifications, the latter is

always insignificant. The final row in Table 2.9 compares these two interaction terms

using a Wald test, showing that for entrepreneurs the WTA-WTP gap is significantly

lower than for managers.16

The second alternative measure of loss aversion based on risky choices corrobo-

rates our earlier findings, too. The scaled up version of the mixed prospect (labelled

‘Loss aversion’) gives the same findings as before.17 In addition, we again find that

entrepreneurs subjectively believe they are more willing to take risks than managers,

while their incentivized lottery choices in the gain domain again reveal no differences

in risk attitude (see the first two columns in Table 2.8). In addition, we now also find

this for the pure loss domain not studied in the first experiment, see the third column

16A potential concern might be that (for whatever reason) the 18% overlap in participants of the twoexperiments biases our conclusions. If we run the Tobit regressions on the 1561 novice participants,we get exactly the same results (in terms of significance of the relevant coefficients).

17Note though that the average number of safe choices in each group is lower than the averagesobserved in the first experiment (see Table 2.4). This can be either due to the higher stakes in thesecond experiment or the lower range of possible values resulting from our capping at €-350 (0 to 8versus 0 to 10).

36

Table 2.9 WTA-WTP gaps (Second Experiment)

(1) (2) (3)

Dep. variable: Price Price Price

WTA 9.652∗∗∗ 9.719∗∗∗ 9.104∗∗∗

[16.35] [16.55] [14.46]

Entrepreneur 1.113∗ 2.026∗∗∗ 2.293∗∗

[1.72] [2.82] [2.57]

Manager 1.519∗ 2.546∗∗∗ 1.829∗

[1.73] [2.74] [1.74]

Entrepreneur x WTA -3.014∗∗∗ -3.173∗∗∗ -3.993∗∗∗

[-3.38] [-3.57] [-3.84]

Manager x WTA -0.970 -0.992 -0.533

[-0.78] [-0.81] [-0.41]

Age -0.211∗ -0.281∗

[-1.69] [-1.92]

Age2 / 100 0.163 0.245

[1.19] [1.51]

Female 0.941∗∗ 1.237∗∗

[2.20] [2.47]

Education -0.831∗∗∗ -0.938∗∗∗

[-3.41] [-3.29]

Experience -0.009 -0.009

[-0.32] [-0.29]

Ln(income) 0.595∗

[1.76]

constant 2.322∗∗∗ 10.15∗∗∗ 5.981

[5.40] [3.59] [1.33]

Obs. 1,931 1,931 1,492

Log Lik. -5,224.6 -5,210.3 -4,072.5

ENT x WTA = MAN x WTA1 0.11 0.09* 0.01**

1 This reports the p-value of the Wald test ‘Entrepreneur x WTA’ = ‘Manager x WTA’.

* Denotes statistical significance at the 10% level; ** at the 5% level; *** at the 1% level.

37

in Table 2.8. Entrepreneurs thus differ from managers (only) when directly trading off

potential gains against potential losses.18 Combining this latter combination of findings

with prospect theory actually lends further support to our finding that entrepreneurs

are less loss averse than managers and employees. Prospect theory predicts that choices

between risky prospects are governed by a combination of utility curvature, subjective

probability weighting and loss aversion. By definition, loss aversion only plays a role for

mixed prospects. In contrast, utility curvature and probability weights affect choices for

all types of prospects. The fact that we observe no differences between entrepreneurs

and managers for the two non-mixed prospects (i.e. the gain domain and the loss

domain), but do observe differences in the mixed domain, strongly suggests that loss

aversion is the driver for why entrepreneurs behave differently.

By making parametric assumptions, some insights can be obtained in the extent

of loss aversion. Assuming (piece wise) linear utility and a common specification of

loss aversion as in Kobberling and Wakker (2005), the choices in the gain and loss

domain identify probability weighting. These can thus be used to isolate the effect of

probability weighting on the mixed prospect. Correcting for individual heterogeneity

in probability weighting in this way, we calculate the inferred value of loss aversion

coefficient λ for each individual (see Appendix F4 in Appendices Chapter 2 for a

more detailed elaboration). The observed differences across occupational groups keep

standing and the median λ’s - equal to 1.71 for entrepreneurs, 1.87 for managers and

1.82 for employees - compare reasonably well with other findings in the literature (see

e.g. the median value of 2.61 found in Abdellaoui et al. (2008), and of 2.25 in Tversky

and Kahneman (1992)).19 Overall we conclude that our main finding from the first

experiment is not just a false positive or due to an arguably confounded measure of

loss aversion. Also for the two alternative loss aversion measures used in the second

experiment we find that, when comparing entrepreneurs with managers, much of the

action comes from loss aversion and not simply risk aversion.

18In Appendix F in Appendices Chapter 2 we comprehensively report the equivalent analyses ofTables 2.1 through 2.7, as well as the heterogeneity checks of the previous subsection, for the secondexperiment. Overall, we obtain the same conclusions as before. Perhaps the most notable additionalfinding is that - within the sample of entrepreneurs - incorporated entrepreneurs are less risk loving inthe loss domain, while founders are more risk loving in the case of unavoidable losses (see AppendixF8). Taken together with the fact that incorporated entrepreneurs are also less risk averse in the gaindomain, this suggests that they have risk preferences that are closer to risk neutrality.

19In line with Booij and Van de Kuilen (2009), we also find that females are signifcantly more lossaverse than males. Unlike them, however, we find no effect of education on loss aversion.

38

2.5 Conclusion

Common wisdom and economic theory alike portray entrepreneurs as a truly distinct

breed. Most notably, the stereotype is that entrepreneurs, as business owning residual

claimants, are more willing and better able to cope with risk and uncertainty. Existing

empirical studies that ask entrepreneurs and non-entrepreneurs to subjectively self-

assess their attitude towards risk and uncertainty, indeed by and large confirm this

conventional wisdom; entrepreneurs take themselves as being more willing to take

risks than non-entrepreneurs are. Other studies that employ incentivized choice-based

measures of risk aversion, however, find little differences between entrepreneurs and the

comparison group employed. These opposing findings immediately raise the question

of whether entrepreneurs’ more positive attitude towards risk is merely a common

(mis)perception, or whether they have truly distinct preferences.

In this paper we report the results from a lab-in-the-field experiment that sheds

light on this matter. Our experiment has a number of distinguishing features. First

of all, it is relatively large in size with 2,288 respondents overall, including 910 en-

trepreneurs. Second, we compare entrepreneurs with two well-defined control groups,

viz. managers and employees. Entrepreneurs and managers are very similar in terms

of background characteristics and arguably also in terms of the professional decisions

and tasks they face, including managing the employees they direct. Yet as residual

claimants only entrepreneurs directly feel the financial consequences of the decisions

they take. Especially this difference is thought to draw people with distinct risk prefer-

ences into entrepreneurship. Differences between entrepreneurs and employees (both in

terms of background characteristics and professional activities) are more pronounced.

Third, we collect a large variety of background characteristics and measures of individ-

ual ‘success’. This allows us to zoom in on particular subsamples, using more stringent

definitions of both entrepreneurs and managers based on being ‘more successful’. Last,

we include both a subjective, survey-based measure of risk attitude, as well as incen-

tivized, choice-based measures of risk related preferences. This allows us to compare

subjective perceptions of risk attitude with objectives measures based on actual choices

with true financial consequences. Besides a standard measure of risk aversion, we also

include measures of loss aversion and ambiguity aversion.

In line with previous studies and conventional wisdom, the entrepreneurs in our

sample on average perceive themselves as being more risk tolerant than the other re-

spondents. This not only holds with respect to the employees in our sample, but also in

regard to the more comparable control group of managers. Based on the incentivized

choice-based measure of risk aversion, however, entrepreneurs are equally risk averse

as managers (with employees being significantly more risk averse). The different per-

39

ceptions of entrepreneurs and managers thus cannot be explained by differences in risk

aversion as narrowly defined by economists. Rather, our results show that these differ-

ent perceptions mainly result from significant differences in attitudes towards losses;

managers are significantly more loss averse than entrepreneurs are (with employees

in turn being more loss averse than managers, although not significantly). The three

groups do not differ in terms of ambiguity aversion. These findings are largely inde-

pendent of the definition of who is an entrepreneur and who is a manager. If anything,

limiting the sample to more successful entrepreneurs somewhat strengthens our results.

Moreover, the results presented here also do not appear to be a false positive.

In a large independent addtional experiment across a predominantly new sample of

entrepreneurs, managers and employees, in which we elicit two additional measures of

loss aversion, we find again that all of the actions comes from loss aversion and not from

risk aversion. This result is perhaps most affirmed when examining the WTA-WTP

gaps of the three groups of interest. Consistent with loss aversion in risky choices,

entrepreneurs also appear to have the lowest loss aversion in riskless choices, followed

by managers and employees.

In an effort to reconcile all findings, we find that when self-assessing their ‘willing-

ness to take risks in general’ on a 0 to 10 scale, respondents appear to have a broader

notion of ‘risk’ in mind than the narrow risk aversion measure of economists (assum-

ing experimenter demand effects away). Besides risk aversion, also ambiguity and loss

aversion play an important role in shaping individual perceptions. For the perceived

difference between entrepreneurs and managers, loss aversion turns out to be key. Man-

agers are on average more inclined to avoid losses than entrepreneurs are, leading to a

lower self-assessed willingness to take risks.

Overall we conclude that, when it comes to attitudes towards risk and uncertainty,

entrepreneurs are different but in a rather subtle way. The Merriam-Webster dictionary

website defines an entrepreneur as “ ... a person who starts a business and is willing

to risk loss in order to make money”. In terms of their willingness to risk losses,

entrepreneurs indeed appear to be distinct.

40

Appendices Chapter 2

Appendix A. Literature overview outlining the similarities and differences in e.g. fo-

cus and control groups, methodology, and the general results.

Appendix B. Example of a translated survey invitation.

Appendix C. Regression outputs of within-entrepreneur differences in risk attitude.

Appendix D. Regression outputs when using alternative samples of entrepreneurs

and managers.

Appendix E. Cross-occupational experience and the impact on risk, loss, and ambi-

guity aversion.

Appendix F. Results of the second experiment (December 2014).

Appendix G. Screenshots of the applied methodologies and CDF Plots of all mea-

sures.

41

Ap

pen

dix

A.

Lit

era

ture

Overv

iew

Focu

sC

om

pari

son

Defi

nit

ion

Defi

nit

ion

Reg

ion

Met

hod

Typ

eof

Sam

plin

gT

akes

MP

LIn

centi

viz

edF

Gm

ore

Gro

up

(FG

)G

rou

p(C

G)

Focu

sG

rou

pC

om

pari

son

surv

eylo

sses

pro

ne

to

Gro

up

qu

esti

on

sin

toacc

ept

acc

ou

nt

un

cert

ain

ty

Risk

Bro

ckh

au

sS

tart

-up

Bu

sin

ess

Ma

jor

ow

ner

Bu

sin

ess

Un

ited

SH

yp

oth

etic

al

RN

oN

oN

oN

o

(1980)

entr

epre

neu

rm

an

ager

san

dm

an

ager

man

ager

sS

tate

sri

sk-t

akin

g

(n=

31)

(n=

62)

of

ab

usi

nes

s

ven

ture

Hu

llet

al.

Ow

ner

-B

usi

nes

sA

per

son

wh

oB

usi

nes

sU

nit

edS

Per

son

ality

NR

No

No

No

Yes

(1980)

man

ager

ssc

hool

org

an

izes

an

dsc

hool

alu

mn

iS

tate

sin

ven

tory

(n=

57)

alu

mn

im

an

ages

aqu

esti

on

s

(n=

250)

bu

sin

ess

un

der

takin

g

Cair

d(1

991)

Ow

ner

-P

rofe

ssio

nal

Inn

ovati

ve

Tea

cher

s,tr

ai-

Un

ited

SN

ot

NR

No

No

No

Yes

man

ager

sgro

up

sb

usi

nes

sn

ers,

nu

rses

,S

tate

savailab

le

(n=

73)

(n=

189)

ow

ner

-man

ager

civil

serv

ants

,

wh

ota

kes

cler

ical

train

-

calc

ula

ted

risk

see

s,le

ctu

rers

Beg

ley

Fir

mB

usi

nes

sF

ou

nd

er,

Bu

sin

ess

Un

ited

SH

yp

oth

etic

al

NR

No

No

No

Yes

(1995)

fou

nd

ers

Man

ager

sru

nn

ing

ayou

ng

man

ager

sS

tate

sri

sk-t

akin

g

(n=

114)

(n=

114)

com

pany

or

des

irin

gto

run

ah

igh

-gro

wth

com

pany

Koh

(1996)

Entr

e-N

ot-

Ch

ara

cter

isti

cs:

Do

not

have

Un

ited

SH

yp

oth

etic

al

NR

No

No

No

Yes

pre

neu

rially

entr

epre

-n

eed

for

chara

cter

isti

csS

tate

sri

sk-t

akin

g

incl

ined

neu

rially

ach

ievem

ent

of

focu

sgro

up

MB

Ain

clin

edlo

cus

of

contr

ol,

stu

den

tsM

BA

pro

pen

sity

to

(n=

22)

stu

den

tsta

ke

risk

,

(n=

32)

tole

ran

ceof

am

big

uit

y,et

c

Note

s.FG,fo

cusgro

up;CG,compariso

ngro

up;S,su

rvey;E,experiment;

R,ra

ndom;NR,nonra

ndom;M

PL,M

ultip

lePriceList.

42

Ap

pen

dix

A.

Lit

era

ture

Overv

iew

(cont’

d)

Focu

sC

om

pari

son

Defi

nit

ion

Defi

nit

ion

Reg

ion

Met

hod

Typ

eof

Sam

pli

ng

Takes

MP

LIn

centi

viz

edF

Gm

ore

Gro

up

(FG

)G

rou

p(C

G)

Focu

sG

rou

pC

om

pari

son

surv

eylo

sses

pro

ne

to

Gro

up

qu

esti

on

sin

toacc

ept

acc

ou

nt

un

cert

ain

ty

Risk

Sara

svath

yE

ntr

epre

neu

rsB

an

ker

sF

ou

nd

ers

of

Ban

ker

sU

nit

edS

Hyp

oth

etic

al

NR

No

No

No

Yes

(1998)

part

icip

ati

ng

part

icip

ati

ng

com

pan

ies

part

icip

ati

ng

Sta

tes

risk

-takin

g

ined

uca

tion

insa

me

com

pan

ies

insa

me

pro

gra

med

uca

tion

part

icip

ati

ng

educa

tion

(n=

4)

pro

gra

min

edu

cati

on

pro

gra

m

(n=

4)

pro

gra

m

Ste

wart

etal.

Entr

epre

neu

rsC

orp

ora

teS

mall

Corp

ora

teU

nit

edS

Bip

ola

rN

RN

oN

oN

oY

es

(1999)

(n=

428)

Man

ager

sb

usi

nes

sM

an

ager

sS

tate

squ

esti

on

s

(n=

239)

ow

ner

Van

Pra

ag

Cu

rren

tly

Nev

erse

lf-

Ind

ivid

uals

Nev

erse

lf-

Net

her

-S

Inte

rvie

ws

RN

oN

oN

oY

es

an

dC

ram

eror

ever

emp

loyed

wh

ost

art

aem

plo

yed

lan

ds

(2001)

self

-b

efore

new

bu

sin

ess

bef

ore

emp

loyed

(n=

1,5

05)

or

pu

rch

ase

(n=

258)

an

exis

tin

gon

e

Uu

sita

loS

elf-

Not

self

-S

elf-

Not

self

-F

inla

nd

SA

bilit

yan

dR

No

No

No

Yes

(2001)

emp

loyed

emp

loyed

emp

loym

ent

emplo

yed

per

soan

lity

(n=

428)

(n=

428)

test

s

Cra

mer

etal.

Sel

f-N

ever

Sel

f-N

ever

Net

her

-S

Hyp

oth

etic

al

RN

oN

oN

oY

es

(2002)

emp

loyed

self

-em

plo

ym

ent

self

-la

nd

sri

sk-t

akin

g

(n=

330)

emp

loyed

emp

loyed

(n=

1,5

67)

Els

ton

etal.

Fu

ll-t

ime

Fu

ll-t

ime

Entr

epre

neu

rship

Rep

ort

an

Un

ited

E-

NR

No

No

Yes

No

(2006)

Entr

epre

neu

rsE

ntr

epre

neu

rsas

sole

bu

sin

ess

entr

epre

neu

rial

Sta

tes

(n=

42)

(n=

38)

focu

san

dn

oven

ture

an

d

ad

dit

ion

al

sou

rce

ad

dit

ion

al

of

inco

me

full-t

ime

job

s

Dja

nkov

etE

ntr

epre

neu

rsN

on

-O

wn

eror

co-

Not

aow

ner

or

Ch

ina

SH

yp

oth

etic

al

Str

ati

fied

No

No

No

Yes

al.

(2006)

(n=

414)

entr

epre

neu

rsow

ner

of

aco

-ow

ner

of

ari

sk-t

akin

gR

S

(n=

561)

bu

sin

ess

bu

sin

ess

≥5

FT

Es

≥5

FT

Es

Note

s.FG,fo

cusgro

up;CG,compariso

ngro

up;S,su

rvey;E,experiment;

R,ra

ndom;NR,nonra

ndom;M

PL,M

ultip

lePriceList.

43

Ap

pen

dix

A.

Lit

era

ture

Overv

iew

(cont’

d)

Focu

sC

om

pari

son

Defi

nit

ion

Defi

nit

ion

Reg

ion

Met

hod

Typ

eof

Sam

plin

gT

akes

MP

LIn

centi

viz

edF

Gm

ore

Gro

up

(FG

)G

rou

p(C

G)

Focu

sG

rou

pC

om

pari

son

surv

eylo

sses

pro

ne

to

Gro

up

qu

esti

on

sin

toacc

ept

acc

ou

nt

un

cert

ain

ty

Risk

Dja

nkov

etE

ntr

epre

neu

rsN

on

-O

wn

er/

man

ager

Not

Ru

ssia

SH

yp

oth

etic

al

Str

ati

fied

No

No

No

No

al.

(2007)

(n=

400)

entr

epre

neu

rsof

ab

usi

nes

sow

ner

/m

an

ager

risk

-takin

gR

S

(n=

550)

wit

h≥

6of

ab

usi

nes

s

emp

loyee

sw

ith≥

6

emp

loyee

s

Mack

oan

dE

ntr

epre

neu

rsS

tud

ents

Bu

sin

ess

Stu

den

tsP

ola

nd

E-

NR

No

No

Yes

No

Tysk

a(2

009)

(n=

40)

(n=

86)

ow

ner

Calien

do

etIn

div

idu

als

Indiv

idu

als

Sel

f-In

div

idu

als

Ger

many

SH

yp

oth

etic

al

RS

No

No

No

Yes

al.

(2010)

wh

otr

an

sfer

wh

ore

main

emp

loyed

wh

ore

main

risk

-takin

g

into

self

-em

plo

yed

emp

loyed

emp

loym

ent

(n=

8,5

61)

(n=

147)

San

dri

etal.

Entr

epre

neu

rsS

tud

ents

Fou

nd

ers

of

Stu

den

tsG

erm

any

E-

NR

Yes

Yes

Yes

No

(2010)

(n=

15)

(n=

84)

pri

vate

firm

s

Lis

tan

dE

ntr

epre

neu

rsS

tud

ents

CE

OS

tud

ents

Cost

aE

-N

RN

oN

oY

esN

o

Maso

n(n

=29)

(n=

101)

Ric

a

(2011)

Bu

rmei

ster

-E

ntr

epre

neu

rsS

tud

ents

Ind

ivid

uals

wh

oS

tud

ents

Ger

many

E-

NR

Yes

No

No

No

Lam

pet

al.

(n=

25)

(n=

29)

start

an

ew

(2012)

bu

sin

ess

Gra

ham

etU

SC

EO

sN

on

-US

US

CE

Os

Non

-US

Un

ited

SH

yp

oth

etic

al

NR

Yes

No

No

Yes

al.

(2013)

(n=

1,0

17)

exec

uti

ves

exec

uti

ves

an

dS

tate

sri

sk-t

akin

g

an

dC

FO

sC

FO

san

d

(n=

1,2

76)

Eu

rop

e

Holm

etal.

Entr

epre

neu

rsP

op

ula

tion

Ow

ner

/fo

un

der

Pop

ula

tion

Ch

ina

E-

RN

oY

esY

esN

o

(2013)

(n=

700)

at

larg

eof

afi

rmat

larg

e

(n=

200)

Note

s.FG,fo

cusgro

up;CG,compariso

ngro

up;S,su

rvey;E,experiment;

R,ra

ndom;NR,nonra

ndom;M

PL,M

ultip

lePriceList.

44

Ap

pen

dix

A.

Lit

era

ture

Overv

iew

(cont’

d)

Focu

sC

om

pari

son

Defi

nit

ion

Defi

nit

ion

Reg

ion

Met

hod

Typ

eof

Sam

pli

ng

Takes

MP

LIn

centi

viz

edF

Gm

ore

Gro

up

(FG

)G

rou

p(C

G)

Focu

sG

rou

pC

om

pari

son

surv

eylo

sses

pro

ne

to

Gro

up

qu

esti

on

sin

toacc

ept

acc

ou

nt

un

cert

ain

ty

Loss

Gach

ter

Entr

epre

neu

rsM

an

ager

s-

-A

ust

ria,

ER

Yes

Yes

Yes

n/a

etal.

(2010)

of

larg

eof

larg

eG

erm

any

Ger

man

car

Ger

man

car

an

d

manu

fact

ure

rsm

anu

fact

ure

rsS

wit

zer-

(n=

65)

(n=

147)

lan

d

Gra

ham

etU

SC

EO

sN

on

-US

US

CE

Os

Non

-US

Un

ited

SH

yp

oth

etic

al

NR

Yes

No

No

Yes

al.

(2013)

(n=

1,0

17)

exec

uti

ves

exec

uti

ves

and

Sta

tes

risk

-takin

g

an

dC

FO

sC

FO

san

d

(n=

1,2

76)

Eu

rop

e

Ambiguity

Sch

ere

Fir

mM

an

ager

sF

irm

fou

nd

ers

Man

ager

sU

nit

edS

Sta

tem

ents

NR

No

No

No

Yes

(1982)

fou

nd

ers

(n=

65)

Sta

tes

(n=

52)

Koh

(1996)

Entr

e-N

ot-

Ch

ara

cter

isti

cs:

Do

not

have

Un

ited

SH

yp

oth

etic

al

NR

No

No

No

Yes

pre

neu

rially

entr

epre

-n

eed

for

chara

cter

isti

csS

tate

sri

sk-t

akin

g

incl

ined

neu

rially

ach

ievem

ent

of

focu

sgro

up

MB

Ain

clin

edlo

cus

of

contr

ol,

stu

den

tsM

BA

pro

pen

sity

to

(n=

22)

stu

den

tsta

ke

risk

,

(n=

32)

tole

ran

ceof

am

big

uit

y

self

-con

fid

ence

an

d

inn

ovati

ven

ess

Mack

oan

dE

ntr

epre

neu

rsS

tud

ents

Bu

sin

ess

Stu

den

tsP

ola

nd

E-

NR

No

No

No

Yes

Tysk

a(2

009)

(n=

40)

(n=

86)

ow

ner

Holm

etal.

Entr

epre

neu

rsP

op

ula

tion

Ow

ner

/fo

un

der

Pop

ula

tion

Ch

ina

E-

RN

oN

oN

oN

o

(2013)

(n=

700)

at

larg

eof

afi

rmat

larg

e

(n=

200)

Note

s.FG,fo

cusgro

up;CG,compariso

ngro

up;S,su

rvey;E,experiment;

R,ra

ndom;NR,nonra

ndom;M

PL,M

ultip

lePriceList.

45

Appendix B. Example Survey Cover Letter (translated)

Dear relation of ACE,

Since its establishment in 2006, the Amsterdam Center for Entrepreneurship (ACE)

has been conducting high-quality research in the field of entrepreneurship. We aim to

continue this ambition into the future.

That’s why we have initiated a new large-scale study in collaboration with Synpact

and VNO-NCW De Baak, which explores differences in decision-making between en-

trepreneurs, managers and employees. We would greatly appreciate it if you would

be willing to participate in our new unique study, which includes making choices that

have real financial consequences. While your participation will predominantly be an

important contribution to science, the results of this research will also be used to de-

velop training material for entrepreneurs, managers and employees.

Our questionnaire is online and will take no longer than 20 minutes of your time. De-

pending on your decisions and luck, you can win an amount up to €675 if you are

selected as a prize winner. To avoid any conflicts of interest, a civil-law notary will

monitor the drawing of the prize winners, and will make sure that the draw obliges

with all legal requirements.

We are very enthusiastic about the value of such insights and we strongly believe that

the study outcomes can also be beneficial for you. We will therefore offer interested

respondents the opportunity to receive a free individual report containing the main

results of this study. However, for this report to be really valuable we need the partic-

ipation of many people. That is, we need you.

We look forward to hearing from you. To participate please click on the link below:

https://uvafeb.qualtrics.com/SE/?SID=SV 2mnigp2XTFh3xXL

Can we kindly request you to finalize the survey before October 16th? Since VNO-

NCW De Baak, Synpact and ACE are jointly sending out this survey, it might be that

cross-postings will occur. Our sincere apologies for this in advance. We only require

your participation once.

46

Appendix C. Risk, Loss, and Ambiguity Aversion Within Entrepreneurs

(1) (2) (3) (4)

Dep. variable: Risk Risk Loss Ambiguity

Aversion Aversion Aversion Aversion

Survey (S) or Experimental (E) (S) (E) (E) (E)

i) Incorporated (n = 446) -0.187∗∗∗ -0.116∗ -0.067 0.118

[-2.70] [-1.66] [-0.95] [1.59]

ii) Above median no. of FTE (n = 401) -0.169∗∗ -0.005 -0.001 0.097

[-2.39] [-0.07] [-0.02] [1.28]

iii) Above median ent. income (n = 377) -0.207∗∗∗ -0.035 -0.003 -0.052

[-2.86] [-0.51] [-0.04] [-0.71]

iv) Founder (n = 757) -0.061 -0.050 -0.113 -0.137

[-0.72] [-0.59] [-1.33] [-1.41]

v) In survival phase -0.014 0.026 0.039 -0.162∗

(firm age ≤ 5 years, n = 347) [-0.18] [0.32] [0.48] [-1.87]

Control variables YES YES YES YES

This table reports risk, loss, and ambiguity aversion differences within entrepreneurs, including all

controls but without income (which is another proxy for success). Significance at the 10% level is

denoted by *, 5% by **, and 1% by ***, with t-statistics reported in parentheses. Standard errors are

robust.

47

Appendix D1. Risk, Loss, and Ambiguity Aversion of Entrepreneurs vs.

Others

(1a) (1b) (2a) (2b) (3a) (3b)

Dep. variable: Risk Risk Loss Loss Ambiguity Ambiguity

Aversion Aversion Aversion Aversion Aversion Aversion

Survey (S) or Exp. (E) (E) (E) (E) (E) (E) (E)

Entrepreneur -0.191∗∗∗ -0.161∗∗∗ -0.168∗∗∗ -0.199∗∗∗ 0.063 -0.003

[-4.23] [-2.90] [-3.70] [-3.61] [1.34] [-0.05]

Age -0.025∗ -0.034∗∗ -0.015 -0.009 0.008 0.012

[-1.82] [-2.14] [-1.12] [-0.58] [0.64] [0.79]

Age2 / 100 0.028∗ 0.036∗∗ 0.017 0.012 -0.013 -0.016

[1.83] [2.08] [1.13] [0.71] [-0.89] [-1.00]

Female 0.058 0.014 0.074 0.038 -0.018 -0.014

[1.23] [0.26] [1.62] [0.69] [-0.37] [-0.26]

Education 0.022 0.030 0.035

[0.70] [0.93] [1.08]

Experience 0.003 -0.001 -0.004

[1.05] [-0.20] [-1.25]

Ln(income) -0.025 -0.070∗∗ 0.053

[-0.78] [-2.15] [1.56]

constant 2.401∗∗∗ 2.830∗∗∗ 1.549∗∗∗ 2.123∗∗∗ 0.977∗∗∗ 0.329

[8.19] [6.57] [5.67] [5.06] [3.44] [0.75]

Obs. 2,288 1,805 2,288 1,805 2,288 1,805

Log lik. -5,055.4 -3,970.4 -4,920.8 -3,865.8 -4,919.7 -3,841.1

This table reports risk, loss, and ambiguity aversion of entrepreneurs and others (managers and

employees combined), including controls. Significance at the 10% level is denoted by *, 5% by **, and

1% by ***, with t-statistics reported in parentheses. Standard errors are robust.

48

Appendix D2. Risk, Loss, and Ambiguity Aversion of Entrepreneurs and Man-

agers in Firms ≤ 15 yrs

(1) (2) (3) (4)

Dep. variable: Risk Risk Loss Ambiguity

Aversion Aversion Aversion Aversion

Survey (S) or Experimental (E) (S) (E) (E) (E)

Entrepreneur (n = 641) -0.686∗∗∗ -0.315∗∗∗ -0.212∗∗∗ -0.010

[-8.21] [-3.98] [-2.66] [-0.12]

Manager (n = 81) -0.230∗∗ -0.195 0.055 -0.186

[-2.09] [-1.63] [0.44] [-1.46]

Age 0.038∗∗ -0.018 0.014 0.005

[1.99] [-1.02] [0.77] [0.28]

Age2 / 100 0.028 0.023 -0.018 -0.008

[1.38] [1.15] [-0.87] [-0.42]

Female 0.148∗∗ 0.001 0.033 -0.028

[2.45] [0.02] [0.53] [-0.44]

Education 0.081∗∗ -0.006 0.068∗ 0.011

[2.04] [-0.15] [1.85] [0.30]

Experience 0.004 -0.003 0.001 -0.005

[1.17] [-0.67] [0.22] [-1.18]

Ln(income) -0.104∗∗ 0.028 -0.079∗∗ 0.053

[-2.50] [0.70] [-2.05] [1.29]

constant -2.327∗∗∗ -2.104∗∗∗ -1.728∗∗∗ -0.516

[-4.24] [-4.11] [-3.39] [-0.97]

Obs. 1,342 1,342 1,342 1,342

Log lik. -2,509.4 -2,958.9 -2,890.6 -2,881.5

ENT = MAN1 < 0.01*** 0.30 0.03** 0.15

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

This table reports risk, loss, and ambiguity aversion of entrepreneurs and managers in firms ≤ 15

years, and employees, including controls. Significance at the 10% level is denoted by *, 5% by **, and

1% by ***, with t-statistics reported in parentheses. Standard errors are robust.

49

Appendix D3. Risk, Loss, and Ambiguity Aversion of Entrepreneurs and Man-

agers in Firms ≤ 25 FTE

(1) (2) (3) (4)

Dep. variable: Risk Risk Loss Ambiguity

Aversion Aversion Aversion Aversion

Survey (S) or Experimental (E) (S) (E) (E) (E)

Entrepreneur (n = 808) -0.633∗∗∗ -0.283∗∗∗ -0.159∗∗ 0.001

[-8.69] [-3.86] [-2.23] [0.00]

Manager (n = 53) -0.134 -0.243∗ 0.126 0.316∗∗

[-0.89] [-1.69] [0.90] [2.00]

Age 0.037∗∗ -0.024 -0.006 0.023

[2.20] [-1.38] [-0.38] [1.40]

Age2 / 100 -0.027 0.028 0.005 -0.028

[-1.47] [1.47] [0.27] [-1.55]

Female 0.144∗∗ -0.011 0.028 -0.015

[2.44] [-0.18] [0.46] [-0.25]

Education 0.075∗∗ -0.001 0.059∗ 0.033

[2.09] [-0.00] [1.65] [0.91]

Experience -0.003 -0.001 0.003 -0.005

[-0.98] [-0.18] [0.76] [-1.31]

Ln(income) -0.106∗∗∗ 0.017 -0.065∗ 0.025

[-2.65] [0.45] [-1.70] [0.65]

constant -2.357∗∗∗ -2.295∗∗∗ -1.974∗∗∗ -0.323

[-4.56] [-4.67] [-4.06] [-0.64]

Obs. 1,434 1,434 1,434 1,434

Log lik. -2,689.2 -3,156.0 -3,084.8 -3,061.3

ENT = MAN1 < 0.01*** 0.77 0.04** 0.04**

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

This table reports risk, loss, and ambiguity aversion of entrepreneurs and managers in firms ≤ 25

FTEs, and employees, including controls. Significance at the 10% level is denoted by *, 5% by **,

and 1% by ***, with t-statistics reported in parentheses. Standard errors are robust.

50

Appendix E1. Cross-Occupational Experience of Entrepreneurs, Managers and

Employees

Entrepreneurs Managers Employees

(n = 910) (n = 397) (n = 981)

% with Managerial Experience in the past 70.7 - -

Level of past man. Experience (scale: 1-5) 1.72 - -

% that is also Employee now 9.0 - -

% that is also Entrepreneur now - 16.9 10.1

Level of current ent. experience (scale: 1-8) - 1.81 1.33

% with entrepreneurial experience in the past - 12.1 8.9

Level of past ent. experience (scale: 1-8) - 2.33 1.71

The ‘level of managerial experience’ is measured based on a question about the number of directly re-

porting subordinates when and if entrepreneurs were managers beforehand. The answering categories

that we coded 1 to 5, respectively, are: 2-5 // 6-10 // 11-25 // 26-50 // More than 50. The ‘level of

entrepreneurial experience’ measure is based on the categorized answers to managers and employees

how many fulltime equivalent people they employed when they were entrepreneurs. This question was

posed only to those who had been entrepreneurs in the past. Answer categories were: 0 // 1-4 //

5-10 // 11-25 // 26-100 // 101-250 // 251-1,000 // More than 1,000 employees. The first answer (0)

corresponds with a value of 1, the second answer (1-4) with a value of 2, and so on.

51

Appendix E2. Cross-Occupational Experience and Risk, Loss, and Ambiguity

Aversion

(1) (2) (3)

Dep. variable: Risk Loss Ambiguity

Aversion Aversion Aversion

Survey (S) or Experimental (E) (E) (E) (E)

Entrepreneur -0.351∗∗∗ -0.257∗∗∗ 0.062

[-4.30] [-3.27] [0.75]

Entrepreneur x Also employee 0.198 -0.165 0.021

[1.64] [-1.25] [0.15]

Entrepreneur x Level of past mgmt experience 0.032 0.036 -0.053∗

[1.24] [1.36] [-1.80]

Manager -0.264∗∗∗ -0.008 -0.031

[-3.16] [-0.10] [-0.34]

Manager x Also entrepreneur 0.068 -0.056 0.060

[0.47] [-0.45] [0.42]

Manager x Level of past ent. experience 0.004 -0.017 0.074

[0.08] [-0.37] [1.22]

Employee x Also entrepreneur -0.127 -0.126 -0.092

[-1.09] [-1.25] [-0.81]

Employee x Level of past ent. experience 0.005 0.011 -0.044

[0.21] [0.37] [-1.53]

Control variables YES YES YES

Obs. 1,805 1,805 1,805

Log lik. -3,967.6 -3,863.3 -3,837.4

ENT = MAN1 0.31 < 0.01*** 0.33

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

This table reports risk, loss, and ambiguity aversion of entrepreneurs, managers and employees, in-

cluding controls for cross-occupational experiences and interactions. Control variables are the same as

in Table 2.5. Significance at the 10% level is denoted by *, 5% by **, and 1% by ***, with t-statistics

reported in parentheses. Standard errors are robust.

52

Appendix F1. Descriptive Statistics of the Measures of Risk Aversion and Loss

Aversion (Second Experiment)

Panel A: Means Obs. Mean Median St. dev. Min. Max.

Risk Aversion

- Survey measure1 1,931 3.40 3 1.70 0 10

- Experimental measure (gain domain) 1,931 5.60 6 2.43 0 10

- Experimental measure (loss domain) 1,931 3.81 4 2.20 0 10

Loss Aversion 1,931 3.30 4 2.21 0 8

Lambda (λ) 1,931 4.48 1.71 12.17 0.06 264

Willingness to Accept (WTA) 983 € 10.22 9 € 7.31 0 21

Willingness to Pay (WTP) 948 € 5.54 5 € 5.84 0 21

Panel B: Correlations Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or

Experimental (E) (S) (E) (E) (E) (E)

Risk Aversion

- Survey measure1 -

- Experimental measure (gain domain) 0.23 *** -

- Experimental measure (loss domain) 0.08 *** 0.09 *** -

Loss Aversion 0.24 *** 0.28 *** 0.17 *** -

Lambda (λ) -0.04 -0.21 *** -0.33 *** 0.20 *** -

Willingness to Accept (WTA) -0.05 ** -0.04 -0.05 ** 0.03 0.00

Willingness to Pay (WTP) -0.06 *** -0.04 -0.01 -0.04 ** 0.05 **

1 Reverse coded measure of “Willingness to take risks”

53

Appendix F2. Descriptives of Variables Used for Stricter Definitions (Second

Experiment)

Entrepreneurs Managers Employees

(n = 697) (n = 265) (n = 969)

Panel A: General Panel A: General

Age 49.11 Age 47.22 43.88

Female (dummy) 0.28 Female (dummy) 0.32 0.44

Education (highest level): Education (highest level):

- High School 5% - High School 2% 11%

- Lower vocational degree 14% - Lower vocational degree 4% 33%

- College education 46% - College education 42% 37%

- University education 35% - University education 52% 19%

Experience (years) 14.27 Experience (years) 14.00 21.03

Income 1) €99,490 Income 1) €97,586 €38,066

Panel B: Entrepreneur characteristics Panel B: Manager characteristics

Founder 82% CEO 16% -

Business taken over 18% General Manager 62% -

Joined firm within 5 yrs 0% Project Manager 22% -

Panel C: Firm age and legal structure Panel C: Firm age and size

Start-up phase (0 - 3 yrs) 20% Firm age ≤ 5 yrs 4% 6%

Survival phase (0 - 5 yrs) 38% Firm age 6 - 50 yrs 49% 55%

Firm age > 50 yrs 47% 39%

Incorporated 44% Firm size ≤ 25 FTE 11% 12%

Sole propriotership 44% Firm size 26 - 1000 FTE 50% 48%

Other 12% Firm size > 1000 FTE 39% 40%

Panel D: Management level Panel D: Management level

No. of FTE in own firm: Direct reports:

Less than 2 47% 2 - 5 38% -

2 - 5 23% 6 - 10 29% -

6 - 10 10% 11 - 25 22% -

11 - 25 10% 26 - 50 8% -

26 - 50 4% More than 50 3% -

More than 50 6%

1) For income, the no. of observations drops to: 442 (entrepreneurs), 233 (managers), and 817 (employees).

54

Appendix F3. Background Characteristics of Entrepreneurs, Managers, and

Employees (Second Experiment)

Entrepreneurs Managers Employees

(n = 697) (n = 265) (n = 969)

Age 49.11 a,b 47.22 b,c 43.88 a,c

Female (dummy) 0.28 a 0.32 c 0.44 a,c

Education (highest degree): d,e e,f d,f

- High School 5% 2% 11%

- Lower intermediate vocational degree 14% 4% 33%

- College education 46% 42% 37%

- University education 35% 52% 19%

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

d) Significant difference between entrepreneurs and employees at the 5% level (Kolmogorov-Smirnov test)

e) Significant difference between entrepreneurs and managers at the 5% level (Kolmogorov-Smirnov test)

f) Significant difference between managers and employees at the 5% level (Kolmogorov-Smirnov test)

55

Appendix F4. Raw Differences in Risk Aversion and Loss Aversion (Second

Experiment)

Risk Risk Risk Loss Willingness Willingness Lambda

Aversion Aversion Aversion Aversion to Accept to Pay

(gain) (loss) (WTA) (WTP) (λ)

Survey (S) or

Experimental (E) (S) (E) (E) (E) (E) (E) (E)

Correspondence

with exp. one exact exact - times 50 - - -

Entrepreneurs 2.88 a,b 5.16 a 3.86 2.68 a,b € 9.46 a,b € 5.73 3.37 a,b

Managers 3.33 b,c 5.09 c 3.97 3.51 b € 11.56 b € 6.21 5.36 b

Employees 3.79 a,c 6.07 a,c 3.74 3.70 a € 10.53 a € 5.27 4.88 a

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

Explanation Of The Final Column In Appendix F4

In the main text we note that, by making parametric assumptions, the choices in the non-

mixed and mixed domains can be used to obtain individual specific estimates of the extent

of loss aversion. Here we elaborate on how these inferred values of loss aversion coefficient λ

are calculated.

Using a common utility specification (cf. Kobberling and Wakker 2005), a respondent is

indifferent between accepting and rejecting a lottery that gives a 50% chance of wining G

and a 50% chance of losing L when (see equation (8) in Abdellaoui et al. 2008):

δ+u(G) + δ−λu(L) = u(0) = 0

where δ+ and δ− represent probability weighting (of p = 12) in the gain and loss domain,

u(G) and u(L) utility curvature in the gain and loss domain, and λ > 0 the loss aversion coef-

ficient. Careful elicitation methods have been developed that allow parameter-free elicitation

of these different components (cf. Abdealloui 2000, Abdellaoui et al. 2007, and Abdellaoui et

al. 2008). Unfortunately, these methods are rather time consuming and as such unsuitable

for our subject pool. But if one is willing to make the simplifying assumption that utility is

linear for the moderate amounts of money involved,20 we can use the three lottery choices to

20Wakker and Deneffe (1996) and Rabin (2000) argue that for small to moderate amounts of money,utility is approximately linear. Studies that measure utility curvature under prospect theory therefore

56

obtain an unconfounded measure of loss aversion. Note that for u(x) = x the loss aversion

parameter follows from:

λ =δ+

δ−×[−GL

]

With linear utility, δ+ follows from the certainty equivalent of the lottery in the gain

domain (winning €300 with 50% chance). Similarly, δ− follows from the point of indifference

for the lottery in the loss domain (losing €300 with 50% chance). A weight δ+ below one

half then reflects risk aversion for gains, while δ− below one half reflects risk loving for losses.

With the ratio δ+

δ− inferred from the choices in the gain and loss domain and by using (2)

above, loss aversion coefficient λ follows from the loss L that makes the respondent indifferent

in the mixed domain (i.e. indifferent between winning €300 with 50% chance and losing €Lwith 50% chance, versus getting 0 for sure). This yields an individual measure of loss aversion

coefficient λ corrected for individual heterogeneity in probability weighting.

The observed differences across occupational groups (see the final column in Appendix

F4) keep standing. The average loss aversion coefficient is somewhat higher than the ones

typically observed in the literature. This is mainly due to the fact that for some subjects the

correction for probability weighting δ+

δ− takes an unrealistic high value. A better comparison

is therefore given by the median values of λ reported in the main text (1.71 for entrepreneurs,

1.87 for managers and 1.82 for employees), which compare reasonably well with other findings

in the literature (see e.g. the median value of 2.61 found in Abdellaoui et al. 2008, and of

2.25 in Tversky and Kahneman 1992).

typically rely on prospects with considerable amounts of money (often in the order of magnitude of acouple of thousands euros), see e.g. Abdellaoui et al. (2007), Abdellaoui et al. (2008), and Booijand Van der Kuilen (2009).

57

Appendix F5. Risk Aversion and Loss Aversion of Entrepreneurs, Managers and

Employees (Second Experiment)

(1) (2) (3) (4) (5)

Dep. variable: Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or

Experimental (E) (S) (E) (E) (E) (E)

Entrepreneur -0.593∗∗∗ -0.392∗∗∗ 0.139∗ -0.282∗∗∗ -1.566∗

[-7.51] [-5.06] [1.80] [-3.50] [-1.66]

Manager -0.339∗∗∗ -0.400∗∗∗ 0.128 -0.046 0.209

[-3.56] [-4.56] [1.47] [-0.54] [0.22]

Age 0.009 0.022 0.022 0.008 -0.197

[0.57] [1.43] [1.35] [0.47] [-0.95]

Age2 / 100 -0.002 -0.021 -0.030∗ -0.017 0.259

[-0.11] [-1.20] [-1.68] [-0.83] [1.11]

Female 0.114∗∗ 0.0806 -0.182∗∗∗ 0.221∗∗∗ 2.005∗∗∗

[1.97] [1.35] [-3.14] [3.73] [2.81]

Education 0.073∗∗ 0.004 0.092∗∗∗ 0.080∗∗ -0.104

[2.37] [0.11] [2.78] [2.40] [-0.25]

Experience 0.001 -0.005 0.004 0.002 -0.026

[0.24] [-1.42] [1.04] [0.58] [-0.65]

Ln(income) -0.031 -0.127∗∗∗ -0.063∗ -0.084∗∗ 0.023

[-0.74] [-3.31] [-1.65] [-2.17] [0.05]

constant 2.124∗∗∗ 2.824∗∗∗ 1.369∗∗∗ 1.736∗∗∗ 7.875

[3.96] [5.83] [2.82] [3.31] [1.22]

Obs. 1,492 1,492 1,492 1,492 1,492

Log lik. -2,690.1 -3,138.5 -3,128.4 -2,988.9 -5,915.0

ENT=MAN1 < 0.01*** 0.91 0.89 < 0.01*** 0.08 *

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

58

Appendix F6. Relationship between Subjective and Objective Measures (Second

Experiment)

(1) (2)

Dep. variable: Risk Risk

Aversion Aversion

Survey (S) or Experimental (E) (S) (S)

Risk Aversion, gain (E) 0.176∗∗∗ 0.151∗∗∗

[7.90] [6.90]

Risk Aversion, loss (E) 0.037 0.050∗∗

[1.63] [2.25]

Loss Aversion (E) 0.178∗∗∗ 0.144∗∗∗

[7.79] [6.28]

Entrepreneur -0.409∗∗∗

[-8.73]

Manager -0.199∗∗∗

[-2.90]

constant < 0.001 0.175∗∗∗

[0.00] [5.63]

Obs. 1931 1931

Log lik. -2653.6 -2618.1

This table reports the output of running the same regressions as in Table 2.6 in the main text. The

only difference is that the variables are now standardized since the new loss aversion measure does

not run from 0 to 10 (like all risk aversion measures), but from 0 to 8. Significance at the 10% level

is denoted by *, 5% by **, and 1% by ***, with t-statistics reported in parentheses. Standard errors

are robust.

59

Appendix F7. Differences in Risk Attitude using Stricter Definitions (Second

Experiment)

(1) (2) (3) (4) (5)

Dep. variable: Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or Experimental (E) (S) (E) (E) (E) (E)

Panel A: Subsets of Entrepreneurs

i) Incorporated (n = 306) -0.580 a,b -0.429 a 0.195 -0.870 a,b -1.615 b

[-5.36] [-4.13] [1.88] [-3.65] [-1.27]

ii) Above median no. of FTE (n = 310) -0.619 a,b -0.394 a 0.217 a -0.596 a,b -0.229

[-6.42] [-4.15] [2.30] [-2.98] [-0.21]

iii) Above median ent. income (n = 189) -0.495 a,b -0.249 a 0.150 -0.998 a,b -1.344 b

[-3.95] [-2.20] [1.25] [-3.62] [-0.89]

iv) Founder (n = 626) -0.618 a,b -0.415 a 0.005 -0.647 a,b -0.001

[-7.22] [-5.03] [0.06] [-3.53] [-0.00]

v) In survival (firm ≤ 5 yrs, n = 221) -0.518 a,b -0.365 a 0.133 -0.517 a,b -1.236

[-6.08] [-4.46] [1.62] [-3.62] [-0.75]

vi) Not in surv. (firm > 5 yrs, n = 476) -0.484 a,b -0.357 a 0.171 a -0.522 a,b -1.780 b

[-5.51] [-3.40] [2.08] [-2.70] [-1.58]

β(Entrepreneur) in Appendix F5: -0.593 a,b -0.392 a 0.139 -0.282 a,b -1.566 b

Panel B: Subsets of Managers

vii) CEO or general manager (n = 230) -0.344 b,c -0.408 c 0.060 -0.238 b 0.090 b

[-3.41] [-4.40] [0.67] [-1.10] [0.27]

viii) CEO (n = 46) -0.386 b,c -0.485 c 0.095 -0.216 0.477

[-2.16] [-2.95] [0.57] [-1.31] [0.24]

ix) Above median dir. reports (n = 168) -0.350 b,c -0.405 c 0.071 -0.143 b 1.634 b

[-3.13] [-4.18] [0.68] [-0.59] [1.50]

x) Above median man. income (n = 183) -0.370 b,c -0.466 c 0.126 -0.053 b 1.494 b

[-3.09] [-4.21] [1.22] [-0.21] [1.29]

xi) Manager in a firm > 15 yrs (n = 272) -0.343 b,c -0.407 c 0.077 -0.171 b 0.527 b

[-3.40] [-4.51] [0.86] [-0.79] [0.61]

β(Manager) in Appendix F5: -0.339 b,c -0.400 c 0.128 -0.046 b 0.209 b

Panel C: Combinations of A&B

i) vs. viii); p-values Wald tests 0.12 0.74 0.46 0.85 0.43

ii) vs. ix); p-values Wald tests 0.02 0.80 0.14 0.02 0.01

iii) vs. x); p-values Wald tests 0.09 0.47 0.63 < 0.01 0.05

60

Appendix F8. Risk and Loss Aversion Within Entrepreneurs (n = 697, Second

Experiment)

(1) (2) (3) (4)

Dep. variable: Risk Risk Risk Loss

Aversion Aversion Aversion Aversion

(gain) (loss)

Survey (S) or Experimental (E) (S) (E) (E) (E)

i) Incorporated (n = 306) 0.087 -0.145∗ 0.175∗∗ -0.008

[1.04] [-1.77] [2.19] [-0.09]

ii) Above median no. of FTE (n = 310) 0.062 -0.093 0.068 -0.015

[0.08] [-1.16] [0.86] [-0.18]

iii) Above median ent. income (n = 377) 0.025 -0.083 0.129 -0.110

[0.26] [-0.98] [1.44] [-1.19]

iv) Founder (n = 626) -0.246∗∗ 0.048 -0.191∗∗ -0.138

[-2.46] [0.46] [-2.04] [-1.41]

v) In survival phase (firm ≤ 5 yrs, n = 221) 0.022 -0.024 -0.140 -0.131

[0.24] [-0.25] [-1.47] [-1.38]

Control variables YES YES YES YES

This table is the equivalent of Appendix C and reports risk and loss aversion differences within

entrepreneurs, including all controls but without income (which is another proxy for success). Sig-

nificance at the 10% level is denoted by *, 5% by **, and 1% by ***, with t-statistics reported in

parentheses. Standard errors are robust.

61

Appendix F9. Risk Aversion and Loss Aversion of Entrepreneurs vs. Others

(Second Experiment)

(1) (2) (3) (4) (5)

Dep. variable: Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or

Experimental (E) (E) (E) (E) (E) (E)

Entrepreneur -0.435∗∗∗ -0.206∗∗∗ 0.080 -0.261∗∗∗ -1.662∗∗

[-6.46] [-3.23] [1.24] [-3.83] [-1.99]

Age 0.003 0.015 0.024 0.008 -0.193

[0.17] [0.94] [1.50] [0.42] [-0.93]

Age2 / 100 0.001 -0.018 -0.031∗ -0.016 0.258

[0.04] [-1.01] [-1.73] [-0.82] [1.10]

Female 0.110∗ 0.076 -0.181∗∗∗ 0.221∗∗∗ 2.007∗∗∗

[1.91] [1.28] [-3.12] [3.72] [2.83]

Education 0.056∗ -0.016 0.098∗∗∗ 0.078∗∗ -0.094

[1.81] [-0.47] [2.99] [2.36] [-0.23]

Experience 0.005 0.001 0.002 0.003 -0.029

[1.62] [0.09] [0.60] [0.80] [-0.72]

Ln(income) -0.089∗∗ -0.195∗∗∗ -0.041 -0.092∗∗∗ 0.059

[-2.34] [-5.40] [-1.14] [-2.59] [0.18]

constant 2.841∗∗∗ 3.662∗∗∗ 1.093∗∗ 1.836∗∗∗ 7.425

[5.66] [7.93] [2.43] [3.78] [1.37]

Obs. 1,492 1,492 1,492 1,492 1,492

Log lik. -2,696.6 -3,147.6 -3,129.4 -2,989.0 -5,915.0

This table is the equivalent of Appendix D1 and reports risk and loss aversion of entrepreneurs and

others (managers and employees combined), including controls. Significance at the 10% level is denoted

by *, 5% by **, and 1% by ***, with t-statistics reported in parentheses. Standard errors are robust.

62

Appendix F10. Risk Aversion and Loss Aversion of Entrepreneurs and Managers

in Firms ≤ 15 yrs (Second Experiment)

(1) (2) (3) (4) (5)

Dep. variable: Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or

Experimental (E) (S) (E) (E) (E) (E)

Entrepreneur (n = 536) -0.632∗∗∗ -0.395∗∗∗ 0.164∗ -0.365∗∗∗ -1.679∗

[-7.35] [-4.69] [1.92] [-4.15] [-1.66]

Manager (n = 51) -0.363∗∗ -0.268 0.203 0.338∗∗∗ 0.297

[-2.17] [-1.61] [1.23] [3.06] [0.16]

Age 0.008 0.015 0.025 -0.005 -0.233

[0.46] [0.90] [1.48] [-0.27] [-1.14]

Age2 / 100 0.001 -0.014 -0.035∗ -0.001 0.003

[0.04] [-0.76] [-1.85] [-0.05] [1.36]

Female 0.151∗∗ 0.109 -0.216∗∗∗ 0.208∗∗∗ 1.584∗∗

[2.38] [1.64] [-3.38] [3.21] [2.17]

Education 0.050 -0.005 0.090∗∗ 0.094∗∗∗ -0.126

[1.47] [-0.15] [2.49] [2.60] [-0.30]

Experience 0.001 -0.004 0.006 0.003 -0.029

[0.00] [-0.96] [1.43] [0.69] [-0.64]

Ln(income) -0.001 -0.098∗∗ -0.074∗ -0.077∗ -0.202

[-0.03] [-2.40] [-1.81] [-1.87] [-0.41]

constant 1.888∗∗∗ 2.592∗∗∗ 1.381∗∗∗ 1.915∗∗∗ 10.97∗

[3.30] [5.11] [2.71] [3.46] [1.71]

Obs. 1,196 1,196 1,196 1,196 1,196

Log lik. -2,153.4 -2,519.5 -2,535.2 -2,380.2 -4,637.2

ENT = MAN1 0.10* 0.43 0.81 < 0.01*** 0.28

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

This table is the equivalent of Appendix D2.

63

Appendix F11. Risk Aversion and Loss Aversion of Entrepreneurs and Managers

in Firms ≤ 25 FTE (Second Experiment)

(1) (2) (3) (4) (5)

Dep. variable: Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or

Experimental (E) (S) (E) (E) (E) (E)

Entrepreneur (n = 631) -0.636∗∗∗ -0.385∗∗∗ 0.160∗∗ -0.228∗∗∗ -1.480

[-7.68] [-4.82] [2.04] [-2.74] [-1.60]

Manager (n = 87) -0.162 -0.375∗∗∗ 0.078 -0.031 -0.048

[-1.21] [-3.22] [0.65] [-0.27] [-0.03]

Age 0.008 0.019 0.027 0.003 -0.213

[0.50] [1.17] [1.59] [0.14] [-1.08]

Age2 / 100 0.000 -0.001 -0.038∗∗ -0.013 0.292

[0.00] [-0.97] [-2.03] [-0.64] [1.33]

Female 0.125∗∗ 0.096 -0.203∗∗∗ 0.209∗∗∗ 1.604∗∗

[2.03] [1.49] [-3.24] [3.32] [2.32]

Education 0.068∗∗ -0.007 0.088∗∗ 0.087∗∗ -0.154

[2.05] [-0.19] [2.55] [2.47] [-0.39]

Experience 0.001 -0.004 0.006 0.006 -0.024

[0.17] [-1.16] [1.47] [1.45] [-0.55]

Ln(income) -0.035 -0.137∗∗∗ -0.063 -0.095∗∗ -0.239

[-0.78] [-3.33] [-1.55] [-2.27] [-0.50]

constant 2.173∗∗∗ 2.962∗∗∗ 1.265∗∗ 1.950∗∗∗ 10.88∗

[3.80] [5.74] [2.51] [3.55] [1.74]

Obs. 1,288 1,288 1,288 1,288 1,288

Log lik. -2,317.0 -2,696.0 -2,713.5 -2,580.8 -4,965.8

ENT = MAN1 < 0.01*** 0.93 0.49 0.09* 0.31

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

This table is the equivalent of Appendix D3.

64

Appendix F12. Cross-Occupational Experience of Entrepreneurs, Managers and

Employees

Entrepreneurs Managers Employees

(n = 697) (n = 265) (n = 969)

% with Managerial Experience in the past 70.6 - -

Level of past man.agerial Experience (scale: 1-5) 2.48 - -

% that is also Employee now 12.5 - -

% that is also Entrepreneur now - 11.7 7.8

Level of current ent. experience (scale: 1-8) - 1.42 1.27

% with entrepreneurial experience in the past - 22.2 8.2

Level of past ent. experience (scale: 1-8) - 1.86 1.65

This table is the equivalent of Appendix E1. The ‘level of managerial experience’ is measured based

on a question about the number of directly reporting subordinates when and if entrepreneurs were

managers beforehand. The answering categories that we coded 1 to 5, respectively, are: 2-5 // 6-10

// 11-25 // 26-50 // More than 50. The ‘level of entrepreneurial experience’ measure is based on the

categorized answers to managers and employees how many fulltime equivalent people they employed

when they were entrepreneurs. This question was posed only to those who had been entrepreneurs

in the past. Answer categories were: 0 // 1-4 // 5-10 // 11-25 // 26-100 // 101-250 // 251-1,000 //

More than 1,000 employees. The first answer (0) corresponds with a value of 1, the second answer

(1-4) with a value of 2, and so on.

65

Appendix F13. Cross-Occupational Experience and Risk Aversion and Loss Aver-

sion (Second Experiment)

(1) (2) (3) (4) (5)

Dep. variable: Risk Risk Risk Loss Lambda

Aversion Aversion Aversion Aversion (λ)

(gain) (loss)

Survey (S) or

Experimental (E) (S) (E) (E) (E) (E)

Entrepreneur -0.589∗∗∗ -0.415∗∗∗ 0.207∗∗ -0.224∗∗ -2.785∗∗∗

[-6.50] [-4.79] [2.34] [-2.39] [-2.69]

Entrepreneur x Also employee -0.113 0.107 -0.286∗∗ -0.0600 1.321

[-0.94] [0.80] [-1.97] [-0.44] [0.80]

Entrepreneur x Level of past -0.004 0.001 -0.038 -0.046 0.919∗∗∗

mgmt experience [-0.10] [0.03] [-1.19] [-1.26] [2.78]

Manager -0.330∗∗∗ -0.402∗∗∗ 0.113 -0.099 -0.814

[-3.23] [-4.21] [1.21] [-1.07] [-0.65]

Manager x Also entrepreneur 0.239 -0.213 0.023 0.365∗∗ 7.292∗∗∗

[1.00] [-1.04] [0.12] [2.19] [3.09]

Manager x Level of past -0.107 0.029 -0.033 -0.035 1.589∗∗

ent. experience [-1.23] [0.68] [-0.52] [-0.66] [2.11]

Employee x Also entrepreneur -0.798 0.042 0.027 -0.510 -1.922

[-1.61] [0.08] [0.05] [-1.03] [-0.32]

Employee x Level of past 0.383 -0.107 -0.006 0.254 1.015

ent. experience [1.53] [-0.39] [-0.02] [0.96] [0.33]

Control variables YES YES YES YES YES

Obs. 1,492 1,492 1,492 1,492 1,492

Log lik. -2,687.0 -3,137.5 -3,124.7 -2,953.3 -5,789.7

ENT = MAN1 < 0.01*** 0.89 0.35 0.24 0.10*

1 This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

This table is the equivalent of Appendix E2.

66

Appendix G. Screenshots of the Methodologies and CDF Plots of All Measures

Figure G1: Measure of Risk Aversion

Figure G2: Measure of Loss Aversion

67

Figure G3: Measure of Ambiguity Aversion

Figure G4: CDFs per Measure per Group

68

Figure G5: Measure of Willingness to Accept (WTA)

Figure G6: Measure of Willingness to Pay (WTP)

69

70

Chapter 3

Optimism & Overconfidence

This chapter is based on Koudstaal, M., R. Sloof, and C.M. van Praag (2015), “Are

Entrepreneurs More Optimistic and Overconfident Than Managers and Employees?”,

working paper.

3.1 Introduction

This study aims at testing whether, in what respect, and to what extent entrepreneurs

are more optimistic and overconfident than managers and employees. Conventional

wisdom has it that especially entrepreneurs are among the most optimistic and over-

confident types, and that self-selection into entrepreneurship would be promoted by

such optimism and/or overconfidence.1 Moreover, optimism and overconfidence have

also been put forward as a main explanation for why people persist in entrepreneurship

despite the fact that earnings are lower and riskier on average than in paid employment

(Hamilton, 2000; Moskowitz and Vissing-Jorgensen, 2002).2 Optimism and overconfi-

dence might also impact the level of success of the firm and its way of financing (Bitler

et al., 2005; Puri and Robinson, 2006; and Landier and Thesmar, 2009).

1For optimism, see e.g. Krueger et al. (2000); Hmieleski and Baron (2009); and Puri and Robinson(2013). For overconfidence, see e.g. Abdelsamad and Kindling (1978); Cooper et al. (1988); Kahnemanand Lovallo (1993); Busenitz and Barney (1997); Camerer and Lovallo (1999); Arabsheibani et al.(2000); Simon et al. (2000); Astebro (2003); Lovallo and Kahneman (2003); Fraser and Greene (2006);Puri and Robinson (2006); Crane and Crane (2007); Puri and Robinson (2007); Koellinger et al.(2007); Koellinger (2008); Trevelyan (2008); Landier and Thesmar (2009); Cassar (2010); Ucbasaranet al. (2010); Rietveld et al. (2013); Bengtsson and Ekeblom (2014); Dawson et al. (2014); and Dawsonet al. (2015).

2Alternative explanations for this puzzle include higher risk appetites (e.g., Hvide and Panos,2014), non pecuniary benefits deriving from higher levels of autonomy and control (e.g., Blanchflowerand Oswald, 1998; Hamilton, 2000; Hurst and Pugsley, 2011; and Puri and Robinson, 2013), differentselection and treatment effects of personality traits (e.g. Hamilton et al., 2014), and status concerns(e.g., Parker and Van Praag, 2010). See also Astebro et al. (2014) for an overview of the (behavioral)roots of entrepreneurship.

71

Interestingly, many of these results have also been documented in the literature

on (top) managers. That is, managers have also been found to be more optimistic

and overconfident than others, and a number of empirical studies have showed that

these behavioural traits can account for some of the heterogeneity in corporate policies

(Bertrand and Schoar, 2003; Malmendier and Tate, 2005, 2009; and Graham et al.,

2013). Various theoretical explanations have been put forward to rationalize these

findings, including selection into and out of management, slightly overconfident man-

agers leading to less manager-shareholder conflicts, and overconfident employees having

a higher probability of being promoted to CEO (Goel and Thakor, 2008; Hackbarth,

2008; and Campbell et al., 2011). Van den Steen (2004) provides an alternative reason

why managers (and entrepreneurs) are more overconfident than regular employees. In

his model, overconfidence increases in the number of actions an agent can choose from.

Therefore, “An agent who can choose his own projects will be more optimistic than one

who gets assigned his projects. This may be one reason why entrepreneurs often seem

more overoptimistic than regular employees. It also implies that restricting a manager’s

degree of freedom may reduce her bias.” (p. 1144).

Surprisingly, very little is known about the levels of optimism and overconfidence

of entrepreneurs vis-a-vis managers. Hence, are optimism and overconfidence truly

unique traits of entrepreneurs? Or do these traits rather pertain to strategic decision-

makers in general? And finally, how do the arguably more successful managers and

entrepreneurs compare to each other?

To answers these questions, we provide an encompassing assesment of the level

and kind of optimism and overconfidence of entrepreneurs relative to managers and

employees. Our large lab-in-the-field experiment includes two well-defined survey mea-

sures of optimism, i.e. dispositional optimism (Scheier and Carver, 1985, and Scheier

et al., 1994) and attributional style (Seligman, 2000), and two well-defined incentivized

measures of overconfidence, i.e. overestimation of one’s own ability (e.g., Moore and

Healy, 2008) and overestimation of a future stock market closing price (cf. Bengtsson

and Ekeblom, 2014).

Overall, our study has three distinguishing features relative to the literature. First

and foremost, measuring and comparing an entrepreneur’s (optimistic or pessimistic)

attributional style is novel, and we find it to be a key differentiating factor between

entrepreneurs and managers. The only entrepreneurship study we know of that also

includes the attributional style measure is Krueger et al. (2000). They use a version of

Seligman (2000)’s measure of learned optimism to explain entrepreneurial intentions

among 97 students. In contrast, our sample consists of 1,391 established entrepreneurs

and managers, hence those individuals who have realized their intentions. A second

distinguishing feature of our study is that we compare entrepreneurs with two spe-

72

cific groups, i.e. managers and employees, rather than just the public at large (see

also Koudstaal et al., 2015). Entrepreneurs and managers may be relatively similar

to each other, both being responsible for strategic decisions and for hiring and di-

recting their personnel (see also Table 1.1 in Chapter 1). In addition to these more

functional similarities, our sample also reveals that entrepreneurs and managers are re-

markably similar in terms of their background characteristics. The differences between

entrepreneurs and managers might therefore be more informative on what really makes

an entrepreneur different in terms of their optimism and overconfidence. Employees

are nevertheless an interesting category to compare entrepreneurs and managers to,

and also allow relating our study to previous studies that compare entrepreneurs with

‘others’ (e.g. Puri and Robinson, 2013). Finally, the third distinctive aspect of our

study is the large samples of entrepreneurs, managers and employees, which allows us

to create meaningful subsamples of (more successful) entrepreneurs and managers and

to test the robustness of our findings to alternative definitions of these occupational

groups.3

Based on a sample of 875 entrepreneurs, 516 managers and 667 employees (so n =

2,058), we find that entrepreneurs differ from managers and employees in their dispo-

sitional optimism and in their attributional style. Regarding dispositional optimism,

entrepreneurs are significantly more optimistic than managers, who in turn are more

optimistic than employees. We further show that 58% of entrepreneurs and 54% of

managers are very optimistic, while only 32% of the employees are so. In terms of

attributional style, entrepreneurs appear to be especially unique in their attitude to-

wards dealing with bad events. They take these as more specific and less pervasive -

thus a more optimistic approach - than managers and employees do. These findings

confirm the stereotype that occupations with a high failure rate require an optimistic

explanatory style to persist (Kahneman, 2011). In the area of overconfidence, we find

that entrepreneurs and managers are alike. They both overestimate their own abilities

as well as a future stock market closing price more than employees do, but they do not

differ from each other.

The heterogeneity and other checks yield two additional notable findings. First, we

find that the differences between entrepreneurs and managers in dispositional optimism

and attributional style disappear when limiting the sample of managers to the arguably

3In the baseline sample we take as ‘entrepreneur’ someone who founded, inherited or has taken overa company that s/he is currently (co-)managing and of which s/he has at least 5% of the shares. Thestricter definitions of entrepreneurship we employ focus on those who are arguably more successfuland thus more similar to the ‘Schumpeterian’ entrepreneur. These are the ones with an incorporatedfirm, earning above median income or having an above median number of employees. Managers inthe overall sample are defined as employees in firms not started up by the respondent, and havingat least two direct reports under their responsibility. Here stricter definitions are based on being theCEO, or having above median income or above median number of direct reports. Employees are thepeople who work in organizations and do not belong to the groups of entrepreneurs and managers.

73

more ‘successful’ ones (e.g. CEOs). Optimism and success turn out to be even more

positively correlated for managers than for entrepreneurs, and this might therefore

explain the disappearance of differences between entrepreneurs and managers when

considering solely CEOs. Second, we find that the results on overconfidence are largely

robust to the use of different sample definitions, with two exceptions. First, when

using the definitions of entrepreneurs and managers of Busenitz and Barney (1997),

i.e., business founders who recently started their businesses and managers in firms with

more than 10,000 employees, we find that entrepreneurs do overestimate themselves

more than managers. The second exception in the overestimation results arises when

we compare entrepreneurs in young and small firms to managers in fairly identical

firms with arguably similar information availability. Again we find that entrepreneurs

are more prone to overestimate their own ability than managers. Therefore, the ques-

tion arises to what extent information availability really explains the differences (cf.

Busenitz and Barney, 1997) or whether it is self-selection in the sample of entrepreneurs

(cf. Landier and Thesmar, 2009; Van den Steen, 2011).

Unfortunately our study is purely descriptive and thus unable to identify cause and

effect. In other words, we are unable to detect the source of differences in dispositional

optimism and attributional style between entrepreneurs and managers, i.e. whether

it is an ‘endowment’ or ‘investment’. Similarly, little can be learned from our study

about the potential causal relationship between optimism and success. A number of

existing studies are informative in this regard, though. Longitudinal studies suggest

that dispositional optimism, attributional style and overestimation are relatively stable

traits throughout adult life (see e.g. Burns and Seligman, 1989; Carver et al., 2010; and

Dawson et al., 2014). Dawson et al. (2014) explicitly study whether overestimation

is a trait of future entrepreneurs or whether it is developed during entrepreneurship.

They find stronger evidence for overestimation being a cause than a consequence of

entrepreneurship, but the latter is not zero either. Regarding the potential correlation

between optimism and success, some studies find it to be positive while others find

a non-monotonic relationship.4 By and large the collective evidence seems to suggest

that moderate levels of optimism are good, but extreme levels lower performance.

In what follows, we discuss the data, design and measurement choices in Section

3.2. In Section 3.3 we present the descriptive statistics, while we describe the empirical

findings in Section 3.4. Section 3.5 provides a discussion and conclusion.

4For a positive relationship between optimism and success, see e.g. Seligman and Schulman (1986),Segerstrom (2007), Solberg Nes et al. (2009), and Kaniel et al. (2010). For a negative relationshipbetween being too optimistic or overconfident and success, see e.g. Lowe and Ziedonis (2006), Puriand Robinson (2007), Hmieleski and Baron (2009), Dawson et al. (2014), and Dawson et al. (2015).Apart from better socioeconomic status (like education and income) and broader social networks,optimism has also been found to be positively associated with psychological well-being, better copingwith adverse events and physical health. See Carver et al. (2010) for an informative overview.

74

3.2 Design, measurement and sampling

3.2.1 Measurement of optimism

A variety of definitions and measures of optimism exist in the social sciences literature.

In our study, we rely on two of them: dispositional optimism (Scheier and Carver, 1985

and Scheier et al., 1994) and an optimistic attributional style (Seligman, 2000). We se-

lect these two because they are among the main optimism measures within psychology

(see Peterson, 2000, for an overview), and also among the main psychological measures

of optimism used within economics (see e.g., Krueger et al., 2000; Puri and Robinson,

2006; Hmieleski and Baron, 2009; and Graham et al., 2013, for applications).5 Fur-

thermore, both dispositional optimism and an optimistic attributional style are found

to be relatively stable over time as well as across situations, contexts, and cultures

(see e.g. Burns and Seligman, 1989; Schulman et al., 1993; Scheier et al., 1994; Giltay

et al., 2006; and Fisher and Chalmers, 2008). Finally, the attributional style test also

bears the advantage of being hard to ‘beat’, i.e. test-takers cannot fake optimal re-

sponses (Schulman et al., 1987). The latter is important given that some entrepreneurs

seem to fill out survey questions in line with what they view as being expected from

entrepreneurs (see e.g., Chapter 2). We will now discuss each of the two measures in

detail.

3.2.1.1 Dispositional optimism

Dispositional optimism is the global expectation that good things will be plentiful in the

future and bad things will be scarce (e.g., Scheier and Carver, 1985; Scheier et al., 1994;

and Peterson, 2000). It is measured using the brief 10-item self-report questionnaire

of Scheier et al. (1994), which is also referred to as the Revised Life Orientation Test

(LOT-R), see Figure C1 in Appendix C of Appendices Chapter 3. Among the 10 items

in the LOT-R, three of them are associated with positive expectations (1, 4 and 10)

and three with negative expectations (3, 7 and 9). The remaining four statements are

filler items (2, 5, 6 and 8). Each statement can be answered with either “Strongly

disagree” (0 points), “Disagree” (1 point), “Neutral” (2 points), “Agree” (3 points)

or “Strongly agree” (4 points). To obtain a score for optimism, the scores of all six

non-filler items are added, where the items with negative expectations are reversely

coded. Overall, participants thus obtain a minimum score of 0 points and a maximum

score of 24 points. A participant who answers “Neutral” on all items ends up with a

score of 12.

5An exception is the comparative optimism measure used by Ucbasaran et al. (2010).

75

3.2.1.2 An optimistic attributional style

In contrast to dispositional optimism, focusing more on evaluations of future events,

an optimistic attributional style measures an individual’s explanatory style of past

life events (Buchanan and Seligman, 1995). Optimists are referred to as those who

believe that good events will persist (i.e. are “permanent”) and will extend to other

areas (i.e. are “pervasive”). Bad events are, by contrast, regarded as impermanent

and non-pervasive. Unsurprisingly, for pessimists the opposite is true: they believe

good events to be non-persistent (temporary) and non-pervasive and bad events as

permanent and pervasive. Seligman (2000) further clarifies the aforementioned by

considering an example with two accountants who are redundant in their firm (p. 90).

Both are looking for a new job, but feel depressed due to their sacking. One of them,

however, keeps on going to the gym, stays healthy and remains a loving family member

at home, while the other falls apart and catastrophizes. The latter is what Seligman

typifies as pessimistic behavior, while the former is what he deems the optimistic side

of the same coin.

To measure the level of an optimistic attributional style, we use the 32-item op-

timism test from Seligman (2000), see also Figure C2 in Appendix C of Appendices

Chapter 3. The test generates scores on four variables: PmG (permanence of good

events), PvG (pervasiveness of good events), PmB (permanence of bad events) and

PvB (pervasiveness of bad events). We use the same scoring rule as Seligman (2000).

Furthermore, the sum of PmG and PvG minus the sum of PmB and PvB gives an

indication of the level of optimism in one’s attributional style (range: -16 until 16). A

high score is associated with optimism, as good events are believed to be caused by fac-

tors that are permanent and universal, whereas bad events are explained by temporary

and specific causes. Conversely, low scores on attributional style are associated with

pessimism: pessimists believe bad events to be caused by factors that are permanent

and universal, whereas good events are temporary and specific.

3.2.2 Measurement of overconfidence

The two measures of overconfidence that we use focus on overestimation, or the be-

havioral bias where one can be shown to be too optimistic due to the availability of an

objective ‘right’ estimate. While the general term ‘overconfidence’ has often been used

to describe this bias, Astebro et al. (2014) rightly point out that “multiple measures

and definitions across empirical studies have made it hard to pin down the precise bias

that may be behind entrepreneurship” (p. 58). In earlier work, Moore and Healy (2008)

have therefore made a subtle distinction between overestimation, overplacement and

76

overprecision, even though these terms might seem observationally equivalent. In this

paper we will focus on incentivized measures of overestimation of one’s own ability and

a non-ability related quantity (stock index development).6

3.2.2.1 Overestimation of one’s own ability

The first measure of overconfidence is measured as the difference between a participant’s

assessments of their own performance on ten Raven test questions with a varying level

of difficulty and the actual number of correct answers. Trivia questions or Raven test

questions are generally used to obtain a measure of overestimation (see e.g. Lichtenstein

et al., 1982; Moore and Healy, 2008; and Herz et al., 2014). Before answering the 10

Raven test questions, we provided an example to get participants accustomed with the

task. After all 10 questions had been answered, participants had to indicate the number

of correct answers.7 The incentive was as follows: a correct answer (i.e. forecasted =

actual) was rewarded with €100 and an incorrect answer with €0.

The number of correct answers itself was also measured and used as a proxy for

intelligence in the set of control variables. This is possible because we selected as the

first five Raven puzzles the one prescribed by Bilker et al. (2012), which are shown to

have a correlation of 0.95 with the actual score on the full 60-item Raven Standard

Progressive Matrices (RSPM) test. The last five Raven puzzles were selected from the

shortened 12-item Raven test of Arthur and Day (1994), which are somewhat harder to

answer correctly and therefore create - in combination with the first five - some extra

dispersion across participant scores, especially at the high end.

3.2.2.2 Overestimation of a future stock market closing price

In line with Bengtsson and Ekeblom (2014), our second measure of overconfidence is

based on forecasts of a future stock market closing price. We consider this measure

to be complementary to the previous measure in the sense that this one is uncorre-

lated with the individual’s own life or work situation (Bengtsson and Ekeblom, 2014).

However, the flip side of this advantage is that it also introduces two drawbacks; first,

the realization is similar for everybody and second, there might be a larger scope for

unobserved differences in e.g. financial skills. We are therefore mainly interested if

6Incentivized behavioral measures of overoptimism about the future have been used before by e.g.Weinstein (1980), Dunning et al. (2004), Ben-David et al. (2013) and Bengtsson and Ekeblom (2014).

7A potential critique of this setup might be that more (less) able participants are less (more)likely to overestimate themselves. However, this criticism may be less relevant in our particular case.Only 31 out of 2,058 participants answered either zero or all questions correctly, and only 2 of theseparticipants correctly forecasted that.

77

overestimation of own ability extends to situations that are less under an individual’s

control.

To measure overestimation of a future stock market closing price, participants are

requested to provide a 3-month forecast of the value of the AEX index (the Dutch stock

market index composed of companies that trade on NYSE Euronext Amsterdam). The

measure of overestimation subtracts from this forecast the actual value on the date

for which the forecast was requested (September 1, 2014). To help participants, we

disclosed the rounded AEX closing price of May 6th, which was 397. As an incentive,

selected prize winners were rewarded with €100 if their estimate was within 10 points

from the actual closing price, and €0 otherwise.

3.2.3 Sampling

The data collection uses the same approach as described in Chapter 1, which is a

combination of a lab-in-the-field experiment and an online survey. The procedure to

reach large samples of entrepreneurs, managers and employees was also similar. Again,

for reaching volume in the entrepreneur sample, we teamed up with “Synpact”, a

company that has access to more than 15,000 entrepreneurs in the Netherlands. All of

these entrepreneurs received an invitation to participate in the online research and a

link to the questionnaire and experiments. The control group of managers was again

drafted from the large and highly reputed training center De Baak, which is part of

the largest and influential employers organization in the Netherlands (“VNO-NCW,

MKB-Nederland”). We approached 4,131 managers in their files.8 Finally, the same

invitation and survey were sent to a sample of 7,500 employees, recruited via a Dutch

market research agency with access to over 70,000 Dutch employees.

All invitations were sent out to all groups on May 7th 2014, with the explicit

mentioning of a response time of 14 days at maximum. A reminder was sent out after

7 days. 875 entrepreneurs, 516 managers and 667 employees completed the survey.

Similar to Chapter 2, the response rates were thus in the range of 5 - 15% and in line

with expectations of Synpact and De Baak who regularly send out qualitative surveys to

their database on their own. A comparison of respondents’ background characteristics

in this research wave and the one gathered in November 2013 (see Chapter 2) yields

that the distributions of background variables among entrepreneurs, managers and

employees are generally similar across the two rounds.

8Note that this number is lower than in Chapter 1 due to cleansing of the database by The Baak.

78

Definition of subsamples

The qualifying characteristics for inclusion in the entrepreneur sample were similar

as in Chapter 2. That is; ‘entrepreneurs’ are all people who have founded, inherited

or taken over a company that they are currently (co-)managing. We also classified

participants as ‘entrepreneurs’ when they currently (co-)manage a company which they

joined within 5 years after start-up and of which they have obtained at least 5% of

the company shares.9 ‘Managers’ are all people who are employed by an organization

that they did not start up themselves and have at least two subordinates for whom

they are directly responsible. Project managers also classify in case of overall project

responsibility and at least two direct reporting lines. People belong to the group of

‘employees’, finally, if they are employed by an organization and do not belong to the

first two groups. Participants who were eligible for multiple subsamples were instructed

to select the one generating most of their income.

3.2.4 Incentives

Respondents were requested to first complete the two unincentivized parts (on disposi-

tional optimism and attributional style) and then fill out the two incentivized parts (ten

Raven test questions and the AEX 3 month forecast). All participants were first in-

formed about the general setup of the survey and the incentives structure. Instructions

also included examples and practice rounds to get acquainted with the experimental

setup, which differed from many of the surveys that entrepreneurs and managers nor-

mally fill out. Overall, an average respondent spent 23 minutes on the survey, including

possible breaks. Ex ante, the estimated average earnings per winning respondent were

around €300. Participants could earn a maximum of €450, which consisted of a fixed

fee of €250 and two times €100 that could be earned in the two incetivized parts.

Given budget limitations and the rather high income levels of the participants in our

survey, we chose to pay out a substantial (instead of very small) amount to a few

(instead of all) randomly selected participants. In doing so, we follow e.g. Gneezy

and Rustichini (2000) and Laury (2006) who show that this should produce similar

results as when paying smaller amounts to all participants. In total, we selected 25

prize winners. This was clearly communicated at the beginning of our survey. Hence,

ex post the chance to be paid out was 1/83, but this was unknown to the participants

(and ourselves) ex ante. To further foster trust and truthful reporting, we assigned

the selection of prize winners and all random draws in the experiments to a civil-law

notary who also monitored a legitimate course of the payouts.

9The Dutch tax authority considers a five percent ownership to be a substantial interest.

79

3.3 Descriptive statistics

Panel A of Table 3.1 outlines the descriptive statistics of the measures of optimism and

overconfidence we employ: dispositional optimism, attributional style, overestimation

of own ability and the forecast of the 3-month AEX closing price (which was 414).

Panel B shows the correlations between these measures. We find that individual scores

on all measures vary substantially (Panel A), and that their intercorrelations are pos-

itive but low (Panel B), see also Isaacowitz and Seligman (2001), thus suggesting that

the four measures pick up complementary aspects of optimism.

Table 3.1 Descriptive Statistics of the Optimism and Overconfidence Measures

Panel A: Means Observations Mean St. dev. Min. Max.

Optimism 2,058 16.96 4.08 0 24

- Dispositional optimism

- Attributional style 2,058 2.02 3.35 -15 13

Overconfidence

- Overest. own ability 2,058 0.74 1.65 -6 8

- Overest. AEX 3M closing price 2,058 1.92 32.79 -404 313

Panel B: Correlations Dispositional Attributional Over-

optimism Style estimation

Own ability

Attributional style 0.25 *** -Overest. own ability 0.01 0.05 ** -

Overest. AEX 3M closing price 0.06 *** 0.04 0.02

* Denotes statistical significance at the 10% level; ** at the 5% level; *** at the 1% level.

Table 3.2 shows the statistics of the background characteristics that will define stricter

subsamples of entrepreneurs and managers (see also Chapter 2). The first Panel (A)

shows the income distribution of each of the three samples, using the answer cate-

gories from the survey. Entrepreneurs are over-represented in both tails of the income

distribution relative to managers and employees, consistent with previously obtained

evidence, e.g. Hamilton (2000). The average levels of entrepreneurial and managerial

incomes are similar. For both groups, the median income category is €50,001-€75,000,

while the median employee income falls in the category €25,001-€50,000 (note that

the modal income was €33,000 in The Netherlands in 2014).

80

Table 3.2 Descriptives of Variables to Define Sample Splits

Entrepreneurs Managers Employees

(n = 875) (n = 516) (n = 667)

Panel A: Income Panel A: Income

< €25,000 24% < €25,000 6% 24%

€25,001 - €50,000 21% €25,001 - €50,000 22% 58%

€50,001 - €75,000 16% €50,001 - €75,000 29% 13%

€75,001 - €125,000 23% €75,001 - €125,000 32% 5%

€125,001 - €200,000 10% €125,001 - €200,000 8% 0%

€200,001 - €300,000 3% €200,001 - €300,000 1% 0%

€300,001 - €400,000 1% €300,001 - €400,000 1% 0%

> €400,000 2% > €400,000 1% 0%

Panel B: Entrepreneur characteristics Panel B: Manager characteristics

Founder 80% CEO 14% -

Business taken over 17% General Manager 71% -

Joined firm within 5 yrs 3% Project Manager 15% -

Panel C: Firm characteristics Panel C: Firm characteristics

Start-up phase (0 - 3 yrs) 15% Firm age ≤ 5 yrs 4% 3%

Survival phase (0 - 5 yrs) 31% Firm age 6 - 50 yrs 54% 59%

Firm age > 50 yrs 42% 38%

Incorporated 52% Firm size ≤ 25 FTE 12% 14%

Sole propriotership 37% Firm size 26 - 1000 FTE 54% 53%

Other 11% Firm size > 1000 FTE 34% 33%

Panel D: Management level (FTE) Panel D: Management level (direct reports)

Less than 2 43% 2 - 5 41% -

2 - 5 23% 6 - 10 28% -

6 - 10 9% 11 - 25 17% -

11 - 25 13% 26 - 50 5% -

26 - 50 4% More than 50 2% -

More than 50 8%

81

Panel B further describes the sample of entrepreneurs and managers that partici-

pated in our experiment. 80% of the entrepreneurs are firm founders, 17% of the firms

were acquired by takeover, and the remaining 3% of the entrepreneurs have joined ex-

isting firms within 5 years after start-up and acquired a minimum of 5% of its shares.

In later analyses, we will restrict the sample of entrepreneurs to align with part of

the entrepreneurship literature (e.g. Begley, 1995; Busenitz and Barney, 1997; Sandri

et al., 2010; and Holm et al., 2013). Likewise, for managers, we will analyze subsamples

of CEOs (14%), project managers (15%), and general managers (71%).

Panel C describes the types of firm that entrepreneurs, managers and employees

work in, a basis for the analysis of further relevant subgroups (see Section 3.2). 15%

(31%) of the entrepreneurs are currently managing firms in the start-up (survival) phase

(the definition of entrepreneurs used by, for instance Brockhaus (1980) and Busenitz

and Barney (1997)), whereas the rest is beyond that phase (the definition of Holm

et al., 2013), We will also restrict the sample of entrepreneurs to those (52%) who

are incorporated (see Levine and Rubinstein, 2013). The right handside of Panel C

depicts the age and size distributions of the firms that managers and employees work

for. They are rather similar, but different from the age and size distributions of en-

trepreneurial firms that are younger and smaller. As an additional heterogeneity check

in Section 3.2 we will therefore limit the sample of managers to the ones of small and

young firms, respectively, to form an arguably more ‘entrepreneurial’ benchmark group.

Table 3.3 Background Characteristics

Entrepreneurs Managers Employees

(n = 875) (n = 516) (n = 667)

Age 48.71 a 47.66 c 43.13 a,c

Female (dummy) 0.27 a 0.27 c 0.47 a,c

Education (highest degree): d,e e,f d,f

- High School 6% 4% 7%

- Lower intermediate vocational degree 12% 12% 34%

- College education 45% 38% 39%

- University education 37% 46% 20%

IQ (scale 0-10) 5.93 a 5.93 c 5.24 a,c

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

d) Significant difference between entrepreneurs and employees at the 5% level (Kolmogorov-Smirnov test)

e) Significant difference between entrepreneurs and managers at the 5% level (Kolmogorov-Smirnov test)

f) Significant difference between managers and employees at the 5% level (Kolmogorov-Smirnov test)

82

Panel D of Table 3.2 shows the distribution of the span of control of entrepreneurs and

managers in our sample. 17% of the entrepreneurs in our sample have zero employees

and 43% have at most one. As a final heterogeneity check we shall therefore also limit

the sample of entrepreneurs and managers to those with an above median span of

control (cf. e.g. Tag et al., 2013).

Table 3.3 compares the background characteristics of the three subsamples. As

announed in the Introduction, entrepreneurs and managers are similar in terms of the

most commonly used background characteristics. The only exception this time is that

managers have a significantly higher average degree of education than entrepreneurs.

Employees are different in terms of all background characteristics.

3.4 Results

3.4.1 Main results

Table 3.4 first shows the means of the each of the two measures of optimism and

overconfidence for each of the three groups of interest. Starting with dispositional

optimism in the first column, we find that entrepreneurs have the highest average

score of all groups. Employees follow at some distance, while managers end up in

between (but closer to entrepreneurs). Note that the average score for employees is

close to the 14.33 - 15.15 interval reported by Scheier et al. (1994). Further (unreported)

descriptives reveal that 58% of entrepreneurs, 54% of the managers, and 32% of the

employees can be classified as ‘very optimistic’ (i.e. have a score of 18 or more), which

in the case of managers is comparable to the 54% observed for European CEOs in

Graham et al. (2013).10 All percentages are significantly different from each other at

the 5% level.

The second column of Table 3.4 shows the means of the attributional style measure.

The pattern here is similar as in column 1, although managers now end up closer to

employees than to entrepreneurs. Again, all measured differences between the three

groups are significant. Hence, entrepreneurs deal with past events in a significantly

more optimistic way than managers, who in turn are more optimistic than employees.

The last two columns of Table 3.4 pertain to overconfidence and show the mean values

of the incentivized overestimation measures. Entrepreneurs overestimate themselves

most of all three groups followed by managers and employees (3rd column). The differ-

ence between entrepreneurs and employees is significant (5% level), while the difference

10Note that these percentages still seem much lower than what is found for US CEOs (Graham et al.,2013) and what seems to be the case for US entrepreneurs (Hmieleski and Baron, 2009). AlthoughHmieleski and Baron (2009) do not report actual percentages, their average LOT-R score suggeststhat US entrepreneurs are more optimistic than the entrepreneurs who have participated in this study.

83

between entrepreneurs and managers is not. Note that on average all groups overes-

timate themselves. When examining the overestimation measure related to the stock

market forecast, this picture is reinforced (4th column). Entrepreneurs and managers

exhibit more overestimation than employees, but not than each other. The latter result

is based on the winsorized part of the distribution (i.e. the 99% of the distribution

which excludes the ends of the tails), to avoid a large role of outliers.11

Table 3.4 Raw Differences in Optimism and Overconfidence

Optimism Overconfidence

Dispositional Attributional Overestimation Overestimation

optimism Style Own ability AEX 3Mclosing price

(n = 2,058) (n = 2,058) (n = 2,058) (n = 2,017)

Entrepreneurs 17.87 a,b 2.68 a,b 0.84 a 2.99 a

Managers 17.52 b,c 1.86 b,c 0.82 c 4.27 c

Employees 15.32 a,c 1.28 a,c 0.55 a,c -0.39 a,c

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

In Table 3.5 we analyze the differences in dispositional optimism and attributional

style using standard regression analyses. We start discussing the results for disposi-

tional optimism, the most widely used measure of optimism in academic economics

research (see e.g. Krueger et al., 2000; Puri and Robinson, 2006; Hmieleski and Baron,

2009; and Graham et al., 2013). The results reinforce what we learned from Table

3.4; entrepreneurs are most inclined to be optimistic about the future and about the

fact that bad events will be scarce. Managers on the other hand possess this attitude

slightly less than entrepreneurs, but still more than employees. Some of the control

variables show significant coefficients, too. For age, we find that people tend to be-

come more optimistic until they reach the age of 52, after which it decreases again.

Furthermore, women are more optimistic than men12, people with higher levels of IQ

11This effectively implied dropping all answers corresponding to a very unrealistic expected returnof approximately plus or minus 25% in 3 months.

12So far, evidence for a relationship between gender and optimism has been mixed. Many studiesfind no difference (e.g., Fischer and Leitenberg, 1986; Scheier et al., 1994; and Puskar et al., 1999),some more optimistic females like we do (e.g. Collard and Reynolds, 2004; Yazdipour, 2010), whilethe opposite is found by e.g. Stipek et al. (1981). Given that women are also more likely to experiencea depression (see e.g. Piccinelli and Wilkinson, 2000, for a review), The Economist (2010) concludedthat this either suggested that “women are more likely to experience more extreme emotions”, or “thata few women are more miserable than men, while most are more cheerful”.

84

Tab

le3.5

Pro

bit

Regre

ssio

ns

on

Dis

posi

tion

al

Op

tim

ism

an

dA

ttri

bu

tion

al

Sty

le(1

)(2

)(3

)(4

)(5

)(6

)

Dep

.va

riab

le:

Dis

pos

itio

nal

Dis

pos

itio

nal

Dis

posi

tion

al

Att

rib

.A

ttri

b.

Att

rib

.

opti

mis

mop

tim

ism

op

tim

ism

Sty

leS

tyle

Sty

le

TotalScore

Optimism

Pessimism

Totalscore

Score

on

Score

on

Goodeven

tsBadeven

ts

Entr

epre

neu

r0.

442∗∗

∗0.

401∗

∗∗0.3

26∗∗

∗0.1

79∗∗

0.0

46

-0.2

12∗∗

∗

[5.8

8][5

.31]

[4.3

5]

[2.4

9]

[0.6

5]

[-2.9

3]

Man

ager

0.27

9∗∗

∗0.

239∗∗

∗0.1

88∗

∗0.0

01

-0.0

36

-0.0

48

[3.7

9][3

.22]

[2.5

3]

[0.0

2]

[-0.4

8]

[-0.6

4]

Age

0.06

0∗∗

∗0.

032∗

0.0

70∗∗

∗0.0

16

0.0

03

-0.0

19

[3.3

6][1

.72]

[3.9

4]

[0.9

5]

[0.1

5]

[-1.0

9]

Age

2/

100

-0.0

58∗∗

∗-0

.028

-0.0

67∗

∗∗-0

.012

0.0

04

0.0

19

[-2.

97]

[-1.

42]

[-3.4

6]

[-0.6

3]

[0.2

0]

[1.0

4]

Fem

ale

0.20

3∗∗∗

0.12

1∗∗

0.2

19∗∗

∗0.0

87

0.1

54∗

∗∗0.0

24

[3.5

9][2

.14]

[3.8

8]

[1.5

3]

[2.7

1]

[0.4

3]

Ed

uca

tion

0.14

3∗∗

∗0.

051

0.1

64∗

∗∗0.0

35

-0.1

24∗∗

∗-0

.160∗∗

∗

[4.3

4][1

.53]

[4.9

6]

[1.0

7]

[-3.7

6]

[-5.0

1]

Exp

erie

nce

-0.0

02-0

.001

-0.0

04

-0.0

09∗∗

∗-0

.006∗

0.0

07∗∗

[-0.

59]

[-0.

19]

[-1.2

6]

[-2.7

6]

[-1.7

8]

[2.0

2]

IQ0.

046∗

∗∗0.

017

0.0

59∗

∗∗-0

.007

-0.0

42∗

∗∗-0

.033∗∗

[3.2

4][1

.17]

[4.1

2]

[-0.4

7]

[-2.8

7]

[-2.3

3]

Ln

(in

com

e)0.

198∗

∗∗0.

177∗

∗∗0.1

70∗∗

∗0.1

34∗

∗∗0.1

08∗∗

∗-0

.083∗∗

[5.9

0][5

.26]

[4.9

7]

[4.0

9]

[3.2

6]

[-2.5

6]

con

stan

t-1

.089

∗∗-0

.159

-1.4

95∗

∗∗1.3

69∗

∗∗2.1

51∗∗

∗4.6

59∗∗

∗

[-2.

03]

[-0.

32]

[-3.1

2]

[2.6

2]

[4.4

7]

[10.0

9]

Ob

s.1,

691

1,69

11,6

91

1,6

91

1,6

91

1,6

91

Log

lik.

-4,5

29.2

-3,5

73.0

-3,6

98.6

-4,3

46.0

-3,6

96.4

-3,7

58.9

EN

T=

MA

N1)

<0.

01**

*<

0.01

***

0.0

3**

<0.0

1***

0.1

8<

0.0

1***

1)

Th

isro

wre

port

sth

ep

-valu

esof

Wald

test

sonβ

(Entr

epre

neu

r)=β

(Man

ager

).

85

and education tend to be more optimistic and experience is insignificantly associated

with dispositional optimism. When we decompose the dispositional optimism measure

into an “optimistic” part (i.e. add the scores on items 1,4, and 10) and a “pessimistic”

part (i.e. add the scores on items 3,7, and 9), it shows that the main results remain

standing, see columns 2-3. In columns 4-6, the attributional style measure is further

explored. We discuss the total measure in column 4, the sum of the factors the ‘per-

manence of good event’s (PmG) and ‘pervasiveness of good events’ (PvG) in column

5 (“Score on good events”), and finally the sum of the factors ‘permanence of bad

events’ (PmB) and ‘pervasiveness of bad events’ (PvB) in column 6 (“Score on bad

events”). Column 4 shows that controlling for a large set of background characteristics

in a probit regression does not change the picture painted in Table 3.4 (raw means).

Entrepreneurs have the most optimistic attributional style, all else equal. Managers

are similar upon including controls. When distinguishing between good and bad events

in columns 5 and 6 we see that entrepreneurs are different from the rest only because

of their differential attributional style related to bad events. Unreported regressions

show that the the significant difference that we find in column 6 largely pertains to the

difference in the score on PvB, or the pervasiveness of bad events. Hence, entrepreneurs

are distinct in what Seligman (2000) describes as: “Some people can put their troubles

neatly into a box and go about their lives even when one important aspect of it - their

job, for example, or their love life, is crumbling” (p. 90).

We now turn to discussing the regression results on overconfidence. The probit

results in the first column of Table 3.6 show that both entrepreneurs and managers

overestimate themselves more than employees do when estimating how many out of ten

Raven puzzles they have solved correctly. However, we do not find that the difference

between an ‘average’ entrepreneur and an ‘average’ manager reaches significance (the

p-value of the Wald test βENT =βMAN is 0.62).13

Column 2 shows the regression output on stock market predictions, the other mea-

sure of overconfidence. Again, both entrepreneurs and managers are more likely to

overestimate the future stock market than employees, but to a similar extent. Columns

3 and 4 repeat the analyses performed in (1) and (2) using as outcome measures dum-

mies that are one for correct forecasts and zero otherwise.14 While one might have

expected based on columns 1 and 2 that entrepreneur and managers are less realistic

13We obtain similar results when we use a different methodology cf. Dawson et al. (2014), seeAppendix A. Overall, our results thus appear inconsistent with the findings of Busenitz and Barney(1997), who use different samples of entrepreneurs and managers. We will explore this issue furtherin the next section.

14 A correct forecast in column (4) is a stock market forecast between 404 and 424 (i.e. less than

10 points off the realization of 414).

86

Table 3.6 Probit / OLS Regressions on Overestimation and Correctness

(1) (2) (3) (4)

Dep. variable: Over- Ove- Correct Correct

estimation estimation estimation estimation

(YES=1; NO=0) (YES=1; NO=0)

Own ability AEX 3M Own ability AEX 3M

closing price closing price

Regression type: ordered OLS probit probit

probit

Entrepreneur 0.398∗∗∗ 3.983∗∗ 0.211∗∗ 0.162∗

[5.53] [2.53] [2.08] [1.79]

Manager 0.367∗∗∗ 4.409∗∗∗ 0.220∗∗ 0.072

[4.77] [2.78] [2.14] [0.76]

IQ -0.337∗∗∗ -0.301

[-20.39] [-1.00]

Forecast -0.041∗∗ -0.005∗∗∗

[-2.11] [-3.51]

Age -0.010 0.074 0.106∗∗∗ 0.003

[-0.58] [0.18] [3.93] [0.14]

Age2 / 100 0.024 -0.173 -0.123∗∗∗ 0.008

[1.21] [-0.39] [-4.18] [0.35]

Female -0.099∗ -0.759 -0.084 -0.136∗

[-1.76] [-0.59] [-1.07] [-1.90]

Education 0.095∗∗∗ 0.549 -0.005 -0.004

[2.81] [0.78] [-0.03] [-0.09]

Experience 0.003 0.089 0.001 -0.010∗∗

[0.90] [1.23] [0.23] [-2.46]

Ln(income) -0.059∗ 0.889 0.032 0.106∗∗

[-1.70] [1.20] [0.70] [2.53]

constant 5.587∗∗∗ -10.41 3.114∗∗∗ 0.449

[10.06] [-1.00] [4.66] [0.65]

Obs. 1,691 1,663 1,691 1,663

Log lik. -2,903.0 -7,485.4 -1,123.0 -1,121.6

ENT=MAN 1) 0.62 0.74 0.89 0.26

1) This row reports the p-values of Wald tests on β(Entrepreneur) = β(Manager).

* Denotes significance at the 10% level, ** at the 5% level, and *** at the 1% level.

87

in their forecasts, the opposite holds true. We find entrepreneurs and managers to be

the two best performing groups, while employees follow at some distance. A closer

examination of this result is referred to Section 3.4.3. We will now further explore the

impact of using alternative definitions of entrepreneurs and managers.

3.4.2 Heterogeneity checks

We first rerun the main regressions of Tables 3.5 and 3.6 on a subsample of en-

trepreneurs and managers in young and small firms, see Table 3.7, thereby accounting

for the very different distributions within the two groups of firm size and age, see Table

3.2. Panel A (young firms), Panel B (small firms) and Panel C (start-ups) show that

most of the main outcomes of Section 3.4.1 extend to entrepreneurs and managers in

young and/or small firms, although for managers some of the p-values turn higher than

0.10, likely due to smaller samples. The only notable additional finding is that arguably

‘entrepreneurial’ managers generally do not differ from entrepreneurs in their attribu-

tional style, but they do in their lower overestimation of their own ability. We find this

result in both Panels A and B. Apparently entrepreneurs in younger and/or smaller

firms seem to be more prone to overestimation than a control group of managers who

work for comparable firms. In Panel C, we also compare founders of start-up firms

(<3 years old) with managers in firms with more than 10,000 employees, cf. Busenitz

and Barney (1997). Contrary to our result in Table 3.6, the difference between en-

trepreneurs and managers in overestimation of own ability now becomes significant at

the 5% level. Taken together with Table 3.7, the data seem to be consistent with the

view that different types are attracted to entrepreneurship (cf. Landier and Thesmar,

2009) rather than information availability or feedback being the main driver of the

difference (cf. Busenitz and Barney, 1997). Interestingly, we do not obtain different

conclusions when we look at the probabilities of being correct (analogous to the second

part of Table 3.6).

We further examine the impact of using (stricter) definitions of entrepreneurs and

managers in Table 3.8. Using the variation illustrated in Table 3.2, Panel A first ana-

lyzes with the following definitionsof entrepreneurs: (i) entrepreneurs with an incorpo-

rated firm, thereby mainly excluding the own-account self-employed, (ii) entrepreneurs

with an above median number of fulltime equivalent employees in their company, (iii)

entrepreneurs with above median incomes, (iv) entrepreneurs that have founded their

business, instead of obtaining it through takeover or buy-in, and (v) entrepreneurs in

the survival phase (firm age ≤ 5 years). For managers and employees we employ the

original samples in Panel A. Its last line shows the results of Tables 3.5 and 3.6 again.

Note that each coefficient is obtained in a separate regression. The panel shows a clear

88

Table 3.7 Optimism and Overconfidence in Young and Small Firms

(1) (2) (3) (4)

Dependent variable: Dispositional Attrib. Over- Over-

optimism Style estimation estimation

Own ability AEX 3M

closing price

Subset A: Entrepreneurs and Managers

in Firms < 15 yrs, all Employees

Entrepreneur (n = 571) 0.487∗∗∗ 0.320∗∗∗ 0.437∗∗∗ 4.901∗∗∗

[5.08] [3.78] [4.97] [2.70]

Manager (n = 90) 0.264∗ 0.172 0.240∗ 6.048∗∗

[1.87] [1.25] [1.96] [2.01]

ENT = MAN 1) 0.09 * 0.26 0.09 * 0.70

Subset B: Entrepreneurs and Managers

in Firms < 25 FTEs, all Employees

Entrepreneur (n = 779) 0.495∗∗∗ 0.276∗∗∗ 0.368∗∗∗ 3.541∗∗

[5.96] [3.59] [4.74] [2.10]

Manager (n = 40) 0.153 0.209 0.066 -4.030

[1.22] [1.19] [0.43] [-1.31]

ENT = MAN 1) < 0.01 *** 0.69 0.03 ** 0.01 **

Panel C: Entrepreneurs in firms < 3 yrs,

Managers in firms ≥ 10,000 FTE,

all Employees

Entrepreneur (n = 126) 0.480∗∗∗ 0.602∗∗∗ 0.575∗∗∗ 4.044∗

[3.80] [4.53] [4.25] [1.67]

Manager (n = 56) 0.156 0.231 0.185 2.064

[0.98] [1.44] [1.23] [0.62]

ENT = MAN 1) 0.06 * 0.04 ** 0.02 ** 0.58

* Denotes significance at the 10% level, ** at the 5% level, and *** at the 1% level.

89

Table 3.8 Differences in Optimism and Overconfidence using Stricter Definitions

(1) (2) (3) (4)

Dependent variable: Dispositional Attrib. Over- Over-

optimism Style estimation estimation

Own ability AEX 3M

Panel A: Subsets of Entrepreneurs

i) Incorporated (n = 461) 0.472 a,b 0.196 a,b 0.412 a 3.647

[4.90] [2.27] [4.66] [1.85]

ii) Above median no. of employees (n = 413) 0.452 a,b 0.194 a,b 0.390 a 2.297

[4.87] [2.32] [4.57] [1.24]

iii) Above median ent. income (n = 367) 0.386 a 0.153 a,b 0.449 a 4.033

[3.80] [2.02] [2.82] [1.81]

iv) Founder (n = 700) 0.453 a,b 0.202 a,b 0.421 a 5.025 a

[5.69] [2.97] [5.62] [2.94]

v) In survival phase (firm ≤ 5 yrs, n = 277) 0.609 a,b 0.227 a,b 0.542 a 2.320

[5.44] [3.16] [5.15] [0.67]

vi) Not in survival (firm > 5 yrs, n = 616) 0.355 a 0.159 a,b 0.358 a 3.223

[4.72] [2.06] [4.68] [1.91]

β(Entrepreneur) in Tables 3.5 & 3.6: 0.442 a,b 0.179 a,b 0.398 a 3.983 a

Panel B: Subsets of Managers

vii) CEO or general manager (n = 437) 0.289 b,c -0.003 b 0.384 c 3.925 c

[5.74] [-0.04] [4.81] [2.38]

viii) CEO (n = 71) 0.442 c 0.339 c 0.522 c 5.223

[4.04] [2.54] [3.37] [1.80]

ix) Above median no. dir. reports (n = 270) 0.376 c 0.199 c 0.359 c 4.394 c

[4.15] [2.16] [3.72] [2.30]

x) Above median man. income (n = 198) 0.448 c 0.280 c 0.408 c 2.110

[4.54] [2.66] [3.74] [1.01]

xi) Manager in firm > 15 yrs old (n = 427) 0.299 b,c 0.053 b 0.392 c 3.901 c

[3.96] [0.07] [4.85] [2.36]

β(Manager) in Tables 3.5 & 3.6: 0.279 b,c 0.001 b 0.367 c 4.409 c

Panel C: Combinations of A&B

i) vs. viii); p-values Wald tests 0.56 0.87 0.18 0.51

ii) vs. ix); p-values Wald tests 0.10 0.21 0.52 0.50

iii) vs. x); p-values Wald tests 0.66 0.75 0.41 0.46

a) Significant difference between entrepreneurs and employees (5% level), b) Significant difference between entrepreneurs

and managers (5% level), and c) Significant difference between managers and employees (5% level).

90

pattern consistent with the findings in Tables 3.5 and 3.6. Whatever definition of the

entrepreneur is used, entrepreneurs assess themselves as more optimistic than both

managers and entrepreneurs. Also the overestimation measures reveal largely the same

outcomes, except that most of the findings on overestimation of the AEX closing price

only turn significant at the 10% level, largely due to higher standard errors.

Panel B of Table 3.8 shows the results when the definition of a manager is varied

while using the complete samples of entrepreneurs and employees. We restrict the

sample of managers to (vii) CEOs or general managers (so without project managers),

(viii) CEOs exclusively, (ix) managers with more than the median number of direct

reports, (x) managers with above median managerial income, and (xi) managers in

firms that are older than 15 years old. The stricter definitions create samples of more

successful managers and managers that can reasonably be expected to be more different

from entrepreneurs than average, such as the ones employed in older firms. Again, the

last line of the panel shows the benchmark results for managers taken from Tables 3.5

and 3.6. In contrast to Panel A, the main results change when considering the samples

of (v) CEOs only, (vi) managers with above median direct reports, and (vii) managers

with above average income, whereas the results do not change for subsamples (vii)

and (xi). The difference between entrepreneurs and managers has vanished. Hence,

successful managers and entrepreneurs both stand out from employees in their higher

level of optimism.

Finally, in Panel C we test alternative definitions against each other. We compare (i)

entrepreneurs of incorporated firms with CEOs, (ii) entrepreneurs and managers with

larger spans of control, and (iii) entrepreneurs and managers with higher than median

incomes. We find no differences between successful entrepreneurs and managers.

In Appendices A-B in Appendices Chapter 3, we perform two additional tests.

Appendix A deals with the potential concern that the three identified categories are not

entirely mutually exclusive. For instance, entrepreneurs might also be employees or vice

versa, and managers and employees might have taken their entrepreneurial experiences

from the past into their current jobs. Appendix A1 shows some descriptives of the

information on past work experiences across the different occupational categories that

we gathered through the survey. 72% of the entrepreneurs in the sample have been a

manager before, while 9% of the entrepreneurs is currently also a salaried employee in

another firm. Moreover, 8% of the managers and of the employees owns a business on

the side, whereas 11% of the managers and 9% of the employees have had one in the

past. To cope with these ‘grey’ areas, we re-run the main regressions of Tables 3.5 and

3.6 including dummy variables for all the potential cross-occupational areas. Appendix

A2 shows the results: almost none of these controls reach significance, and our main

conclusions keep standing.

91

As a final heterogeneity check we explore variation in optimism within the sam-

ples of entrepreneurs and managers (Appendix B). More specifically, we explore the

correlation between optimism and specific (crude) measures of performance. The re-

sults indicate that entrepreneurs with higher scores on dispositional optimism are more

likely to be incorporated or to have an above median number of employees or income

(see Panel A, columns (1a), (2a), and (3a)). In Panel B we find that the effect of

attributional style is positive as well, although the coefficient only reaches significance

in the case of incorporated entrepreneurs. Moving down to Panels C and D, we find

largely insignificant differences between more and less successful entrepreneurs in their

likelihood of overestimating their own ability or being correct about it.

For managers, we find similar patterns as for entrepreneurs, see columns (1b), (2b),

and (3b). Those with higher scores on dispositional optimism and attributional style

are significantly more likely to be a successful manager (e.g. a CEO or a manager

with an above median number of direct reports or income). Again, success does not

differentiate between managers who are more and less likely to provide correct or too

high estimates of their own ability or the stock market index. Overall, we conclude

that dispositional optimism, attributional style and success appear especially positively

correlated for managers, and to a slightly lesser extent for entrepreneurs.

As a final check, we have analyzed an alternative measure of optimism using the

data of Koudstaal et al. (2015) and the theoretical framework of Andersen et al. (2014).

It shows in unreported regressions that entrepreneurs and non-entrepreneurs differ in

their “probability optimism”, which is the tendency to perceive an objective probabil-

ity of winning a positive amount to be larger than an equal probability of losing the

same amount. Examining the data, and in line with much of the previous, we find that

both entrepreneurs and managers are more “probability optimistic” than employees,

but do not differ from each other. Hence, compared to employees, they both view the

probability of winning a positive amount more optimistically than the probability of

losing the same amount.

3.5 Conclusion

Many of us have a brighter view on life than is warranted by reality. However, en-

trepreneurs are believed to be even more optimistic and overconfident than others.

Why would one opt for entrepreneurship, with uncertain outcomes that are varying

over time, and low on average? This choice may be explained by entrepreneurs hold-

ing (over-)optimistic beliefs. They would overestimate their probability of survival,

neglect the quality of the competition, and overestimate the market for their product

or service. In more direct tests between entrepreneurs and the population at large,

92

the empirical evidence indeed suggests that entrepreneurs are not only more optimistic

than others but also more prone to overconfidence. Yet, at the same time a different

strand of literature shows that optimism and overconfidence are also behavioral traits

prevalent among corporate managers such as CEOs and CFOs. Our analysis of the

differences between entrepreneurs and managers in optimism and overconfidence can

point out whether or not these are unique traits of entrepreneurs, or whether these

characteristics pertain to strategic decision-makers in general. In other words, might a

certain degree of optimism not only be required for entrepreneurship but also to climb

the corporate ladder?

We have explored this question by means of a lab-in-the-field experiment among

substantial groups of entrepreneurs, managers and employees (n = 2,058). We used

two well-known measures of optimism, i.e. dispositional optimism and attributional

style, and two well-known measures of overconfidence, i.e. overestimation of one’s own

ability and overestimation of a future stock market closing price. All measures test for

slightly different sides of optimism and overconfidence. In that sense, we aim to provide

an encompassing assessment of differences in optimism and overconfidence between the

three groups of interest. Besides that, we believe we contribute to the literature by

being the first to test the attributional style of entrepreneurs and managers, and more in

general by testing optimism among such large samples of entrepreneurs and managers.

The benefit of large samples proves to be particularly high when we explore the impact

of alternative (and stricter) definitions of an entrepreneur and/or a manager. In the

entrepreneurship literature, for instance, there is a debate going on about the definition

of an entrepreneur.

The results indicate that entrepreneurs are more optimistic than managers and em-

ployees in their dispositional optimism and their attributional style. Concerning the

latter, we do not find that entrepreneurs are significantly more optimistic about the

permanence and pervasiveness of good events, but we do find significant differences

when we examine attitudes towards dealing with bad events. Here it shows that en-

trepreneurs are more optimistic than both managers and employees, thus suggesting

that they are more resilient in the face of setbacks than the other two groups. Fur-

thermore, the two measures of overconfidence indicate that both entrepreneurs and

managers are more prone than employees to overestimate their own ability or a future

stock market closing price.

Our heterogeneity and robustness checks demonstrate that the main findings are

hardly sensitive to the definitions used, with two exceptions. First, entrepreneurs in

young and/or small firms do seem to overestimate themselves more than managers

who work for comparable firms, in contrast to our general finding that entrepreneurs

and managers are not different in this respect. Second, when we restrict the sample

93

of managers to the arguably more ‘successful’ ones (e.g. CEOs), we find that the gaps

between entrepreneurs and managers on the two optimism measures vanish. Our anal-

yses indicate that optimism and success are even more positively correlated within the

sample of managers than within the sample of entrepreneurs. These same analyses also

show that the more successful entrepreneurs and managers are not more overconfident

than their less successful peers.

So do you have to be an optimist to be an entrepreneur, as suggested in one of

the quotes in Chapter 1? The answer appears to be yes. Moreover, our evidence

also points out that entrepreneurs are not the only ones who are so optimistic and

overconfident. Especially the more successful managers are so, too. Our findings thus

show that; “The people who have the greatest influence on the lives of others are likely

to be optimistic and overconfident...” (Kahneman, 2011, p. 256). More specifically,

we find that entrepreneurs and managers are more overconfident than employees in

general, whereas entrepreneurs and the most successful managers are more optimistic

than others.

The fact that entrepreneurship and general management require an optimistic atti-

tude is interesting from a managerial perspective as well. However, one needs to note

that high level general managers will often work in management teams, probably con-

sisting of CFOs and other functions. The same may be true, but to a lesser extent, for

entrepreneurs. Future research might address the role of optimism and overconfidence

of individuals working in teams and how a more pessimistic and less overconfident

team member (for instance the CFO) might offset or strengthen the role of the CEO’s

optimism on team performance.

94

Appendices Chapter 3

Appendix A. Cross-occupational experience and the impact on optimism.

Appendix B. Correlations of optimism with success.

Appendix C. Survey screenshots.

95

Appendix A1. Cross-Occupational Experience of Entrepreneurs, Managers andEmployees

Entrepreneur Manager Employee

(n = 875) (n = 516) (n = 667)

% with Managerial Experience in the past 71.7 - -

Level of past man. experience (scale: 1-5) 1.72 - -

% that is also Employee now 9.0 - -

% that is also Entrepreneur now - 7.9 8.0

Level of current ent. experience (scale: 1-8) - 1.46 1.40

% with entrepreneurial experience in the past - 11.0 9.3

Level of past ent. experience (scale: 1-8) - 2.41 1.53

The ‘level of managerial experience’ is measured based on a question about the number of directly re-

porting subordinates when and if entrepreneurs were managers beforehand. The answering categories

that we coded 1 to 5, respectively, are 2-5 // 6-10 // 11-25 // 26-50 // More than 50. The ‘level of

entrepreneurial experience’ measure is based on the categorized answers to managers and employees

how many fulltime equivalent people they employed when they were entrepreneurs. This question was

posed only to those who had been entrepreneurs in the past. Answer categories were: 0 // 1-4 //

5-10 // 11-25 // 26-100 // 101-250 // 251-1,000 // More than 1,000 employees. The first answer (0)

corresponds with a value of 1, the second answer (1-4) with a value of 2, and so on.

96

Appendix A2. Cross-Occupational Experience and Optimism

(1) (2) (3) (4) (5) (6)

Dep. variable: Dispositional Attrib. Over- Over- Correct Correct

optimism Style estimation estimation

Own ability AEX 3M Own ability AEX 3M

closing price closing price

Entrepreneur 0.489∗∗∗ 0.203∗∗ 0.396∗∗∗ 3.332∗ 0.327∗∗∗ 0.234∗∗

[5.60] [2.41] [4.79] [1.89] [2.72] [2.20]

- Also employee 0.018 -0.150 -0.200 2.367 -0.084 -0.182

(YES=1; NO=0) [0.13] [-1.39] [-1.49] [0.73] [-0.48] [-1.18]

- Past mgmt exp. 0.037 -0.003 -0.044∗ -0.497 0.071∗ 0.034

(=0 if none) [1.23] [-0.01] [-1.72] [-0.81] [1.88] [1.02]

Manager 0.254∗∗∗ 0.001 0.397∗∗∗ 5.413∗∗∗ 0.191 0.033

[3.22] [0.01] [4.80] [3.07] [1.63] [0.33]

- Also entrepreneur 0.001 -0.075 0.168 -4.797∗ 0.079 0.269

(YES=1; NO=0) [0.01] [-0.58] [1.10] [-1.66] [0.37] [1.36]

- Past ent. exp. 0.046 0.069 0.028 -0.179 0.037 0.031

(=0 if none) [1.04] [1.45] [0.53] [-0.16] [0.51] [0.42]

Employee

- Also entrepreneur 0.038 0.025 -0.010 2.545 -0.456 0.296

(YES=1; NO=0) [0.25] [0.17] [-0.06] [1.04] [-1.63] [1.52]

- Past ent. exp. -0.014 0.013 0.078∗∗∗ 0.282 0.002 -0.056

(=0 if none) [-0.46] [0.41] [2.59] [0.30] [0.04] [-1.24]

Control variables YES YES YES YES YES YES

Obs. 1,691 1,691 1,691 1,663 1,691 1,663

Log lik. -4,527.8 -4,344.3 -2,933.5 -74,83.2 -831.5 -1,113.0

ENT = MAN 1) <0.01 ** 0.01 ** 0.99 0.23 0.22 0.05 *

1) This row reports the p-values of Wald tests on β(Entrepreneur) = β(Manager).

This table reports the measures of optimism of entrepreneurs, managers and employees, but now

including controls for cross-occupational experiences and interactions. All variables have been defined

in Appendix Table B.1 or before. Significance at the 10% level is denoted by *, 5% by **, and 1% by

***, with t-statistics reported in parentheses. Standard errors are robust.

97

Ap

pen

dix

B.

Corr

ela

tion

sof

Op

tim

ism

wit

hS

uccess

wit

hin

Entr

ep

ren

eu

rsh

ipan

dM

an

agem

ent

Su

cces

sm

easu

res

Dep

.va

riab

le:

Inco

rpor

ated

CE

OA

bov

em

ed.

Ab

ove

med

.A

bov

em

ed.

Ab

ove

med

.

emp

loye

esd

irec

tre

port

sin

com

ein

com

e

(1a)

(1b

)(2

a)

(2b

)(3

a)

(3b

)

Pan

elA

Dis

pos

itio

nal

0.04

4∗∗

∗0.0

48∗∗

∗0.0

23∗

0.0

28∗

0.0

46∗∗

∗0.0

52∗∗

∗

opti

mis

m[3

.57]

[2.8

2]

[1.9

2]

[1.7

5]

[3.7

3]

[2.9

6]

βEN

T=βM

AN

1)

0.48

0.3

40.2

4

Pan

elB

Att

rib

.st

yle

0.02

9∗∗

0.0

53∗

∗0.0

10

0.0

48∗

∗∗0.0

07

0.0

52∗

∗

[2.0

2][2

.54]

[0.7

2]

[2.6

0]

[0.4

9]

[2.5

7]

βEN

T=βM

AN

1)

0.26

0.0

5**

0.0

2**

Pan

elC

Ove

rest

imat

ion

0.01

30.0

66

0.0

15

0.0

12

-0.0

04

0.0

52

(ow

nab

ilit

y)

[0.3

9][1

.27]

[0.4

7]

[0.2

8]

[-0.1

1]

[1.1

4]

βEN

T=βM

AN

1)

0.49

0.7

30.4

7

Pan

elD

Ove

rest

imat

ion

0.00

20.0

04

0.0

02

0.0

01

0.0

04∗

∗0.0

01

(AE

Xfo

reca

st)

[1.1

5][1

.61]

[1.4

5]

[0.5

8]

[2.4

1]

[0.2

5]

βEN

T=βM

AN

1)

0.56

0.4

70.1

1

Con

trol

sfo

rag

e,ge

nd

er,

YE

SY

ES

YE

SY

ES

YE

SY

ES

edu

cati

onan

dIQ

1)

Th

isro

wre

port

sth

ep

-valu

esof

Wald

test

sonβ

(Entr

epre

neu

r)=β

(Man

ager

).

98

Appendix C. Survey screenshots

Figure C1: Dispositional Optimism

Figure C2: Attributional Style

99

100

Chapter 4

Intuitive versus Contemplative

Decision-Making

This chapter is based on Koudstaal, M., R. Sloof, and C.M. van Praag, “Are En-

trepreneurs More Intuitive Than Managers and Employees? Evidence From a Lab-in-

the-Field Experiment”, working paper.

4.1 Introduction

Executive intuition and heuristics are considered important elements in entrepreneurial

and managerial decision-making.1 Interestingly, a large number of studies have been

performed within this area, but only a few of them have compared entrepreneurs and

managers directly to each other. Moreover, a variety of different measures of intuition

have been used in the literature, either being ‘objective’ or ‘subjective’ in nature.

The aim of our study is therefore twofold: (i) to try to establish how entrepreneurs,

managers and employees differ from each other on four separate measures of intuition

and (ii) to verify whether the results are consistent across these four different measures.

To motivate why entrepreneurs, managers and employees might vary in their deci-

sion-making styles, it is important to identify potential factors that could call for dif-

ferent decision-making styles. Following Busenitz and Barney (1997), entrepreneurs

are expected to lean more on heuristics than managers due to the uncertain nature of

their business environment and the difference in the availability of useful information

(see also Miller and Friesen, 1984). Testing their prediction empirically, Busenitz and

Barney (1997) find that entrepreneurs are indeed more likely to use the representa-

tiveness heuristic than managers in large organizations. Similar findings related to

1For entrepreneurs, see e.g. Allinson and Hayes (1996), Busenitz and Barney (1997), and Allinsonet al. (2000). For managers, see. e.g. Simon (1987), Behling and Eckel (1991), Parikh (1994), Hayashi(2001), and Sadler-Smith and Shefy (2004).

101

intuition have been reported by Allinson et al. (2000), although they also find that

entrepreneurs are not more intuitive than senior managers and executives.

Whether managers are also expected to be less intuitive than employees is a question

with no clear-cut answer. On the one hand, one could expect managers to be more

contemplative than employees given that most organizations still tend to rely on some

form of rational analysis in its decision-making (Sadler-Smith and Shefy, 2004), thus

requiring the more contemplative types to sort into management positions. A similar

hypothesis could be derived from the suggestive evidence that more rational and less

biased individuals appear to perform better within organizations (e.g., Biais and Weber,

2009; Choi et al., 2014). On the other hand, there is also evidence suggesting that an

enhanced “gut feeling” is a necessary trait to climb especially the last part of the

corporate ladder (see e.g. Parikh (1994) and Hayashi (2001), but note the critique of

Bonabeau (2003) to the latter).2 Taken together, we thus expect an average manager

to be more contemplative than employees, while top managers might in turn be more

intuitive than the average manager.

Following Kahneman (2011), our first measure of intuition builds on a recent strand

of literature by Rubinstein (2007, 2013, 2016), who assumes a typology where some

individuals think ‘slow’ and behave contemplative, while others think ‘fast’ and behave

intuitive. Rubinstein (2016) finds empirical support for his typology among a large

global sample of students, showing a positive relationship between response times and

the level of contemplativeness. In the psychological literature, these notions of fast and

slow thinking have been part of many dual-process theories (for a review, see e.g. Evans,

2008). One such example is the Cognitive-Experiential Self-Theory (CEST) of Epstein

(2003), which is based on the idea that we use two independently operating systems:

one that is fast, automatic and emotional (“the intuitive-experiential system”) and one

that is deliberate, slow and logical (“the analytical-rational system”).3 According to

Epstein (2003), these two sytems work best in tandem and interact with each other

either sequentially or simultaneously. The two forms of information processing have

been extensively studied empirically, thereby mainly using the Faith in Intuition (FI)

2Note that the applied methodologies of all these papers vary. The study by Choi et al. (2014) forinstance uses objective benchmarks, while the studies by Parikh (1994) and Hayashi (2001) lean onsubjective assessments of the use of intuition by managers. Moreover, some studies regard ‘intuition’as quick decision-making in new situations where more rigourous thinking would have been possibleas well (as in our study), while other studies view intuition more in the sense of experience in dealingwith similar situations (in line with the chess player example of Simon, 1987) or in dealing with newsituations where rigourous analyses do not seem to be possible (cf. Hayashi, 2001).

3While other theories in this field come in different forms, they all generally seem to agree that oneprocessing mechanism is fast, non-conscious and intuitive (also referred to as “System 1”) while theother is slow, controlled and conscious (“System 2”). See also Thaler and Shefrin (1981), Stanovichand West (2000), and Kahneman (2011). Note that generally most thoughts and actions are consideredintuitive in this sense (Gilbert, 1989, Gilbert, 2002, and Wilson, 2002).

102

and Need for Cognition (NFC) scales of e.g. Epstein et al. (1996).4 In addition to

the more ‘objective’ measure of Rubinstein (2007, 2013, 2016), we will therefore also

measure intuition using these two well-grounded FI and NFC scales.

We believe that this study has two main contributions. First of all, we expand on

Rubinstein (2007, 2013, 2016) by investigating his typology among large samples of

established entrepreneurs, managers and employees (n = 1,931), hence those people

who actively participate in the labor market. An additional benefit of having such

large samples is that we are also able to verify the robustness of our findings using

other definitions of an ‘entrepreneur’ or a ‘manager’ (for a discussion on this, see e.g.

Parker, 2009; Levine and Rubinstein, 2013; and Tag et al., 2013). Second, we test the

consistency of the results on the two sets of measures. In other words, do the ‘objective’

measures (i.e. choice behavior and response times) lead to qualitatively similar results

as the ‘subjective’ measures (i.e. Faith in Intuition and Need for Cognition)?

The main results are as follows. First, we find that entrepreneurs make significantly

more intuitive choices than managers, but not than employees. Surprisingly, however,

the three occupational groups do not differ in their response times once we control for

general background characteristics. Second, on the subjective measures we find that

entrepreneurs have the highest score on Faith in Intuition, while managers have the

highest score on Need for Cognition. Overall, the results on the subjective measures

thus seem consistent with the objective measures, but also somewhat more pronounced.

Furthermore, to better understand the mixed results on the objective measures,

we run several additional regressions to test possible explanations that might have

caused these seemingly inconsistent findings. These additional regressions show that

the inconsistency might be explained by a non-homogeneous effect of response times on

contemplativeness. We find that entrepreneurs and managers - even though they might

not spend more time on the experimental games - are relatively more contemplative per

unit of time than employees are. Moreover, controlling for this heterogeneity, we find

that entrepreneurs are not only more intuitive than managers, but also than employees

4Overall, NFC has been found to be positively associated with e.g. increased academic achievement,self-esteem, openness to experience, and conscientiousness and negatively related to e.g. depression,neuroticism, overgeneralization, and naive optimism (Epstein et al., 1996; Pacini et al., 1998; Paciniand Epstein, 1999; Sladek et al., 2010; and Norris and Epstein, 2011). Faith in Intuition (FI) onthe other hand has been positively linked to e.g. creativity, spontaneity, agreeableness, and favorableinterpersonal relationships, but also to superstitious beliefs, naive optimism, and stereotypical thinking(Epstein et al., 1996; Pacini and Epstein, 1999; Kemmelmeier, 2010; Norris and Epstein, 2011; andSagiv et al., 2014). Furthermore, a positive relationship has been discovered between FI and (i) the useof the availability heuristic (Danziger et al., 2006), (ii) the likelihood of providing suboptimal responsesto the “Linda problem” (Toyosawa and Karasawa, 2004), and (iii) reliance on the representativenessheuristic (Alos-Ferrer and Hugelschafer, 2012). Overall, despite being quite useful in providing fastand ‘close to optimal’ answers in many general situations, a higher Faith in Intuition might thus alsolead to an increased susceptability to making systematic errors (Tversky and Kahneman, 1974; Colinand Loewenstein, 2004).

103

whenever their response times are low. Finally, we also test the robustness of our

findings using other (stricter) definitions of an entrepreneur and/or a manager. The

results generally indicate that the differences remain significant.

Due to its descriptive nature, our study cannot disentangle between personality

traits (i.e. ‘nature’) and environmental effects (i.e. ‘nurture’). Hence, it could therefore

be that the uncertain environments that entrepreneurs are dealing with rationalize

more intuitive decision-making in the absense of any historic information (cf. e.g.

Busenitz and Barney, 1997). Conversely, for managers the reverse might be true,

especially for those who work in firms with long histories and with a lot of historical

data available. Alternatively, if there is a predominance of one of the two decision-

making styles, as suggested by Kahneman (2011, p. 48), our findings could be the

result of different types of individuals sorting into different types of occupations with

different skill requirements.

In what follows, Section 4.2 will first discuss the measurement and sampling of our

study. Section 4.3 provides the descriptive statistics of our sample, while Section 4.4

reports the empirical findings. Section 4.5 concludes.

4.2 Measurement and sampling

Both in economics and psychology, intuitive and contemplative behavior has often been

studied. In this study, we will rely on both a recent strand of literature by Rubinstein

(2007, 2013, 2016), which is based on response times in strategic games, as well as on

the standard practice in psychology, which is based on the Cognitive-Experiential Self-

Theory (CEST) of Epstein (e.g., Epstein et al., 1996; Pacini et al., 1998; and Pacini and

Epstein, 1999). One of the main differences between the two methodologies is that the

latter is based on self-assessment tests, and can thus be regarded as more ‘subjective’,

while the former is based on objectively verifiable intuitive and contemplative choices,

and is thus more ‘objective’. Below we will describe the two types of measures in detail.

4.2.1 Objective measures of decision-making

For our objective measures we rely on the recent work of Rubinstein (2007, 2013,

2016), who uses a large student database to construct a what he calls Contemplative

Index (CI). As the basis for this index, Rubinstein uses ten games adapted from game

theory which all have intuitive choices (receiving a score of 0) and contemplative ones

104

(receiving a score of 1).5 Scoring is thus done such that the more contemplative choices

a subject provides, the higher the index, and conversely, the more intuitive choices that

are given, the lower the index. A particularly attractive feature of this index is that

it is ‘objective’, i.e. all individual games allow for an objective identification of the

intuitive and contemplative choices.6 In addition, Rubinstein (2016) also finds that

the Contemplative Index (CI) has predictive power when examining subject behavior

in other games than the ten used for the construction of the CI.

For our purposes, we ideally would have liked to copy the entire setup of Rubinstein

(2016) and have all participants play all ten games. However, due to the limited time

available among our working participants, we did not have sufficient time to do so.

Instead, we therefore aimed to select three games that could serve as a decent proxy

for the Contemplative Index.7 In selecting these three games we used the following

five criteria: i) the game setup remains close to the entrepreneur / manager world, ii)

the game can be well-understood in an online survey, iii) the game can be played in

a relatively short period of time, iv) the game clearly differentiates between intuitive

and contemplative thinkers (i.e. agreement rates are high), and v) the game contains

no obvious relevant confounding factors.

Applying these criteria, we ultimately opted to provide participants with the fol-

lowing three games: #1 The One-Shot Chain Store Game (Selten 1978), #2 Hotelling’s

Main Street Game (Hotelling 1929) and #3 The Two-contests Game (Huberman and

Rubinstein, 2000). A more detailed description of each game can be found below. For

convenience we have marked the intuitive choices with an (I) and the contemplative

choices with a (C). In line with Rubinstein (2016), we recorded the response times for

each participant on each of these three games.

#1: The One-shot Chain Store Game (based on Selten, 1978)In your neighborhood, there is one grocery store and one tailor. At the moment, theprofits of the grocery store owner are around $120,000 per year while the tailor’s profitsare only $50,000 per year. The tailor asks your advice about whether to change hisshop into a grocery store. He figures that if the grocer does not respond aggressivelyto the new competition, each of them will earn about $70,000 per year. On the otherhand, if the grocer does respond aggressively and starts a price war, then the earningsof each store will be reduced to about $25,000 per year. What is your advice to thetailor? [ENTER (C) or NOT ENTER (I)].

5All ten games that are used are standard in economics and include e.g. Hotelling’s main streetgame (Hotelling, 1929).

6However, to further strengthen his case, Rubinstein (2016) also tested for agreement rates aboutthe intuitive and contemplative choices among 17 graduate students.

7Initially we had four games in our original survey, but decided to drop one of the games (“Relyingon an other player’s rationality”) based on the updated version of Rubinstein (2016). In this newversion, he reports that the agreement rate about the intuitive and contemplative choice among the17 graduate students is low for this particular game.

105

#2: Hotelling’s Main Street Game (based on Hotelling, 1929)Imagine you are the manager of a chain of cafes competing with two other similarchains. Each of you is about to rent a shop in one of the 7 new identical huge apart-ment buildings standing along a beach strip. Once each of you knows exactly where theother two competitors locate it will be too late to move to another location. You expectthat the customers (the residents in the 7 buildings) will not distinguish between thethree cafes and will pick the one which is closest to their home. In which building (anumber between 1 and 7) will you locate your cafe? [1 (I), 2 (C), 3 (C), 4 (I), 5 (C), 6(C) or 7 (I)].

#3: The Two-contests Game (based on Huberman and Rubinstein, 2000)8

Imagine you are participating in a game with over 200 participants worldwide. Eachparticipant chooses to compete in one of two contests. In contest A, each contestantguesses the outcomes of 20 coin flips (heads or tails). In contest B, each contestantguesses the outcomes of 20 rolls of a die (i.e., each of the twenty guesses is a number1, 2, 3, 4, 5 or 6). Each contest will be conducted independently. In each contest, youwill be competing against people who, like you, chose that contest. After the guessesof all the participants are collected, a computer will simulate a series of 20 coin flips forcontest A and a series of 20 rolls of a die for contest B. The winner of each contest willbe the person with the most correct guesses. (In the case of a tie, the winner will bechosen by a lottery among those with the most correct guesses.) I choose to participatein: [CONTEST A (I) or CONTEST B (C)].

4.2.2 Subjective measures of decision-making

Our subjective measures of intuitive and contemplative decision-making were taken

from Epstein et al. (1998), who examine the two independently operating information

processing systems using a short form of their Rational-Experiental Inventory. This in-

ventory is based on the theoretical Cognitive-Experiential Self-Theory (Epstein, 1973),

which distuiguishes between two cognitive styles: the intuitive-experiential style, as

measured by Faith in Intuition (FI), and the analytical-rational style, as measured by

Need for Cognition (NFC). Overall, the short form of the Rational-Experiental Inven-

tory consists of 10 items:9

Faith in Intuition (FI)

- I trust my initial feelings about people.

- I believe in trusting my hunches.

- My initial impressions of people are almost always right.

8See Figure B1 in Appendix B of Appendices Chapter 4 for a screenshot of how the games werepresented to the participants in our survey.

9See Figure B2 in Appendix B of Appendices Chapter 4 for a graphical representation.

106

- When it comes to trusting people, I can usually rely on my “gut feelings”.

- I can usually feel when a person is right or wrong even if I can’t explain how I know.

Need for Cognition (NFC)

- I don’t like to have to do a lot of thinking. (R)10

- I try to avoid situations that require thinking in depth about something. (R)

- I prefer to do something that challenges my thinking abilities rather than some-

thing that requires little thought.

- I prefer complex to simple problems.

- Thinking hard and for a long time about something gives me little satisfaction.

All 10 items were rated on a 5-point Likert scale ranging from “Strongly disagree”

to “Strongly agree”. With the exception of the reversely scored items, “Strongly dis-

agree” was assigned 1 point, “Disagree” 2 points, “Neutral” 3 points, “Agree” 4 points,

and “Strongly agree” 5 points. To obtain overall scores for FI and NFC, all relevant

individual scores were added up. Hence, the maximum (minimum) scores for both FI

and NFC were 25 (5).

4.2.3 Sampling

Our sampling procedure was similar to Chapter 1 and aimed to collect the responses

of as many individuals from the working population as possible. In doing so, we

collaborated with the same three business partners to gain substantial reach.

The invitations to participate in our online survey were sent to 15,000 entrepreneurs,

4,131 managers and 8,000 employees on December 16th, 2014. All groups had 14 days

to respond, and a reminder was sent after 7 days. The survey was completed by 697

entrepreneurs, 265 managers and 969 employees (n = 1,931). Response rates were thus

between 5 - 12%, in line with earlier experiences. Furthermore, comparing the samples

of entrepreneurs, managers and employees across the different research rounds reveals

that the sampled selections are rather similar in every round. For instance, when we

compare the samples in this paper with the first experiment of Koudstaal et al. (2015),

we find no differences in gender and education for each of the occupational groups and

only small differences in age. One might conjecture that this may be the result of many

repeat participants, but we find that this applies to only 18% of the sample.

10(R) indicates reversely scored.

107

Default definitions of entrepreneurs, managers and employees

Similar to Koudstaal et al. (2015), the following definitions were used. Entrepreneurs

were definied as all people who have founded, inherited or taken over a company that

they are currently (co-) managing. Also individuals who obtained firm ownership

over a company within 5 years after start-up and who are currently its (co-)manager

were classified as entrepreneur. A manager was someone who had at least two direct

reporting lines in an organization that was not started up by him/herself. Project

managers who have overall responsibility for their projects and at least two direct

reporting lines were also considered as managers. Furthermore, employees were those

individuals who were employed by an organization that they did not start up themselves

and who did not have at least two direct reports. Finally, the category ‘Other’ consisted

of all invididuals who belonged to neither of these categories. These individuals (n

= 82) were dropped from the final sample, leaving us with the 1,931 observations

mentioned before.

4.2.4 Incentives

The online survey that we sent out to all participants contained four parts. In this

chapter, we report the results of Parts 2 and 3, which included the measures as described

in Subsections 4.2.2 and 4.2.3.11 Note that the ordering in the survey was the same

as in these two Subsections; that is, we first asked participants to make choices in

the selected games from Rubinstein (2016) and then asked them to what extent they

agreed/disagreed with the ten statements related to Faith in Intuition and Need for

Cognition.

Before filling out each of these two parts, all participants were instructed that they

would receive €200 in their accounts after completing Part 2 (the selected games) and

€175 after completing Part 3 (the ten statements). The main reason for including this

€375 was not to incentivize Parts 2 and 3, but rather to ensure that all participants

could really lose money in the incentivized gambles in Parts 1 and 4.12 As the fixed

payment in Parts 2 and 3 did not vary with the choices made, these choices could

therefore be considered hypothetical (in line with Rubinstein, 2016).

11The other two parts of the survey, Parts 1 and 4, were related to loss aversion and are furtherdescribed in Chapter 2.

12See also Chapter 2. Note that instead of paying out every participant, we opted to pay out 20randomly drawn prize winners. For these prize winners, we added up all gains and losses in Parts 1and 4 and added/subtracted this number to/from the €375 gained in Parts 2 and 3.

108

4.3 Descriptive statistics

In Panel A of Table 4.1, we first describe the main background characteristics of our

final sample. In line with our earlier research waves, we find that entrepreneurs and

managers are on average quite comparable to each other, while employees are somewhat

younger and more likely to be female. On average, employees also have a lower degree

of education and earn less. Furthermore, for entrepreneurs, the key take-aways from

Panels B - D are that they are mostly business founders, equally likely to be incorpo-

rated or not, and most likely to be in the start-up phase. The average entrepreneurial

firm in our sample consists of 17 employees, but the distribution is highly skewed to

the right. The opposite holds true for the firms that managers and employees work for,

which are generally old and/or large. Finally, we also find that the ‘average’ manager

is typically a general manager rather than a CEO or project manager, and is most

likely to have responsibility over 2-5 employees.

In Table 4.2, we turn to the descriptive statistics of both the objective and subjec-

tive measures. The first set of rows in Panel A of Table 4.2 refers to the objectives

measures, while the second set of rows in Panel A describes the statistics of the subjec-

tive measures. First examining the total number of contemplative choices, we observe

that most individuals in our sample chose intuitively. Participants on average made

0.87 contemplative choices on the three games, while the median is 1 out of 3.13 The

average response time is close to 200s, which translates to a bit more than one minute

per game. Note that there is quite a large variation in response times, though, with

the lowest response time being almost 9s while the highest one is over 1,400s. Finally,

the last two rows of Panel A show that the average values for Faith in Intuition and

Need for Cognition are 18.92 and 17.74, respectively, and the participant values seem

to cover almost the entire feasible range of 5 - 25.

Panel B of Table 4.2 reports the (Pearson) correlations between the four measures

of intuition. As expected, and in line with Rubinstein (2016), we find that response

times are positively associated with contemplative thinking. Also the correlations with

the subjective measures have the expected signs; a higher score on Faith in Intuition

is negatively associated with the Contemplative Index (CI) and response times, while

the reverse holds for Need for Cognition. Finally, the correlation between Faith in

Intuition and Need for Cognition is -0.12, which is slightly higher than the -0.07 (n.s.)

found by Epstein et al. (1996) among a sample of 184 participants.

13For a direct comparison between the intuitive and contemplative choices per game (as in Rubin-stein, 2016), see Appendix A.

109

Table 4.1 Background Characteristics of Entrepreneurs, Managers and Employees

Entrepreneurs Managers Employees

(n = 697) (n = 265) (n = 969)

Panel A: General Panel A: General

Age 49.11 Age 47.22 43.88

Female (dummy) 0.28 Female (dummy) 0.32 0.44

Education: Education:

- High School 5% - High School 2% 11%

- Lower voc. degree 14% - Lower voc. degree 4% 33%

- College education 46% - College education 42% 37%

- University education 35% - University education 52% 19%

Experience (years) 14.27 Experience (years) 14.00 21.03

Income 1) €99,490 Income 1) €97,586 €38,066

Panel B: Entrepreneur characteristics Panel B: Manager characteristics

Founder 82% CEO 16% -

Business taken over 18% General Manager 62% -

Joined within 5 yrs 0% Project Manager 22% -

Panel C: Firm age and legal structure Panel C: Firm age and size

Start-up phase (0 - 3 yrs) 20% Firm age ≤ 5 yrs 4% 6%

Survival phase (0 - 5 yrs) 38% Firm age 6 - 50 yrs 49% 55%

Firm age > 50 yrs 47% 39%

Incorporated 44% Firm size ≤ 25 FTE 11% 12%

Sole propriotership 44% Firm size 26 - 1000 FTE 50% 48%

Other 12% Firm size > 1000 FTE 39% 40%

Panel D: Management level Panel D: Management level

No. of FTE in own firm: Direct reports:

0 18% 2 - 5 38% -

1 29% 6 - 10 29% -

2 - 5 23% 11 - 25 22% -

6 - 10 10% 26 - 50 8% -

11 - 25 10% More than 50 3% -

26 - 50 4%

More than 50 6%

1) For income, the number of observations drops to: 442 (entrepreneurs), 233 (managers), and 817 (employees).

110

Table 4.2 Descriptive Statistics

Panel A: Means Obs. Mean Median St. dev. Min. Max.

Objective measures

- CI (total no. of contemplative choices) 1,931 0.87 1 0.82 0 3

- Response Time 1,931 196.84s 162.90s 134.66s 8.84s 1,406.19s

Subjective measures

- Faith in Intuition (FI) 1,931 18.92 19 2.78 7 25

- Need for Cognition (NFC) 1,931 17.74 18 3.08 7 25

Panel B: Correlations CI Response Faith in Need for

(= total no. of con- Time Intuition Cognition

templative choices) (FI) (NFC)

Objective measures

- CI (total no. of contemplative choices) -

- Response Time 0.14 *** -

Subjective measures

- Faith in Intuition (FI) -0.05 ** -0.08 ** -

- Need for Cognition (NFC) 0.08 *** 0.11 *** -0.12 *** -

Significance at the 10% level is denoted by *, 5% by **, and 1% by ***.

4.4 Results

4.4.1 Main results

Table 4.3 provides an overview of the raw differences between entrepreneurs, managers

and employees. The numbers indicate that entrepreneurs make significantly less con-

templative choices than managers (0.84 versus 1.04), but not than employees (0.84).

Similar results are also obtained when we use the Kolmogorov-Smirnov test for equality

of distribution functions, although the differences between managers and the other two

groups now only turn significant at the 10% level (both p-values are 0.06).

111

Table 4.3 Raw Differences in Decision-Making

Entrepreneurs Managers Employees

(n = 697) (n = 265) (n = 969)

Objective measures

- CI (total no. of contemplative choices) 0.84 b 1.04 b,c 0.84 c

- Response Time 209.85s a 209.99s c 183.89s a,c

Subjective measures

- Faith in Intuition (FI) 19.11 a,b 18.61 b 18.86 a

- Need for Cognition (NFC) 17.93 a,b 18.55 b,c 17.38 a,c

The superscripts report significant differences from making pairwise within-column comparisons, in the following way:

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

Surprisingly, the raw differences in response times are not fully in line with the find-

ings on the Contemplative Index. We find that managers have a significantly higher

response time than employees, as expected, but we also find that the same holds true

for entrepreneurs vis-a-vis employees. In fact, the difference between entrepreneurs and

managers is not significant, so we cannot conclude that entrepreneurs spend a lower

amount of time on the three games than managers. The latter conclusion does not

change when we test for differences in the distributions of response times (p-value of

the K-S test is equal to 0.71). We will therefore return to this point in Section 4.4.2.

The last two lines in Table 4.3 illustrate the differences on Faith in Intuition and

Need for Cognition. It shows that entrepreneurs have the highest Faith in Intuition,

followed by both managers and employees. On the Need for Cognition scale, the pattern

is somewhat different. Here we find that managers have the highest score, followed by

entrepreneurs and then employees. The outcomes of the Kolmogorov-Smirnov tests

lead to a different conclusion than the two-sample t-tests, though. We find that the

differences in Need for Cognition are again only signficant at the 10% level, while the

differences in Faith in Intuition do not reach signficance at all (i.e. all p-values exceed

0.10).

Table 4.3 reports raw differences, but does not take differences in background char-

acteristics into account. It is however important to do so, as Table 4.1 shows that

entrepreneurs and managers are typically better educated and more likely to be male

than employees. We therefore next consider the occupational differences in a regression

framework.

112

Table 4.4 depicts the results when running ordered probit regressions on the four

different measures of decision-making styles. Column (1a) first confirms that en-

trepreneurs and employees are indeed equally intuitive, while managers are more con-

templative if background characteristics are left out. Adding control variables in col-

umn (1b) does not change this result; we find that the coefficient of the dummy variable

“Manager” is somewhat lower, but still significant at the 5% level. Furthermore, we

also find that younger, male, and higher educated participants tend to behave more

contemplative (in line with e.g. Choi et al., 2014), while experience and having Eco-

nomics as the field of study do not have a significant effect.

Columns (2a) and (2b) report the regression outputs when we take response time

as the dependent variable. Similar as in Table 4.3, Column (2a) illustrates that en-

trepreneurs and managers have a significant higher response time than employees.

When we include the background characteristics in column (2b), we find that older,

male and less experienced participants as well as those with Economics as their field

of study have higher response times.

Finally, the last two sets of columns in Table 4.4 relate to Faith in Intuition and

Need for Cognition. In line with the two-sample t-tests in Table 4.3, column (3a)

indicates that entrepreneurs have the highest Faith in Intuition, while column (4a)

shows that managers stand out in their Need for Cognition. When we include the

controls for background characteristics in columns (3b) and (4b), these results keep

standing. We only find that the difference between entrepreneurs and managers on

Need for Cognition is not significant anymore (p-value is 0.22).

Taking all columns together, there are several puzzling features in Table 4.4. For

instance, why do we find that entrepreneurs and managers differ in their Contemplative

Index and Faith in Intuition, but not in their response times? And similarly, why do

we find that managers differ from employees in their Contemplative Index and Need

for Cognition, but not in their response times? We will deal with this in the next

Subsection.

4.4.2 Heterogeneous effects of response times on contempla-

tiveness

To gain further insight into the seemingly mixed findings on the objective measures, we

further explore the mismatch between the Contemplative Index and response times in

Table 4.5. Here we report the average Contemplative Index per response time quartile

per occupational category (in fact, we separate the lower quartile, the interquartile

113

Tab

le4.4

Pro

bit

Regre

ssio

ns

on

Decis

ion

-Makin

gO

bje

ctiv

em

easu

res

Su

bje

ctiv

em

easu

res

(1a)

(1b

)(2

a)

(2b

)(3

a)

(3b

)(4

a)

(4b

)

Dep

.va

riab

le:

CI

CI

Res

p.

Res

p.

FI

FI

NF

CN

FC

Tim

eT

ime

Entr

epre

neu

r-0

.031

-0.0

5925.9

6∗∗

∗8.0

62

0.0

87∗

0.2

03∗

∗∗0.1

84∗

∗∗0.0

73

[-0.

58]

[-0.

87]

[3.7

8]

[0.9

3]

[1.7

5]

[3.2

2]

[3.6

6]

[1.1

5]

Man

ager

0.25

3∗∗

∗0.

197∗∗

26.1

1∗∗

∗7.5

38

-0.0

90

0.0

29

0.3

83∗

∗∗0.1

62∗∗

[3.3

0][2

.30]

[2.8

9]

[0.7

4]

[-1.2

8]

[0.3

7]

[5.7

6]

[2.1

6]

Age

-0.0

08∗∗

0.6

86∗

-0.0

06∗

∗-0

.009∗∗

∗

[-2.

27]

[1.6

5]

[-1.9

6]

[-2.9

4]

Fem

ale

-0.1

77∗∗

∗-1

9.4

2∗∗∗

0.2

70∗

∗∗-0

.169∗∗

∗

[-3.

34]

[-3.0

9]

[5.5

4]

[-3.4

6]

Ed

uca

tion

0.07

0∗∗

5.7

29

-0.0

91∗∗

∗0.3

20∗∗

∗

[2.1

9][1

.46]

[-3.2

9]

[11.0

7]

Eco

n0.

022

25.0

6∗∗

0.1

46∗∗

-0.0

39

(Y=1,N=0)

[0.3

2][2

.48]

[2.2

6]

[-0.6

5]

Exp

erie

nce

-0.0

01-0

.902∗∗

0.0

02

0.0

01

[-0.

20]

[-2.2

7]

[0.6

1]

[0.2

1]

con

stan

t0.

267∗

∗∗0.

506∗∗

∗183.9

∗∗∗

163.2

∗∗∗

3.2

75∗

∗∗3.6

46∗

∗∗3.2

06∗

∗∗3.0

09∗∗

∗

[7.0

0][3

.44]

[45.9

4]

[8.9

0]

[11.4

4]

[11.3

2]

[11.1

6]

[9.0

8]

Ob

s.1,

931

1,93

11,9

31

1,9

31

1,9

31

1,9

31

1,9

31

1,9

31

Log

lik.

-2,2

22.6

-2,2

06.2

-12197.7

-12181.8

-4602.2

-4570.7

-4865.0

-4775.4

EN

T=

MA

N1)

<0.

01∗∗

∗<

0.01

∗∗∗

0.9

90.9

60.0

2∗∗

0.0

2∗∗

<0.0

1∗∗

∗0.2

21)

Th

isre

port

sth

ep

-valu

eof

the

Wald

test

‘Entr

epre

neu

r’=

‘Man

ager

’.

Not

es:

the

cate

gor

ical

vari

able

‘ed

uca

tion

’h

asb

een

sum

mari

zed

into

on

eva

riab

lein

stea

dof

usi

ng

ase

tof

du

mm

ies.

Th

eed

uca

tion

vari

ab

leta

kes

on

the

valu

e0

ifth

eh

igh

est

atta

ined

leve

lis

hig

hsc

hool

orlo

wer

,1

ifse

con

dary

edu

cati

on

isob

tain

edat

ah

igh

erle

vel,

2if

ap

art

icip

ant

has

coll

ege

edu

cati

on

and

3if

the

par

tici

pan

th

asa

un

iver

sity

deg

ree.

Eco

nis

an

indic

ato

rva

riab

lew

hic

his

1if

the

fiel

dof

stu

dy

was

Eco

nom

ics.

Exp

erie

nce

mea

sure

sth

eye

ars

ofex

per

ien

ce.

Sig

nifi

can

ceat

the

10%

level

isd

enot

edby

*,5%

by

**,an

d1%

by

***,w

ith

t-st

ati

stic

sre

port

edin

pare

nth

eses

.S

tand

ard

erro

rsare

rob

ust

.

114

Table 4.5 Raw Differences in Decision-Making (continued)

CI CI CI P-value of Cuzick’s test

Response Time quartile: Bottom 25% 25 - 75% Upper 25% for within-row trends

Entrepreneurs (n = 697) 0.64 a,b 0.73 b 1.12 a 0.04

Managers (n = 265) 0.89 b 0.96 b 1.26 c 0.01

Employees (n = 969) 0.80 a 0.80 0.98 a,c 0.04

The superscripts report significant differences from making pairwise within-column comparisons, in the following way:

a) Significant difference between entrepreneurs and employees at the 5% level (two-sample t-test)

b) Significant difference between entrepreneurs and managers at the 5% level (two-sample t-test)

c) Significant difference between managers and employees at the 5% level (two-sample t-test)

range and the upper quartile).14 The results indicate that there is an upward trend in

the average Contemplative Indices when we move from the lower response time quartile

to the upper one (i.e. all p-values of the Cuzick test are below 0.05). Interestingly,

Table 4.5 also indicates that the three occupational groups seem to differ in their

“starting points” and the steepness of their “learning curves”. In particular, we find

that entrepreneurs make significantly more intuitive choices than employees when we

focus on the lowest 25% of the response times, while the opposite holds when we

zoom in on the highest 25%. Also the gap between entrepreneurs and managers closes

when we move from the lower quartile to the upper one, and the raw difference turns

insignificant when we examine the highest 25% of response times (p-value is 0.12).

The raw differences in Table 4.5 suggest that entrepreneurs have a stronger intuitive

prior, i.e. are more likely (than managers and employees) to choose the intuitive option

when response times are low. Yet the effect of taking more time to think one’s choices

over seems larger as compared to (especially) employees. We further explore this in

our earlier Probit regression framework explaining the CI score. Taking column (1b)

in Table 4.4 as a starting point, we add Response Time and its interactions with the

two respective occupational dummies “Entrepreneur” and “Manager” as additional

regressors. Column (1) in Table 4.6 reports the results. We indeed observe that the

“Entrepreneur” dummy is significantly negative, again pointing at entrepreneurs having

14This separation is made based on the full sample statistics. This implies that all participants whodecide quicker than 120.88s are in the lower quartile, all those who take more than 230.76s to decideare in the upper quartile, and all others that are in between are in the interquartile range. Notethat these bounds seem to coincide well with the 144.28s (243.77s) when adding up the three medianresponse times in Appendix A for the three intuitive (contemplative) choices. Hence, the respondentsin the lower (upper) quartile can reasonably be expected to be more intuitive (contemplative).

115

Table 4.6 Additional Probit Regressions Using the Other Measures

(1) (2) (3) (4)

Dependent variable: CI CI CI CI

Entrepreneur -0.251∗∗ -0.063 -0.056 -0.252∗∗

[-2.32] [-0.93] [-0.82] [-2.35]

Manager -0.015 0.195∗∗ 0.195∗∗ -0.014

[-0.10] [2.27] [2.27] [-0.09]

Response Time / 100 0.057∗ -0.018 0.031 -0.044

[1.82] [-0.20] [0.20] [-0.27]

Entrepreneur x Response Time / 100 0.088∗∗ - - 0.094∗∗

[2.20] - - [2.42]

Manager x Response Time / 100 0.098∗ - - 0.100∗

[1.71] - - [1.83]

Education x Response Time / 100 0.043 - -

[1.52] - -

Faith in Intuition (FI) 0.012 0.011

[0.70] [0.64]

Need for Cognition (NFC) -0.018 -0.018

[-1.35] [-1.28]

Faith in Intuition x Response Time / 100 -0.011 -0.010

[-1.57] [-1.48]

Need for Cognition x Response Time / 100 0.015∗∗∗ 0.016∗∗∗

[2.98] [2.79]

Constant 0.415∗∗∗ 0.591∗∗∗ 0.469 0.573

[2.65] [2.60] [1.10] [1.33]

Control variables YES YES YES YES

Obs. 1,931 1,931 1,931 1,931

Log lik. -2,185.9 -2,186.5 -2,182.4 -2,179.1

ENT=MAN 1) 0.10∗ < 0.01∗∗∗ < 0.01∗∗∗ 0.09∗

ENT x Resp. Time = MAN x Resp. Time 2) 0.84 - - 0.91

1) This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

2) This reports the p-value of the Wald test ‘Ent. x Response Time / 100’ = ‘Man. x Response Time / 100’.

Significance at the 10% level is denoted by *, 5% by **, and 1% by ***.

116

a stronger intuitive prior. The interaction terms between Response Time and the

two occupational dummies are significantly positive, revealing that entrepreneurs and

managers are relatively more contemplative per unit of time than employees. They

however do not differ in this respect from each other (p-value is: 0.84).15

Because entrepreneurs and managers vary in a number of background characteristics

from employees, the finding that they are relatively more contemplative per unit of

time may potentially be driven by these differences in their background. A plausible

conjecture for instance might be that especially the higher educated are more likely to

deviate from making the a priori intuitive choice when they take more time to carefully

think about the decision. Column (2) explores this potential explanation. Instead of

interaction terms with occupation, it includes the interaction between Response Time

and Education. This interaction appears to be insignificant, hence we cannot conclude

that higher educated participants are relatively more contemplative per unit of time. A

similar explanation could be that, instead of education, it is rather those participants

with a higher Need for Cognition or a higher Faith in Intuition who are different.

Column (3) reports the results when these explanatory variables and their interactions

with Response Time are included. Those with a higher Need for Cognition appear to

be relatively more contemplative per unit of time; the interaction with Response Time

is significantly positive. In column (4) we therefore include both the interactions with

occupation and the interactions with FI and NFC. We find that both explanations

remain valid, although the coefficients in column (4) differ only marginally from the

ones in column (1). Hence, altogether, Table 4.6 suggests that entrepreneurs, managers

and employees differ in their “starting points” as well as the steepness of their “learning

curves”.

4.4.3 Robustness checks

In this Subsection, we examine the impact of applying stricter definitions of entrepre-

neurs and managers. Panel A in Table 4.7 first exploits the variation in background

characteristics of entrepreneurs, see also Table 4.1. We test the impact of restricting

the sample to: (i) entrepreneurs with an incorporated firm, thereby mainly excluding

the own-account self-employed, (ii) entrepreneurs with an above median number of

fulltime equivalent employees in their company, (iii) entrepreneurs with above median

incomes, (iv) entrepreneurs that have founded their business, instead of obtaining it

through takeover or buy-in, and (v) entrepreneurs in the survival phase (i.e. firm age

≤ 5 years). Note that every coefficient reported in Table 4.7 results from a separate

15To control for potential non-linear effects of response times on contemplativeness, we have alsoincluded a quadratic term of Response Time and its interaction with the occupational dummies in thesame regression as in column (1). None of the (unreported) coefficients however reach significance.

117

Table 4.7 Differences in Decision-Making Styles using Stricter Definitions

(1) (2) (3) (4) (5)

Dependent variable: CI Resp.Time FI NFC CI

Regression similar as in: T4.4 (1b) T4.4 (2b) T4.4 (3b) T4.4 (4b) T4.6 (4)

Panel A: Subsets of Entrepreneurs

i) Incorporated (n=306) -0.131 b 9.765 0.271 a,b 0.037 -0.332 a,b

[-1.44] [0.86] [3.17] [0.43] [-2.39]

ii) Above med. no. of FTE (n=310) -0.078 b 4.522 0.263 a,b -0.009 -0.305 a,b

[-0.94] [0.42] [3.34] [-0.11] [-2.32]

iii) Above med. ent. income (n=189) -0.052 b -9.977 0.249 a,b 0.072 -0.295

[-0.52] [-0.88] [2.56] [1.03] [-1.85]

iv) Founder (n=626) -0.019 b 20.47 a 0.230 a,b 0.049 -0.350 a,b

[-0.54] [2.15] [3.28] [0.68] [-2.54]

v) In survival phase (n=221) -0.131 b 9.765 0.271 a,b 0.049 -0.265 a,b

[-1.44] [0.86] [3.17] [0.72] [-2.23]

β(Entrepreneur) in Tables 4.4&4.6: -0.059 b 8.062 0.203 a,b 0.073 -0.252 a

Panel B: Subsets of Managers

vii) CEO or general manager (n=230) 0.201 b,c -4.738 0.047 b 0.193 c 0.057

[2.52] [-0.44] [0.55] [2.38] [0.69]

viii) CEO (n=46) 0.129 -13.09 -0.048 0.229 -0.621

[0.77] [-0.80] [-0.32] [1.41] [-1.79]

ix) Above med. dir. reports (n=168) 0.250 b,c -4.738 0.066 b 0.148 -0.092

[2.48] [-0.44] [0.71] [1.70] [-0.45]

x) Above med. man. income (n=183) 0.129 -3.302 0.111 0.154 -0.146

[1.26] [-0.28] [1.25] [1.76] [-0.86]

xi) Manager in firm > 15 yrs (n=272) 0.219 b,c 11.41 0.042 b 0.152 c 0.072

[2.41] [1.06] [0.51] [2.03] [0.97]

β(Manager) in Tables 4.4&4.6: 0.197 b,c 7.538 0.029 b 0.162 c -0.014

Panel C: Combinations of A&B

i) vs. viii); p-values Wald tests 0.41 0.23 0.05 0.43 0.63

ii) vs. ix); p-values Wald tests < 0.01 0.17 0.03 0.22 0.26

iii) vs. x); p-values Wald tests 0.25 0.68 0.41 0.73 0.60

Control variables YES YES YES YES YES

The superscripts report significant differences from making pairwise within-column comparisons, in the following way:

a) Significant difference between entrepreneurs and employees at the 5% level

b) Significant difference between entrepreneurs and managers at the 5% level

c) Significant difference between managers and employees at the 5% level

118

regression. Moreover, the first four columns test the robustness of the findings in Table

4.4, while the final column tests the robustness of the findings in Table 4.6.

The results in Panel A show that our findings in Tables 4.4 and 4.6 are robust to

using stricter definitions of an entrepreneur. In other words, we find that all differences

between entrepreneurs and the other two occupational groups remain significant when

using the stricter definitions as in i) - v). We also find that all of the (unreported)

interaction terms between Response Time and the stricter definitions of an entrepreneur

remain significantly different from zero, and that they do not differ from the interaction

term between Response Time and Manager.

Panel B of Table 4.7 shows the regression outputs when we use alternative defi-

nitions for managers. We restrict the sample of managers to: (vii) CEOs or general

managers (so without project managers), (viii) CEOs exclusively, (ix) managers with

more than the median number of direct reports, (x) managers with an above median

managerial income, and (xi) managers in firms that are older than 15 years old.

The results show that the main findings of Table 4.4 now only seem to be robust at

the 10% significance level and in just three out of the five cases. For CEOs and man-

agers with an above median income, the coefficients in column (1) are not only lower

than in Table 4.4, but also insignificantly different from the ones for entrepreneurs and

employees. While the latter might be explained by the smaller sample sizes, the former

could be consistent with the suggestions of Parikh (1994) and Hayashi (2001) that

top managers are required to have an enhanced ‘business intuition’ when climbing the

corporate ladder. Furthermore, and in line with Panel A, we find that the results of

Table 4.6 are robust to using stricter definitions of a manager. In fact, all (unreported)

interaction terms between Response Time and the other definitions of a manager re-

main significantly different from zero, and insigificantly different from the interaction

term between Response Time and Entrepreneur.

4.5 Conclusion

Decision-making styles have been an interest of many studies both in economics and

psychology. While most studies in psychology have relied on the Faith in Intuition and

Need for Cognition scales, the economics literature has largely focused on contempla-

tiveness in a game theoretical framework. In a recent stream of work of Rubinstein

(2007, 2013, 2016), there has been a growing interest in using response times as another

proxy of contemplativeness. In other words, the “fast thinkers” are expected to be less

contemplative and thus more intuitive than the “slow thinkers”. The main difference

between these aforementioned measures is that the former set of measures (i.e. Faith

119

in Intuition and Need for Cognition) is based on statements and therefore more ‘sub-

jective’, while the latter set of measures (i.e. response times and choice behavior in

experimental games) is more ‘objective’.

Using these two sets of measures, this paper tests (i) how entrepreneurs, managers

and employees differ in their decision-making styles and (ii) whether these findings are

consistent across the two sets of measures. Our final sample consists of 697 established

entrepreneurs, 265 managers, and 969 employees (hence n = 1,931).

The results on the objective measures first show that entrepreneurs make more

intuitive choices than managers, but not than employees. Surprisingly, however, the

three occupational groups do not differ in their response times when we include controls

for age, gender, education, and experience. Second, on the subjective measures we find

that entrepreneurs score highest on Faith in Intuition, while Managers score highest on

Need for Cognition. Taken together, the results on the subjective measures thus seem

consistent with the objective measures, but also somewhat more pronounced.

Furthermore, to reconcile the mixed findings on the objective measures, we run

an addtitional set of regressions to verify what might have caused this inconsistency.

We find it to be most likely that there is a heterogenous effect of response times,

especially for entrepreneurs and managers. In other words, while entrepreneurs and

managers might not have higher response times than employees, they are relatively

more contemplative per unit of time. Taking this heterogeneity into account, it shows

that entrepreneurs are not only more intuitive than managers, but also more than

employees when their response times are low (and hence when they do not capitalize

on their higher effectiveness).

Finally, we also test the robustness of the aforementioned results to the use of

stricter definitions of an entrepreneur and/or a manager. We find that the differences

between the three occupational groups remain significant in most of the additional

specifications. Only when we restrict the sample of managers to CEOs and managers

with an above median income do we find that some of the differences lose significance.

While this might be attributable to smaller sample sizes, it may also be that the

decision-making styles of top managers do converge somewhat to entrepreneurs and

employees.

120

Appendices Chapter 4

Appendix A. Raw Differences in Decision-Making Between Instinctive and Contem-

plative Types (Objective Measure).

Appendix B. Survey screenshots.

121

APPENDIX A. Raw Differences in Decision-Making Between Instinctive and

Contemplative Types (Objective Measure)

Obs. % Median Mean Significance level

Response Response of the difference in

Time Time Response Times

(I) = instinctive, (C) = contemplative (p-value, t-test)

- Game #1

— Response Time of “Enter” (C) 350 18% 47.70s 70.10s < 0.01 ***

— Response Time of “Not enter” (I) 1,581 82% 38.27s 47.69s

- Game #2

— Response Time of “nr 1/4/7” (I) 1,329 69% 56.69s 76.71s < 0.01 ***

— Response Time of “nr 2/3/5/6” (C) 602 31% 72.95s 93.19s

- Game #3

— Response Time of “Coin” (I) 1,205 62% 49.32s 59.69s < 0.01 ***

— Response Time of “Die” (C) 726 38% 57.63s 80.48s

Significance at the 10% level is denoted by *, 5% by **, and 1% by ***.

For each individual game we report the percentages of participants who chose the intuitive

and the contemplative answer, respectively, as well as the median and mean response times

for each of these two categories. The final column indicates the significance level of the differ-

ence between the mean response times (in a two sample t-test). The results confirm that the

more contemplative types are indeed the ‘slower’ thinkers as suggested by Kahneman (2011)

and Rubinstein (2016). We find that not only their mean response times are different, but

also their median response times, thus alleviating the concern that it might be just a very

few participants who think about their answers for a prolonged time. Furthermore, all of

the results presented here also turn out to be robust when cutting off the fastest 5% of the

responses (see also Rubinstein, 2016) or when cutting off both the fastest 5% as well as the

slowest 5% of the responses.16 Hence, our findings also do not seem to be the mere outcome

of partcipants rushing through the survey.

16In all three individual games, the lower 5% cutoff level was approximately 17 secondswhile the upper 5% cutoffs varied from 114 to 188 seconds.

122

Appendix B. Survey screenshots

Figure B1: Example game (Objective Measure)

Figure B2: Decision-Making Style (Subjective Measure)

123

124

Chapter 5

Preference for Invididual versus

Team Incentives

This chapter is based on Rosendahl Huber, L., E. Czibor, and M. Koudstaal, “Risks,

Gains and Autonomy: An Experimental Analysis of Preferences for Joining Teams”,

working paper.

5.1 Introduction

Different incentive schemes lead to different output levels not only through the incentive

effect but also through sorting. A famous example is the study by Lazear (2000)

showing that half of the productivity gains at Safelite Glass may be attributed to the

self-selection of more productive agents into the company after the introduction of a

piece rate incentive scheme (see also Leuven et al. (2011) who disentangle the incentive

and sorting effect of tournaments). Nowadays, many firms rely on teamwork, or use

some sort of team-based reward structure (Hamilton et al., 2003; Bandiera et al., 2013).

The productivity of the teams and the success of the use of team incentives strongly

depend on the sorting that will occur, that is, on the characteristics of the individuals

that self-select into the organization offering these types of incentives Hamilton et al.

(2003). Although the effectiveness of team remuneration schemes has been extensively

researched (see e.g. van Dijk et al., 2001; Hamilton et al., 2003; Bandiera et al., 2013;

and Danilov et al., 2013), most papers on team decision-making or team production

take the composition of the team as a given, and do not take the drivers and the effect

of sorting into team incentives into account. Our aim is to contribute to the literature

on team incentives by studying the actual participation decision of individuals.

Team-based incentive schemes affect different aspects of the individual’s decision

problem: they influence both the expected monetary gains and the uncertainty asso-

125

ciated with the payoffs. The participation decision is therefore governed by the indi-

vidual’s ability, confidence, risk preferences and beliefs about the potential partners.1

The extent to which these factors matter for sorting depends largely on the design

of the team remuneration scheme. Most papers focusing on self-selection into teams

model the team option as a simple revenue sharing contract, e.g. an equal split of

the pooled total output of the members (Dohmen and Falk, 2011; Baker and Mertins,

2013; Herbst et al., 2015). A few studies acknowledge potential team efficiency gains

by adding some pre-defined, automatic mark-up on top of the joint output (Cooper

and Saral, 2013; Kuhn and Villeval, 2014). A distinguishing feature of real life team

production, however, is the possibility for synergies: gains from the team option are

ex ante unknown and depend on the degree of complementarities between the team

members. The study discussed in this chapter addresses this issue by using a team

production function that allows potential synergy gains.

Moreover, in real life team situations the potential gains from joint production

often come at a price: team members have to partially give up their authority and

make joint decisions. Even when the instrumental value of the decision rights is not

too high (in the sense that the joint choice is not far from the individual’s optimum),

the loss of power resulting from joint decision-making could still make the team choice

unattractive for some (Fehr et al., 2013; Bartling et al., 2014; Sloof and von Siemens,

2014). The existing evidence suggests that people differ in the intrinsic value they

attach to decision rights. Self-employed, for instance, are claimed to value control

more than non-entrepreneurs (Benz and Frey, 2008; Reynolds and Curtin, 2008). It

is unclear, however, whether they are motivated by a desire for payoff autonomy (i.e.

they prefer to retain an independent control over their own outcomes, see Owens et al.,

2014) or authority (i.e. the right to make decisions, see Bartling et al., 2014). Our

study intends to separate the impact of these two drivers on the choice to enter teams.

To summarize, this chapter aims to contribute to the existing economics literature

on team preferences in two ways. The first objective is to shed light on the factors

that influence the choice between individual and team incentives in a setting where

the teamwork includes potential synergy gains. We test for heterogeneous choices by

individual characteristics such as age, gender, education, income and occupational cat-

egories, and see if these heterogeneities are attributable to differences in performance,

beliefs or preferences. Secondly, we want to determine the importance of payoff auton-

omy and riskiness for the self-selection into teams that involve both joint production

and joint decision-making. To this end, we specifically designed two team treatments

reflecting these different team characteristics. We conduct a large scale lab-in-the-field

1Teyssier (2008) and Kuhn and Villeval (2014) show that inequity aversion could also play a rolein the choice to select team remuneration schemes. This factor is not in the focus of our study.

126

experiment in which individuals are randomly assigned to one of the two team treat-

ments, combining a controlled experimental environment with the flexibility and reach

allowed by an online survey. Our sample contains 1,164 individuals (entrepreneurs,

manager and employees) and is very diverse in terms of age, education, experience and

income.

The experiment consists of four incentivized parts. In the first part, we measure the

productivity level of each individual in a real-effort task (i.e. solving Raven matrices

adopted from Raven et al., 2003). In the second part, we elicit participants’ risk pref-

erences following Gneezy and Potters (1997). In Part 3, we use the BDM mechanism

(Becker et al., 1964) to elicit participants’ willingness to pay for the team incentive

scheme. In the fourth part of the experiment we measure the beliefs of each partic-

ipant regarding his/her own performance, and the performance and risk preferences

of a random survey participant. These four incentivized parts are followed by a short

questionnaire to collect information about the rationale behind the team choice and

some background characteristics. At the beginning of the survey we randomly assign

each individual to one of two team treatments conditions. In the Baseline treatment

the team option only involves joint production (as in e.g. Cooper and Saral, 2013;

Kuhn and Villeval, 2014). In the second (Joint Decision) treatment, the team option

includes both joint production and joint decision-making. The comparison between

the willingness to pay for the team option in the two treatments allows us to examine

if and to what extent the loss of autonomy in decision-making affects self-selection into

teams.

We analyze the determinants of team choice by first focusing on the subsample

of participants in the Baseline treatment (joint production without joint investment).

Contrary to the findings from previous laboratory experiments (Kocher et al., 2006;

Teyssier, 2008; Dargnies, 2012), we find that the willingness to be in a team is unrelated

to task performance, i.e. in our study low performers do not prefer team pay more than

high performers. To understand why we observe no adverse selection in this setting, we

evaluate several factors that potentially affect the choice for team incentives. In line

with previous studies (Dohmen and Falk, 2011; Kuhn and Villeval, 2014) we find that

relative and absolute confidence and risk aversion all influence the selection into team

incentives, in the expected directions. All else equal, confidence in one’s own absolute

performance decreases the willingness to join a team. Higher relative performance

expectations also lead to lower bids for the team option. Moreover, we find that risk

aversion has a negative impact on the willingness to be in a team. Thus, similar to the

results from a recent study by Baker and Mertins (2013), we find that the uncertainty

about (the performance and the effort of) the teammate is more important than the

reduction in the exposure to individual idiosyncratic shocks.

127

Interestingly, we find that education is strongly positively correlated with prefer-

ences for team incentives (even when controlling for actual and guessed performance).

That is, all else equal, higher educated individuals in our sample bid significantly more

for the team option. This result is surprising given that low educated people on average

perform worse on the task and could thus gain even more from the team option than

higher educated respondents. Looking further into this positive education effect, we

find that the negative impact of risk aversion on the preference for teams decreases

with the education level: the higher a participant’s level of education, the weaker the

link between her risk preferences and her team sorting decision. This suggests that the

level of education is correlated with the way people evaluate team incentives. While

lower-educated individuals tend to concentrate on the risky aspect of team pay, those

with a higher level of education focus more on the potential gains and synergies from

teamwork. This interpretation of our findings stems from self-reported explanations of

participants for their team bids and is also confirmed by a regression analysis of the

impact of risk preferences on the bids for the team option. Given that the experimental

task (i.e. solving Raven puzzles) is a proxy for cognitive ability, our results are consis-

tent with the idea that education (above and beyond its impact on task performance)

affects preferences for teams by changing the way people weigh the associated risks

and gains. This brings to mind the finding of Heckman et al. (2013) that an education

intervention may improve life outcomes through its impact on non-cognitive skills and

motivation even when IQ is unchanged. An alternative interpretation of the “educa-

tion effect” we observe is that people who self-select into obtaining high education are

inherently different from others in the way they assess situations with (strategic) risks

and potential synergy gains.

In order to measure the effect of payoff riskiness on the willingness to be in a team,

we compare the team choices between the Baseline and the Joint Decision treatments.

The results show that the response to shared decision rights is heterogeneous with

occupational categories. We find that managers’ and employees’ preferences for team

pay are not significantly affected by the inclusion of joint decision-making in the team

option. Entrepreneurs, on the other hand, are averse to team decision-making but if

and only if they expect their potential teammate to make different investment choices

than themselves. This result is in line with the findings of Masclet et al. (2009) who

demonstrate that self-employed people have a stronger preference than employees for

making decisions individually instead of in a team.2 While previous studies demon-

strate the non-pecuniary value attached to decision rights (e.g. Fehr et al., 2013), our

study provides no evidence for a positive willingness to pay for authority per se. This

2Relatedly, Reynolds and Curtin (2008) find, using survey data from the Panel Study of En-trepreneurial Dynamics II, that entrepreneurs are to a large extent motivated by a preference forautonomy.

128

difference in findings might be explained by the fact that in our setting the power to

make a decision was shared with, not delegated to the other party: even in case of joint

decision-making, each individual member retained a large influence over the choice that

was selected.

The remainder of this chapter is structured as follows. In Section 5.2 we provide an

overview of the related literature. Section 5.3 describes the experimental context and

the design, while Section 5.4 shows the descriptive statistics. Results are presented in

Section 5.5. In Section 5.6 we summarize and conclude.

5.2 Related literature

As we have briefly discussed in the Introduction, limited research has been conducted

to understand the factors underlying the preferences for individual and team incentives.

Table 5.1 provides an overview of the existing studies on sorting into teams. A few

insights emerge from the summary.

First, none of the reviewed papers analyze team settings that involve joint pro-

duction as well as joint decision-making. These two aspects have only been studied

separately so far. Moreover, none of the experimental studies on selection into team

remuneration schemes model team production with a scope for synergies: the major-

ity of studies impose an equal revenue sharing rule where individuals can only benefit

from joint work if they are teamed up with a more productive partner than themselves.

Such settings necessarily result in adverse selection into teams. Even studies that ac-

knowledge the possibility of team efficiency gains do so in a rather artificial way, by

adding an automatic mark-up over the sum of individual earnings. The production

function suggested in our study, by making synergies possible but not certain, is a

novel addition to the existing literature.3 Finally, most of the abovementioned papers

study a ‘traditional’ subject pool of university students, and even those that include

non-student participants have a rather limited number of observations.4 Our large and

diverse sample allows a more extensive analysis of the impact of experience, education

and occupational categories on team preferences.

In the following, we review in more detail the papers that are closest related to our

study. In particular, we focus on papers that analyze the choice between an individual

or a team piece rate scheme, and those that address preferences for individual versus

3A drawback of this design choice is that the size of the potential gains from teamwork are a prioriunknown, and we do not observe participants’ beliefs about the size of these gains. While this featurecomplicates the analysis we decided for it because in our opinion it models more accurately real lifeteam situations.

4In the paper by Masclet et al. (2009) the total number of participants is 144, and less than half ofthem are non-students. Cooper and Saral (2013) study a sample of 184 individuals, 44 of whom are(full- or part-time) self-employed.

129

Tab

le5.1

Overv

iew

of

the

Team

Decis

ion

-Makin

gL

itera

ture

Data

Sam

ple

Join

tR

eal

Tou

rnam

ent

a)

Join

tIn

tera

ctio

nb)

Au

tom

ati

cF

ree-

rid

ing

d)

Mu

tual

sou

rce

Pro

du

ctio

nE

ffort

dec

isio

na)

EA

c)

Con

sent

e)

Ham

ilto

net

al.

(2003)

Fie

ldF

act

ory

""

""

""

work

ers

Koch

eret

al.

(2006)

Lab

ora

tory

Stu

den

ts"

""

Tey

ssie

ret

al.

(2008)

Lab

ora

tory

Stu

den

ts"

""

Masc

let

etal.

(2009)

Lab

ora

tory

Stu

den

ts,

""

sala

ried

work

ers,

self

-em

plo

yed

Doh

men

etal.

(2011)

Lab

ora

tory

Stu

den

ts"

""

"

Darg

nie

s(2

012)

Lab

ora

tory

Stu

den

ts"

""

both

f)"

Coop

eran

dS

ara

l(2

013)

Lab

-in

-th

e-fi

eld

Stu

den

ts,

""

""

(on

lin

esu

rvey

)sa

lari

edw

ork

ers,

self

-em

plo

yed

Baker

an

dM

erti

ns

(2013)

Lab

ora

tory

Stu

den

ts"

""

"

Ku

hn

an

dV

ille

val

(2014)

Lab

ora

tory

Stu

den

ts"

"b

oth

both

both

Her

bst

etal.

(2015)

Lab

ora

tory

Stu

den

ts"

""

both

Th

isst

ud

yL

ab

-in

-th

e-fi

eld

Entr

epre

neu

rs,

""

""

"

(on

lin

esu

rvey

)M

an

ager

s,

Em

plo

yee

s

a)

Are

team

sco

mp

etin

gagain

stea

choth

er?

b)

Isth

ere

act

ual

inte

ract

ion

bet

wee

nth

ete

am

mem

ber

s?

c)

Isth

ere

an

auto

mati

ceffi

cien

cyad

vanta

ge

inco

rpora

ted

into

the

team

op

tion

?(T

he

alt

ern

ati

ve

isa

sim

ple

split

of

the

poole

djo

int

ou

tpu

t)

d)

Isth

ere

ap

oss

ibilit

yfo

rfr

ee-r

idin

gin

the

team

opti

on

?

e)

Does

team

form

ati

on

requ

ire

mu

tual

con

sent

from

all

team

mem

ber

s?

f)T

he

nota

tion

‘both

’in

dic

ate

sst

ud

ies

that

incl

ud

etr

eatm

ents

wit

han

dw

ith

ou

tth

egiv

enfe

atu

res

130

team decision-making. The paper by Hamilton et al. (2003) is among the first to

report about self-selection into team incentives. Although the main focus of their field

experiment is the effect of team composition on team performance, the authors also

provide evidence on the characteristics of those who are the first to join the teams once

the new incentive system is introduced. The task studied in this paper (i.e., sewing

together pieces of garment in a factory) is primarily based on joint production and

does not contain any specific elements of joint decision-making. The team setting in

this experiment provides a scope for synergies. Contrary to the theoretical predictions

on adverse selection, Hamilton et al. (2003) find that the high-productivity workers are

among the first to join the team.5

One of the first studies to look at sorting into teams in a controlled laboratory ex-

periment is the paper by Kocher et al. (2006) who study the preferences for individual

or group decision-making using a beauty-contest game. In their setting there are mon-

etary gains to be expected from teaming up. They find that 60% of individuals in their

sample prefer to work in a team. There is also some indication for adverse selection.

In order to measure the drivers of team preferences, the authors focused on the costs

and benefits from the joint decision-making process.6 Higher profits turned out to be

the most important reason for choosing the team option. They found that the endoge-

nously formed teams performed significantly better than the individuals that decided

to play alone. However, they also found that the individuals in both settings were very

satisfied with their choice, despite the lower earnings for the individual players. Hence,

the authors argue that individuals are on average willing to pay a price for autonomy

in decision-making.7

Kuhn and Villeval (2014) conduct a lab experiment to study gender differences in

the choice between individual and team-based incentives. The participants have to

perform a real-effort task for a piece rate. The experiment contains two different team

treatments, both with equal revenue sharing but neither with joint decision-making.

5There are two factors that could be confounding the self-selection into the teams in this fieldexperiment. First, as pointed out by Kocher et al. (2006), the team production framework is introducedover a period of three years, with the purpose that at the end of this period everybody is workingin teams. Thus, the self-selection is not entirely voluntary. It could be that the high ability workersare the first to realize that they do not have a choice in the end and thus decide to opt for the teamoption at an early stage in order to secure their job. Secondly, the team option in the field experimentinvolved slightly higher piece rates. Hamilton et al. (2003) discuss that this increased intensity ofincentives could also act as a confound.

6As the main drivers underlying this decision, the participants were offered a choice between twoexplanations: “I want to act as an individual (in a team), because I want (do not want) to decidealone” and “I want to act as an individual/in a team because in this way I can earn more” (Kocheret al., 2006, p.263).

7Masclet et al. (2009), studying decision-making in groups in a lottery-choice experiment, alsoaddress the question of selection. Their main result is that risk averse individuals then to bid moreto avoid joint decision-making. They also find that self-employed participants tend to have a higherwillingness to decide alone.

131

In their baseline team treatment the piece rate is the same as in the individual setting,

while in the efficiency advantage treatment (EA) there is an automatic 10% markup

on the team pay.8 As expected, due to the equal revenue sharing, the authors find

evidence for adverse selection both in the baseline and in the EA treatment for both

men and women. The increase in willingness to join the team between the baseline and

the EA treatment is much larger for men than for women, thus eliminating the gender

gap that was observed in the baseline treatment.9

Cooper and Saral (2013) conduct an online (lab-in-the-field) real-effort experiment

on a diverse sample in terms of socio-economic characteristics. The aim of their paper

is to test for differences in team preferences and free-riding tendencies between the dif-

ferent sub-groups. The set up of the experiment is such that self-regarding participants

should weakly prefer the team option over the individual option. The results show that

full-time entrepreneurs have a significantly stronger preference to be alone compared

to all other employment categories. The results provide some evidence for adverse

selection, i.e., individuals that performed better in the first part of the experiment bid

(slightly) more to be alone. However, they find no differences in free-riding among the

different occupational groups. Moreover, even though the experimental set-up did not

involve any joint decision-making, participants cite a fear for the loss of control or a

preference for self-reliance as a reason for choosing to play alone.

Finally, Baker and Mertins (2013) investigate how risk influences the sorting into

two different variable payment schemes, i.e., individual and team piece rates. The

authors look at two types of risks that are associated with team pay. First, there is the

risk of being matched with a low productivity co-worker, which would then decrease

pay off. On the other hand, being matched with another worker reduces the risk from

individual idiosyncratic shocks, such as luck, motivation or distraction, and thus may

positively influence the desire to be in a team. The results from this experiment show

that both types of risk indeed influence the selection into team piece rates: higher

idiosyncratic risk is shown to increase the probability of choosing the team pay option,

whereas the risk of being teamed up with a low ability worker reduces the probability

of sorting into team pay. Because the marginal effect size and the significance of the

strategic risk component is larger than the idiosyncratic risk component, the former

seems to be more important for the sorting decision.

8Each individual first performs the task twice: once alone and once the under the team paymentscheme. Then each individual can choose between a team or individual incentives three times. First,when matched with the performance of the partner from the individual round. Second, when matchedwith the performance of the partner from the team incentive round, and finally, matched with theperformance from the partner in that round, but only if both choose to be in a team (i.e., mutualconsent).

9Dohmen and Falk (2011) also consider the impact of gender on sorting into team incentivesschemes where the alternative is a flat wage option.

132

Other topics that have been studied more extensively in relation to self-selection

into teams are social preferences such as cooperation (Kosfeld and Siemens, 2009; Dur

and Sol, 2010), inequality aversion (Teyssier, 2008) and in-group favoritism (Herbst

et al., 2015). Given the design of the experiment and the focus of our paper, we do not

discuss them further here.

5.3 Context and Design

5.3.1 Context

To recruit participants for our study, we used the same channels as in the previous three

waves (see also Chapters 1-4). We contacted entrepreneurs with the help of “Synpact”,

a large organizer of entrepreneurship events. Managers were contacted via “De Baak”,

a highly reputed training institute for managers. For the recruitment of employees, we

collaborated with a Dutch market research agency. The invitations for the survey were

sent out by e-mail on March 24, 2015, followed by a reminder after 7 days. The survey

was open to respondents for 14 days. In total, close to 25,000 potential participants

received an invitation and 1,164 individuals (400 entrepreneurs, 155 managers and 609

employees) completed our survey.10

Many respondents in the sample have very high income. Therefore, the relatively

low earnings used in traditional laboratory experiments with student subjects were

unlikely to provide proper incentives in our case. Instead, we decided to use very high

payoffs and only pay out a subsample of our respondents. We therefore randomly se-

lected 20 prizewinners from among all participants who completed the survey.11 These

prizewinners received the total amount they had earned in the survey. The payment

structure was communicated very clearly to the participants at the beginning of our

survey. Furthermore, to foster trust, the drawing of prizewinners was performed by a

civil-law notary.

Prizewinners earned on average €330,58 with a minimum of €148 and a maximum

of €785. The ex post chance of being paid out was approximately 1/58, but this

was unknown to the participants (and ourselves) beforehand. However, to alleviate

the concern that participants might hold different beliefs about the likelihood of being

a prizewinner, we informed the participants at the beginning of the survey that the

chance of being paid out had been approximately 1/100 in earlier research waves.

10We conducted a pilot study among employees between March 11 and March 18, 2015 and received192 complete responses. The aim of this pilot was to test the length of the survey and to comparedifferent calibrations. Answers from the pilot study are not included in our main analysis.

11Such an approach is common in the literature and should produce similar results as when payingout all participants (see e.g. Gneezy and Rustichini (2000) and Laury (2006)).

133

Participants took on average 13 minutes to complete the survey that was designed

and pre-tested to take approximately 15 minutes, suggesting that they took the tasks

and choices seriously and read the instructions carefully. The original survey was

conducted in Dutch.

5.3.2 Design

Our experiment contained four incentivized parts: a production phase, an investment

phase, a choice between an individual or a team incentive scheme, and an evaluation

phase; followed by an unincentivized background questionnaire. To analyze the impact

of joint decision-making on preferences for team incentives, respondents were randomly

assigned to one of two treatment conditions: the Baseline and the Joint decision

treatment that differed from each other in whether the team option entailed a potential

compromise in a decision situation.12 The details of each part of the experimental

design and the two treatment conditions are described below.

Production phase (individual)

The first part of the experiment entailed a real-effort task. Participants were presented

with 10 puzzles from the Raven Advanced Progressive Matrices (see Raven et al., 2003)

and were asked to solve as many of them as possible within a time frame of 10 minutes.

This task required participants to complete puzzles consisting of three rows of three

figures where the bottom-right figure was missing (see Figure C1 in Appendix C of

Appendices Chapter 5). Raven test questions of varying difficulty are not uncommon

as a production task in experimental economics (see e.g. Herz et al., 2014). One of the

main benefits of these puzzles for our study is that it is hard to find the correct solution

on the Internet, which is a potential hazard when using an online survey. Moreover,

performance on this task provides a proxy for cognitive ability, as Raven matrices are

developed to serve as a “culture-free IQ test” (Herz et al., 2014, p.5).

To get participants acquainted with the set-up, we first provided them with an

example question without any time limit and the general instructions as in Raven

et al. (2003). Once participants indicated that they were ready, they were directed to

the next page containing all 10 Raven puzzles one below another, as well as a timer

showing the time remaining from the total 10 minutes. Participants were free to decide

12We performed a stratified randomization by gender and occupational category to ensure thatwe can analyze these subsamples separately. Our design contained also a within-subject element.Participants were asked to make a choice between individual and team remuneration schemes in twosubsequent scenarios. The scenarios differed from each other in the characteristics of the potentialteammate. In this paper we only discuss the first scenario where the teammate was randomly drawnfrom among all participants of the survey. Since the second scenario was only introduced after par-ticipants made their choice in the first one, their answers in the first scenario are unaffected.

134

the order in which they solved the puzzles and they could go back and forth between

the puzzles within the set time limit. After the allotted time was over, respondents

were automatically directed to the next page (it was also possible to move on to the

next page before the time was up). Participants faced individual piece rate incentives of

€40 per correctly solved puzzle. There was no money deducted for incorrect answers.

Participants were informed that they would receive feedback on the number of puzzles

they solved correctly at the very end of the survey.

Investment phase (individual)

Part 2 measured participants’ risk preferences. More specifically, following Gneezy and

Potters (1997) we asked participants what share of their Part 1 earnings (0-100%) they

were willing to invest in the following risky gamble:

• a 2/3 (67%) chance that you lose the money you invested

• a 1/3 (33%) chance that you win two and a half times the amount you invested

(on top of your investment).

Subjects made their investment choice using a slider, as shown in Figure C2 in Ap-

pendix C of Appendices Chapter 5. Gneezy and Potters (1997) let their subjects make

this investment choice several times in order to measure (myopic) loss aversion. In our

design participants only answer this question once, and we use this investment choice

as a proxy for their risk preferences. While this measure is not able to differentiate

between risk loving and risk neutral subjects (the expected return on the gamble is pos-

itive, so already a risk neutral subject should invest everything), it is a simple, quick

and easy-to-understand method for measuring different degrees of risk aversion.13 In

this paper, we calculate risk aversion as the share that the participant did not invest

in the risky bet (i.e. 100% share invested).

Team option

In Part 3, the key element of our experiment, participants were offered a choice between

an individual and a team renumeration scheme. In particular, respondents had to

decide whether they wanted to keep their individual piece rate earnings from Part

1 or whether they wanted to form a team with another survey participant instead.

Note that by asking participants to choose incentive schemes ex post for their Part

1 performance, we have eliminated the possibility of free-riding. In our opinion, the

13This method has been used by e.g. Dreber et al. (2011) and Charness and Gneezy (2012) tomeasure risk aversion. Also, since risk aversion is in general considered to be a stable personalitytrait, we assume our measurement of risk preferences to be unaffected by participants’ performancein Part 1. We revisit this assumption in Section 5.4.

135

Table 5.2 Summary of the Two Treatments

Team option Randomization by:

Joint Production Joint Investment

Treatment 1 (“No Loss of Control”) yes no gender + occupation

Treatment 2 (“Loss of Control”) yes yes gender + occupation

elimination of this potential confound makes the design cleaner and the results easier

to interpret. Moreover, while many theoretical papers on teamwork emphasize the

incentives to free ride, empirical studies often find little evidence for such practice in

real teams (Hamilton et al., 2003; Baker and Mertins, 2013; Herbst et al., 2015).

As we have mentioned before, respondents were randomly assigned to one of the

two treatment conditions that determined the content of the team option. Table 5.2

provides an overview of the design.

In the Baseline treatment, the team option only influenced the earnings from the real

effort task but not the investment decision from Part 2. The Joint Decision treatment,

on the other hand, entailed both joint production and a joint investment decision. In

this treatment the share invested in the risky bet was determined as the unweighted

average of the two teammates’ individual choices in Part 2. Hence, the team option

in this treatment entailed a potential compromise, with shared decision rights in the

investment choice and the possibility of the being moved away from one’s individual

utility-maximizing risk exposure.14

A distinguishing feature of our design is that we model team incentives by intro-

ducing conditional efficiency gains. More specifically, payoffs in the team option were

determined by the following production function with complementarities (the same in

both treatment conditions):

P teami = P team

j =∑10

n=1[max{Ini , Inj }],

where i and j denote the two team members, n = 1, .., k, .., 10 represents the ques-

tion numbers from Part 1 and Iki and Ikj are indicator variables showing whether partic-

ipant i and j solved question k correctly. This production function allows participants

to benefit from teaming up even with a less able partner provided that there are com-

plementarities between their outputs, i.e. that their correct answers do not completely

overlap.

14In our design, there is no scope for the ‘wisdom of the crowds’: since the decision is relatedto individual preferences, there is no ‘correct’ answer. Team decision-making thus does not helpthe members to achieve a more efficient outcome: individual choice in our setting is always weaklypreferred to the group choice. We do not model the bargaining process either: the compromise thatresults from the joint decision-making is always the unweighted average of the two members’ individualchoices.

136

Instead of showing participants the production function, we gave them the following

explanation for team earnings: “You get €40 for each puzzle that either you or your

teammate solved correctly in Part 1. Therefore your earnings in the team option are

always at least as high as in the individual option, and higher in case your teammate

solved more/different puzzles correctly than you did”. To ensure that the overall set-

up was clear to all participants in both treatments, we provided them with numerical

examples on how the team option could affect their earnings and investment decisions

(see Figures C3 and C4 in Appendix C in Appendices Chapter 5). After the example,

we elicited our main measure of interest, i.e. the willingness to pay for the team option.

Instead of a binary choice between the individual and the team option, we elicited

the willingness to pay for the team option by means of the BDM mechanism (Becker

et al., 1964) which allows us to obtain a continuous measure of team preferences in

an incentive-compatible manner. Specifically, we gave each respondent an endowment

of €50 that they could either keep or use to bid for the possibility to be in a team.

Participants were informed that the actual price of the team option would be randomly

drawn from the interval [€1, €50] at the notary after the survey was closed. Teams

were formed when both potential team members submitted a bid that was at least as

high as the actual price. Team formation thus required mutual consent. We instructed

participants that their teammate would be randomly drawn from the total sample

of survey respondents. To fix beliefs, we explicitly mentioned that the teammate is

equally likely to belong to either of the three occupational categories (entrepreneur,

manager or employee). Participants received no feedback at any point in the survey

about the identity, performance or bid of their potential teammate. Bids only had to

be paid in case subjects actually formed a team. Participants were reminded that it

was in their best interest to report their preferences truthfully.15

The team option in our setting did not involve an actual interaction between the

teammates. This design choice was mostly due to practical constraints imposed by our

data collection method: respondents of our online survey did not necessarily work on

the questionnaire at the same time, so real-time interaction or communication would

not have been possible. As the literature overview in Table 5.1 shows, it is common

to study sorting into team incentive schemes without allowing respondents to interact

with each other. Moreover, there are also examples from real life that resemble the way

we modeled teamwork. In the world of open-source software, developers often work

individually and remotely on issues, submit their solutions, and the best suggestion gets

15In this setting the team option weakly dominates the individual option in terms of expectedpayoffs, and bidding zero always ensured that a respondent is payed on the individual basis, so we didnot allow negative bids. In our analysis we account for the potential left- and right-censoring imposedby our elicitation technique that restricts the willingness to pay between the boundaries of €0 and€50 by estimating tobit models.

137

accepted.16 Similarly, in case of international scientific cooperations, the parties often

already have ideas or preliminary results at the time when they decide to cooperate

and pool their resources for a better final outcome.

Evaluation phase

In Part 4 of the survey we measured several factors that may influence the choice for the

team option. Based on the existing literature, we identified three main candidates: (1)

beliefs about own performance, (2) beliefs about the potential teammate’s performance,

and (3) value attached to decision rights (both instrumental and non-pecuniary). We

therefore asked participants to submit their guesses for the following three questions:

1. The number of puzzles they solved correctly in Part 1;

2. The number of puzzles a random other survey respondent solved correctly in Part 1;

3. The share a random other survey respondent chose to invest in the risky gamble in

Part 2.

All three evaluation questions were incentivized. The participants received €20 when

their answer to the first question was correct, and €20 when their answer to the second

question was correct. Finally, they could earn another €20 when their estimate in the

third question was less than five percentage points away from the true value.

Background questionnaire

After the four incentivized parts, the final part of the survey included a questionnaire

to gain insight into respondents’ choices and to collect some background characteris-

tics. First, we asked participants to explain their choice in Part 3 by selecting from a

list of possible explanations the option(s) they found most applicable:

• I believed the team option could increase my earnings.

• I did not want to take too much risk.

• I thought I solved more puzzles correctly than other participants.

• I have calculated the expected gains.

• I wanted to be responsible for my earnings and not depend on others.

• I don’t trust someone I don’t know.

16An example is the Linux kernel development process, see e.g.http://techblog.aasisvinayak.com/linux-kernel-development-process-how-it-works/.

138

• It was just a guess.

• I followed my intuition.

The list was based on the most common answers from the pilot survey (see footnote

(10)) where respondents answered an open-ended question explaining their bid for

the team option. Respondents then answered background questions specific to their

occupational categories.17 All respondents were asked to report the years of work

experience they had, and to select the income category they belonged to (with the

option to keep this information private).

5.4 Descriptive statistics

This section provides an overview of the data collected in our survey and the most

important variables used in the analysis. Table 5.3 shows the descriptive statistics for

the total sample (columns (1) and (2)) and for the two treatment groups separately

(columns (3)-(4) and (5)-(6)). Panel A describes the sample in terms of background

characteristics while Panel B introduces the survey outcomes.

Table 5.3 suggests that the randomization worked well. Comparing the two treat-

ment groups, we find no significant difference in terms of background characteristics

(confirmed by two-sample t-tests and Kolmogorov-Smirnov tests). Panel A further

shows that participants in our sample are on average 46.34 years old (the standard

deviation is 11.14) and that 40% of our respondents are female. Respondents are most

likely to have a college degree, but there is substantial variation in education levels.

Participants have on average 18.74 years of work experience. The modal (gross) income

category is €25,001 - 50,000 per annum (as a comparison, the gross modal income was

€33,500 in 2014 in The Netherlands, CPB, 2014).

Panel B of Table 5.3 shows the descriptive statistics for the survey outcomes. The

mean bid for the team option is 27.46 with a standard deviation of 16.08 (a more

detailed description of this outcome measure will be provided in the Results section).

Participants solved on average 5.01 out of 10 puzzles. Figure B1 in Appendix B of

Appendices Chapter 5 depicts the distribution of the number of correct answers per

participant and shows a large variance in puzzle performance. Reassuringly, only a very

17Entrepreneurs reported the legal structure of their companies, whether they were founders, thenumber of their employees and the share they owned in their companies. Managers reported whetherthey were general or project managers, whether they were the CEOs of their organization and thenumber of their direct reports. Using this information we could specify groups of entrepreneurs andmanagers according to various (stricter) definitions used in the literature (see also Koudstaal et al.(2015).

139

Table 5.3 Descriptive Statistics

Total sample Treatment 1 Treatment 2

(n = 1,164) (n = 588) (n = 576)

mean std. dev. mean std. dev. mean std. dev.

Panel A: Background

characteristics

Age 46.34 11.14 46.69 11.20 45.98 11.07

Female (dummy) 0.40 0.49 0.40 0.49 0.40 0.49

Education (highest degree):

- High school 11% 12% 11%

- Lower vocational degree 24% 22% 26%

- College education 41% 41% 40%

- University 24% 25% 23%

Work experience (years) 18.74 11.56 18.86 11.75 18.62 11.40

Income 1)

- ≤ €25,000 25% 26% 24%

- €25,001 - €50,000 40% 41% 40%

- €50,001 - €75,000 16% 16% 17%

- €75,001 - €125,000 12% 10% 13%

- €125,001 - €200,000 4% 4% 4%

- €200,001 - €300,000 1% 1% 1%

- €300,001 - €400,000 1% 1% 0%

- > €400,000 1% 1% 1%

Occupational category:

- Entrepreneur 35% 34% 34%

- Manager 13% 14% 13%

- Employee 52% 52% 53%

Panel B: Main variables

WTP for the team (0-50) €27.46 €16.08 €28.00 €16.32 €26.91 €15.83

Actual correct answers (0-10) 5.01 2.25 5.06 2.27 4.97 2.22

Est. own cor. answers (0-10) 5.73 1.98 5.67 2.03 5.78 1.92

Est. other cor. answers (0-10) 5.54 1.48 5.54 1.49 5.53 1.47

Risk aversion (0-100) 53.72 27.06 54.26 27.79 53.17 26.31

Est. other risk aversion (0-100) 52.89 19.29 52.75 19.23 53.04 19.36

1) For income, the number of observations drops to n = 779 (total sample), n = 381 (T1), and n = 398 (T2).

Note: */**/*** indicates a significant difference at the 10%/5%/1%-level between Treatment 1 and Treatment 2.

small fraction of respondents have zero correct answers, suggesting that participants

took the task seriously and exerted effort. The right hand side of Figure B1 presents

for each puzzle the share of respondents who solved the given question correctly. We

see that there was a substantial difference in difficulty between the puzzles: while some

140

questions were solved by close to 80% of the participants, this rate drops below 25%

for other questions. Column (1) of Table A1 in Appendix A of Appendices Chapter

5 demonstrates the importance of demographic variables for the performance on the

Raven puzzles. All else equal, older respondents were less successful in our task, while

more educated individuals solved significantly and substantially more puzzles. Even

after controlling for age, gender and education, entrepreneurs and managers performed

significantly better on the task (compared to employees).

Panel B of Table 5.3 further shows participants’ guesses for their own performance

as well as that of their potential teammate. The mean guess for the number of own

correct answers was 5.73, while respondents on average estimated that their potential

teammates solved 5.54 puzzles correctly. Figure B2 in Appendix B of Appendices

Chapter 5 sheds more light on the different aspects of individuals’ confidence. On

the left hand side of Figure B2, we see the distribution of overestimation (i.e. the

difference between own actual and guessed absolute performance (Moore and Healy,

2008) among participants: the modal answer is 1 and the distribution is shifted to

the right, suggesting that the majority of participants overestimate their performance.

The right hand side of Figure B2 compares participants’ guesses for their own and

their partner’s number of correct answers. Respondents are most likely to anticipate

no performance difference. Among those who do predict a gap the majority expects to

be better than a randomly chosen other survey participant.

Panel B of Table 5.3 also discusses outcomes related to risk preferences. We find

that the average percentage invested in the risky gamble is 46.28% (not reported),

which corresponds to a ‘risk aversion’ of 53.72 (with a standard deviation of 27.06)

on scale of 0 to 100. On average, participants’ guesses for the potential teammate’s

investment behavior is close to the respondents’ own choice, but its variance is lower

(the mean is 52.89 with a standard deviation of 19.29). The left hand side of Figure

B3 in Appendix B of Appendices Chapter 5 shows the distribution of the investment

choices and confirms that the majority of respondents are risk averse. The right hand

side of Figure B3 shows the estimated difference between own and partner’s guessed

investment choice. The majority of respondents believe there is no difference, and

hardly anyone expects a gap larger than 50 percentage points. Finally, column (2)

of Table 5.3 analyzes the relationship between demographics and risk aversion. It

confirms the findings of Eckel and Grossman (2008) and Charness and Gneezy (2012)

that women are more averse to risk than men. It further shows that both entrepreneurs

and managers take more risk than employees (consistent with the results of Chapter 2).

Age and education do not seem to play a role. Reassuringly, we find that risk aversion

is unaffected by actual or perceived performance in the puzzle task, supporting the

141

claim that our risk elicitation technique was not confounded by the outcomes in the

preceding task.

Before analyzing in more detail participants’ bids for the team option, we review the

statements participants selected as explanations for their team bids in the background

questionnaire at the end of the survey. The most popular answer was emphasizing the

monetary benefits of the team option (“I believed the team option could increase my

earnings”), selected by 39.3% of the respondents. Participants also frequently cited

following their intuition (38.9%) and trying to avoid taking too much risk (23.6%).

About a fifth of the respondents expressed a preference for payoff autonomy (“I wanted

to be responsible for my earnings and not depend on others”), while 14.1% based their

decision on the belief that they performed better than others. Reassuringly, only

about 10 percent of respondents indicated that they made their team bid at random

(“It was just a guess”). The statements about calculating the expected gains and

about not trusting strangers were the least frequently chosen, by about 5 percent of

the respondents each.18

5.5 Results

This section presents our results on team preferences. For the analysis of the different

factors that influence participants’ willingness to pay for the team option we only con-

sider the Baseline treatment (involving joint production but no joint decision-making).

This enables a more straightforward comparison of our findings with results of other

studies that focus exclusively on team production. To estimate the effect of joint

decision-making on preferences for team pay, we then compare the willingness to pay

for the team option between the Baseline and the Joint Decision treatments.

5.5.1 Determinants of team choice in the Baseline treatment

Figure 5.1 gives an overview of participants’ willingness to pay for the team option

in the Baseline treatment. The mean bid in this treatment group is 28.00 with a

standard deviation of 16.32. Little more than 10% of the participants chose to bid zero

and 18% was willing to pay the maximum possible amount, €50 for the team option.

Even though bids were reported using a slider, respondents were still inclined to choose

“round” numbers, i.e. multiples of five and especially ten. We see no indication for

participants being biased by the slider’s default setting: €25, the default option is only

the fifth most common answer.

18The above shares do not add up to 100% since participants were allowed to select more than oneexplanation. The shares reported in the text are calculated based on the responses of participants inthe Baseline treatment. Answers are very similar in the Joint Decision condition as well.

142

Figure 5.1. Distribution of bids for the team option in the Baseline

Table 5.4 displays pairwise correlation coefficients between participants’ willingness

to pay for the team option and variables we suspect might influence the bids: actual

and guessed number of correct answers, the partner’s estimated performance, risk aver-

sion, gender, age, education and occupation categories.19 We find that only a few of

these factors are significantly correlated with preferences for team pay: beliefs about

the potential teammate’s performance are positively associated with team bids, while

higher risk aversion corresponds to a lower willingness to pay. Among the demographic

characteristics, age and education show up significant. At the same time, many of these

covariates are highly significantly correlated with each other. We therefore continue to

analyze each of these factors to estimate their impact on team preferences in isolation.

We first focus on the effect of task performance on sorting into the team pay option.

Table 5.4 suggests that neither actual nor guessed performance is significantly related

to the team bids. This result is puzzling given that low-performing participants have

much more to gain from the team option. Figure 5.2 illustrates this point by plotting

participants’ actual and optimal bids against their true (Panel A) and guessed (Panel B)

number of correct answers. The optimal bids represent the ex ante expected gains from

the team option for a perfectly informed participant.20 Unsurprisingly, the optimal bids

are highest for low-performing respondents and decrease with the number of correct

19In our survey we have also measured participants’ work experience (in years). This measure ishighly correlated with age, and is missing for 34 respondents, so we decided not to include it inour analysis and focus only on age. All the results presented in this section are robust to includingexperience instead of age in the analyses.

20Optimal bids account for synergy possibilities: expected gains are calculated by assessing foreach puzzle and for each potential teammate whether the given question was solved correctly by theteammate but not by the participant herself. Optimal bids are constructed assuming everyone whobids €25 or higher is a potential teammate. Weights are used to ensure that the likelihood of beingteamed up with a manager, entrepreneur or employee is the same.

143

Tab

le5.4

Corr

ela

tion

Matr

ix

WT

PP

uzz

les

Pu

zzle

sP

art

ner

’sp

uzz

les

Ris

kF

emale

Age

Ed

uca

tion

Entr

epre

neu

rM

an

ager

for

team

corr

ect

(act

ual)

corr

ect

(gu

ess)

corr

ect

(gu

ess)

Aver

sion

WT

Pfo

rte

am

-

Pu

zzle

sco

rrec

t(a

ctu

al)

-0.0

14-

Pu

zzle

sco

rrec

t(g

ues

s)-0

.022

0.61

4**

*-

Part

ner

’sp

uzz

les

0.16

2**

*0.

204

***

0.44

4***

-

corr

ect

(gu

ess)

Ris

kA

ver

sion

-0.2

27**

*-0

.040

-0.1

02**

-0.0

56

-

Fem

ale

-0.0

54-0

.010

-0.1

11***

0.0

21

0.1

56

***

-

Age

0.07

9*

-0.1

56**

*-0

.050

-0.0

02

-0.0

42

-0.1

01

**

-

Ed

uca

tion

0.13

2**

*0.

332

***

0.20

0***

0.0

18

-0.0

82

**

-0.0

21

0.0

32

-

Entr

epre

neu

r0.

065

0.08

3**

0.15

3***

0.0

32

-0.1

34

***

-0.2

17

***

0.2

76

***

0.1

73

***

-

Man

ager

0.02

30.

205

***

0.12

2***

0.0

50

-0.0

67

-0.0

16

0.0

35

0.2

55

***

-0.2

85

***

-

Sig

nifi

can

ceat

the

10%

level

isd

enot

edby

*,5%

by

**,

and

1%

by

***.

144

Figure 5.2. Optimal versus actual team bids by actual (left)and guessed (right) number of correct puzzles

puzzles. Actual bids, however, do not follow the same pattern: participants with a

low (guessed) score do not pay more for the chance to be teamed up. As a result, we

find no evidence for adverse selection in our setting: the sorting decision does not seem

related to performance.

To separate the impact of true performance and confidence, we turn to a regression

framework. Table 5.5 reports results from tobit models explaining the willingness to

pay for the team option in the Baseline treatment. Column (1) confirms the finding

that the actual number of correct answers does not affect the decision to join a team.

It also shows that given true performance, confidence in ability matters: participants

who guess they solved more puzzles correctly bid significantly less for the team op-

tion, ceteris paribus. Relative performance beliefs are also important determinants

of the team choice: those respondents who expect their potential partner to answer

more questions correctly bid more. A unit increase in the partner’s estimated score

is associated with an approximately €3 increase in the predicted willingness to pay,

corresponding to about one fifth of a standard deviation.

Previous studies identified risk aversion as an important factor in the sorting deci-

sion into teams. According to Table 5.4, risk preferences are strongly correlated with

the team bids also in our setting. Column (2) of Table 5.5 shows that this result is ro-

bust to controlling for performance and confidence: more risk averse participants have

a significantly lower willingness to pay for the team option. The effect is sizable: all

else equal, participants who invest zero in the risky bet are predicted to bid on average

€18.6 less for the team than those who choose to invest 100% of their earnings in the

risky lottery in Part 2.21 Our results suggest that participants perceive the team op-

21Could the result that risk preferences are important determinants of team choice be merely a “sideeffect” of the particular elicitation method that we used to measure team preferences? As discussedin Section 5.3, participants received €50 which they could use to bid for the team option. The designthus involved a choice between keeping the “safe endowment” and investing (some share of) it in the

145

Table 5.5 Tobit Regressions on the Willingness to Pay for the Team (Baseline)

(1) (2) (3) (4)

Dep. variable: WTP WTP WTP WTP

for team for team for team for team

Puzzles correct (actual) 0.335 0.412 0.055 0.091

[0.65] [0.83] [0.10] [0.17]

Puzzles correct (guess) -1.427∗∗ -1.737∗∗∗ -1.835∗∗∗ -1.889∗∗∗

[-2.26] [-2.82] [-2.98] [-3.05]

Partner’s puzzles 3.014∗∗∗ 3.023∗∗∗ 3.190∗∗∗ 3.097∗∗∗

correct (guess) [4.29] [4.43] [4.69] [4.54]

Risk Aversion -0.186∗∗∗ -0.173∗∗∗ -0.176∗∗∗

[-5.64] [-5.23] [-5.26]

Female -0.867 -0.847

[-0.47] [-0.44]

Age 0.082 0.092

[1.02] [1.06]

Education 3.175∗∗∗ 3.319∗∗∗

[3.30] [3.28]

Entrepreneur 1.349

[0.56]

Manager -0.090

[-0.03]

Income NO NO NO YES

Obs. 588 588 588 588

Log lik. -2,061.9 -2,046.2 -2,039.7 -2,037.1

ENT=MAN 1) - - - 0.62

1) This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

Significance at the 10% level is denoted by *, 5% by **, and 1% by ***, with t-statistics reported inparentheses. Standard errors are robust.

risky team option. In a pilot study we compared this calibration with a different elicitation techniquewhere respondents received no endowment but had to use their survey earnings to bid for the teamoption. There was no significant difference in bids between the two calibrations in the pilot survey.Moreover, risk aversion was found to be an important predictor of team choice in the “no-endowment”version as well. We thus believe the result that risk aversion matters substantially for sorting intoteams is a general finding and is not driven by the specifics of our design.

146

tion as more risky than the individual piece rate. This is in line with the conclusions

of Baker and Mertins (2013) who find that the additional strategic risk associated with

team pay matters more for the sorting decision than the reduction in the idiosyncratic

components of risk.

One of the goals of our study is to explore how personal characteristics such as

age, gender and level of education affect preferences to join teams. We therefore add

these variables to our model explaining team choices (controlling for performance,

beliefs and risk preferences). The resulting estimates are presented in column (3) of

Table 5.5. Age does not seem to affect team bids: the marginally significant negative

correlation observed in Table 5.4 disappears when we include other covariates in the

regression.

Ex ante predictions for the impact of gender are ambiguous. On the one hand,

women are less confident: even though there is no gender difference in actual task

performance, female participants’ guesses for the number of puzzles they solved cor-

rectly is lower (see Table 5.4). This lower confidence should increase their bids for the

team option. On the other hand, their higher risk aversion (see Table A1, Appendix

A of Appendices Chapter 5) is predicted to decrease their willingness to participate

in teams. Indeed, the raw correlation we observe between gender and team bids in

Table 5.4 is insignificant. This is in line with the findings of Kuhn and Villeval (2014)

who show no gender gap in sorting into teams in the presence of efficiency advantages

associated with the team option. When we control for relative performance beliefs

and risk aversion, we also find no gender difference in team bids (column (3) of Table

5.5), suggesting that in our sample female respondents do not differ from men in their

inherent ‘taste’ for team remuneration schemes.

Existing research on sorting into teams has not analyzed the influence of educa-

tion, a factor we find has a substantial influence on sorting into teams. Column (3)

of Table 5.5 shows that above and beyond its impact on task performance, education

has a significant, large and positive effect on the willingness to pay for the team op-

tion.22 Column (4) confirms that this effect is not driven by the correlation between

education and occupational categories: including indicator variables for entrepreneurs

and managers, the estimated coefficient for education remains virtually unchanged.

It is interesting to note that entrepreneurs and managers do not seem to differ from

employees in their team preferences once we control for differences in ability, beliefs

and demographic characteristics.23 Self-reported explanations do not indicate a larger

22Results are very similar when instead of estimating a linear effect of education we include indicatorvariables for the different levels of education.

23We find no difference in the willingness to pay for the team option when, instead of the broadoccupational categories, we make a distinction between founders and entrepreneurs who have takenover/inherited their companies, or ‘people managers’ and project managers.

147

desire for control among the self-employed: entrepreneurs in our sample were not any

more likely than non-entrepreneurs to select the statement “I wanted to be responsible

for my earnings and not depend on others” in the background questionnaire. This is

in contrast with the findings of Cooper and Saral (2013) whose results indicate that

self-employed have a stronger preference to work alone than others. Column (4) fur-

ther shows that the positive association between education and team bids is robust to

controlling for income differences between the high- and low-educated.24

Lastly, we explore in more detail the role that education plays in shaping preferences

for teams. To gain some insight into the respondents’ decision-making process, we

analyze in more detail the explanations they gave for their team bids. Figure 5.3

compares the frequencies with which respondents with different levels of education

chose the explanations related to risks (“I did not want to take too much risk”) or

potential gains (“I believed the team option could increase my earnings”). The figure

Figure 5.3. Self-reported explanations for team bids by education levels

indicates a pattern: the higher a respondent’s level of education, the less likely she is

to mention the former and the more likely she is to select the latter explanation. (The

differences in frequencies are (marginally) significant when we compare respondents

with either a college or university degree to those with primary/secondary or voca-

tional education.) This finding suggests that higher educated participants are more

likely to base their choice on the possible gains from teamwork, whereas the lower ed-

ucated individuals tend to focus more on risk considerations. Results in Table A2 in

Appendix A of Appendices Chapter 5 support this interpretation by showing that the

24Note that only 381 participants in the Baseline treatment reported their income category. In theregression presented in Table 5.5, we treated non-respondents as the omitted category. Our resultsare similar when we focus only on the subsample who disclosed their income.

148

impact of risk aversion on team bids is heterogeneous with education. While risk pref-

erences have a large impact on the willingness to pay for teams for respondents with

primary/secondary education, they seem not to affect the choices of participants with

a university degree at all.25 As a result, participants with a lower level of education -

who on average perform worse on the task than those with more education - miss out

on the sizable potential benefits that the team option entails for them.

5.5.2 Comparison of team choice between the treatments

In this section we analyze the effect of joint decision-making on respondents’ willing-

ness to pay for the team option by contrasting the bids between the two treatment

conditions: the Baseline where the team option only entails joint production and the

Joint Decision treatment where the team option affects both the earnings from the

real-effort task and the investment decision. Figure 5.4 depicts the distribution of bids

in both treatments. In the Joint Decision treatment, respondents bid on average 26.91

(st. dev. 15.83) which is only slightly lower than the mean bid of 28.00 in the Base-

line. A simple comparison by means of a t-test shows no difference between the two

treatments (p-value= 0.248). A Kolmogorov-Smirnov test does not reject the equality

of the two distributions, either (p-value= 0.187).

Figure 5.4. Comparison of WTP betwen the two treatments

25To exclude the possibility that our findings are driven by lower educated people being confusedabout the specifics of the team option or the bidding procedure, we re-estimate the model presented incolumn (1) of Table B2 (Appendix A of Appendices Chapter 5) on the subsample of participants in theBaseline treatment who are most likely to have understood the scheme. In particular, we omit thoseparticipants who selected “It was just a guess” as an explanation for their bids at the end of the survey(we have 79 such participants in the Baseline treatment). In the resulting subsample we replicate thefinding that the impact of risk is heterogeneous with respect to education. We therefore argue thatmisunderstanding is unlikely to drive the different choices of low- and high-educated participants.

149

Table 5.6 Tobit Regressions on the Willingness to Pay (Both Treatments)

(1) (2) (3) (4) (5)

Dep. variable: WTP WTP WTP WTP WTP

for team for team for team for team for team

Sample: Full Full Entrepreneurs Managers Employees

Puzzles correct (actual) 0.547 0.515 0.656 1.855 0.327

[1.55] [1.46] [1.03] [1.44] [0.75]

Puzzles correct (guess) -2.244∗∗∗ -2.152∗∗∗ -0.371 -4.223∗∗∗ -2.649∗∗∗

[-5.04] [-4.84] [-0.43] [-2.74] [-4.92]

Partner’s puzzles 2.635∗∗∗ 2.590∗∗∗ 0.987 2.960 3.268∗∗∗

correct (guess) [5.39] [5.29] [1.02] [1.64] [5.66]

Risk Aversion -0.189∗∗∗ -0.191∗∗∗ -0.233∗∗∗ -0.004 -0.222∗∗∗

[-7.87] [-7.97] [-5.33] [-0.06] [-6.97]

Female -1.571 -1.650 -0.687 0.673 -2.594

[-1.20] [-1.26] [-0.26] [0.17] [-1.64]

Age 0.089 0.086 0.263∗∗ 0.187 0.019

[1.52] [1.47] [2.19] [0.82] [0.28]

Education 1.240∗ 1.298∗ 1.271 0.975 1.002

[1.78] [1.87] [1.01] [0.37] [1.18]

Entrepreneur 0.461 0.519

[0.30] [0.34]

Manager -0.753 -0.711

[-0.36] [-0.35]

Joint Decision Treatment -1.289 1.932 3.944 -1.502 1.508

[-1.05] [1.07] [1.16] [-0.27] [0.67]

Abs. Diff. RA 0.028 0.086 -0.075 -0.008

[0.60] [1.01] [-0.55] [-0.14]

Joint Decision Treatment -0.171∗∗ -0.304∗∗ -0.025 -0.110

x Abs. Diff. RA [-2.48] [-2.44] [-0.13] [-1.22]

Constant 27.50∗∗∗ 26.90∗∗∗ 16.91∗ 17.87 32.97∗∗∗

[6.10] [5.76] [1.66] [1.00] [5.90]

Obs. 1,163 1,163 400 155 608

Log lik. -4,063.9 -4,059.8 -1,361.2 -531.4 -2,148.0

ENT=MAN 1) 0.55 0.54 - - -

1) This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

150

Table 5.6 analyzes the impact of joint choice on team bids in a regression framework.

Column (1) shows that even after controlling for potential differences in performance,

beliefs, preferences and personal characteristics between the two groups, the bids for

teams are not significantly different in the Joint Decision treatment. The fact that the

inclusion of covariates does not change our results is unsurprising given that partic-

ipants were randomly assigned to the treatment conditions and the two groups were

balanced in terms of observables, as shown in Table 5.3.

Having found no overall effect of the treatment conditions, we check whether the

response to joint decision-making depends on respondents’ guesses about the instru-

mental value of the right to decide. In particular, we test whether those respondents

who predict a large gap between their own and their teammate’s investment choice

(and consequently expect the joint choice to be far away from their own) bid less in the

Joint Decision treatment. Column (2) of Table 5.6 shows that this is indeed the case:

the greater the predicted (absolute) difference in risk taking between the teammates,

the less appealing the team option in the Joint Decision treatment is compared to

the Baseline. Respondents in our sample thus have a clear preference for keeping the

decision rights in cases where they believe the team’s choice would be different than

their individual optimum.

Inspired by studies claiming that entrepreneurs have a particularly strong need for

power, authority and control, we analyze whether including joint decision-making in the

team option has a particularly strong negative impact for entrepreneurs. In columns

(4)-(6) of Table 5.6 we assess the impact of the Joint Decision treatment condition

and its interaction with the predicted difference between investment choices separately

for the three occupational categories. We find that the result of column (2) is driven

entirely by the entrepreneurs in our sample: employees and managers, irrespective of

their beliefs about their partner’s risk preferences, do not differ in their willingness

to pay for the team option between the two treatments. Entrepreneurs, on the other

hand, respond to a predicted gap between their own and their teammate’s investment

choice by placing significantly lower bids in the Joint Decision treatment than in the

Baseline.

5.6 Summary and conclusion

Our study reports results from a large-scale lab-in-the-field experiment analyzing the

sorting decision into teams. We replicate several findings from related lab studies,

using a larger and more diverse subject pool and a different modeling of teamwork.

We confirm that absolute and relative confidence are important determinants of the

willingness to pay for the team option (Kuhn and Villeval, 2014; Herbst et al., 2015)

151

also in a setting where team production is not implemented as an equal split of the

pooled output but entails potential synergy gains. We further confirm that strategic

risk resulting from uncertainties about the teammate’s performance has a large impact

on the selection into teams (Baker and Mertins, 2013) even when the possibility of

free riding is eliminated by design. Moreover, we show that the result of no gender

gap in team choices in the presence of efficiency gains (Kuhn and Villeval, 2014) per-

sists also in a setting where such gains are not automatic but are conditional on the

complementarities between the teammates.

Our results highlight that selection into teams is related to participants’ level of

education, and that entrepreneurs respond differently to joint decision-making than

managers and employees. These findings would not have been possible to obtain with

a sample of college students typically participating in laboratory experiments. Thus,

our paper demonstrates the added value of studying a ‘non-traditional’ sample, i.e.

participants with diverse socio-economic backgrounds. In this aspect our study is

similar to the work of e.g. Harbaugh et al. (2002) who analyze risk preferences in

different age groups or Kocher et al. (2006) who focus on the relationship between age

and trust.

A novelty of our paper is the finding that education is an important predictor

for individuals’ participation decision in teams. Controlling for differences in task

performance (which can be viewed as a proxy for IQ), confidence and risk preferences,

higher education is associated with a greater willingness to pay for the team option. We

find suggestive evidence that this heterogeneity is explained by differences in evaluating

the team option: while participants with higher levels of education tend to primarily

consider the potential gains from team pay, lower-educated respondents focus more on

the risks associated with teams. As a consequence, lower educated people miss out on

the sizable efficiency gains that the team option entails. It is important to note that

education does not affect risk preferences: educational attainment is uncorrelated with

the share of earnings participants invest in the risky bet. It is the extent to which risk

preferences matter for the selection into teams that is affected by education: the same

level of risk aversion leads to a greater reduction in bids for the team option for the

low- than the high-educated respondents in our sample. These inferences are consistent

with the conclusions from the Perry Preschool program where education (in the form

of an early childhood intervention) did not have a lasting effect on IQ but it still lead

to improved life outcomes through its impact on non-cognitive skills (Heckman et al.,

2013). Our conjectures also bring into mind the results of Choi et al. (2014) who find

higher educated people make ‘higher quality’, i.e. more rational decisions than those

with lower levels of education. We wish to emphasize that our data does not allow us

tell whether education changes individuals’ decision-making process or whether those

152

who obtain higher education are innately different from others in the way they evaluate

situations involving strategic risk and synergy gains. We find this question to represent

an interesting avenue for future research.

Our study also assesses the influence of shared decision rights on the self-selection

into team incentive schemes. We show that respondents are heterogeneous in their

response to a potential compromise in decision-making. In particular, we find that

entrepreneurs are averse to joint decision-making when they predict that it moves

them away from their individual optimal choice. This results supports claims that

entrepreneurs have a greater desire for control than non-entrepreneurs (Benz and Frey,

2008; Reynolds and Curtin, 2008; Masclet et al., 2009). We do not reproduce the

finding that people attach a non-pecuniary value to decision rights per se (Fehr et al.,

2013; Bartling et al., 2014): even among entrepreneurs, shared decisions only decrease

the willingness to join teams in cases when the partner’s choice is expected to deviate

substantially from one’s own optimum. Those who predict no difference between their

own and their partner’s investment decision bid the same for the team option in both

treatments. We speculate that this is due to decisions being shared, not delegated. In

our setting, instead of an obvious loss of power, a compromise is implemented, and the

resulting team choice is still strongly influenced by each member’s individual choice.

153

Appendices Chapter 5

Appendix A. Additional tables.

Appendix B. Additional graphs.

Appendix C. Survey screenshots.

154

Appendix A1. Regressions on Task Performance and Risk Aversion

(1) (2)

Dep. variable: Puzzles Risk

correct (actual) Aversion

Sample: Full Full

Female 0.029 5.420∗∗∗

[0.22] [3.26]

Age -0.035∗∗∗ -0.039

[-5.81] [-0.56]

Education 0.496∗∗∗ -1.207

[7.24] [-1.42]

Entrepreneur 0.704∗∗∗ -9.558∗∗∗

[4.77] [-5.16]

Manager 1.209∗∗∗ -9.594∗∗∗

[6.39] [-3.57]

Puzzles correct (actual) 0.441

[0.95]

Puzzles correct (guess) -0.772

[-1.20]

Partner’s puzzles correct (guess) -0.508

[-0.76]

Constant 4.834∗∗∗ 66.31∗∗∗

[13.71] [12.86]

Obs. 1,163 1,163

Log lik. -2,512.4 -5,445.7

ENT=MAN 1) < 0.01 *** 0.99

1) This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

155

Appendix A2. Tobit Regressions on the Willingness to Pay (Baseline)

(1) (2) (3) (4) (5)

Dep. variable: WTP WTP WTP WTP WTP

for team for team for team for team for team

Education: All Primary/ Vocational College University

Secondary

Puzzles correct (actual) 0.081 1.101 -0.261 -0.631 1.175

[0.16] [0.90] [-0.35] [-0.79] [0.68]

Puzzles correct (guess) -1.904∗∗∗ -0.798 -0.459 -1.647∗ -4.504∗∗∗

[-3.10] [-0.48] [-0.46] [-1.73] [-2.69]

Partner’s puzzles 3.269∗∗∗ 4.838∗∗∗ 2.104∗ 3.376∗∗∗ 2.838

correct (guess) [4.83] [3.29] [1.88] [3.15] [1.52]

Risk Aversion -0.446∗∗∗ -0.274∗∗∗ -0.217∗∗∗ -0.211∗∗∗ -0.020

[-4.42] [-3.05] [-4.07] [-3.82] [-0.26]

Female -0.553 -5.311 -2.467 -2.634 7.818∗

[-0.30] [-1.25] [-0.79] [-0.84] [1.71]

Age 0.075 -0.161 0.188 0.091 0.123

[0.89] [-0.78] [1.51] [0.67] [0.53]

Entrepreneur 0.426 1.699 -3.331 -0.130 3.287

[0.20] [0.29] [-0.92] [-0.04] [0.55]

Manager -1.307 -1.478 -7.278 0.096 -2.935

[-0.45] [-0.13] [-0.80] [0.02] [-0.47]

Education -2.069

[-0.99]

Education x RA 0.095∗∗∗

[2.87]

Constant 33.77∗∗∗ 22.69∗ 25.08∗∗∗ 30.50∗∗∗ 31.09∗

[4.10] [1.73] [2.72] [3.18] [1.93]

Obs. 588 73 128 242 145

Log lik. -2,035.4 -251.9 -461.4 -834.0 -460.3

ENT=MAN 1) 0.55 - - - -

1) This reports the p-value of the Wald test ‘Entrepreneur’ = ‘Manager’.

156

Appendix B. Additional tables and graphs

Figure B1: Distribution of correct answers per participant (L) and per puzzle (R)

Figure B2: Absolute (L) and relative (R) performance guesses

Figure B3: Investment in the risky gamble (L) and difference between teammates in

shares invested (R)

157

Appendix C. Survey screenshots

Figure C1: Example of a Raven puzzle

Figure C2: Measure of Risk Aversion

158

Figure C3: The Baseline Team Option

Figure C4: The Joint Decision-Making Team Option

159

160

Chapter 6

Summary

This dissertation reports the results of four lab-in-the-field experiments which examine

the differences between entrepreneurs and other occupational groups. Each of these

experiments deals with a different element of decision-making, being either: risk &

uncertainty (Chapter 2), optimism & overconfidence (Chapter 3), intuitive versus con-

templative decision-making (Chapter 4), and preferences for individual versus team

incentives (Chapter 5). In this chapter, the results of these experiments are summa-

rized.

Chapter 2 is based on the paper “Risk, Uncertainy and Entrepreneurship: Evidence

from a Lab-in-the-Field Experiment”, co-authored with Mirjam van Praag and Ran-

dolph Sloof and forthcoming in Management Science. The results of this chapter first

confirm that entrepreneurs indeed perceive themselves as less risk averse than man-

agers and employees, in line with common wisdom. However, when using experimental

incentivized measures, the differences are subtler. We find that entrepreneurs are only

unique in their lower degree of loss aversion, but not in their risk or ambiguity aversion.

These results again show up in a replication study using a predominantly new sample,

thus further affirming our initial conclusions. Furthermore, in an effort to reconcile the

contrasting empirical evidence on subjective and objective measures of risk, we show

that perceived risk attitude is not only correlated to risk aversion, but also to loss and

ambiguity aversion. Apparently economists use a more narrow definition of risk than

the forces that drive the behavior of (risk taking) entrepreneurs.

Two other important elements of decision-making are optimism and overconfidence.

In Chapter 3, which is based on the paper “Are Entrepreneurs More Optimistic and

Overconfident than Managers and Employees?”, again co-authored with Mirjam van

Praag and Randolph Sloof, we further explore this topic. First examining the existing

theoretical and empirical literature, it shows that entrepreneurs are believed and found

to be more optimistic and overconfident than others, but so are top managers. How-

ever, it is relatively unknown how these two groups actually compare to each other.

161

We therefore aim to fill this gap by means of studying optimism and overconfidence

among entrepreneurs, managers and employees. The results of this study indicate

that entrepreneurs are more optimistic than others in their dispositional optimism and

their attributional style when bad events occur. Hence, entrepreneurs hold more op-

timistic views of the future and deal more positively with past bad events than all

others. Furthermore, the two incentivized measures of overestimation yield that both

entrepreneurs and managers are more prone to overconfidence than employees. Only

if we zoom in on start-up entrepreneurs (i.e. firm age < 3 years) do we find that en-

trepreneurs also overestimate themselves more than managers. But in all other cases,

they behave similarly. Finally, exploiting the various success measures in our survey

(e.g. incorporated business owner or CEO), we find that successful entrepreneurs and

managers are more optimistic but not more overconfident than their less successful

peers. Overall, it seems that optimism and overconfidence are traits of entrepreneurs

as well as of (successful) decision-makers in general.

The third element of decision-making that we study is decision-making style, which

is further discussed in Chapter 4 based on the paper “Are Entrepreneurs More Intu-

itive Than Managers and Employees? Evidence From a Lab-in-the-Field Experiment”,

which is also co-authored with Mirjam van Praag and Randolph Sloof. In this chapter,

we test how how contemplative entrepreneurs, managers and employees are using two

well-known subjective measures taken from the psychology literature and two objective

measures based on response times and the number of contemplative choices made. The

results show that entrepreneurs make less contemplative choices than managers, but

not than employees. Surprisingly, however, we do not find that the three occupational

groups differ in their response times. One conjecture for this inconsistency might be

that the effect of response time is heteregeneous and we indeed find that entrepreneurs

and managers are relatively more contemplative per unit of time than employees. The

same regressions also reveal that entrepreneurs are in fact more intuitive than others

whenever their response times are low. The two subjective measures show a similar

pattern; entrepreneurs score highest on Faith in Intuition, while managers score highest

on Need for Cognition. We therefore conclude that entrepreneurs mostly differ in their

decision-making styles from managers, and somewhat less so from employees.

The final topic of this dissertation is the preference for individual vis-a-vis team

incentives. This is further elaborated on in Chapter 5, which is based on the paper

“Risks, Gains and Autonomy: An Experimental Analysis of Preferences for Joining

Teams”, co-authored with Laura Rosendahl Huber and Eszter Czibor. In this study, we

rely on an incentivized lab-in-the-field experiment with two treatments: one “baseline”

treatment where only payoff autonomy is at stake and one “joint investment” treatment

where both payoff autonomy and payoff riskiness are at stake. The results of the

162

former treatment indicate that education and optimistic beliefs about the teammate’s

performance have a positive impact on the willingness to pay for team incentives, while

risk aversion and overconfidence have a negative impact. Occupational category, ability

and gender on the other hand do not signicantly affect the willingness to pay. Hence,

we do not find that entrepreneurs in general pay more or less than other occupational

groups. Furthermore, when including the data of the “joint investment” treatment,

it surprisingly shows that the loss in decision rights does not necessarily lead to a

lower willingness to pay for team incentives. Further analysis however reveals that

entrepreneurs are the exception to this rule, but if and only if they perceive their

potential teammate’s risk appetite as too different from their own. Overall, the latter

might thus help further explaining (i) in which specific cases entrepreneurs prefer to

work alone and (ii) why venture teams often have homophilic risk preferences.

163

164

Bibliography

Abdellaoui, M. 2000. Parameter-free elicitation of utility and probability weighting

functions. Management Science 46(11) pp. 1497–1512.

Abdellaoui, M., H. Bleichrodt, O. l’Haridon. 2008. A tractable method to measure

utility and loss aversion under prospect theory. Journal of Risk and Uncertainty

36(3) pp. 245–266.

Abdellaoui, M., H. Bleichrodt, Corina Paraschiv. 2007. Loss aversion under prospect

theory: A parameter-free measurement. Management Science 53(10) pp. 1659–1674.

Abdelsamad, M.H., A.T. Kindling. 1978. Why small businesses fail. SAM Advanced

Management Journal 43(2) pp. 24–32.

Allinson, C.W., E. Chell, J. Hayes. 2000. Intuition and entrepreneurial behavior. Eu-

ropean Journal of Work and Organizational Psychology 9 pp. 31–43.

Allinson, C.W., J. Hayes. 1996. The cognitive style index: A measure of intuition-

analysis for organizational research. Journal of Management Studies 33(1) pp. 119–

135.

Alos-Ferrer, C., S. Hugelschafer. 2012. Faith in intuition and behavioral biases. Journal

of Economic Behavior & Organization 84(1) pp. 182–192.

Andersen, S., A. Di Girolamo, G.W. Harrison, M.I. Lau. 2014. Risk and time pref-

erences of entrepreneurs: evidence from a Danish field experiment. Theory and

Decision 77 pp. 341–357.

Arabsheibani, G.R., D.E. De Meza, J. Maloney, B. Pearson. 2000. And a vision

appeared unto them of a great profit: evidence of self-deception among the self-

employed. Economics Letters 67(1) pp. 35–41.

Arthur, W., D.V. Day. 1994. Development of a short form for the Raven Advanced

Progressive Matrices test. Educational and Psychological Measurement 54(2) pp.

394–403.

165

Astebro, T. 2003. The return to independent invention: Evidence of unrealistic op-

timism, risk seeking or skewness loving? The Economic Journal 113(484) pp.

226–239.

Astebro, T., H. Herz, R. Nanda, R.A. Weber. 2014. The behavioral economics of

entrepreneurship. Journal of Economic Perspectives 28(3) pp. 49–70.

Azrieli, Y., C.P. Chambers, P.J. Healy. 2012. Incentives in experiments: a theoretical

analysis. Working Paper, Ohio State University .

Baker, A., V. Mertins. 2013. Risk-sorting and preference for team piece rates. Journal

of Economic Psychology 34 pp. 285–300.

Bandiera, R., I. Barankay, I. Rasul. 2013. Team incentives: Evidence from a firm level

experiment. Journal of the European Economic Association 11(5) pp. 1079–1114.

Bartling, B., E. Fehr, H. Herz. 2014. The Intrinsic Value of Decision Rights. Econo-

metrica 82 pp. 2005–2039.

Beauchamp, J., D. Cesarini, M. Johannesson. 2012. The psychometric properties of

measures of economic risk preferences. mimeo .

Becker, G.M., M.H. DeGroot, J. Marschak. 1964. Measuring utility by a single-response

sequential method. Behavioral Science 9(3) pp. 226–232.

Begley, T.M. 1995. Using founder status, age of firm, and company growth rate as the

basis for distinguishing entrepreneurs from managers of small businesses. Journal of

Business Venturing 10 pp. 249–263.

Behling, O., N.L. Eckel. 1991. Making sense out of intuition. Academy of Management

Executive 5(1) pp. 46–54.

Ben-David, I., J.R. Graham, C.R. Harvey. 2013. Managerial miscalibration. Quarterly

Journal of Economics 128(4) pp. 1547–1584.

Bengtsson, O., D. Ekeblom. 2014. The bright but right view? New evidence on en-

trepreneurial optimism. LUND working paper .

Benz, M., B.S. Frey. 2008. The value of doing what you like: Evidence from the self-

employed in 23 countries. Journal of Economic Behavior & Organization 68(3-4)

pp. 445–455.

Bertrand, Marianne, Antoinette Schoar. 2003. Managing with style: The effect of

managers on firm policies. Quarterly Journal of Economics 118(4) pp. 1169–1208.

166

Biais, B., M. Weber. 2009. Hindsight bias, risk perception, and investment performance.

Management Science 55(6) pp. 1018–1029.

Bilker, W.B., J.A. Hansen, C.M. Brensinger, J. Richard, R.E. Gur, R.C. Gur. 2012.

Development of abbreviated nine-item forms of the Raven’s Standard Progressive

Matrices Test. Assessment 19(3) pp. 354–369.

Bitler, M.P., T.J. Moskowitz, A. Vissing-Jorgensen. 2005. Testing agency theory with

entrepreneur effort and wealth. Journal of Finance 60(2) pp. 539–576.

Blanchflower, D.G., A. Oswald. 1998. What makes an entrepreneur? Journal of Labor

Economics 16 pp. 26–60.

Block, J., P. Koellinger. 2009. I can’t get no satisfaction - necessity entrepreneurship

and procedural utility. Kyklos 62(2) pp. 191–209.

Bolle, F. 1990. High reward experiments without high expenditure for the experi-

menter? Journal of Economic Psychology 11(2) pp. 157–167.

Bonabeau, E. 2003. Don’t trust your gut. Harvard Business Review 81(5) pp. 116–123.

Bonin, H., T. Dohmen, A. Falk, D. Huffman, U. Sunde. 2007. Cross-sectional earnings

risk and occupational sorting: The role of risk attitudes. Labour Economics 14(6)

pp. 926–937.

Booij, A.S., G. Van de Kuilen. 2009. A parameter-free analysis of the utility of money

for the general population under prospect theory. Journal of Economic Psychology

30(4) pp. 651–666.

Borghans, L., A. Duckworth, J. Heckman, B. Ter Weel. 2008. The economics and

psychology of personality traits. Journal of Human Resources 43(4) pp. 972–1059.

Brachert, M., W. Hyll. 2014. On the stability of preferences: Repercussions of en-

trepreneurship on risk attitudes. SOEPpaper No. 667 .

Brockhaus, R.H. 1980. Risk taking propensity of entrepreneurs. The Academy of

Management Journal 23 pp. 509–520.

Buchanan, G.M., M.E.P. (Eds) Seligman. 1995. Explanatory Style. Hillsdale, New

Jersey: Lawrence Erlbaum Associates.

Burmeister, K., C. Schade. 2007. Are entrepreneurs’ decisions more biased? An ex-

perimental investigation of the susceptibility to status quo bias. Journal of Business

Venturing 22(3) pp. 340–362.

167

Burmeister-Lamp, K., M. Levesque, C. Schade. 2012. Are entrepreneurs influenced

by risk attitude, regulatory focus or both? An experiment on entrepreneurs’ time

allocation. Journal of Business Venturing 27(4) pp. 456–476.

Burns, M.O., M.E.P. Seligman. 1989. Explanatory style across the life span: Evidence

for stability over 52 years. Journal of Personality and Social Psychology 56(3) pp.

471–477.

Busenitz, L.W., J.B. Barney. 1997. Differences between entrepreneurs and managers

in large organizations: biases and heuristics in strategic decision-making. Journal of

Business Venturing 12 pp. 9–30.

Caird, S. 1991. The enterprise tendency of occupational groups. International Small

Business Journal 9(4) pp. 75–91.

Caliendo, M., F. Fossen, A. Kritikos. 2009. Risk attitudes of nascent entrepreneurs:

New evidence from an experimentally-validated survey. Small Business Economics

32(2) pp. 153–167.

Caliendo, M., F. Fossen, A. Kritikos. 2010. The impact of risk attitudes on en-

trepreneurial survival. Journal of Economic Behavior & Organization 76(1) pp.

45–63.

Camerer, C.F., D. Lovallo. 1999. Overconfidence and excess entry: An experimental

approach. American Economic Review 89 pp. 306–318.

Campbell, T.C., M. Gallmeyer, S.A. Johnson, J. Rutherford, B.W. Stanley. 2011. CEO

optimism and forced turnover. Journal of Financial Economics 101(3) pp. 695–712.

Cantillon, R. 1755. Essai Sur la Nature General . London Macmillan.

Carver, C.S., M.F. Scheier, S.C. Segerstrom. 2010. Optimism. Clinical Psychology

Review 30(7) pp. 879–889.

Cassar, G. 2010. Are individuals entering self-employment overly optimistic? An em-

pirical test of plans and projections on nascent entrepreneur expectations. Strategic

Management Journal 31(8) pp. 822–840.

Charness, G., U. Gneezy. 2012. Strong evidence for gender differences in risk taking.

Journal of Economic Behavior & Organization 83(1) pp. 50–58.

Choi, S., S. Kariv, W. Muller, D. Silverman. 2014. Who is (more) rational? American

Economic Review 104(6) pp. 1518–1550.

168

Colin, C., G. Loewenstein. 2004. Behavioral economics: Past, present, future. Prince-

ton: Princeton University Press.

Collard, J., C. Reynolds. 2004. Leadership Gender and Culture in Education. Open

University Press.

Cooper, A.C., C.Y. Woo, W.C. Dunkelberg. 1988. Entrepreneurs’ perceived chances

of success. Journal of Business Venturing 3 pp. 97–108.

Cooper, D.J., K.J. Saral. 2013. Entrepreneurship and team participation: An experi-

mental study. European Economic Review 59 pp. 126–140.

CPB. 2014. Table main economic indicators 2012-2015.

http://www.cpb.nl/cijfer/kerngegevenstabel-2012-2015-voor-het-centraal-

economisch-plan-2014. accessed: 2015-07-08. .

Cramer, J. S., J. Hartog, N. Jonker, C. M. Van Praag. 2002. Low risk aversion en-

courages the choice for entrepreneurship: an empirical test of a truism. Journal of

Economic Behavior & Organization 48(1) pp. 29–36.

Crane, F., E. Crane. 2007. Dispositional optimism and entrepreneurial success. The

Psychologist-Manager Journal 10(1) pp. 13–25.

Cubitt, R.P., C. Starmer, R. Sugden. 1998. On the validity of the random lottery

incentive system. Experimental Economics 1(2) pp. 115–131.

Danilov, A., T. Biemann, T. Kring, D. Sliwka. 2013. The dark side of team incentives:

Experimental evidence on advice quality from financial service professionals. Journal

of Economic Behavior & Organization 93(C) pp. 266–272.

Danziger, S., S. Moran, V. Rafaely. 2006. The influence of ease of retrieval on judgment

as a function of attention to subjective experience. Journal of Consumer Psychology

16(2) pp. 191–195.

Dargnies, M.-P. 2012. Men too sometimes shy away from competition: The case of

team competition. Management Science 58(11) pp. 1982–2000.

Dawson, C., D.E. De Meza, A. Henley, G.R. Arabsheibani. 2014. Entrepreneurship:

Cause and consequence of financial optimism. Journal of Economics & Management

Strategy 23(4) pp. 717–742.

Dawson, C., D.E. De Meza, A. Henley, G.R. Arabsheibani. 2015. The power of (non)

positive thinking: Self-employed pessimists earn more than optimists. IZA discussion

paper No. 9242 .

169

DeMartino, R., R. Barbato. 2003. Differences between women and men MBA en-

trepreneurs: exploring family flexibility and wealth creation as career motivators.

Journal of Business Venturing 18(6) pp. 815–832.

Ding, X., J. Hartog, Y. Sun. 2010. Can we measure individual risk attitudes in a

survey? IZA discussion paper No. 4807 .

Djankov, S., Y. Qian, G. Roland, E. Zhuravskaya. 2006. Who are China’s en-

trepreneurs? American Economic Review Papers and Proceedings 96(2) pp. 348–352.

Djankov, S., Y. Qian, G. Roland, E. Zhuravskaya. 2007. What makes a successful

entrepreneur? evidence from Brazil. Working Paper No. 0104, Center for Economic

and Financial Research .

Dohmen, T., A. Falk. 2011. Performance pay and multidimensional sorting: Produc-

tivity, preferences, and gender. American Economic Review pp. 556–590.

Dohmen, T., A. Falk, D. Huffman, U. Sunde. 2010. Are risk aversion and impatience

related to cognitive ability? American Economic Review 100(3) pp. 1238–1260.

Dohmen, T., A. Falk, D. Huffman, U. Sunde, J. Schupp, G. G. Wagner. 2007. The

measurement and stability of risk attitudes. mimeo .

Dohmen, T., A. Falk, D. Huffman, U. Sunde, J. Schupp, G. G. Wagner. 2011. Individual

risk attitudes: Measurement, determinants and behavioral consequences. Journal of

the European Economic Association 9(3) pp. 522–550.

Dreber, A., D. Rand, N. Wernerfelt, J. Garcia, M. Vilar, J. Lum, R. Zeckhauser. 2011.

Dopamine and risk choices in different domains: Findings among serious tournament

bridge players. Journal of Risk and Uncertainty 43(1) pp. 19–38.

Dunning, D., C. Heath, J.M. Suls. 2004. Flawed self-assessment implications for health,

education, and the workplace. Psychological Science in the Public Interest 5(3) pp.

69–106.

Dur, R., J. Sol. 2010. Social interaction, co-worker altruism, and incentives. Games

and Economic Behavior 69(2) pp. 293–301.

Eckel, C. C., P. J. Grossman. 2008. Forecasting risk attitudes: An experimental study

using actual and forecast gamble choices. Journal of Economic Behavior & Organi-

zation 68 pp. 1–17.

Ellsberg, D. 1961. Risk, ambiguity, and the savage axioms. Quarterly Journal of

Economics 75 pp. 643–669.

170

Elston, J.A., G.W. Harrison, E.E. Rutstrom. 2006. Experimental economics, en-

trepreneurs and the entry decision. Working Paper No. 6, University of Central

Florida .

Epstein, S. 1973. The self-concept revisited or a theory of a theory. American Psychol-

ogist 28(5) pp. 404–416.

Epstein, S. 2003. Cognitive-experiential self-theory of personality. Handbook of psy-

chology .

Epstein, S., R. Pacini, V. Denes-Raj, H. Heier. 1996. Individual differences in intuitive-

experiential and analytical-rational thinking styles. Journal of Personality and Social

Psychology 71(2) p. 390.

Epstein, S., R. Pacini, P Norris. 1998. The rational-experiential inventory, long form.

University of Massachsetts at Amherst: Unpublished inventory .

Evans, D.S., B. Jovanovic. 1989. An estimated model of entrepreneurial choice under

liquidity constraints. Journal of Political Economy 97 pp. 808–827.

Evans, J.S.B.T. 2008. Dual-processing accounts of reasoning, judgment, and social

cognition. Annual Review of Psychology 59 pp. 255–278.

Fairlie, R.W., H.A. Krashinsky. 2012. Liquidity constraints, household wealth, and

entrepreneurship revisited. Review of Income and Wealth, International Association

for Research in Income and Wealth, 58(2) pp. 279–306.

Fehr, E., L. Goette. 2007. Do workers work more if wages are high? Evidence from a

randomized field experiment. American Economic Review 97(1) pp. 298–317.

Fehr, E., H. Herz, T. Wilkening. 2013. The lure of authority: Motivation and incentive

effects of power. American Economic Review 103(4) pp. 1325–1359.

Filippin, A., P. Crosetto. 2014. A reconsideration of gender differences in risk attitudes.

IZA discussion paper No. 8184 .

Fischer, M., H. Leitenberg. 1986. Optimism and pessimism in elementary school-aged

children. Child development 57 pp. 241–248.

Fisher, R., A. Chalmers. 2008. Is optimism universal? A meta-analytical investigation

of optimism levels across 22 nations. Personality and Individual Differences 45 pp.

378–382.

171

Fouarge, D., B. Kriechel, T. Dohmen. 2014. Occupational sorting of school graduates:

The role of economic preferences. Journal of Economic Behavior & Organization

106 pp. 335–351.

Fox, C.R., A. Tversky. 1995. Ambiguity aversion and comparative ignorance. Quarterly

Journal of Economics 110(3) pp. 585–603.

Fraser, S., F. Greene. 2006. The effects of experience on entrepreneurial optimism and

uncertainty. Economica 73(290) pp. 169–192.

Gachter, S., E.J. Johnson, A. Herrmann. 2010. Individual-level loss aversion in riskless

and risky choices. CeDEx discussion paper series, No. 2010-20 .

Gilbert, D.T. 1989. Thinking Lightly About Others: Automatic Components of the

Social Inference Process . in James S. Uleman and John A. Bargh, eds., Unintended

Thought., Englewood Cliffs, NJ: Prentice-Hall, pp. 189-211.

Gilbert, D.T. 2002. Inferential Correction, chap. in Thomas Gilovich, Dale Griffin and

Daniel Kahneman, eds., Heuristics and Biases. New York: Cambridge University

Press, pp. 167–184.

Giltay, E.J., M. Kamphuis, S. Kalmijn, F. Zitman, D. Krombout. 2006. Dispostional

optimism and the risk of cardiovascular death: the Zutphen Elderly Study. Archives

of Internal Medicine 166 431–436.

Gneezy, U., A. Pietrasz. 2014. When half of the men are more competitive than all

women. ASSA discussion paper, Philadelphia .

Gneezy, U., J. Potters. 1997. An experiment on risk taking and evaluation periods.

Quarterly Journal of Economics pp. 631–645.

Gneezy, U., A. Rustichini. 2000. Pay enough or don’t pay at all. Quarterly Journal of

Economics 115(3) pp. 791–810.

Goel, A.M, A.V. Thakor. 2008. Overconfidence, CEO selection, and corporate gover-

nance. Journal of Finance 63(6) pp. 2737–2784.

Graham, J.R., C.R. Harvey, M. Puri. 2013. Managerial attitudes and corporate actions.

Journal of Financial Economics 109 pp. 103–121.

Hackbarth, D. 2008. Managerial traits and capital structure decisions. Journal of

Financial and Quantitative Analysis 43(04) pp. 843–881.

Hall, R.E., S.E. Woodward. 2010. The burden of the nondiversifiable risk of en-

trepreneurship. American Economic Review 100(3) pp. 1163–1194.

172

Hamilton, B.H. 2000. Does entrepreneurship pay? An empirical analysis of the returns

to self-employment. Journal of Political Economy 108(3) pp. 604–631.

Hamilton, B.H., J.A. Nickerson, H. Owan. 2003. Team incentives and worker hetero-

geneity: An empirical analysis of the impact of teams on productivity and partici-

pation. Journal of Political Economy 111(3) pp. 465–497.

Hamilton, B.H., N.W. Papageorge, N. Pande. 2014. The Right Stuff? Personality and

Entrepreneurship. mimeo .

Harbaugh, W. T., K. Krause, L. Vesterlund. 2002. Risk attitudes of children and adults:

Choices over small and large probability gains and losses. Experimental Economics

5(1) pp. 53–84.

Harrison, G.W., M.I. Lau, E. Rutstrom. 2007. Estimating risk attitudes in Denmark:

A field experiment. The Scandinavian Journal of Economics 109(2) pp. 341–368.

Hayashi, A.M. 2001. When to trust your gut. Harvard Business Review pp. 59–65.

Heckman, J., R. Pinto, P. Savelyev. 2013. Understanding the mechanisms through

which an influential early childhood program boosted adult outcomes. American

Economic Review 103(6) pp. 2052–2086.

Henrekson, M., T. Sanandaji. 2014. Small business activity does not measure en-

trepreneurship. Proceedings of the National Academy of Sciences pp. 1760–1765.

Herbst, L., K.A. Konrad, F. Morath. 2015. Endogenous group formation in experi-

mental contests. European Economic Review 74 pp. 163–189.

Herz, H., D. Schunk, C. Zehnder. 2014. How do judgmental overconfidence and overop-

timism shape innovative activity? Games and Economic Behavior 83 pp. 1–23.

Hmieleski, K.M., R.A. Baron. 2009. Entrepreneurs’ optimism and new venture perfor-

mance: A social cognitive perspective. Academy of Management Journal 52(3) pp.

473–488.

Holm, H.J., S. Opper, V. Nee. 2013. Entrepreneurs under uncertainty: An economic

experiment in China. Management Science 59(7) pp. 1671–1687.

Holt, C.A., S.K. Laury. 2002. Risk aversion and incentive effects. American Economic

Review 92(5) pp. 1644–1655.

Holtz-Eakin, D., D. Joulfaian, H.S. Rosen. 1994a. Entrepreneurial decisions and liq-

uidity constraints. RAND Journal of Economics 25 334–347.

173

Holtz-Eakin, D., D. Joulfaian, H.S. Rosen. 1994b. Sticking it out: entrepreneurial

survival and liquidity constraints. Journal of Political Economy 102 53–75.

Horowitz, J.K., K.E. McConnell. 2002. A review of WTA/WTP studies. Journal of

Environmental Economics and Management 44(3) pp. 426–447.

Hotelling, H. 1929. Stability in competition. The Economic Journal 39(153) pp. 41–57.

Huberman, G., A. Rubinstein. 2000. Correct belief, wrong action and a puzzling gender

difference. mimeo .

Hull, D., J. Bosley, G. Udell. 1980. Renewing the hunt for heffalump: Identifying

potential entrepreneurs by personality characteristics. Journal of Small Business

Management 18(1) pp. 11–18.

Hurst, E., B.W. Pugsley. 2011. What do small businesses do? Tech. rep., National

Bureau of Economic Research, No. w17041.

Hvide, H.K., J. Møen. 2010. Lean and hungry or fat and content? Entrepreneurs’

wealth and start-up performance. Management Science 56(8) pp. 1242–1258.

Hvide, H.K., G.A. Panos. 2014. Risk tolerance and entrepreneurship. Journal of

Financial Economics 111(1) pp. 200–223.

Isaacowitz, D.M., M.E.P. Seligman. 2001. Is pessimistic explanatory style a risk factor

for depressive mood among community-dwelling older adults? Behaviour Research

and Therapy 39 pp. 255–272.

Kahneman, D. 2011. Thinking, Fast and Slow . Farrar, Straus and Giroux, New York.

Kahneman, D., J.L. Knetsch, R.H. Thaler. 1990. Experimental tests of the endowment

effect and the Coase Theorem. Journal of Political Economy 98(6) pp. 1325–1348.

Kahneman, D., D. Lovallo. 1993. Timid choices and bold forecasts: A cognitive per-

spective on risk taking. Management Science 39(1) pp. 17–31.

Kahneman, D., A. Tversky. 1979. Prospect theory: An analysis of decision under risk.

Econometrica 47 pp. 263–291.

Kahneman, D., A. Tversky. 1984. Choices, values, and frames. American Psychologist

39(4) pp. 341–350.

Kaniel, R., C. Massey, D.T. Robinson. 2010. The importance of being an optimist.

Evidence from labor markets. NBER Working Paper No. w16328 .

174

Kemmelmeier, M. 2010. Authoritarianism and its relationship with intuitive-

experiential cognitive style and heuristic processing. Personality and Individual Dif-

ferences 48(1) pp. 44–48.

Kihlstrom, R.E., J.-J. Laffont. 1979. A general equilibrium entrepreneurial theory

of firm formation based on risk aversion. Journal of Political Economy 87(4) pp.

719–748.

Kirzner, I. 1973. Competition and Entrepreneurship. University of Chicago Press.

Knight, F.H. 1921. Risk, uncertainty, and profit . Houghton Mifflin Company (Boston).

Kobberling, V., P.P. Wakker. 2005. An index of loss aversion. Journal of Economic

Theory 122(1) pp. 119–131.

Kocher, M., S. Strauß, M. Sutter. 2006. Individual or team decision-making: causes and

consequences of self-selection. Games and Economic Behavior 56(2) pp. 259–270.

Koellinger, P. 2008. Why are some entrepreneurs more innovative than others? Small

Business Economics 31(1) pp. 21–37.

Koellinger, P., M. Minniti, C. Schade. 2007. ‘I think I can, I think I can’: overconfidence

and entrepreneurial behaviour. Journal of Economic Psychology 28 pp. 502–527.

Koh, H. C. 1996. Testing hypotheses of entrepreneurial characteristics: a study of

Hong Kong MBA students. Journal of Managerial Psychology 11 pp. 12–25.

Kosfeld, M., F.A. Siemens. 2009. Worker self-selection and the profit from cooperation.

Journal of the European Economic Association 7(2-3) pp. 573–582.

Koudstaal, M., R. Sloof, C.M. van Praag. 2015. Risk, uncertainty and entrepreneurship:

Evidence from a large lab-in-the-field experiment. Management Science (forthcom-

ing) .

Krueger, N.F., M.D. Reilly, Carsrud A.L. 2000. Competing models of entrepreneurial

intentions. Journal of Business Venturing 15(5-6) pp. 411–432.

Kuhn, P., M.C. Villeval. 2014. Are women more attracted to co-operation than men?

The Economic Journal (125) pp. 115–140.

Landier, A., D. Thesmar. 2009. Financial contracting with optimistic entrepreneurs.

The Review of Financial Studies 22(1) pp. 117–150.

Laury, S.L. 2006. Pay one or pay all: Random selection of one choice for payment.

Andrew Young School of Policy Studies Research Paper Series No. 06-13 .

175

Lazear, E.P. 2000. Performance pay and productivity. American Economic Review

90(5) pp. 1346–1361.

Lee, S.Y. 2015. Entrepreneurs, managers and inequality. Working Paper Series, Uni-

versity of Mannheim .

Leuven, E., H. Oosterbeek, J. Sonnemans, B. Van der Klaauw. 2011. Incentives ver-

sus sorting in tournaments: Evidence from a field experiment. Journal of Labor

Economics 29(3) pp. 637–658.

Levesque, M., C. Schade. 2005. Intuitive optimizing: Experimental findings on time

allocation decisions with newly formed ventures. Journal of Business Venturing

20(3) pp. 313–342.

Levine, R., Y. Rubinstein. 2013. Smart and illicit: Who becomes an entrepreneur and

does it pay? NBER Working Paper No. w19276 .

Lichtenstein, S., B. Fischhoff, L.D. Phillips. 1982. Calibration of probabilities: The

state of the art to 1980, Kahneman D., Slovic P., Tversky A., Judgment under

uncertainty: Heuristics and biases, 1982, 306-334.

Lindquist, M., J. Sol, C.M. van Praag. 2015. Why do entrepreneurial parents have

entrepreneurial children? Journal of Labor Economics 33(2) pp. 269–296.

Lonnqvist, J.E., M. Verkasalo, G. Walkowitz, P.C. Wichardt. 2014. Measuring individ-

ual risk attitudes in the lab: task or ask? An empirical comparison. CESifo Working

Paper Series .

Lovallo, D., D. Kahneman. 2003. Delusions of success. Harvard Business Review 81(7)

pp. 56–63.

Lowe, R.A., A.A. Ziedonis. 2006. Overoptimism and the performance of entrepreneurial

firms. Management Science 52(2) pp. 173–186.

Macko, A., T. Tyszka. 2009. Entrepreneurship and risk taking. Applied Psychology

58(3) pp. 469–487.

Malmendier, U., G. Tate. 2005. CEO overconfidence and corporate investment. Journal

of Finance 60(6) pp. 2661–2700.

Malmendier, U., G. Tate. 2009. Superstar CEOs. Quarterly Journal of Economics

124(4) pp. 1593–1638.

Marshall, A. 1930. Principles of Economics . London Macmillan.

176

Masclet, D., N. Colombier, L. Denant-Boemont, Y. Loheac. 2009. Group and indi-

vidual risk preferences: A lottery-choice experiment with self-employed and salaried

workers. Journal of Economic Behavior & Organization 70(3) pp. 470–484.

Miller, D., P.H. Friesen. 1984. Organizations: A Quantum View.. Englewood Cliffs,

NJ: Prentice Hall.

Miner, J.B., N.S. Raju. 2004. Risk propensity differences between managers and en-

trepreneurs and between low- and high-growth entrepreneurs: A reply in a more

conservative vein. Journal of Applied Psychology 89(1) pp. 3–13.

Moore, D.A., P.J. Healy. 2008. The trouble with overconfidence. Psychological Review

115(2) pp. 502–517.

Moskowitz, T.J., A. Vissing-Jorgensen. 2002. The returns to entrepreneurial invest-

ment: A private equity premium puzzle? American Economic Review 92(4) pp.

745–778.

Norris, P., S. Epstein. 2011. An experiential thinking style: Its facets and relations

with objective and subjective criterion measures. Journal of Personality 79(5) pp.

1043–1080.

Owens, D., Z. Grossman, R. Fackler. 2014. The control premium: A preference for

payoff autonomy. American Economic Journal: Microeconomics 6(4) pp. 138–161.

Pacini, R., S. Epstein. 1999. The relation of rational and experiential information

processing styles to personality, basic beliefs, and the ratio-bias phenomenon. Journal

of Personality and Social Psychology 76(6) p. 972.

Pacini, R., F. Muir, S. Epstein. 1998. Depressive realism from the perspective of

cognitive-experiential self-theory. Journal of Personality and Social Psychology 74(4)

1056.

Parikh, J. 1994. Intuition: The new frontier of management . Oxford: Blackwell

Business.

Parker, S.C. 2009. The Economics of Entrepreneurship. University Press, Cambridge.

Parker, S.C., C.M. Van Praag. 2010. Group status and entrepreneurship. Journal of

Economics & Management Strategy 19(4) pp. 919–945.

Peterson, C. 2000. The future of optimism. American psychologist 55(1) pp. 44–55.

Piccinelli, M., G. Wilkinson. 2000. Gender differences in depression. critical review.

The British Journal of Psychiatry 177(6) pp. 486–492.

177

Puri, M., D. Robinson. 2006. Who are entrepreneurs and why do they behave that

way? Duke University Working Paper .

Puri, M., D.T. Robinson. 2007. Optimism and economic choice. Journal of Financial

Economics 86 pp. 71–99.

Puri, M., D.T. Robinson. 2013. The economic psychology of entrepreneurship and

family business. Journal of Economics & Management Strategy 22(2) pp. 423–444.

Puskar, K.R., S.M. Sereika, J. Lamb, K. Tusaie-Mumford, T. McGuinness. 1999. Op-

timism and its relationship to depression, coping, anger, and life events in rural

adolescents. Issues in Mental Health Nursing 20(2) pp. 115–130.

Rabin, M. 2000. Risk aversion and expected-utility theory: A calibration theorem.

Econometrica 68 pp. 1281–1292.

Raven, J., J.C. Raven, J.H. Court. 2003. Manual for RavenOs Progressive Matrices

and Vocabulary Scales. Harcourt Assessment..

Reynolds, P.D., R.T. Curtin. 2008. Business creation in the united states: Panel

study of entrepreneurial dynamics ii. initial assessment. Foundations and Trends in

Entrepreneurship 4(3) pp. 155–307.

Rietveld, C.A., P.J.F. Groenen, P. Koellinger, M.J.H.M. van der Loos, A.R. Thurik.

2013. Living forever: Entrepreneurial overconfidence at older ages. ERIM working

paper .

Rubinstein, A. 2007. Instinctive and Cognitive Reasoning: A Study of Response Times.

The Economic Journal 117 pp. 1243–1259.

Rubinstein, A. 2013. Response time and decision making: A “free” experimental study.

Judgment and Decision Making 8 pp. 540–551.

Rubinstein, A. 2016. A typology of players: between instinctive and contemplative.

Quarterly Journal of Economics (forthcoming) .

Sadler-Smith, E., E. Shefy. 2004. The intuitive executive: understanding and applying

’gut feel’ in decision-making. The Academy of Management Executive 18(4) pp.

76–91.

Sagiv, L., A. Amit, D. Ein-Gar, S. Arieli. 2014. Not All Great Minds Think Alike:

Systematic and Intuitive Cognitive Styles. Journal of Personality 82(5) pp. 402–

417.

178

Sahm, C. 2012. How much does risk tolerance change? Quarterly Journal of Finance

2(4).

Sandri, S., C. Schade, O. Musshoff, M. Odening. 2010. Holding on for too long? An

experimental study on inertia in entrepreneurs’ and non-entrepreneurs’ disinvestment

choices. Journal of Economic Behavior & Organization 76(1) pp. 30–44.

Sarasvathy, D.K., H. Simon, L. Lave. 1998. Perceiving and managing business risks:

Differences between entrepreneurs and bankers. Journal of Economic Behavior &

Organization 33(2) pp. 207–222.

Say, J.-B. 1803. A Treatise on Political Economy or the Production, Distribution and

Consumption of Wealth. A.M. Kelley Publishers, New York.

Scheier, M.F., C.S. Carver. 1985. Optimism, coping, and health: Assessment and

implications of generalized outcome expectancies. Health Psychology 4 pp. 219–247.

Scheier, M.F., C.S. Carver, M.W. Bridges. 1994. Distinguishing optimism from neu-

roticism (and trait anxiety, self-mastery, and self-esteem): A reevaluation of the Life

Orientation Test. Journal of Personality and Social Psychology 67 pp. 1063–1078.

Schere, J.L. 1982. Tolerance of ambiguity as discriminating variable between en-

trepreneurs and managers. Academy of Management Best Paper Proceedings 42

pp. 404–408.

Schmalz, M.C., D.A. Sraer, D. Thesmar. 2013. Household collateral and entrepreneur-

ship. NBER Working Paper No. w19680 .

Schulman, P., D. Keith, M.E.P. Seligman. 1993. Is optimism heritable? a study of

twins. Behaviour Research and Therapy 31(6) pp. 569–574.

Schulman, P., M.E.P. Seligman, D. Amsterdam. 1987. The attributional style ques-

tionnaire is not transparent. Behaviour Research and Therapy 25 pp. 391–395.

Segerstrom, S.C. 2007. Optimism and resources: Effects on each other and on health

over 10 years. Journal of Research in Personality 41(4) pp. 772–786.

Seligman, M.E.P. 2000. Authentic Happiness . The Free Press.

Seligman, M.E.P., P. Schulman. 1986. Explanatory style as a predictor of produc-

tivity and quitting among life insurance agents. Journal of Personality and Social

Psychology 50 pp. 832–838.

Selten, R. 1978. The chain store paradox. Theory and Decision 9(2) pp. 127–159.

179

Simon, H.A. 1987. Making management decisions: the role of intuition and emotion.

Academy of Management Executive 1(1) pp. 57–65.

Simon, M., S.M. Houghton, K. Aquino. 2000. Cognitive biases, risk perception, and

venture formation: How individuals decide to start companies. Journal of Business

Venturing 15(2) pp. 113–134.

Skriabikova, O.J., T.J. Dohmen, B. Kriechel. 2014. New evidence on the relationship

between risk attitudes and self-employment. Labour Economics 30 pp. 176–184.

Sladek, R.M., M.J. Bond, P.A. Phillips. 2010. Age and gender differences in preferences

for rational and experiential thinking. Personality and Individual Differences 49(8)

pp. 907–911.

Sloof, R., F. von Siemens. 2014. Illusion of Control and the Pursuit of Authority.

CESifo Working Paper Series 4764, CESifo Group Munich .

Solberg Nes, L., D.R. Evans, S.C. Segerstrom. 2009. Optimism and college retention:

Mediation by motivation, performance, and adjustment. Journal of Applied Social

Psychology 39 pp. 1887–1912.

Stanovich, K.E., R.F. West. 2000. Advancing the rationality debate. Behavioral and

Brain Sciences 23(05) pp. 701–717.

Starmer, C., R. Sugden. 1991. Does the random-lottery incentive system elicit true

preferences? An experimental investigation. American Economic Review pp. 971–

978.

Stewart Jr., W. H., P. L. Roth. 2001. Risk propensity differences between entrepreneurs

and managers: a meta-analytic review. Journal of Applied Psychology 86(1) pp.

145–153.

Stewart Jr., W. H., W.E. Watson, J.C. Carland, J.W. Carland. 1999. A proclivity

for entrepreneurship: A comparison of entrepreneurs, small business owners, and

cooperate managers. Journal of Business Venturing 14(2) pp. 189–214.

Stipek, D.J., M.E. Lamb, E.F. Zigler. 1981. Opti: A measure of children’s optimism.

Educational and Psychological Measurement 41 pp. 131–143.

Swalm, R.O. 1966. Utility theory (insights into risk taking). Harvard Business Review

44(6) pp. 123 – 136.

Tag, J., T. Astebro, P. Thompson. 2013. Hierarchies, the small firm effect, and en-

trepreneurship: Evidence from Swedish microdata. IFN Working Paper No. 954

.

180

Teyssier, S. 2008. Experimental evidence on inequity aversion and self-selection between

incentive contracts. GATE working paper (0710).

Thaler, R.H., H.M. Shefrin. 1981. An economic theory of self-control. Journal of

Political Economy 89 pp. 392–406.

The Economist. 2010. The U-bend of life (print edition of December 18, page 33).

Toyosawa, J., K. Karasawa. 2004. Individual differences on judgment using the ratio-

bias and the linda problem: Adopting CEST and Japanese version of REI. Japanese

Journal of Social Psychology 20(2) pp. 85–92.

Trautmann, S.T., G. Van der Kuilen. forthcoming. Attitudes towards Ambiguity . in:

G. Keren and G. Wu (eds.), Blackwell Handbook of Judgment and Decision Making,

Chapter 11.

Trevelyan, R. 2008. Optimism, overconfidence and entrepreneurial activity. Manage-

ment Decision 46(7) pp. 986–1001.

Tversky, A., D. Kahneman. 1974. Judgment under uncertainty: Heuristics and biases.

Science 185(4157) pp. 1124–1131.

Tversky, A., D. Kahneman. 1992. Advances in prospect theory: Cumulative represen-

tation of uncertainty. Journal of Risk and Uncertainty 5(4) pp. 297–323.

Ucbasaran, D., P. Westhead, M. Wright, M. Flores. 2010. The nature of entrepreneurial

experience, business failure and comparative optimism. Journal of Business Ventur-

ing 25(6) pp. 541–555.

Uusitalo, R. 2001. Homo entreprenaurus? Applied Economics 33(13) pp. 1631–1638.

Van den Steen, E.J. 2004. Rational overoptimism (and other biases). American Eco-

nomic Review 94(4) pp. 1141–1151.

Van den Steen, E.J. 2011. Overconfidence by Bayesian-rational agents. Management

Science 57(5) pp. 884–896.

van Dijk, F., J. Sonnemans, F. van Winden. 2001. Incentive systems in a real effort

experiment. European Economic Review 45(2) pp. 187–214.

Van Praag, C. M., J. S. Cramer. 2001. The roots of entrepreneurship and labour

demand: individual ability and low risk aversion. Economica 68 pp. 45–62.

Van Praag, C.M. 1999. Some classic views on entrepreneurship. The Economist 147(3)

pp. 311–335.

181

Wakker, P. P. 2005. Formalizing reference dependence and initial wealth in Rabin’s

calibration theorem. Working Paper, Erasmus University Rotterdam .

Wakker, P. P. 2010. Prospect theory: For risk and ambiguity . Cambridge University

Press, Cambridge.

Wakker, P.P., D. Deneffe. 1996. Eliciting von Neumann-Morgenstern utilities when

probabilities are distorted or unknown. Management Science 42(8) pp. 1131–1150.

Weinstein, N.D. 1980. Unrealistic optimism about future life events. Journal of Per-

sonality and Social Psychology 39(5) p. 806.

Willebrands, D., J. Lammers, J. Hartog. 2012. A successful businessman is not a

gambler; Risk attitude and business performance among small enterprises in Nigeria.

Journal of Economic Psychology 22(2) pp. 342–354.

Wilson, T.D. 2002. Strangers to Ourselves: Discovering the Adaptive Unconscious .

Cambridge, MA: Harvard University Press.

Yazdipour, R. 2010. Advances in Entrepreneurial Finance: With Applications from

Behavioral Finance and Economics . Springer.

182

Samenvatting (Summary in Dutch)

Dit proefschrift beschrijft de resultaten van vier zogenoemde ‘lab-in-the-field’ exper-

imenten naar de verschillen tussen ondernemers en andere beroepsgroepen. Elk van

deze experimenten is gericht op een specifiek element van besluitvorming, zijnde: risico

& onzekerheid (Hoofdstuk 2), optimisme en zelfoverschatting (Hoofdstuk 3), intuıtieve

of contemplatieve besluitvorming (Hoofdstuk 4) en de voorkeur voor individuele of

teambeloning (Hoofdstuk 5). Hieronder volgt een samenvatting van de resultaten.

Hoofdstuk 2 is gebaseerd op het paper “Risk, Uncertainy and Entrepreneurship:

Evidence from a Lab-in-the-Field Experiment”, dat is geschreven samen met Mirjam

van Praag en Randolph Sloof, en dat binnenkort wordt gepubliceerd in Management

Science. De resultaten van dit hoofdstuk laten allereerst zien dat ondernemers zichzelf

als minder risico avers beschouwen dan managers en werknemers, in lijn met de al-

gemene wijsheid op diverse internet sites and blogs. Als we risicohouding echter objec-

tief meten met behulp van experimentele ‘incentivized’ maatstaven zijn de verschillen

subtieler. We vinden dat ondernemers alleen uniek zijn in hun lagere aversie voor ver-

liezen, maar niet in hun houding ten aanzien van risico of onzekerheid. Bij de replicatie

van deze studie vinden we precies hetzelfde als hiervoor omschreven, hetgeen verdere

ondersteuning biedt voor onze eerdere conclusie. Daarnaast hebben we aan de hand

van extra analyses getracht om de incongruente resultaten op de subjectieve en ob-

jectieve maatstaven met elkaar te kunnen verenigen. Uiteindelijk vinden we dat voor

onze respondenten de subjectieve maatstaf van risico niet alleen samenhangt met de

objectieve maatstaf voor risico, maar ook met hun houding ten aanzien van verliezen

en onzekerheid. Blijkbaar gebruiken economen dus een andere (nauwere) definitie van

risico dan ondernemers.

Twee andere belangrijke elementen van besluitvorming zijn optimisme en zelfover-

schatting. In Hoofdstuk 3, dat gebaseerd is op het paper “Are Entrepreneurs More

Optimistic and Overconfident than Managers and Employees?”, dat wederom samen

is geschreven met Mirjam van Praag en Randolph Sloof, onderzoeken we dit verder.

Op basis van de bestaande theoretische en empirische literatuur lijkt het dat zowel

ondernemers als managers optimistischer zijn dan werknemers en ook meer lijden aan

zelfoverschatting. Gek genoeg zijn ze echter op dit vlak nog niet eerder met elkaar

183

vergeleken. Wij proberen daarom dit gat in de literatuur te verkleinen door onderne-

mers, managers en werknemers te vergelijken op deze twee aspecten. De resultaten van

deze studie laten zien dat ondernemers optimischer zijn dan managers en werknemers in

hun dispositionele optimisme en hun optimistischere houding na slechte gebeurtenissen.

Op de twee meer objectieve ‘incentivized’ maatstaven van zelfoverschatting vinden we

echter dat zowel ondernemers als managers meer aan lijden aan zelfoverschatting dan

werknemers. De enige uitzondering wordt gevormd door start-up ondernemers (d.w.z.

die werken voor een eigen bedrijf dat niet ouder is dan drie jaar), die qua zelfover-

schatting ook verschillen van managers. Maar voor alle andere type ondernemers en

managers vinden we exact hetzelfde zoals hierboven is beschreven. Tot slot kijken we

in deze studie ook nog naar het effect van de verschillende succesmaatstaven die we

hebben. We vinden dat succesvolle ondernemers en managers (bijvoorbeeld onderne-

mers met een B.V. of managers die CEO’s zijn) optimistischer zijn dan hun minder

succesvolle lotgenoten maar niet meer lijden aan zelfoverschatting. Er zijn echter geen

significante verschillen meer tussen deze succesvolle ondernemers en managers. Over

het geheel genomen lijken optimisme en zelfoverschatting dus niet zozeer unieke eigen-

schappen van ondernemers - zoals door menig theoretisch en empirisch paper wordt

verondersteld -, maar meer een eigenschap van (succesvolle) besluitvormers.

Het derde element van besluitvorming dat we onderzoeken is de besluitvormingsstijl.

Hierover valt meer te lezen in het paper “Are Entrepreneurs More Intuitive Than Man-

agers and Employees? Evidence From a Lab-in-the-Field Experiment”, dat wederom is

geschreven met Mirjam van Praag en Randolph Sloof. In dit hoofdstuk testen we hoe

ondernemers, managers en werknemers van elkaar verschillen in hun contemplativiteit.

We doen dit aan de hand van twee bekende subjectieve maatstaven uit de psycholo-

gische literatuur en twee objectieve maatstaven die gebaseerd zijn op responstijden en

het aantal contemplatieve keuzes in een aantal experimentele spellen van Rubinstein

(2016). De resultaten laten zien dat ondernemers wel intuıtiever zijn dan managers,

maar niet dan werknemers. Verrassend genoeg vinden we echter een ander patroon in

de responstijden, ook al zouden deze een redelijke voorspellende waarde moeten hebben

voor het aantal contemplatieve keuzes. Een mogelijke verklaring voor deze mismatch is

dat de relatieve effecten van tijd anders zijn voor de drie groepen en onze vervolgana-

lyses laten inderdaad zien dat ondernemers en managers relatief contemplatiever zijn

per eenheid van tijd. Als we deze heterogeniteit meenemen in onze eerdere regressies

vinden we dat ondernemers ook intuıtiever zijn dan werknemers zo lang ze niet lang

nadenken over hun antwoord. De twee subjectieve maatstaven schetsen een eenduidig

beeld; ondernemers scoren namelijk het hoogst op ‘Faith in Intuition’, terwijl managers

het hoogste scoren op ‘Need for Cognition’. Over het geheel genomen verschillen on-

184

dernemers dus met name van werknemers in hun besluitvormingsstijl, en in mindere

mate van managers.

Het laatste onderwerp van deze dissertatie is de voorkeur voor individuele of team-

beloning. Hierover gaat Hoofdstuk 5, dat gebaseerd is op het paper “Risks, Gains

and Autonomy: An Experimental Analysis of Preferences for Joining Teams”, dat

geschreven is met Laura Rosendahl Huber en Eszter Czibor. In deze studie leunen

we op een zogenoemd ‘incentivized lab-in-the-field’ experiment met twee variaties: een

“basis” variatie waarbij alleen de autonomie over de uitbetaling op het spel staat en

een “gezamenlijke investering” variatie waarbij zowel de autonomie als het risico van de

uitbetaling op het spel staat. De resultaten van eerstgenoemde variatie laten allereerst

zien dat respondenten bereid zijn om meer te betalen voor teambeloning als hun oplei-

ding hoger is en als ze optimistische verwachtingen hebben van hun teamgenoot. Een

hogere aversie tegen risico’s en zelfoverschatting leiden daarentegen tot een voorkeur

voor individuele beloning. Factoren als geslacht, IQ en beroepsgroep hebben verder

geen invloed. Als we de data van de andere variatie toevoegen in de regressie-analyses,

vinden we gek genoeg dat respondenten niet per se minder betalen voor teambeloning,

ondanks dat deze optie nu potentieel risicovoller is en daardoor potentieel minder

aantrekkelijk. We vinden dat er echter een uitzondering is op deze regel: onderne-

mers die namelijk denken dat hun teamgenoot teveel verschilt in hun risicohouding

van henzelf kiezen dan nu eerder voor individuele beloning. Laatstegenoemde resul-

taat kan een verklaring zijn voor het feit dat: (i) ondernemers in sommige situaties

graag alleen werken en (ii) venture teams vaak homogene risico voorkeuren hebben.

185

This dissertations is based on the following studies:

Chapter 2:

Risk, Uncertainy and Entrepreneurship: Evidence from a Lab-in-the-Field Experiment

Co-authors: Randolph Sloof and Mirjam van Praag

Published as: “Risk, Uncertainty and Entrepreneurship: Evidence from a Lab-in-The-

Field Experiment”, Management Science (forthcoming)

Contribution of the doctoral candidate: He is the main contributor of this study. He

was responsible for initiating the new research project, the design of the (follow-up)

survey and the econometric analyses. He also wrote substantial parts of this paper.

Chapter 3:

Are Entrepreneurs More Optimistic and Overcondent than Managers and Employees?

Co-authors: Randolph Sloof and Mirjam van Praag

Published as: Tinbergen Institute Discussion Paper 15-124/VII

Contribution of the doctoral candidate: He is the main contributor of this study. He

was responsible for initiating the new wave, the design of the survey and the econo-

metric analyses. He also wrote substantial parts of this paper.

Chapter 4:

Are Entrepreneurs More Intuitive Than Managers and Employees? Evidence From a

Lab-in-the-Field Experiment

Co-authors: Randolph Sloof and Mirjam van Praag

Unpublished manuscript

Contribution of the doctoral candidate: He is the main contributor of this study. He

was responsible for initiating the new wave, the design of the survey and the econo-

metric analyses. He also wrote substantial parts of this manuscript.

Chapter 5:

Risks, Gains and Autonomy: An Experimental Analysis of Preferences for Joining

Teams

Co-authors: Laura Rosendahl Huber and Eszter Czibor

Unpublished manuscript

Contribution of the doctoral candidate: He is one of the main contributors of this study.

He and his co-authors jointly formulated the research question and the design of the

survey. He was responsible for the implementation of the survey, the entire sampling

and payout process, and wrote substantial parts of this manuscript.

186

The Tinbergen Institute is the Institute for Economic Research, which was founded

in 1987 by the Faculties of Economics and Econometrics of the Erasmus University

Rotterdam, University of Amsterdam and VU University Amsterdam. The Institute

is named after the late Professor Jan Tinbergen, Dutch Nobel Prize laureate in eco-

nomics in 1969. The Tinbergen Institute is located in Amsterdam and Rotterdam.

The following books recently appeared in the Tinbergen Institute Research Series:

607 T.C.A.P. GOSENS, The Value of Recreational Areas in Urban Regions

608 L.M. MARC, The Impact of Aid on Total Government Expenditures

609 C.LI, Hitchhiking on the Road of Decision Making under Uncertainty

610 L. ROSENDAHL HUBER, Entrepreneurship, Teams and Sustainability: a

Series of Field Experiments

611 X. YANG, Essays on High Frequency Financial Econometrics

612 A.H. VAN DER WEIJDE, The Industrial Organization of Transport

Markets: Modeling pricing, Investment and Regulation in Rail and Road

Networks

613 H.E. SILVA MONTALVA, Airport Pricing Policies: Airline Conduct, Price

Discrimination, Dynamic Congestion and Network Effects

614 C. DIETZ, Hierarchies, Communication and Restricted Cooperation in

Cooperative Games

615 M.A. ZOICAN, Financial System Architecture and Intermediation Quality

616 G. ZHU, Three Essays in Empirical Corporate Finance

617 M. PLEUS, Implementations of Tests on the Exogeneity of Selected

Variables and their Performance in Practice

618 B. VAN LEEUWEN, Cooperation, Networks and Emotions: Three Essays

in Behavioral Economics

619 A.G. KOPANYI-PEUKER, Endogeneity Matters: Essays on Cooperation

and Coordination

620 X. WANG, Time Varying Risk Premium and Limited Participation in

Financial Markets

621 L.A. GORNICKA, Regulating Financial Markets: Costs and Trade-offs

622 A. KAMM, Political Actors playing games: Theory and Experiments

623 S. VAN DEN HAUWE, Topics in Applied Macroeconometrics

624 F.U. BRAUNING, Interbank Lending Relationships, Financial Crises and

Monetary Policy

625 J.J. DE VRIES, Estimation of Alonso’s Theory of Movements for

Commuting

626 M. POP LAWSKA, Essays on Insurance and Health Economics

187

627 X. CAI, Essays in Labor and Product Market Search

628 L. ZHAO, Making Real Options Credible: Incomplete Markets, Dynamics,

and Model Ambiguity

629 K. BEL, Multivariate Extensions to Discrete Choice Modeling

630 Y. ZENG, Topics in Trans-boundary River sharing Problems and Economic

Theory

631 M.G. WEBER, Behavioral Economics and the Public Sector

632 E. CZIBOR, Heterogeneity in Response to Incentives: Evidence from Field

Data

633 A. JUODIS, Essays in Panel Data Modelling

634 F. ZHOU, Essays on Mismeasurement and Misallocation on Transition

Economies

635 P. MULLER, Labor Market Policies and Job Search

636 N. KETEL, Empirical Studies in Labor and Education Economics

637 T.E. YENILMEZ, Three Essays in International Trade and Development

638 L.P. DE BRUIJN, Essays on Forecasting and Latent Values

639 S. VRIEND, Profiling, Auditing and Public Policy: Applications in Labor

and Health Economics

640 M.L. ERGUN, Fat Tails in Financial Markets

641 T. HOMAR, Intervention in Systemic Banking Crises

642 R. LIT, Time Varying Parameter Models for Discrete Valued Time Series

643 R.H. KLEIJN, Essays on Bayesian Model Averaging using Economic Time

Series

644 S. MUNS, Essays on Systemic Risk

645 B.M. SADABA, Essays on the Empirics of International Financial Markets

646 H. KOC, Essays on Preventive Care and Health Behaviors

647 V.V.M. MISHEVA, The Long Run Effects of a Bad Start

648 W. LI, Essays on Empirical Monetary Policy

649 J.P. HUANG, Topics on Social and Economic Networks

650 K.A. RYSZKA, Resource Extraction and the Green Paradox: Accounting

for Political Economy Issues and Climate Policies in a Heterogeneous

World

651 J.R. ZWEERINK, Retirement Decisions, Job Loss and Mortality

652 M. K. KAGAN, Issues in Climate Change Economics: Uncertainty,

Renewable Energy Innovation and Fossil Fuel Scarcity

653 T.V. WANG, The Rich Domain of Decision Making Explored: The

Non-Triviality of the Choosing Process

188

654 D.A.R. BONAM, The Curse of Sovereign Debt and Implications for Fiscal

Policy

655 Z. SHARIF, Essays on Strategic Communication

656 B. RAVESTEIJN, Measuring the Impact of Public Policies on

Socioeconomic Disparities in Health

189