Social Identity and Incentives

in Workgroups ∗

Jonathan Yeo†

University of Warwick

June 28, 2019

Abstract

Incentive theory has paid relatively little attention to workers’ identities.

In this paper, I conduct the first experiment exploring the relationship be-

tween identity and optimal incentives. I construct workgroups which are either

homogeneous or heterogeneous in members’ identities and compare their pro-

ductivity at a real-effort task under tournaments or team-pay. I find that in

homogeneous workgroups, productivity is higher under team-pay. In hetero-

geneous workgroups, productivity is however similar across incentives. This

is because identities and incentives interact to influence inputs. Generally,

team-pay encourages helping — more so in homogeneous workgroups — while

tournaments encourage personal effort — less so in homogeneous workgroups.

Furthermore, I find that incentives also influence and shape workers’ identities.

Keywords: Social Identity, Minimal Groups, Incentive Structures, Cooper-

ation, Real-effort

∗I thank the College of Humanities, Arts, and Social Sciences (HASS) International PhD Schol-arship (HIPS) at Nanyang Technological University, Singapore for funding in this research.†Department of Economics, University of Warwick, Coventry, CV4 7AL, United Kingdom, e-mail:

j.x.w.yeo@warwick.ac.uk.

1

1 Introduction

There is a large literature in economics on optimal worker incentives. A factor which

has received relatively little attention within this literature — but seems to be im-

portant in practice — is workers’ identities. For instance, at Nucor Steel, where

employees share a strong sense of common identity, group-based incentives comprise

a large fraction of workers’ compensation (66%).1 By comparison, group-based incen-

tives comprise less than 20% of overall earnings at US Steel, where employee identity

is more fragmented (Byrnes and Arndt, 2006).2 Relatedly, unlike individualist coun-

tries (e.g. US), collectivist countries (e.g. Japan) display a strong association between

group-based incentives and organisational performance (Allen et al., 2004).

In this paper, I conduct the first experiment exploring the relationship between

optimal incentives and workers’ identities. I systematically induce identities in the

laboratory by assigning participants to one of two groups using an implementation

similar to Chen and Li (2009). They are then placed in workgroups which are either

homogeneous or heterogeneous in members’ identities. I examine productivity of

each workgroup type under two different incentive schemes: tournament or team-pay.

Participants engage in a task where one’s productivity depends on personal effort as

well as help from fellow members in their workgroup.

Results confirm that identities do influence optimal incentives. I find that in

homogeneous workgroups, productivity is higher under team-pay. In heterogeneous

workgroups, productivity is however similar across incentives. These productivity dif-

ferences can be explained by the interaction between identities and incentives in influ-

encing inputs. Generally, team-pay encourages helping, while tournaments encourage

personal effort. However, team-pay stokes greater help in homogeneous workgroups

while tournaments stoke greater effort in heterogeneous workgroups.

Results also indicate that incentives further influence and shape workers’ identi-

ties. Utilising post experiment survey data, I find that under team-pay, participants

feel greater identification with their workgroup. This hints at a feedback process

1A quote from a frontline supervisor exemplifies this: “At Nucor, we’re not ’you guys’ and’us guys’. It’s ’all of us’ guys. Wherever the bottleneck is, we go there, and everyone works onit.” (Byrnes and Arndt, 2006). Employees even refer to each other as “teammates” (B Arthur,1999). Production employees at Nucor are organised into teams including maintenence workers andsupervisors and have bonuses tied to team production which can be up to 200 percent of their basesalary. This is in addition to earnings from a profit sharing program which further increases theteam component of earnings (Sheridan, 1998; Vasanthi and Chowdary, 2009; Bakshi, 2015).

2Boyd and Gove (2000) mention the dramatic differences of culture at competitors of Nucor, withworkers having an ”us vs them” mentality, fear and distrust.

2

between incentives and identities which might be important in determining optimal

incentives, especially in dynamic settings.

The structure of the paper is as follows. Section 2 summarises the literature

related to the paper. Section 3 describes the experimental design and procedures.

This is followed by a simple model to generate predictions in Section 4. Section 5

discusses the experimental results, and Section 6 concludes.

2 Related Literature

There is an existing experimental literature in economics on identity (see for example

Butler, 2014; Goette et al., 2006, 2012; Heap and Zizzo, 2009; Mcleish and Oxoby,

2007).3 A robust finding of this literature is that identity affects agent’s preferences.

For example, Chen and Li (2009) find that subjects are more charitable and reciprocal

towards members of their ingroup.

Several papers have explored the impact of identity on productivity.4 For in-

stance, Hoff and Pandey (2006) find that, when identity is made salient, individuals

of lower caste perform worse in a maze-solving task. Chen and Chen (2011) show that

individuals exert higher effort in a minimum-effort game when paired with ingroup

members. Kato and Shu (2016) find that common identity makes workers less com-

petitive, while Hamilton et al. (2012) find that common identity stokes teamwork.5

While these papers vary identities, they do not vary incentives. Hence, they do not

speak to this paper’s main question of how identities affect optimal incentives.

More closely related are papers by Bandiera et al. (2005) and Hamilton et al.

(2003) who compare workers’ productivity under a change in incentive regime.6

Bandiera et al. find a cost of relative-pay while Hamilton et al. find a benefit of team

pay when workers are “groupy”. My contribution is twofold. Firstly, I synthesise

3A psychology literature on identity starting with Tajfel et al. (1971) argues that group mem-bership can influence behaviour. More recently, there has been a growing economics literature onidentity beginning with Akerlof and Kranton (2000). See Akerlof and Kranton (2010) for a summary.

4More generally, this is related to the idea of social incentives and job-meaning. See Ashraf andBandiera (2018) and Cassar and Meier (2018) respectively for a review of the related literature.

5Under relative pay, Kato and Shu find that in Chinese textile firms with rural-urban distinctionsamong workers, productivity is higher in the presence of more able out-group workers, but not in-group ones. Under team pay, Hamilton et al. find that workgroups at a garment factory in Napa,California are more productive when composed of a single (Hispanic) ethnicity; Afridi et al. (2018)find similar results in a lab-in-field experiment in India.

6Bandiera et al. examine a switch from relative performance pay to piece rates at a UK farm.Hamilton et al. examine a switch from piece rates to team pay at a US garment factory.

3

their work and show that “groupiness” — whether workers share or lack a common

identity — can affect the optimality of relative-pay versus team-pay.7 To the best of

my knowledge, my study is the first to cleanly demonstrate that optimal incentives

may be influenced by workers’ identities.8,9 Secondly, I measure two distinct inputs:

personal effort and help; this helps provide a clearer picture of the interaction between

identity and incentives in production. I show that under team pay, the benefits of

common identity derive largely from cooperation. Under relative-pay, the costs of

common identity however derive from personal effort.

Finally, there is some theoretical work which speaks to the issues in this paper.

Most related is Huck et al. (2012) who show that a concern for others in the workgroup

can raise the effectiveness of team performance pay — and decrease that of relative

performance pay. Other work discusses how strong identification with an organisa-

tion’s mission can reduce the need for high-powered incentives (Akerlof and Kranton,

2005, 2008; Besley and Ghatak, 2005; Henderson and Van Den Steen, 2015).10

3 Experimental Design

The experiment is designed to examine the effect of identity within workgroups on op-

timal incentives. Participants perform a real-effort task under a 2×2 between-subject

design as in Table 1. On one dimension, the identity composition of workgroups is

varied: subjects are assigned either to a homogeneous workgroup or a (maximally)

heterogeneous workgroup. On the second dimension, the incentive scheme faced by

the workgroup is varied: a tournament scheme or a team-pay scheme.

7How groupiness is measured differs across their papers and mine. In Bandiera et al. (2005), itdepends on the number of co-workers in one’s workgroup. In Hamilton et al., it depends on howearly the workgroup is formed after the introduction of team-pay.

8Blazovich (2013) in the management accounting literature has a design which may in principalexamine this, but is unable to draw conclusive results due to the low sample size.

9Experiments in the management literature which examine how collectivism and individualismimpact group versus individual incentives (Naranjo-Gil et al., 2012; Papamarcos et al., 2007) arealso related to the extent that these traits affect identity in groups (Chatman et al., 2019). Moregenerally, other research has studied the impact of identity on the relative effectiveness of differentmanagement schemes like imposing control (Masella et al., 2014; Riener and Wiederhold, 2016),punishment schemes (Weng and Carlsson, 2015) and reporting structures (Towry, 2003). This paperdiffers from these studies in its focus on pecuniary performance incentives.

10These papers however differ in their modelling of how identification occurs. Akerlof and Kranton(2005, 2008) view identity as being malleable within the organisation. Besley and Ghatak (2005)discuss how assortative matching of workers and organisations naturally occurs under competitivepressure from larger surplus in such matches. Henderson and Van Den Steen (2015) instead describehow workers self select into firms with pro-social purposes due to identity and reputation benefits.

4

Workgroup composition Incentive scheme Sample sizeHomogeneous Tournament 78Homogeneous Team-Pay 78Heterogeneous Tournament 72Heterogeneous Team-Pay 78

Table 1: Treatment Descriptions

A novel real effort task mimicking on-the-job-help at the workplace is utilised.

In particular, it involves both individual effort and cooperative interaction between

workgroup members. Unlike past research on team production which involve in-

duced costs, it allows for the measurement of both personal effort and help decisions

in a more realistic interaction environment. Together with the choice of incentive

treatments, this design speaks to the multi-dimensional nature of production. In par-

ticular, it seeks to address the dichotomy between motivating effort and cooperation

under two commonly debated incentives schemes in the literature.11

Each session consists of two main stages. In the first stage, near-minimal identities

are induced in subjects. In the second stage, subjects play the real-effort task in their

respective treatments. These are elaborated on in the next two subsections.

3.1 Stage 1: Identity Inducement

In the first stage of the experiment, identity is induced in subjects by adopting the

near-Minimal Group Paradigm.12 Subjects review 5 pairs of paintings sequentially,

each pair containing a painting by Paul Klee and another by Wassily Kandinsky.

They are required to select which painting they prefer without any information about

them. Subsequently, subjects are categorised into two equal sized groups: Klee and

Kandinsky according to their relative preferences within the session.13

11Tournament incentives have been described to provide superior incentives to outperform and ex-ert effort while team-pay based on aggregate team performance instead result in free riding (Lazearand Rosen, 1981; Holmstrom, 1982; Bull et al., 1987; Hannan et al., 2008). On the flip-side, tour-nament incentives have been found to encourage negative “side activities” like sabotage (Falk et al.,2008; Carpenter et al., 2010; Charness et al., 2013) while the converse occurs for team-based incen-tives (Drago and Garvey, 1998; Buser and Dreber, 2015; Friebel et al., 2017; Lazear, 2018).

12This was used in Chen and Li (2009) who adapt the procedures in Tajfel and Turner (1979).13Subjects in each session are ranked by the number of times they prefer paintings by Paul Klee.

The top(bottom) half is subsequently assigned to Group Klee(Kandinsky). This allows for a balancedassignment of groups compared to the standard procedure of using absolute preferences. Since most

5

Subsequently, group members complete a joint task in order to build up their

group identity salience. They are shown a final pair of paintings and asked to guess

the artist of each individually after a 5 minute anonymous discussion with their group

members.14 If the majority of their group gets the answer correct, they are awarded

with 80 Experimental Currency Units (ECUs). Success/failure in this task is only

revealed at the end of the experiment to prevent it from influencing identity salience.

3.2 Stage 2: Production in workgroups

In the second stage of the experiment, subjects complete 4 rounds of 6 minutes of a

real-effort task.15 This is done in randomly assigned workgroups of 6 subjects with

partner matching. There are two types of workgroups: Homogeneous workgroups with

all 6 members from Klee/Kandinsky and Heterogeneous workgroups with 3 members

each from Klee and Kandinsky.16

Their individual scores as well as workgroup members’ scores in the task deter-

mine their task payment in each round. In particular, each workgroup is randomly

assigned to one of two incentive schemes (see Table 2) which determines their Task

payment. In the Tournament scheme, piece rates for the round are based on individ-

ual scores and increase with one’s score rank (ties were broken at random). In the

Team-pay scheme, piece rates are based on the average workgroup score; everyone

thus receives the same payment.17 To eliminate effects from feedback about own and

others’ past performance, relevant information about payments is only displayed at

the end of the experiment. This involves the (anonymous) breakdown of scores in the

workgroup under tournaments and the total score of the workgroup under team-pay.

subjects indicated non-familiarity with both artists (on a 7 point Likert scale, average of 1.27 forKlee and 1.45 for Kandinsky), differences between the two procedures should not be great. Balancechecks of baseline characteristics across the groups (unshown) confirm this.

14The initial five pairs of paintings are: 1A Gebirgsbildung, 1924, by Klee; 1B Subdued Glow,1928, by Kandinsky; 2A Dreamy Improvisation, 1913, by Kandinsky; 2B Warning of the Ships,1917, by Klee; 3A Dry-Cool Garden, 1921, by Klee; 3B Landscape with Red Splashes I, 1913, byKandinsky; 4A Gentle Ascent, 1934, by Kandinsky; 4B A Hoffmannesque Tale, 1921, by Klee; 5ADevelopment in Brown, 1933, by Kandinsky; 5B The Vase, 1938, by Klee. The last pair of paintingsare: 6A Monument in Fertile Country, 1929, by Klee, and 6B Start, 1928, by Kandinsky.

15Subjects are not informed of the exact number of rounds to reduce end-game effects.16Subjects only know their own workgroup composition and are not informed of other possible

workgroup compositions to reduce experimenter demand effects.17Note that the payoff structure for both incentives are designed such that for a given performance,

a participant who perceives that individual scores are drawn from the same distribution (i.e. abilityand motivation for effort are the same) has the same expected payoff in both; this allows us to makea direct comparison of productivity between the treatments.

6

Incentive scheme Task Payment Piece-rate(ECU) Information provided

Tournament [1st: 13, 2nd: 11.5, 3rd: 10 Breakdown of4th: 10, 5th: 8.5, 6th: 7] individual scores

(Based on individual score)

Team-Pay [Everyone: 10] Total score(Based on average workgroup score) of workgroup

Table 2: Details of incentive treatments

3.2.1 Real Effort Task with on-the-job help

The real-effort task involves decoding sets of letters into two digit numbers from

a given letter-number grid displayed on their computer screen.18 Subjects have to

correctly complete a set of decoding questions in order to receive a new question set

(with a different letter-number grid). Their score is the total number of question sets

correctly completed during the round.

An element of cooperation is also present in the real-effort task. In particular,

choices of help by workgroup members determine the number of decoding questions

to solve in each question set. Each time a question set is generated, subjects face

either a difficult or easy question set with equal probability. Difficult question sets

contain 7 questions while easy question sets contain 3. In difficult sets, help requests

are automatically sent to a random workgroup member with known group identity.19

If the help request is accepted, the requester’s task becomes easier: the number of

questions is reduced by 3. In return, the helper has to solve 1 extra question.20

Nothing happens if the help request is rejected other than being informed.

Decisions on whether to provide help are elicited via the strategy method at

the beginning of each round. In particular, subjects have to decide for each group

(Klee/Kandinsky), whether to provide help throughout that round. For homogeneous

groups, the out-group decision is elicited hypothetically.

To increase the salience of “leisure” as an alternative to completing the task and

improve the measurement of effort, subjects are given the option of taking paid rests.

18Examples of papers which used this task are Benndorf et al. (2014) and Charness et al. (2013).19Note that they cannot discover the exact identity (within their workgroup) of whom they interact

with because there are ≥ 2 members from each group: this prevents reputation building.20This implies that help is efficient; we take this to represent a situation where workgroup members

perform related tasks where they occasionally encounter problems which would be more efficientlysolved with coordination of effort, thus raising (net) earnings.

7

When submitting their answers, subjects can choose to rest for 5 seconds before the

next question set is displayed. If they do so, they are paid 2.5 ECUs.21 Their Total

payment in each round is hence the sum of their Task and Rest payments.

3.3 Procedures

At the beginning of each session, detailed printed instructions are provided to subjects

and read aloud to them.22 This is followed by a control quiz to test their understanding

of the real-effort task: all questions have to be answered correctly before they can

proceed.23 Subsequently, there are 2 non-incentivised practice rounds: a first round of

2 minutes to acquaint them with the UI, and a second round of 3 minutes to practice

the task, without the help and rest functions.24 The first and second stages of the

experiment as described then follow.

At the end of each session, subjects are asked to complete an online post-experiment

survey covering questions on demographics, academic background, behaviour and

perceptions about their groups and workgroups. A summary of their payoffs is then

shown, with subjects paid for 1 randomly chosen round. The exchange rate is set to

100 ECUs for 2.5 Sterling Pounds.

Participants were invited using the Warwick SONA system and were primarily

students at the university from a variety of majors. In total, 18 sessions were con-

ducted at the Economics Lab at the University of Warwick from Nov 2017 to Nov

2018, for a total of 306 participants (51 workgroups).25 Each session lasted 75 min-

utes on average and participants earned £10 on average including a show up fee of

£3. The study was programmed in z-Tree (Fischbacher, 2007) and preregistered at

the AEA RCT registry (AEARCTR-0002139).

21Using piece rates without an outside option in short term tasks may cause task completion to betoo salient, preventing adjustments to effort. The rate was chosen such that it would be individuallyrational to rest under team-pay, or low expected rank under tournaments. This was based on pilotswhere decoding 1 letter took 3.5 seconds on average and that there is on average 4 to 5 letters todecode in each set. In one session, the rest rate was lowered to 2 ECUs (which also satisfies theabove criteria), but no differences in behaviour were observed; The data is hence pooled for analysis.

22See Appendix B for the instructions provided to participants.23Over 90% of them stated an understanding of ≥ 5 on a scale from 1 to 7, see Figure A.1.24The second practice round also serves to provide a measure of their baseline ability.25Sessions either had 18 or 12 participants. Those with 18 had either 3 heterogeneous workgroups

or 1 heterogeneous and 2 homogeneous workgroups. Those with 12 had either 2 homogeneousworkgroups or 2 heterogeneous workgroups. The treatments conducted were pre-randomised tobalance the number of treatments across sessions. Figure A.1 in Appendix A illustrates that baselinecharacteristics of participants are roughly balanced across treatments.

8

4 Model

In this section, I utilise a simple model to generate predictions about how salience of

workgroup identity salience, as influenced by the treatments, affects the optimality

of tournaments vs team-pay. In this model, individuals generate output within a

workgroup and are paid based on an exogenously chosen incentive scheme.

The production process is as follows: each member chooses 1) ei ≥ 0: effort put

into their own job, and 2) hi ≥ 0: the amount of help provided to the rest of the

workgroup. Personal output depends linearly on own effort and the help provided

by others: qi = ei + α∑

j 6=i hj where α > 0 here reflects the efficiency of help.26 To

simplify notation, I define the set of i′s chosen actions as ai = {hi, ei}. Furthermore,

I denote vectors of the workgroup’s variables in capitals (e.g. A,Q,E,H).

I examine a class of linear incentives where payment depends on own output (qi)

and others’ average output (q−i):27

Fi(Q) = wqi + (b− w)q−i, w ≥ 0, b > 0.

Here, b > 0 is treated as a fixed parameter throughout the model. w can be interpreted

as the degree of competitiveness of the incentive scheme. This covers a range of

incentives from team-pay (w < b) to piece rate (w = b) to tournament-like incentive

schemes (w > b).28 Keeping with our experimental aims, we seek to compare a

tournament (w1) and a team-pay incentive scheme (w0) with w0 < b < w1.

Individual’s choices are determined by their utility structure which comprises material

utility and social utility.29 Social utility is weighed by a factor β — a concern for

others in the workgroup — which reflects the salience of workgroup identity.30

Ui(E,H) = ui(E,H) + βvi(E,H),

β ≥ 0

26This simplifies exposition by ignoring complementarity between help received and effort whichcreates complicated interactions; results should be similar if the complementarity is small.

27In a symmetric equilibrium, average payment per unit output is independent of w.28This is not exactly the same as the experiment where piece rates in tournaments depend on a

ranking structure, but qualitatively, it should have the same implications.29This adapts the model in Huck et al. (2012) which also has material and social utility, but with

the latter depending on deviations from a social ideal. This accounts for preferences to follow socialnorms.

30In the context of our experiment, the main assumption here is that there are greater concerns forothers in a homogeneous workgroup compared to a heterogeneous workgroup. Chen and Li (2009),Chen and Chen (2011) and Mcleish and Oxoby (2007) have a similar modelling assumption. Infootnote (46), I discuss how allowing incentives to affect identity as well gives similar results.

9

Material utility is linear in the material payoff from the exogenous incentive structure

Fi, with help and effort costs being quadratic:

ui(E,H) = Fi(Q)− 12e2i − 1

2h2i

Social utility is instead increasing in the “kindness” which i shows to others:

vi(E,H) = g(∑

j 6=i

[uj(ei, A\ei)− uj(eS, A\ei)

])+g(∑

j 6=i

[uj(hi, A\hi)− uj(hS, A\hi)

]),

g′ > 0, g′′ ≤ 0.

Kindness here is defined as the impact of one’s actions on others’ utility relative to

a reference point where the (unrestricted) selfish action is chosen, holding others’

actions constant.31,32 It is reflected in the bracketed terms within g( ), where eS, hS

refer to the corresponding unrestricted selfish action when β = 0. Notice that social

utility is assumed to be additively separable in kindness of chosen help and effort.

This can be justified if there is no complementarity of these actions in material utility

which is indeed true.33

It is easy to show that the unrestricted selfish amount of effort and help provided

is eS = w, hS = α(b − w), which is intuitive. Compared to individual piece rates,

team-pay (tournaments) elicit lower (higher) effort, due to the positive (negative) ex-

ternalities not being internalised. Help is higher (lower) for team-pay (tournaments)

than in a piece rate for a similar reason.

31This assumption means that for linear production and incentives, but possibly non-linear g, theactions of other members disappears in the optimisation. It helps to simplify the Nash Equilibriumand hence exposition. Removing/changing this reference point should not change theoretical resultsmuch as long as kindness is linear in it.

32The assumption about selfish actions being unrestricted is not essential for the main results,but helps to simplify presentation. It can be interpreted as structural constraints (here being thenon-negativity constraints) not being relevant on how i judges his kindness towards others.

33In the case of linear incentives and production, kindness of each action is indeed separable:uj(E,H)− uj(es, hs, e−i, h−i) = [uj(E,H)− uj(eS , e−i, H)] + [uj(E,H)− uj(hS , h−i, E)].

10

Throughout the model, we also assume the following simple functional form for social

utility: 34

g(z) =

log(z), if z > 0

−∞, if z ≤ 0

Consequently, the solution for the general case (β > 0) is:

ei =

w +√β, if w < b (Team-Pay)

max(w −√β, 0), if w > b (Tournament)

hi = max(α(b− w) +√β, 0)

The following predictions can then be readily observed from comparative statics of

the relative competitiveness of the incentive scheme (w):

Prediction 1. Keeping identity salience fixed, help weakly falls as the relative com-

petitiveness of the incentive increases.

Prediction 2. Keeping identity salience fixed, effort rises as the relative competitive-

ness of the incentive increases.

The comparative statics of identity salience (β) give the following predictions:

Prediction 3. Help rises with identity salience in both team-pay and tournaments.

Prediction 4. Effort rises (falls) with workgroup identity salience in team-pay (tour-

naments). This implies that the effort difference between a fixed tournament and

team-pay incentive (eTourn− eTeamPay) should fall as team identity salience increases.

Notice that in Predictions 3 and 4, while the effects of team identity salience on

effort and help are in the same direction for team-pay, they are opposed in tour-

naments. Since output is the sum of help and effort, output rises with workgroup

identity salience in team-pay but rises at a lower rate (or even falls) in tournaments.35

34Effectively, this means that actions in social utility are not ”judged” by their prices (the incentiveparameters) which might reflect a deontological perspective of sorts.

35Computing the expression for output of each individual, we have:

qi =

w +√β + α(n− 1)[α(b− w) +

√β], if w < b

α(n− 1)[α(b− w) +√β], if w > b, β > w2

w −√

(β) + α(n− 1)[α(b− w) +√β], if w > b, [α(w − b)]2 < β < w2

w −√β, if w > b, β < [α(w − b)]2

11

This gives the following prediction for output:

Prediction 5. Output rises with workgroup identity salience in team-based incentives,

but changes ambiguously in tournament incentives. The output difference between a

fixed tournament and team-based incentive (qTourn−qTeamPay) should fall as workgroup

identity salience increases.

Further, this implies a threshold β where for β > β, team incentives outperform

tournament incentives.36 This threshold increases with the relative competitiveness

of the tournament scheme (i.e. w1−w0), but decreases in the efficiency of help a. The

latter implies that as cooperation becomes more important, it is easier for team-pay

to outperform tournaments, which is intuitive.

The above predictions outline what we should observe assuming that identity

composition of the workgroup influences identity salience. However, whether we ob-

serve a (clean) reversal of the effectiveness of tournament versus team-pay in the

experiment depends on the levels of workgroup identity salience in homogeneous and

heterogeneous workgroups relative to β; in particular we need βhet << β << βhom.

5 Results

In this section, I first present a comparison of productivity under each treatment. This

is followed by an examination of its determinants: 1) help choices and 2) personal

effort. Lastly, I examine how the salience of their workgroup identity is affected by

incentives and workgroup composition. This identity salience is subsequently used in

robustness checks of the previous results.

Analysis here focuses on the balanced round level panel data set (4 Rounds × 306

Participants= 1224 Observations).37 Several common features apply throughout the

analysis. First, in all bar-plots, error bars are 95% confidence intervals. Second, all

regressions control for session fixed effects while errors are clustered at the workgroup

level to control for any dependency of decisions within workgroups.

36β = max([w1−w0

2 (1− α2(n− 1))], [ (w1−w0)−α(n−1)(b−w0)2+α(n−1) ], 0)

This arises because the relative benefit of tournament incentives in incentivising higher effort fallsas workgroup salience rises, while help rises at an equal pace (for this functional form)

37In some rounds, there were some issues with the server which caused some disconnections midwaythrough, we however still have data for the round before the disconnections. Activity for the roundwas aggregated in these cases by adjusting for the fraction of time which the participants wereconnected. Overall, only 19 data points were affected: excluding them does not influence results.

12

5.1 Productivity

In this experiment, an outcome of interest is whether optimal incentives can be influ-

enced by identity within the workgroup. Figure 1 illustrates productivity across treat-

ments as measured by the average number of question sets completed per round.38

Figure 1: Productivity over Treatments

It can be seen that in homogeneous workgroups (left), productivity is higher under

team-pay.39 In heterogeneous workgroups (right) however, productivity is indistin-

guishable under the two incentive schemes. This is consistent with Prediction 5 which

concerns how output changes with identity under each incentive scheme. Next, I per-

form random effects regressions for a more rigorous analysis.

Table 3 presents regressions with standardised productivity as the dependent vari-

able. The results in Model A are consistent with that observed in the figure. It shows

that team-pay outperforms tournaments in homogeneous workgroups, with a 0.3 sd

increase in productivity. The negative interaction term however, indicates that the

positive effect is cancelled out in heterogeneous workgroups. Models B and C show

that help choices and effort alone do not explain the productivity differences. Both

factors are required to be included for estimated effects of treatment variables to be

close to 0 (Model D). These are summarised below.

Result 1. (Identity and Optimal Incentives): In homogeneous workgroups, team-pay

outperforms tournaments. In heterogeneous workgroups, there is no difference between

incentives. Both worker inputs: help and effort matter for this difference.

38The values are adjusted for ability for which there was some imbalance between treatments.39This is significant under a Mann-Whitney test: p-value<0.001

13

Dep var: Standardised Productivity Model A Model B Model C Model DTeam-Pay 0.306*** -0.127 0.444*** 0.014*

(0.076) (0.087) (0.036) (0.009)Heterogeneous -0.089 0.010 -0.108** -0.001

(0.079) (0.081) (0.047) (0.008)Team-Pay × Heterogeneous -0.356*** -0.178 -0.176*** -0.009

(0.105) (0.108) (0.058) (0.009)Avg questions in set -1.019*** -1.007***

(0.041) (0.017)Effort 0.904*** 1.780***

(0.024) (0.058)Avg questions in set × Effort -0.214***

(0.013)Observations 1224 1224 1224 1224

* 0.10 ** 0.05 *** 0.01, Avg questions is an inverse measure of cooperation.All regressions here control for ability, period effects and their interaction.

Table 3: Random Effects Regressions of Productivity

Above, both choices of personal effort and help are shown to matter for produc-

tivity differences across treatments. To further quantify the extent to which each

worker input matters, I first simulate counterfactual productivity for each incentive

type, given all combinations of effort and help displayed in each workgroup type; this

is displayed in Table 4.40 Subsequently, I use these values to estimate the contri-

bution to output differences from variation in effort and help over the treatments.41

This procedure allows for an estimate of their contribution to workgroup productivity,

accounting for non-linearity in production.

Table 5 shows the results of the above exercise. Under team-pay, cooperation

explains a majority of around 7/10 of output differences across incentives while under

tournaments, this drops to around half. Incentives thus interact with identity to

influence the importance of cooperation relative to effort. In the next 2 sub-sections, I

examine in greater detail how identity influences the inputs which consist production.

40I utilise the production function: Productivity=(360×Effort)/(Average questions per set). Effortis the rate at which a question is solved while 360 refers to the total time in seconds for each round.

41For each workgroup composition, with the lowest counterfactual output as a baseline, the ex-planatory power of each factor is calculated as how much a change in effort or help increases outputas a fraction of the difference between the largest and lowest counterfactual output. With the highestcounterfactual output as a baseline, the explanatory power of each factor is calculated as how mucha change in effort or help decreases output as a fraction of the difference between the largest andlowest counterfactual.

14

Input Hom Help, Hom Help, Het Help, Het Help,Incentive Hom Effort Het Effort Hom Effort Het EffortTournament 21.793 21.965 21.571 21.742

(0.189) (0.190) (0.188) (0.183)Team-Pay 23.252 22.746 22.003 21.525

(0.194) (0.188) (0.189) (0.183)

Standard errors approximated by delta method.

Table 4: Simulated counterfactual productivity in each incentive type

Team-Pay TournamentBaseline: Lowest Baseline: Highest Baseline: Lowest Baseline: Highest

Effort 0.277 0.292 0.433 0.438Help 0.708 0.723 0.562 0.567Total 0.985 1.015 0.995 1.005

Total does not sum up to 1 because of some complementarity between the two inputs.

Table 5: Proportion of counterfactual productivity differences explained

5.2 Help Choices

Here, I examine how identity and incentives influence participant’s choices to help and

hence cooperation within groups. Figure 2 illustrates average Out/In-group choices

to help and measures of workgroup cooperation over the 4 rounds.

Examining choices to help (top-left), we can see an in-group bias across all treat-

ments: participants on average choose to help in-group members more. Furthermore,

as expected, choices to help are more prevalent in team-pay compared to tourna-

ments.42 Subsequently, homogeneous workgroups exhibit higher levels of cooperation,

especially under team-pay.43 Members of homogeneous workgroups are more likely

to receive help (top right), give help (bottom left) and have less questions to solve

(bottom right), with bigger effects under team-pay.44 These observations are qualita-

tively in line with Predictions 1 and 3. I next perform a more rigorous analysis using

regressions in a random effects/discrete choice framework.

42It is interesting to note that those in tournaments also felt that their help would be less beneficialfor the team; this may indicate some form of motivated thinking. See Figure A.1. in Appendix A.

43Except for provision of help and average questions under tournament incentives for which dif-ferences are smaller, homogeneous workgroups always have statistically significant higher levels ofcooperation (Mann-Whitney test, p-value<0.05).

44Questions to solve controls for the rate of receiving help requests which is higher for those whomsolve questions slower and thus affects average questions in a way unrelated to rates of giving andreceiving help.

15

Figure 2: Help choices and cooperation over Treatments

Dep var: Chose to help Random Effects Panel Probit Panel LogitIngroup 0.278*** 1.023*** 1.774***

(0.042) (0.130) (0.230)Team-Pay 0.213*** 0.849*** 1.473***

(0.060) (0.250) (0.440)Ingroup × Team-Pay -0.009 0.248 0.450

(0.057) (0.202) (0.363)Homogeneous -0.071 -0.228 -0.384

(0.051) (0.228) (0.401)Ingroup × Homogeneous -0.018 0.001 -0.010

(0.050) (0.172) (0.308)Team-Pay × Homogeneous -0.163** -0.665* -1.205**

(0.078) (0.349) (0.614)Ingroup × Team-Pay × Homogeneous 0.351*** 1.550*** 2.863***

(0.083) (0.450) (0.808)Constant 0.378*** -0.418 -0.729

(0.080) (0.407) (0.697)Observations 2448 2448 2448

* 0.10 ** 0.05 *** 0.01. Out-group decisions for homogenous workgroups are hypothetical.All regressions here control for ability, period effects and their interaction.

Table 6: Regressions of decisions to help

16

In Table 6, I first examine choices to help by participants under several estimation

methodologies. I focus on the effects of whether the help request is from an ingroup

member, the incentive scheme and their interaction. To account for the fact that out-

group contributions are hypothetical in homogeneous workgroups, these 3 factors are

interacted with a dummy for them. In heterogeneous workgroups (first 3 coefficients)

team-pay and sharing common identity improve the likelihood of providing help while

there is no significant interaction between them. In homogeneous workgroups (next 4

coefficients), participants are less likely to help outgroup members, but more likely to

help ingroup members under team-pay although the former is unincentivised. These

results, summarised below, are robust to different estimation methodologies.

Result 2. (Identity, incentives and choices to help) Participants are more likely to

help under Team-Pay. They also display an ingroup bias in choices to help.

Table 7, columns 1 to 3 use random effects regressions to examine the impact of

choices to help on three different measures of cooperation in the workgroup: the prob-

ability of receiving help, giving help and average questions to solve.45 They confirm

that team-pay raises cooperation, while being in a heterogeneous workgroup reduces

cooperation under both incentives. The difference between Team-Pay and Tourna-

ments in heterogeneous workgroups is however significantly negative, indicating a

smaller cooperation gap between incentives in heterogeneous workgroups.46 Round

effects also indicate a fall in cooperation more prevalent in tournament incentives.47

These results are summarised below:

Result 3. (Identity, incentives and cooperation in workgroups) Team-Pay improves

cooperation in workgroups. However, cooperation in workgroups is lower in heteroge-

neous workgroups under both incentives.

Table 7, columns Columns 4 to 5 instead examine participants’ internal and ex-

ternal motivations to help.48 Results are mostly qualitatively similar, though weaker.

45Ability is not controlled for here since it is found to be inconsequential on aggregate earlier.However, the adjusted measure of average questions indirectly controls for this by equalising theprobability of receiving a help request across individuals.

46The model does not capture this as workgroup identity salience is not affected by incentives:amplifying it β by a factor k > 1 when incentives are team-based (w < b0) in the model wouldcapture this and keep comparative statics of workgroup identity salience the same.

47We have excluded interactions with identity as they were found to be inconsequential.48Motivation to help was obtained from a factor analysis (with 2 principal component factors) of

6 survey questions aimed at eliciting motivation to help coming from the self (internal) vs others(external), see Appendix A. Factor loadings of each question in general confirmed this distinction.

17

Dep-Var: Probability: Probability: Avg Qns Motivation: HelpRec Help Give Help (adj) Internal External

Team-Pay 0.478*** 0.482*** -0.469*** 0.495*** 0.672***(0.047) (0.049) (0.049) (0.088) (0.117)

Tournament × Heterogeneous -0.083* -0.086* 0.084* -0.052 0.177(0.043) (0.045) (0.044) (0.131) (0.150)

Team-Pay × Heterogeneous -0.270*** -0.276*** 0.271*** -0.101 -0.299*(0.040) (0.041) (0.041) (0.108) (0.170)

Tournament × 1/Round 0.198*** 0.203*** -0.200***(0.058) (0.058) (0.057)

Team-Pay × 1/Round 0.043 0.056 -0.061*(0.032) (0.035) (0.036)

Constant 0.518*** 0.517*** 4.500*** -0.126 -0.326(0.097) (0.097) (0.094) (0.150) (0.286)

Observations 1224 1224 1224 306 306

* 0.10 ** 0.05 *** 0.01, Columns 1-3 are Random effects regressions, Columns 4-5 are OLS.Avg Qns here adjusts for differences in help requests received. Motivation is standardised.

Table 7: Regressions of measures of cooperation

In general, identity has an insignificant impact on internal compared to external mo-

tivation. However, while members of heterogeneous workgroups face lower external

pressure to help under team-pay, an opposite contradictory effect occurs under tour-

nament incentives, albeit insignificant; this may reflect an intensive margin effect of

having the ability to discriminate help between the in and out-group.

5.3 Personal Effort

In this subsection, I examine how identity and incentives influence choices to exert

personal effort. Figure 3 graphs barplots (and kernel densities) of various measures of

personal effort across the treatments. Here, effort (adjusted for ability) is measured

as the rate of solving a single decoding question within a question set.49

Consistent with a free rider effect and Prediction 2, participants under team-pay

rest more proportionately (top-right) and have lower levels of overall effort compared

to tournaments (top-left/bottom).50 Also, overall effort is higher (lower) in heteroge-

neous workgroups under tournament (team-pay). This is consistent with Prediction

4, although differences are not strong. I examine this in greater detail in following

regressions.

49This is calculated as the total number of decoding questions solved divided by the total timetaken to solve them, including rest.

50Those under team-pay also believe that rest would have smaller negative impacts on their ownpayment. See Figure A.1. in Appendix A.

18

Figure 3: Measures of Personal Effort over Treatments

In Table 8, Columns 1 to 4, I perform random effects regressions of measures

of effort. Pooling across the identity treatments, Columns 1 and 3 show that just

as in Figure 3, effort exerted is reduced under team-pay. They solve questions at a

lower rate (≈1/4 SD) and rest proportionally more. Furthermore, in Column 2, the

difference between the two interaction variables is positive and mildly statistically sig-

nificant. This indicates that the negative effect of Team-Pay relative to Tournaments

is stronger in heterogeneous workgroups, which is consistent with the second part of

Prediction 4. There is however less support for the first part of prediction 4. While

the signs of interaction coefficients in Columns 2 and 4 are as predicted, with those

in heterogeneous workgroups putting in more effort under tournament incentives, but

less under team-pay, the effects are weak especially in the former. I surmise that this

partially reflects a difficulty in measuring effort under piece rates.51

51Even with a choice to rest, the relatively short time scale of the experiment means that the taskmay feel less repetitive and thus participants do not adjust effort that much given that they knowthey are being paid per question set. This might affect the tournament treatment especially.

19

Dep var: Standardised Proportion of Motivation: Match EffortEffort Qns Rested (Overall) (External)

Team-Pay -0.238*** -0.153 0.213*** 0.165* 0.333** 0.394***(0.067) (0.097) (0.064) (0.087) (0.125) (0.124)

Tournament × 0.021 -0.008 0.366** 0.024Heterogeneous (0.090) (0.094) (0.173) (0.137)

Team-Pay × -0.178* 0.106 -0.328* -0.232*Heterogeneous (0.101) (0.100) (0.178) (0.128)

Ability 0.755*** 0.756*** -0.087** -0.087**(0.041) (0.042) (0.043) (0.044)

1/Round -0.662*** -0.662*** 0.112*** 0.112***(0.041) (0.041) (0.034) (0.034)

1/Round × Ability -0.087** -0.087** 0.032 0.032(0.041) (0.041) (0.037) (0.037)

Constant 0.526*** 0.543*** 0.133* 0.119 -0.402** 0.066(0.108) (0.127) (0.073) (0.109) (0.166) (0.118)

Observations 1224 1224 1224 1224 306 306

* 0.10 ** 0.05 *** 0.01, Cols 1-4 are Random effects regressions while Cols 5-6 are OLS.Motivation is standardised.

Table 8: Regressions of effort variables

Table 8, Columns 5 and 6 provide supplementary evidence for the above using

survey data on motivations for effort.52 Column 5 shows that in line with Prediction

4, members of heterogeneous workgroups face more (less) overall pressure to match

others’ effort in tournament (team-pay) incentives. In Column 6, where pressure refers

to that from facing external disapproval, only team-pay is somewhat relevant. This

likely reflects greater concerns for the positive externalities of effort under team-pay.

The results in this section can be summarised as follows:

Result 4. (Identity, incentives and personal effort) Team pay lowers effort exerted.

Under team-pay, effort is somewhat lower in heterogeneous workgroups. Under tour-

naments, there is no significant difference in effort between workgroup types. However,

motivation to match others’ effort is lower (higher) under team-pay (tournaments) in

heterogeneous workgroups.

5.4 Salience of Workgroup Identity

In this subsection, I examine the impacts of the incentives and workgroup compo-

sition on the salience of workgroup identity. This serves as a check for the identity

manipulation and also allows for further robustness checks of results since identity

52See Table A.1 in Appendix A for a description of these motivation variables.

20

salience aligns with (β) in the model. The preferred measure for salience of work-

group identity here is obtained from a (standardised) weighted average of closeness

to members of the Klee/Kandinsky group obtained using an Inclusion of Other in

the Self (IOS) Scale from social psychology.53 Figure 4 plots participants feelings of

closeness towards ingroup and outgroup members and the standardised constructed

measure of identity salience (β).

Figure 4: Workgroup identity salience over Treatments

Participants can be seen to have higher feelings of closeness towards in-group

members (left) which results in workgroup identity salience being higher for homoge-

neous workgroups (right).54 The identity inducement was thus successful and in-line

with the models’ assumptions. Furthermore, there is also an added effect of incentive

schemes on identity.55 I next perform a regression analysis on these outcomes.

In Table 9, columns 1 and 2 look at how incentives and identity of the group

member influence one’s feelings of closeness. Column 1 shows that participants feel

higher levels of closeness towards ingroup members. Team-Pay however increases

53See Table A.1 in Appendix A for a description. This measure minimises the impact of experiencein the different incentive schemes. Other measures obtained from factor analysis of additional surveyquestions allude to their experiences in Stage 2 and may not reflect true ex-ante salience of workgroupidentity. Nevertheless, results do not change much depending on the measure used.

54Interestingly, they also believe that in-group members will be slightly better in performance(one sided t-test of mean>4: p<0.0001), see Figure A.1 in appendix A.

55That ex-post identity salience is influenced by the incentives is consistent with Chen et al. (2012)who find evidence that incentive structures can influence workgroup cohesion and subsequently cre-ative efforts. While our model does not account for these incentive effects on identity, comparativestatics for workgroup identity salience should not change even if team-pay were to amplify teamidentity salience. Comparative statics for incentives should also remain the same if these amplifi-cation effects are small enough. Footnote 35 discusses how a small modification to how workgroupidentity salience is determined would still be consistent with this.

21

Dep var: Feelings of closeness Workgroup Identityto group salience (βi)

Ingroup × Messages Sent 0.075***(0.024)

Ingroup × Messages Received -0.023(0.037)

Ingroup 2.023*** 1.664***(0.128) (0.258)

Team-Pay 0.388*** 0.376*** 0.320*** 0.418***(0.147) (0.143) (0.113) (0.151)

Heterogeneous -0.582*** -0.474***(0.142) (0.179)

Team-Pay × Heterogeneous -0.226(0.229)

Constant 1.804*** 1.784*** 0.047 -0.018(0.186) (0.201) (0.267) (0.275)

Observations 612 609 306 306likelihood -405.5 -405.0

* 0.10 ** 0.05 *** 0.01

Table 9: Regressions of Closeness and Salience of Workgroup Identity

feelings of closeness to any group member. Column 2 includes interactions with the

amount of message sent and received in Stage 1 as a proxy for the amount of social

interaction. Interestingly, it suggests that there is a pure ingroup identity effect

independent of the levels of social interaction in the first stage.

Table 9, columns 3 and 4 examine how the above influences feelings of closeness to

one’s workgroup. It shows being in a heterogeneous team leads to an approximately

0.5 SD drop in workgroup identity salience with Team-Pay having a comparable

positive effect. Weaker identity salience under tournament incentives may be another

explanation for the relatively weaker effects of identity on effort and cooperation

within tournament incentives. The results so far are summarised as follows:

Result 5. (Workgroup composition, Incentives and Workgroup Identity Salience)

Participants have stronger feelings of closeness towards ingroup members. Team-Pay

also strengthens feelings of closeness to any other. This results in stronger workgroup

identity salience in homogeneous workgroups and under Team-Pay.

Finally, using the constructed measure of workgroup identity salience, I check for

robustness of the previous results. This is done by replacing the identity treatment

dummy with actual elicited salience of workgroup identity and checking whether the

comparative statics for identity hold. I utilise an IV estimation procedure where

workgroup composition is used as an instrument for workgroup identity salience.

22

This assumes that workgroup composition only has effects on the chosen inputs and

subsequent output via workgroup identity salience.

Tables 10 and 11 examine whether results on main variables and motivations

(respectively) are robust to replacing the identity treatment with workgroup identity

salience. In general, comparative statics for identity salience are qualitatively in line

with our aforementioned findings which is reassuring. This is consistent with identity

salience being a moderating factor for how workgroup composition affects choices.

Dep var Std.Effort P:Give Help P:Receive help Avg Qns Std.OutputTeam-Pay -0.287*** 0.299*** 0.286*** -0.353*** 0.090

(0.071) (0.054) (0.057) (0.051) (0.067)Tournament×βi 0.001 0.212***

(0.170) (0.082)Team-Pay×βi 0.257* 0.389***

(0.134) (0.059)Tournament×β−i 0.209**

(0.090)Team-Pay×β−i 0.354***

(0.051)Tournament×βall -0.386 0.301

(0.347) (0.631)Team-Pay×βall -0.647*** 1.188***

(0.197) (0.403)Constant 0.564*** 0.538*** 0.552*** 4.632*** 0.117

(0.123) (0.103) (0.110) (0.066) (0.151)Observations 1224 1224 1224 1224 1224

* 0.10 ** 0.05 *** 0.01, βi, β−i, βall are based on self, others and workgroup respectively.Avg Qns is adjusted for help requests received. Controls not shown here.

Table 10: Random effects IV regressions: Main variables

Dep var Motivation: Match Effort Motivation: Help(Overall) (External) (Internal) (External)

Team-Pay 0.030 0.222** 0.431*** 0.425***(0.124) (0.097) (0.090) (0.116)

Tournament×βi -0.552** 0.017 0.116 -0.232(0.270) (0.227) (0.211) (0.280)

Team-Pay×βi 0.491** 0.338* 0.145 0.442*(0.214) (0.174) (0.150) (0.226)

Constant -0.313** 0.097 -0.121 -0.267(0.134) (0.101) (0.145) (0.261)

Observations 306 306 306 306

* 0.10 ** 0.05 *** 0.01, Cluster robust standard errors used.Motivation is standardised.

Table 11: IV regressions: Motivations for Help and Effort

23

6 Discussion

In this paper, I studied how identity influences optimal incentives in workgroups. To

do so, I induced group identities in participants and allocated them to workgroups

with different compositions. They then performed a novel real-effort task involving

on-the-job help under tournament and team-pay incentives. Results were broadly in

line with the predictions of a simple model where individuals cared about the impacts

of one’s actions on others, as weighted by team-identity salience.

Firstly, consistent with the literature, chosen levels of help exhibited in-group bias.

This explained the higher levels of cooperation in homogeneous versus heterogeneous

workgroups under both incentives. Secondly, common identity was found to influ-

ence effort provision and motivation positively under team-pay, but negatively under

tournaments, although results were weaker. The combination of these two effects

led to team-pay being significantly more effective than tournaments in homogeneous

workgroups, but not in heterogeneous ones. Lastly, identity salience of the workgroup

is found to be dependent not only on the identity composition of the workgroup, but

also the incentives faced. Its estimated impacts on individual input choices are also

broadly consistent with the model, lending robustness to the above results.

Overall, the results show that identity can indeed influence optimal incentives. In

real life, external factors like the social dynamics of the locale and/or internal factors

like corporate culture can influence (the perceptions of) common identity amongst

workers. This can then influence the optimality of incentive schemes — even across

firms with similar production processes. This may be of future empirical interest

both independently and in relation to the literature on management and productivity

differences across firms.56

Practically, the results also have implications for organisational design. They sug-

gest that effective management involves a joint choice of incentives and identity, albeit

with possible constraints on the latter.57 More thought thus needs to be put into pur-

posefully designing performance incentives to optimise productivity, accounting for

identity factors. Organisations can engineer identity salience through team alloca-

tion, hiring decisions and/or training procedures while choosing compatible incentive

56Bloom and Reenen (2007, 2010) examine management practices across firms and countries. Des-sein and Prat (2018) review the literature on productivity differences between firms. In particular,the factors examined here would fall under their classification of organizational capital.

57Possible (external) constraints on engineering identity salience include external social dynamics,high turnover, anti-discrimination laws etc.

24

schemes.58 Nucor is one company which has used this to great effect: workers share

a strong common identity and high levels of team-pay are used, perhaps resulting in

one of the best labour productivity levels in the steel industry.59

Nevertheless, it should be emphasised that there is no “one size fits all” recom-

mendation of this joint choice. Our results hint at this: tournaments are optimal in

both workgroup types considering effort alone (less so in homogeneous groups), but

not with cooperation. It is thus important to consider all elements of the production

process as these may influence the relationship between identity and optimal incen-

tives in more nuanced ways. A useful thought process would be to consider for each

incentive, the externalities imposed on others by each worker input, as these would

determine workers’ choices in interaction with identity.60

That incentives here are found to influence perceptions of shared identity and that

optimal incentives themselves depend on it also implies a feedback mechanism which

has several consequences. On one hand, it suggests that initial exogenous differences

in social identities may be amplified, giving rise to a wider spread of incentives. For

example, an initially homogeneous organisation might prefer team-pay — this am-

plifies perceived homogeneity and subsequently the preference for team-pay. On the

other hand, it also suggests that early mistakes in incentive choices might “lock”

organisations into a wrong identity-incentive equilibrium, resulting in possibly large

opportunity costs. Persistent differences between Nucor and other steel companies

are consistent with this (Ghemawat, 1995). Paying more attention to not only incen-

tives, but also (long run) identity salience would be important in helping organisations

avoid these costly adjustments.

58The possibility of influencing workers’ identification is discussed in Akerlof and Kranton (2008)who focus on how monitoring decisions, by affecting perceived organisational identity, can influencethe amount of individual compensation needed.

59This may an (intentional) result of selective hiring, retainment of workers with particular at-tributes and the relatively flat organisational structure (Collins, 2001).

60This is considered in Ashraf and Bandiera (2018) where they review the literature on horizontalrelationships as a source of social incentives in firms.

25

A Survey variables

Figure A.1: Treatment comparison of select survey variables

26

Understand TaskHow well did you understand the instructions and the decoding task in Part 2?[1: Did not understand at all - 7: Understood it very well]

Task InterestPlease tell us how you felt about the decoding task in Part 2? (Groups here refer toKLEE/KANDINSKY)[1: Very boring - 7: Very interesting]

Help benefits the teamWith respect to the decoding task, how do you think YOUR provision of help to otherswill affect the TOTAL score of the team? (4 = no effect on total score)[1: Total score decreases - 7: Total score increases]

Rest benefits myselfWith respect to the decoding task, how do you think taking breaks will affect your totalpayment? (4 = no effect on earnings)[1: Total payment decreases - 7: Total payment increases]

In-Group is betterIn solving a single set of 3 decoding questions, how do you think your own group will per-form relative to the other group on average? (Groups here refer to KLEE/KANDINSKY)[1: Much worse (slower) on average - 7: Much better (faster) on average]

Motivation: Match Effort (Overall)I felt worried about taking breaks during each round because others on my team mightbe putting a lot of effort.I felt that I had to put in my best effort into the decoding task to match the effort ofothers on the team.

Motivation: Match Effort (External)I felt that others on my team would be upset if I did not put in effort into the decodingtask.

Motivation: HelpI felt pressured by *MYSELF* to provide help.I valued being able to help others during the decoding taskI felt that helping others during the task is the right thing to do.I felt pressured by my *TEAMMATES* to provide help.I felt that others on my team valued being able to help others during the decoding task.I felt that others on my team believed that helping others is the right thing to do.

Closeness to each group (IOS scale)Please select the option which best describes your feeling toward the KLEE (KANDIN-SKY) group after Stage 1 of the experiment.

Motivation questions were on a likert scale from 1 (strongly disagree) to 7 (strongly agree).

In motivation to help, the first 3 statements loaded more positively on the first factor while thenext 3 statements loaded more positively on the second factor. These two factors are referred tointernal and external motivation respectively. In motivation to match effort, the two statementsloaded positively on a single factor.

Table A.1: Description of select survey variables

27

Instructions

Welcome to the experiment.

Please remain silent throughout the course of the experiment and refrain from using any

communication devices, otherwise we may be forced to stop the experiment. If you have any

questions at any point, please raise your hand and an experimenter will come over to see you.

In this experiment, there will be 2 stages and you will earn money based on your performance

in each stage. Please read the instructions below carefully.

The experiment will be conducted in an anonymous fashion. You will not be able to discover

others’ exact identities, neither will others be able to discover your exact identity. Rest assured

that your anonymity will be strictly preserved.

In the experiment, your payoffs will be in Experimental Currency Units (ECUs). At the end of the

experiment, your earnings will be converted into Pounds according to the rate: 100ECU: £2.5.

This will be added to your show up fee of £3. Information about your earnings in each stage will

only be provided at the end of the experiment. You will be paid your earnings privately and

confidentially at the end of the experiment after completing a questionnaire.

If you need to write down anything, please use the paper and pen provided. Please do not write

anything on this instruction sheet.

Stage 1

In the first stage, you will be shown five pairs of paintings sequentially. Each pair contains a

painting by Paul Klee and another by Wassily Kandinsky. They are abstract artists from the last

century. You will not be informed of the artist of each painting. For each pair, you will be asked

which painting you prefer. Your preferences relative to others in the session will then be used

to classify you into one of two groups: i.e. Group Klee and Group Kandinsky. This means that

the more times that you have indicated preference for paintings by Klee (Kandinsky) relative to

others, the more likely you will be assigned to Group Klee (Kandinsky).

For easier identification, Group Klee will be represented by the colour blue while Group

Kandinsky will be represented by the colour red throughout the experiment.

Subsequently, you will be shown a final pair of paintings, one by Klee and the other by Kandinsky

for which you have to guess the artist of each painting. To help in answering the question, you

will have a 5 minutes discussion (subject to restrictions) with members of your group (Klee or

Kandinsky); you will be able to refer to the past pairs of paintings during the discussion.

Subsequently, you will be asked to answer individually. Your answer as well as the answers of

others in your group (Klee or Kandinsky) will determine your payment in this stage:

If the majority of your group members get the answer right, then you will obtain a payment

of 80 ECUs.

You will then move on to Stage 2 of the experiment.

B Instructions

28

Stage 2

Team Assignment

In the second stage, you will be randomly allocated to a work-team consisting of 6 members and

play several rounds of a decoding task. Given the random allocation, your team members may

or may not belong to the same group (Klee or Kandinsky) as you. You will be notified of your

work-team’s composition at the beginning of the second stage. Note that your team members

will always be the same in every round.

The Decoding Task

In this task, you will have to solve sets of decoding questions in each round of 6 minutes. There

will be opportunities to help others on your work-team, as well as to take (paid) breaks during

the round. This will be described in detail below.

1) Solving question sets

Each question set involves converting several letters into numbers using a provided table. See

Figure 1.

There will be two different kinds of question sets in the task which occur with equal chance:

1. An easy question set with 3 decoding questions (represented by ~).

2. A difficult question set with 7 decoding questions (represented by !).

To complete a question set, use the provided table (1) to convert the letters into numbers, filling

the answers in the corresponding boxes (2). Then, submit your answers by clicking either of the

submission buttons (3).

Correct answers are denoted by an O while incorrect or incomplete answers are denoted by an

X. If all submitted answers are correct, you will earn 1 point and the next question set (and a

new table) will automatically appear. Your score (4) at the end of the round will thus be the

total number of question sets you have completed correctly. Your time left in the round is

shown on the top right (5).

2) Taking breaks

Depending on which button is used to submit your answers in (3), you can choose to take a break

for 5 seconds before receiving the next set of questions.

If you click the “Submit and Rest” button, and provided your answers are all correct, then you

will receive the next question set after a 5 second rest period where the Task screen will be

temporarily blanked out; this is shown in Figure 2.

For each break taken, you will be paid 2.5 ECUs.

(Note that this rate is equivalent to 1.75 ECUs for 3.5 seconds: the average time taken to convert 1 letter

into a number using the table in past experiments.)

If you click the “Submit” button, and provided your answers are all correct, then you will receive

the next question immediately.

The total number of times rested during the round is shown under your score (6).

(1)

(2)

(3)

(4)

(5)

Figure 2: After clicking the submit and rest button

(6)

Figure 1: The decoding task

3) Help Requests to Teammates

If you receive a difficult question set of 7 questions, you will automatically send a help request

after 3 seconds to a random teammate whose group (Klee or Kandinsky) you will be informed.

See the left side of Figures 3a/b.

Note that you can continue solving the question set before the help request is sent out.

If the teammate accepts your help request, you will be notified and the number of questions

you have to solve will be reduced by 3. Accepting your help request means that your teammate

will have to solve 1 extra question. See the top right of Figures 3a/b.

If the teammate rejects your help request, you will be notified as well, but the number of

questions will not be reduced. Rejecting your help request means that your teammate will not

have any extra questions. See the bottom right of Figures 3a/b.

Figure 3a: Help request to KLEE group

Figure 3b: Help request to Kandinsky group

7 Questions reduced by 3

7 Questions not reduced

by 3

7 Questions reduced by 3

7 Questions not reduced

by 3

4) Providing Help

At the beginning of each round, you will be asked to decide whether you want to provide help

to your teammates during the round (7).

In particular, you will be asked for your helping decision in two cases: 1) when the requester

belongs to the Klee group and 2) when the requester belongs to the Kandinsky group.

See Figure 4.

This decision will then apply to all help requests sent to you during that round. As mentioned,

each help request accepted during a question set means that you have to solve 1 extra question;

this will appear automatically during the question set. See Figure 5.

Figure 4: Deciding on how much help to give

Figure 5: What happens when help requests are accepted

(7)

5) Information about Help requests

For your interest, a provided sidebar on the left (8) will show you the number of times you have

accepted or rejected help requests from your teammates (in each group) during the ongoing

question set. Relevant icons and numbers will pop up once you have received a help request

during the ongoing question set: these are described in Figure 6 below. If help requests are

received during a rest period, you will see these icons immediately after your rest period ends.

Information on help requests to teammates (i.e. the displays in Figure 3) are also located here.

Figure 6: Information on help requests

Help request from

KLEE accepted

Help request from

KANDINSKY accepted

Help request from

KLEE rejected

Help request from

KANDINSKY rejected

(8)

6) Earnings

Your task payment in each round will depend on your performance as well as your teammates’

performance during the round. You will be notified of the exact payment scheme at the

beginning of the second stage.

Note that members of your work-team will receive the exact same payment scheme as you.

Your total payment in each round will be calculated as the sum of your task payment and the

“rest” payment (number of breaks taken × 2.5 ECU).

Out of the several rounds in the second stage, only 1 will be randomly chosen to make up your

final payment.

_____________________________________________________________________________

We have now come to the end of the instructions. There is a hard copy of the instruction sheet

in case you need to refer to it again. If you have any questions please raise your hand and we

will attend to you privately.

If not, we will now have a short quiz to test your understanding of the Stage 2 instructions. You

will have to answer all questions correctly to proceed.

Following the quiz, we will have 2 practice rounds for the Stage 2 decoding task:

In the first practice round, you will have 2 minutes to experience the user interface of the game;

This will be done with simulated partners, so there will not be any real interactions.

In the second practice round, you will have 3 minutes to practice completing the decoding task.

Note that the helping and resting mechanisms will be unavailable during the second practice

round.

After the 2 practice rounds, we will then begin Stage 1 of the experiment proper.

References

Afridi, F., Dhillon, A., Li, S. X., and Sharma, S. (2018). Using Social Connections

and Financial Incentives to Solve Coordination Failure: A Quasi-Field Experiment

in India’ s Manufacturing Sector. IZA Discussion Paper Series, (11521).

Akerlof, G. A. and Kranton, R. E. (2000). Economics and Identity. Quarterly Journal

of Economics, CXV(3):715–753.

Akerlof, G. A. and Kranton, R. E. (2005). Identity and the Economics of Organiza-

tions. Journal of Economic Perspectives, 19(1):9–32.

Akerlof, G. A. and Kranton, R. E. (2008). Identity, Supervision, and Work Groups.

American Economic Review (Papers & Proceedings), 98(2):212–217.

Akerlof, G. A. and Kranton, R. E. (2010). Identity Economics: How our Identities

Shape our Work, Wages, and Well-being.

Allen, R. S., Takeda, M. B., White, C. S., and Helms, M. M. (2004). Rewards

and Organizational Performance in Japan and the United States: A Comparison.

Compensation & Benefits Review, 36(1):7–14.

Ashraf, N. and Bandiera, O. (2018). Social Incentives in Organizations. Annual

Review of Economics, 10(1):439–463.

B Arthur, J. (1999). Explaining variation in human resource practices in US steel

minimills. In Employment practices and business strategy, chapter 1, pages 11–42.

Oxford University Press.

Bakshi, V. (2015). Nucor Revolutionizes the Steel Industry. https://rctom.hbs.org/

submission/nucor-revolutionizes-the-steel-industry/, Accessed 5/2019.

Bandiera, O., Barankay, I., and Rasul, I. (2005). Social Preferences and the Response

to Incentives: Evidence from Personel Data. The Quarterly Journal of Economics,

120(3):917–962.

Benndorf, V., Rau, H. A., and Christian, S. (2014). Minimizing Learning Behavior

in Experiments with Repeated Real-Effort Tasks. Working paper.

Besley, T. and Ghatak, M. (2005). Competition and incentives with motivated agents.

American Economic Review, 95(3):616–636.

35

Blazovich, J. (2013). Team identity and performance-based compensation effects on

performance. Team Performance Management, 19(3/4):153–184.

Bloom, N. and Reenen, J. V. (2007). Measuring and Explaining Management

Practices Across Firms and Countries. The Quarterly Journal of Economics,

122(4):1351–1408.

Bloom, N. and Reenen, J. V. (2010). Why do Management Practices Differ across

Firms and Countries ? Journal of Economic Perspectives, 24(1):203–224.

Boyd, B. K. and Gove, S. (2000). Nucor Corporation and the U.S. steel industry. In

Strategic Management: Competitiveness and Globalization, 4e.

Bull, C., Schotter, A., and Weigelt, K. (1987). Tournaments and Piece Rates: An

Experimental Study. Journal of Political Economy, 95(1):1–33.

Buser, T. and Dreber, A. (2015). The Flipside of Comparative Payment Schemes.

Management Science, 62(9):2626–2638.

Butler, J. V. (2014). Trust, truth, status and identity: An experimental inquiry. B.E.

Journal of Theoretical Economics, 14(1):293–338.

Byrnes, N. and Arndt, M. (2006). The Art of Motivation. In Bloomberg

BusinessWeek. https://www.bloomberg.com /news/articles/2006-04-30/the-art-of-

motivation, Accessed 3/2019.

Carpenter, J., Matthews, P. H., and Schirm, J. (2010). Tournaments and office

politics: Evidence from a real effort experiment. American Economic Review,

100(1):504–517.

Cassar, L. and Meier, S. (2018). Nonmonetary Incentives and the Implications of

Work as a Source of Meaning. Journal of Economic Perspectives, 32(3):215–238.

Charness, G., Masclet, D., and Villeval, M.-C. (2013). The Dark Side of Competition

for Status. Management Science, 60(1):38–55.

Chatman, J. A., Greer, L., Sherman, E., and Doerr, B. (2019). Blurred Lines: How

the Collectivism Norm Operates Through Group Perceived Diversity to Boost or

Harm Group Performance. Organization Science, 30(2):235–259.

36

Chen, C. X., Williamson, M. G., and Zhou, F. H. (2012). Reward system design and

group creativity: An experimental investigation. Accounting Review, 87(6):1885–

1911.

Chen, R. and Chen, Y. (2011). The potential of social identity for equilibrium selec-

tion. The American Economic Review, 101(6):2562–2589.

Chen, Y. and Li, S. X. (2009). Group Identity and Social Preferences. The American

Economic Review, 99(1):431–457.

Collins, J. (2001). Good to great: Why Some Companies Make the Leap...And Others

Don’t. HarperCollins.

Dessein, W. and Prat, A. (2018). Organizational Capital, Corporate Leadership, and

Firm Dynamics.

Drago, R. and Garvey, G. T. (1998). Incentives for Helping on the Job: Theory and

Evidence. Journal of Labor Economics, 16(1):1–25.

Falk, A., Fehr, E., and Huffman, D. (2008). The power and limits of tournament

incentives.

Fischbacher, U. (2007). Z-Tree: Zurich toolbox for ready-made economic experiments.

Experimental Economics, 10(2):171–178.

Friebel, G., Heinz, M., Krueger, M., and Zubanov, N. (2017). Team incentives

and performance: Evidence from a retail chain. American Economic Review,

107(8):2168–2203.

Ghemawat, P. (1995). Competitive Advantage and Internal Organization: Nucor

Revisted. Journal of Economics and Management Strategy, 3(4):685–717.

Goette, L., Huffman, D., and Meier, S. (2006). The impact of group membership on

cooperation and norm enforcement. American Economic Review, 96(2):212–216.

Goette, L., Huffman, D., Meier, S., and Sutter, M. (2012). Competition Between

Organizational Groups : Its Impact on Altruistic and Antisocial Motivations. Man-

agement Science, 58(5):948–960.

37

Hamilton, B. H., Nickerson, J. A., and Owan, H. (2003). Team Incentives and Worker

Heterogeneity: An Empirical Analysis of the Impact of Teams on Productivity and

Participation. Journal of Political Economy, 111(3):465–497.

Hamilton, B. H., Nickerson, J. A., and Owan, H. (2012). Diversity and productivity

in production teams. Advances in the Economic Analysis of Participatory and

Labor-Managed Firms, 13(2012):99–138.

Hannan, R. L., Krishnan, R., and Newman, A. H. (2008). The Effects of Dissemi-

nating Relative Performance Feedback in Tournament and Individual Performance

Compensation Plans. The Accounting Review, 83(4):893–913.

Heap, S. P. and Zizzo, D. J. (2009). The value of groups. American Economic Review,

99(1):295–323.

Henderson, R. and Van Den Steen, E. (2015). Why do firms have purpose? The

firm’s role as a carrier of identity and reputation. In American Economic Review,

volume 105, pages 326–330.

Hoff, K. and Pandey, P. (2006). Discrimination, social identity, and durable inequal-

ities. American Economic Review, 96(2):206–211.

Holmstrom, B. (1982). Moral Hazard in Teams. The Bell Journal of Economics,

13(2):324–340.

Huck, S., Kubler, D., and Weibull, J. (2012). Social norms and economic incentives

in firms. Journal of Economic Behavior and Organization, 83(2):173–185.

Kato, T. and Shu, P. (2016). Competition and social identity in the workplace: Evi-

dence from a Chinese textile firm. Journal of Economic Behavior and Organization,

131:37–50.

Lazear, E. P. (2018). Compensation and Incentives in the Workplace. Journal of

Economic Perspectives, 32(3):195–2014.

Lazear, E. P. and Rosen, S. (1981). Rank-Order Tournaments as Optimum Labor

Contracts. Journal of Political Economy, 89(5):841–864.

Masella, P., Meier, S., and Zahn, P. (2014). Incentives and group identity. Games

and Economic Behavior, 86:12–25.

38

Mcleish, K. N. and Oxoby, R. J. (2007). Identity, Cooperation, and Punishment.

Naranjo-Gil, D., Cuevas-Rodrıguez, G., Lopez-Cabrales, A., and Sanchez, J. M.

(2012). The effects of incentive system and cognitive orientation on teams’ per-

formance. Behavioral Research in Accounting, 24(2):177–191.

Papamarcos, S. D., Latshaw, C., and Watson, G. W. (2007). Individualism-

collectivism and incentive system design as predictive of productivity in a simulated

cellular manufacturing environment. International Journal of Cross Cultural Man-

agement, 7(2):253–265.

Riener, G. and Wiederhold, S. (2016). Team building and hidden costs of control.

Journal of Economic Behavior and Organization, 123:1–18.

Sheridan, J. H. (1998). Tale Of A ’Maverick’. https://www.industryweek.com/

companies-amp-executives/tale-amaverick, Accessed 3/2019.

Tajfel, H., Billig, M. G., Bundy, R. P., and Flament, C. (1971). Social categorization

and intergroup behaviour. European Journal of Social Psychology, 1(2):149–178.

Tajfel, H. and Turner, J. (1979). An Integrative Theory of Intergroup Conflict. The

Social Psychology of Intergroup Relations.

Towry, K. L. (2003). Control in a teamwork environment- The impact of social ties

on the effectiveness of mutual contracts. The Accounting Review, 78(4):1069–1095.

Vasanthi, V. and Chowdary, N. V. (2009). Nucor Corp’ s ‘ Performance-driven’

Organisational Culture: Employee-driven Competitive Advantage ? In IBSCDC

No:OB0015. IBS Case Development Centre.

Weng, Q. and Carlsson, F. (2015). Cooperation in teams: The role of identity,

punishment, and endowment distribution. Journal of Public Economics, 126:25–

38.

39