Emergent Recordkeeping Changes the Course of Economic History

By

Sudipta Basu

John Dickhaut1

Gary Hecht

Ivo Tafkov

Kristy Towry

and

Gregory Waymire

NOTE: This paper is being prepared for submission to a science journal. The rest of the paper reports additional material that would be published online only as a supplement to the print article. This material is included as a separate document.

October 24, 2006Please Do Not Quote

1 Dickhaut is affiliated with the Carlson School of Management at University of Minnesota. All other co-authors are affiliated with the Goizueta Business School of Emory University. We acknowledge financial support of this research from the Carlson School of Management and the Goizueta Business School.

Emergent Recordkeeping Changes the Course of Economic History

Sudipta Basu, John Dickhaut‡, Gary Hecht, Ivo Tafkov, Kristy Towry, & Gregory

Waymire

Physical recordkeeping is a deeply rooted and ubiquitous feature of exchange in

modern economies. Artefacts from 10,000 years ago reveal a “paper trail” that

arose in concert with increases in the scale and complexity of human interaction1-2.

Despite recordkeeping’s pervasiveness, no economic theory implicates its necessity

for economic exchange. Consequently, no scientific study demonstrates

recordkeeping emerges naturally with exchange complexity and little is known

about the economic forces unleashed by this simple, and often overlooked,

economic institution. We demonstrate experimentally that recordkeeping emerges

with complex exchange and then alters fundamentally the history of consummated

exchange. A record of exchange serves as a useful mnemonic aid when people

exchange repeatedly for several periods in a multi-agent setting. Agents act on

better memory of partners’ past cooperation, leading to greater gains from

exchange and increased commitment to and fairness in exchange. Ultimately,

agents’ economic choices are conditioned on a fundamentally different history of

past exchange in an economy in which recordkeeping is possible. Our findings (1)

imply that an ability to keep records increases gains from multilateral repeated

exchange and (2) demonstrate the powerful influence exerted by seemingly simple

economic institutions that are frequently taken for granted.

Goizueta Business School, Emory University, 1300 Clifton Road NE, Atlanta, GA 30322

‡ Carlson School of Management, University of Minnesota, 321 19th Avenue South, Minneapolis, MN 55455

Economic exchange is necessary to establish the extensive human cooperation reflected

in a specialized division of labour3. A puzzle engaging scholars from several fields has

been how to explain the large-scale cooperative networks sustained by humans4. Some

argue that a proclivity for punishment and numerous other features of human

psychology are adapted to facilitate social exchange5. One of these adaptations is the

human brain’s ability to remember and analyze the results of past interactions, which

facilitates avoidance of previously known cheaters6-8. Cooperation is sustained by

punishment or avoidance of those who violate norms of reciprocity and fairness9-12.

Recent work indicates institutions that sanction non-cooperation can extend the

scale and scope of human cooperation13-14. This is consistent with arguments that

emergent institutions vastly expand human capacity to sustain the cooperation that

underlies extensive exchange and division of labour15-17. At the same time, research has

not established the means by which human societies create the cultural analogue of

neuronal memory18-19 and its effects on cooperative economic behaviour. Our

experiment is designed to establish the importance of recordkeeping in sustaining

beneficial exchange when multiple agents interact for several periods. Establishing this

is important since we know that recordkeeping emerged coincidentally with the earliest

complex human cultures1-2.

Our experimental design generates two comparisons. In the first comparison, we

study a fundamental proposition derived from archaeological artefacts. That

proposition is that more recordkeeping will emerge in a complex economy than in a

simple economy. The second comparison asks how complex economies with

recordkeeping differ from economies without recordkeeping. The intent of our design

is to create a laboratory economy in which we can witness the birth of the modern

institution known as “accounting,” and see how our complex economy evolves

differently when its most primitive ancestor has arisen20-23.

2

The building block for our economies is the two-player investment game24-27. In

the original investment game, one investor is paired anonymously with one trustee for

one period of play. Player 1, the investor, receives ten units of experimental currency,

which are referred to as lira. The investor can decide how many lira out of the ten (in

whole numbers) to send to Player 2, the trustee. The investment produces gains from

exchange because it is tripled in route to the trustee. Thus, if the investor invested two

(five) lira, the trustee would receive six (15) lira. The trustee can then decide how

much of the amount received to return to the investor. Assuming self-interested players,

the trustee should keep all that is received, and the investor anticipating the trustee’s

inclination will send nothing. There is now a large body of evidence that suggests that

the investor almost always sends something (often referred to as a form of trusting

behaviour) and the trustee sends enough back so that the investor gets a positive rate of

return28-29. This behaviour is referred to as positive reciprocity.

In our simple experimental economy the one period investment game between an

investor and a trustee is repeated ten times with new endowments of ten lira each time

(subjects are not told how many times the game will be repeated). Repeated

interactions allow each player to use multi-period strategies, and memory of past

outcomes can be useful in planning one’s own strategy and monitoring its performance,

and inferring and responding to partners’ strategies. In our complex economy five

subjects play the role of investor and five play the role of trustee. In each period, every

investor has five separate endowments of 10 lira, one for each trustee. Each investor

can take a different amount from each trustee endowment and send it to the respective

trustee. Therefore, each trustee receives five tripled amounts, one from each of the five

investors. Every trustee then decides, for each investor, how much of the amount

received to send back. Trustees cannot send money to an investor that exceeds the

amount received from that investor in that period. In the complex economy, this

process is repeated ten times with each investor reendowed each time.

3

The simple and complex economies are conducted using the computer software Z-

tree whose windows are displayed on the left hand side of each subject’s computer

screen30. To see if and when subjects choose to keep records, we position a running

program, TextBox, on the right side of the computer screen. In both the complex and

simple economy subjects are told they can type anything they want. (Experimental

instructions are available in the supplemental materials.) Subjects could not use paper

and pencils, so that any external records they kept were recorded electronically.

Figure 1 depicts the interactions between subjects in our complex economy where

each investor (trustee) interacts with five different trustees (investors). We replicated

our complex economy five times. These replications provided data from 50 subjects; 25

played the role of investor and 25 played the role of trustee. Within each replication,

interactions occur in 25 distinct dyads, or 125 in total dyads across all five replications.

We use the same number of subjects in replications of our simple economy (25

investors and 25 trustees). Because each subject interacts in only one dyad, there are 25

dyads that comprise simple economies.

Recordkeeping is expected to allow subjects to better remember past interactions

in the presence of exchange complexity. We hypothesize that, when available,

recordkeeping will be more likely adopted in the complex exchange condition than in

the simple exchange condition. We also hypothesize that greater gains from exchange

will result in the complex condition with recordkeeping than without recordkeeping.

Recordkeeping allows a trustee to build a reputation because investors can better

remember each trustee’s actions. Trustee reputation sustains investor commitment in a

multi-period setting. Consequently, the history of complex exchange will unfold

differently with and without recordkeeping.

Emergent Recordkeeping and Economic Exchange

4

When recordkeeping was possible, thirty-nine subjects in the complex condition kept

records as measured by any marks on their screen compared to only twenty-five in the

simple condition (see Table 1, Panel A). We rejected the null hypothesis that a

participant keeping records was equally likely to come from the complex and simple

conditions using a Binomial test (one-tailed p = 0.052). (Details of all statistical tests

are described in the supplemental materials.) This difference is greater for investors

(one-tailed p = 0.061) than trustees (one-tailed p = 0.292). Because the Binomial test

has low power, we replicated this comparison using a resampling procedure

(bootstrapping) of the frequency of recordkeeping measured in the same manner. The

bootstrap is based on 100,000 resamples with replacement, and corroborates evidence

that the observed differences in recordkeeping frequency across the conditions are

statistically significant (one-tailed p = 0.012 for all subjects, 0.001 for investors only,

and 0.097 for trustees only).

Records can include evaluations of other participants’ intentions and can serve as

a tool to categorize others’ behaviour (e.g., “2 and 4 are cheap”). The content of records

depends on the complexity of the exchange setting (see Table 1, Panel B). Eleven of the

39 subjects who kept records in the complex condition recorded some form of judgment

in their records. None of the subjects who kept records in the simple condition included

such judgmental statements. A Binomial test rejects the null hypothesis (one-tailed p =

0.004) that judgmental recordkeepers were distributed randomly across the simple and

complex conditions when the ex ante probability of a recordkeeper subject being in the

complex condition was 60.9% (i.e., 39/64).

The evidence (see table 1) is consistent with the hypothesis that recordkeeping

supplements memory when agents seek to track the behaviour of multiple trading

partners over several periods. These results show that the relation between

recordkeeping and exchange complexity is foundational.

5

Recordkeeping Increases the Gains from Complex Economic Exchange

We measure a given dyad’s gains from exchange as the amount that a dyad’s total

payoff exceeds the minimum possible payoff as a percentage of the maximum possible

increase they could have attained. The minimum total payoff is 100 lira (when the

minimum investment of zero lira is chosen by the investor in each of the 10 rounds) and

the maximum possible is 300 lira (when the maximum investment of 10 lira is chosen

in every round and is subject to tripling). Thus, our measure ranges from zero percent

when the dyad earns the minimum total of 100 lira to one hundred percent when the

dyad achieves the maximum possible of 300 lira.

Table 2 shows that dyads in the complex condition with recordkeeping average

greater gains from exchange (56.5%) than dyads in the complex condition without

recordkeeping (49.9%). A Binomial test rejects the null hypothesis that the mean gain

was equal across these two conditions (one-tailed p = .039). The 6.6% mean difference

is economically important since it represents a 13.2% increase in gains from exchange

when recordkeeping is possible for the average complex dyad. This evidence indicates

that recordkeeping is a sustainable institution because it enables efficiencies in complex

exchange.

Recordkeeping Promotes Commitment and Reciprocity in Exchange

The results in table 2 are due to investors that commit by making the maximum

investment. Four hundred thirty maximum investments are made by investors in the

recordkeeping condition compared to only 210 in the non-recordkeeping condition (see

6

Figure 2, Panel A). Using a Binomial test, we reject the null hypothesis that a maximum

investment is equally likely in the complex recordkeeping and complex non-

recordkeeping conditions (p < 0 .0001).

Greater commitment by investors under recordkeeping is consistent with trustees’

subsequent behaviour. Trustees in the complex condition with recordkeeping are more

likely to reciprocate by returning “fair” amounts (see Figure 2, Panel B). We define a

fair return decision as one where the trustee evenly splits the amount received with the

investor. We reject the null hypothesis that fair trustee choices are equally likely to

come from the complex recordkeeping and complex no recordkeeping conditions using

a Binomial test (one-tailed p < 0 .0001). This result also holds when fairness is defined

relative to the total pie available (amount received by trustee plus amount retained by

investor).

Recordkeeping promotes commitment and reciprocity, which generate the

conditional cooperation observed in prior experiments and shown to persist in agent-

based simulations24-27, 31-42. Recordkeeping availability increases trustee reciprocity as

reflected in decisions to fairly divide economic gains in a complex exchange setting.

Recordkeeping Alters the History of Exchange

The impact of recordkeeping on the history of consummated exchange is starkly

apparent in the correlations between an investor’s investment and the past average

return on investment (ROI) realized by other investors (see Figure 3, Panel A). ROI is

the money received from a particular trustee as a fraction of the amount invested with

that trustee. These correlations are large and positive in the complex exchange condition

without recordkeeping. They are near zero when recordkeeping is possible. These

differences are startling in two regards. First, the investor has never directly observed

other investors’ ROIs. Second, these correlations differ by an order of magnitude

7

between economies with and without recordkeeping. This pattern suggests that an

investor’s decisions absent recordkeeping are influenced to a far greater degree by

unseen remote events.

These correlation differences could arise if investors with recordkeeping are better

able to clearly remember past interactions with specific trustees. In this case, the

investor’s interactions with different trustees stay separated in memory and are

influenced less by interactions with other trustees. Consistent with this interpretation,

recordkeeping leads to more positive correlations between period 10 investments and

past within-dyad ROI at lags of two or more periods (see Figure 3, Panel B). Also,

correlations between period 10 investments and average lagged ROIs that the investor

has obtained from other trustees are reliably positive (> .20) at lags of four periods or

more without recordkeeping, but not with recordkeeping (see Figure 3, Panel C).43

A test of the null hypothesis that the correlation structures across the

recordkeeping and non-recordkeeping conditions are equal is rejected (one-tailed p

= .004). Overall, the evidence in Figure 3 indicates that the economic history of

exchange itself depends on the possibility and presence of recordkeeping. More broadly,

this supports the view that institutions, in this case recordkeeping, fundamentally alter

the course of economic history15-16.

Conclusions

Our findings have profound implications in two regards. First, our laboratory results

provide direct causal support for the hypothesis suggested by archaeologists that basic

8

technologies for storing transactional data through symbolic artefacts, and subsequently

through writing, exert an important influence on the scale and complexity of human

economic interaction. These findings complement work suggesting that evolved human

institutions play a central role in sustaining the extraordinary level of cooperation and

complex interaction that characterize humans relative to other species13. Thus, the

emergence of an institution like recordkeeping can provide the basic informational

inputs necessary to sustain reciprocity through image and reputation8, 44-45. Future

experiments should consider the joint effect of recordkeeping, communication, and

punishment institutions on the scale and complexity of human economic networks.

Second, our evidence reminds us that essential economic institutions lie at the

lowest level of the institutional scaffold that humans have gradually erected through

culture over thousands of years16. Our evidence highlights the importance of seemingly

simple institutions. These institutions appear trivial and may not be recognized as

important because their ubiquitous nature leads us to take them for granted. One such

institution is the paper trail embodied in artefacts created to signify the consummation

of exchange. We believe that understanding the cultural origins and subsequent

evolution of this institution will contribute significantly to a comprehensive explanation

for the unprecedented level of cooperation in our species4, 12.

9

Methods

We use a repeated version of the investment game in which each dyad plays ten

consecutive rounds of a computerized game. No direct communication between subjects

was allowed and subject names were never revealed. Each experimental session has ten

subjects; five subjects are randomly assigned to be investors and the remaining five are

assigned the role of trustee. Subjects interacted anonymously over a local computer

network, programmed and conducted with the software z-Tree.

The experiments were conducted at the Center for Interuniversity Research and

Analysis on Organization (CIRANO) in Montreal, Canada. CIRANO staff recruited

participants and ran twenty sessions (five sessions for each experimental condition)

during November and December 2005. Subjects were recruited by CIRANO from a

standard subject pool and their identities remain anonymous to the authors. Two

hundred subjects (115 males and 85 females) participated in the experiment. The

average age of the subjects was 25.5 years and 78 (39%) were 26 years of age or older.

One hundred fifteen subjects were graduate students, eighty-two were undergraduates,

and three were non-students. One hundred sixty-four subjects had been subjects in a

prior experiment at CIRANO.

Both Recordkeeping (Possible or Not Possible) and Exchange Complexity

(Simple or Complex) were manipulated, which resulted in four experimental conditions.

Five experimental sessions were for each condition. In the sessions where

recordkeeping is possible, all subjects are provided the software Textbox, which allows

them to type in any marks they desire. Whether a subject kept records was measured by

whether any marks were made on his/her screen at any time during the experimental

session.

10

In a simple exchange condition, each investor interacts with only a single trustee,

and so each session provides data on play by five exchange dyads (with ten subjects) for

ten periods. Subjects are informed only of their direct trading partners’ decisions, and

not those of other players in the session. In a complex exchange condition session, each

investor is simultaneously matched with each of the five trustees and vice versa.

Subjects in the complex conditions are notified only of others’ decisions that affect their

own payoffs. For example, Trustee B1 learns how much money Investor A1 has chosen

to send to him/her, but not how much Investor A1 sent to the other trustees, B2 through

B5. Each complex condition session provides data on play by five investors and five

trustees for ten periods; however, unlike the simple setting, play occurs in 25 exchange

dyads because each subject is making choices within five different dyads.

In each condition-specific experiment-session, the ten participants received and

read written experiment instructions. Participants then took a quiz to ensure sufficient

understanding of experiment instructions. The experiment facilitator checked quiz

answers and resolved discrepancies privately before the beginning of the first round.

The facilitator collected all instructions and quizzes and ensured participants had no

other materials or writing instruments available. A condition-specific, short-form

version of experiment instructions was taped to the wall next to the computer monitor.

Each condition-specific experiment-session included 10 trading periods.

Participants were not informed of the number of periods to mitigate end game effects,

although they were informed via recruiting materials that the experiment would last

approximately two hours. No experiment-session lasted longer than 90 minutes.

Each trading period began with investors deciding how much of 10 units of

experiment-currency (i.e., lira) to invest in the paired trustee(s). In the complex

condition, investors had five separate endowments of ten lira for each of the five

trustees with whom the investor was paired. All investors’ investment decisions were

11

required before trustees received investment information. Similarly, all trustees’

“return” decisions were required before both player types received feedback

information. Investors received feedback information in the form of what each paired-

trustee sent. Trustees received feedback information in the form of what was received

from each paired-investor (i.e., the tripled investment amount). The next trading period

began once all participants confirmed they were finished reviewing feedback.

At the end of the tenth trading period, participants completed a short questionnaire

containing strategy-oriented and demographic questions. Total earned lira was

converted to cash at a rate that varies by experimental condition to equalize the

maximum possible payout per participant across sessions. Since each participant in the

complex setting participates in five dyads each period, we set the conversion rate in the

complex setting at one-fifth that of the simple setting.

12

Endnotes

1 D. Schmandt-Besserat, How Writing Came About (Univ. of Texas Press, Austin, TX, 1996).

2 H. Nissen, P. Damerow, R.K. Englund, Archaic Bookkeeping: Writing and Techniques of Economic

Administration in the Ancient Near East (University of Chicago Press: Chicago, IL, 1993).

3 A. Smith, The Wealth of Nations (Univ. of Chicago Press, Chicago, IL, 1776).

4 E. O. Wilson, Sociobiology, The 25th Anniversary Edition (Belknap Press of Harvard University Press,

Cambridge, MA, 2000), chapter 27.

5 L. Cosmides, J. Tooby, in Evolutionary Psychology Handbook, David Buss, Ed., (Wiley, New York,

NY, 2005).

6 S. Klein, L. Cosmides, J. Tooby, Psychological Review 109, 306 (2002).

7 R. Axelrod, W. Hamilton, Science 211, 1390 (1981).

8 M. Nowak, K. Sigmund, Nature 393, 573 (1998).

9 E. Fehr, S. Gachter, Nature 415, 137 (2002).

10 E. Fehr, E. Fischbacher, Nature 425, 785 (2003).

11 D. J-F. de Quervain et al., Science 305, 1254 (2004).

12 J. Henrich et al., Science 312, 1767 (2006).

13 O. Gureck, B. Irlenbusch, B. Rockenbach, Science 312, 108 (2006).

14 J. Henrich, Science 312, 60 (2006).

15 V. Smith, American Economic Review 93, 465 (2003).

16 D. North, Understanding the Process of Economic Change (Princeton University Press, Princeton, NJ,

2005).

17 An example of a sanctioning institution is a legal code like the Code of Hammurabi; see H. Saggs,

Civilization before Greece and Rome (Yale University Press, New Haven, CT, 1989).

18 M. Donald, Origins of the Modern Mind: Three Stages in the Evolution of Culture and Cognition

(Harvard University Press, Cambridge, MA, 1991).

13

19 Physical records of exchange have also played an important evidentiary role in imposing sanctions

through established law; see, e.g., R. VerSteeg, Early Mesopotamian Law (Carolina Academic Press,

Durham, NC, 2000).

20 H. Hatfield, J. of Accountancy 37, 241 (1924).

21 Y. Ijiri, Theory of Accounting Measurement (American Accounting Association, Sarasota, FL, 1975).

22 J. Demski, Managerial Uses of Accounting Information (Kluwer Publishing, New York, NY, 1993).

23 S. Basu, G. Waymire, Accounting Horizons 20, 201 (2006).

24 J. Berg, J. Dickhaut, K. McCabe, Games and Economic Behaviour 10, 122 (1995).

25 Fehr E., G. Kirchegeister, A. Riedl, European Economic Review 42: 1-34 (1998).

26 J. Van Huyck, R. Battalio, M. Walters, Games and Economic Behavior 10, 143-170 (1995).

27 K. McCabe, M. Rigdon, V. Smith, Economic Behavior and Organizations 52 (2), 277-295 (2003).

28 R. Croson, N. Buchan, American Economic Review 89(2):386-391 (1999).

29 E. Glaeser, E. Laibson, J. Scheinkman, C. Soutter, Quarterly Journal of Economics 155(3), 811-846

(2000).

30 Fischbacher, U. 1999. Z-Tree - Zurich Toolbox for Readymade Economic Experiments - Experimenter's

Manual, Working Paper Number 21, Institute for Empirical Research in Economics, University of Zurich.

31 E. Hoffman, K. McCabe, K. Shachat, V. Smith, Games and Economic Behaviour 7, 346 (1994).

32 E. Fehr, S. Gachter, G. Kirchsteiger, Econometrica 65, 833 (1997).

33 E. Hoffman, K. McCabe, V. Smith, Economic Inquiry 36, 335 (1998).

34 S. Gachter, A. Falk, Scand. J. of Economics 104, 1 (2002).

35 M. Hauser, M. Chen, F. Chen, E. Chuang, Proc. R. Soc. Lond. B 270, 2363 (2003).

36 R. Sethi, E. Somanathan, Journal of Economic Behaviour & Organization 50, 1 (2003).

37 M. Brown, A. Falk, E. Fehr, Econometrica 72, 747 (2004).

38 E. Fehr, U. Fischbacher, TRENDS in Cognitive Science 8, 185 (2004).

39 C. Bicchieri, J. Duffy, G. Tolle, Philosophy of Science 71, 286 (2004).

40 E. Fehr, G. Kirchsteiger, A. Riedl, Qtr Jour. Econ. 108, 437 (1993).

14

41 E. Fehr, U. Fischbacher, S. Gachter, Human Nature 13, 1 (2002).

42 E. Fehr, B. Rockenbach, Current Opinion in Neurobiology 14, 784 (2004).

43 The evidence in figure 3 parallels findings from financial economics that stock prices in developed

economies impound large amounts of firm-specific information, in contrast to less developed economies

where stock prices exhibit strong common market-wide commonalities. See R. Morck and B.Yeung,

World Economics 3 (3): 1-15 (2002.

44 M. Nowak, K. Sigmund, Nature 437, 1291 (2005).

45 G. Bolton, E. Katok, A. Ockenfels, Journal of Public Economics 89, 1457 (2005).

Acknowledgements

Financial support for Basu, Hecht, Tafkov, Towry, and Waymire was provided by the Goizueta School at

Emory University. Financial support for Dickhaut was provided by the Carlson School at the University

of Minnesota. We are grateful for the help and support provided by the staff at CIRANO in Montreal who

helped in running our experiment. We benefited from comments and suggestions by attendees in

presentations of earlier versions of this research at the meetings of the American Accounting Association,

Economic Science Association, and the Human Behavior and Evolution Society. We also benefited from

interactions at workshops presented at the 2006 Mini-Conference on the Foundations of Accounting at the

Goizueta Business School of Emory University, the Emory University Anthropology Department,

University of Arizona, City University of Hong Kong, City University of New York-Baruch, George

Washington University, Interdisciplinary Center for Economic Studies at George Mason University,

University of Iowa, London School of Economics, University of Minnesota, Norwegian School of

Economics, University of Notre Dame, Queen’s University (Canada), Temple University, University of

Texas-Dallas and Tilburg University.

15

Table 1

Adoption and content of records

Panel A Number of recordkeepers when recordkeeping was possible

All Subjects Investors Only Trustees Only

Complex Exchange

39 (61%)

22 (65%)

17 (57%)

Simple Exchange

25 (39%)

12 (35%)

13 (43%)

BothConditions

64 out of 100

34 out of 50

30 out of 50

Binomialp-value

.052 .061 .292

The predicted directional relationship between complex exchange and the adoption of recordkeeping is confirmed in this setting. This table reports that in the complex setting, of the 50 subjects who could keep records, 39 chose to keep records. In the simple exchange setting, only 24 subjects kept records of the 50 agents who could. We report in parentheses the percentage of the recordkeepers in each column category in the complex and simple exchange conditions.

Panel B. The judgmental content of subjects’ records

Records with Qualitative Judgments All RecordkeepersComplex Exchange

11 recordkeepers (100%)

39 recordkeepers(61%)

Simple Exchange

0 recordkeepers(0%)

25 recordkeepers(39%)

BothConditions

11 total 64 total

Binomialp-value

.004

The predicted directional relationship between complex exchange and the inclusion of qualitative judgments in records is confirmed. This table shows that all of the 11 agents who recorded qualitative judgments were in the complex exchange condition. We report in parentheses the percentage of the recordkeepers in each column category in the complex and simple exchange conditions. We tested the null hypothesis that judgmental recordkeepers were distributed randomly across the simple and complex conditions when the ex ante probability of recordkeeper in the complex setting was 60.9% (i.e., 39/64).

16

Table 2

The value of recordkeepingExperimental Condition Mean Efficiency GainsComplex Exchange - Recordkeeping (No Recordkeeping) 56.5% (49.9%)

p = .039

Simple Exchange - Recordkeeping (No Recordkeeping) 53.1% (52.6%)p = .374

The predicted relationship between the availability of recordkeeping and gains from exchange in the complex condition is confirmed. This table demonstrates that in the complex setting recordkeeping is associated with significantly increased mean gains from exchange. No significant differences in mean gains occur in the simple exchange setting. The method of conducting the statistical test is discussed in the supplementary materials.

17

Figure 1

Differences in Interactions Between and Complex Exchange

ConditionsSimple Exchange Complex Exchange

Investors Trustees Investors Trustees

A1 B1 A1 B1

A2 B2 A2 B2

A3 B3 A3 B3

A4 B4 A4 B4

A5 B5 A5 B5

This figure shows that investors interact with only a single trustee in the simple exchange condition and vice versa (see left panel). In the complex condition each investor and trustee play the investment game simultaneously with five trustees and five investors, respectively (see right panel).

18

Figure 2

Recordkeeping, Investor Trust and Trustee Fairness

Panel A Frequency of investor commitment as reflected in maximum investments in the complex exchange condition

Binomial p-value < .0001

A significant relationship between investor commitment and the availability of recordkeeping in the complex exchange condition is confirmed. This graph demonstrates that in the complex setting with recordkeeping, the frequency of maximum investments is over twice as great when recordkeeping is available than when recordkeeping is not possible. The method of conducting the Binomial test is discussed in the supplementary materials.

19

Panel B Percentage of cases where the trustee sends back a fair amount in the complex exchange

condition

Binomial p-value < .0001 for both definitions of fair trustee returns

A significant relationship between trustee fairness and the availability of recordkeeping in the complex exchange condition is confirmed. This graph demonstrates that in the complex setting with recordkeeping, the frequency of fair returns by trustees is over twice as great when recordkeeping is available than when recordkeeping is not possible. The method of conducting the Binomial test is discussed in the supplementary materials.

20

Figure 3Recordkeeping and the Evolved History of Exchange

Panel A Correlation of period 10 investment with other investors’ average past ROI

This graph indicates more positive correlations between period 10 investments and past average ROI received by other investors when recordkeeping is not possible in a complex exchange condition. The more positive correlations for the no recordkeeping setting suggest that the spillover effect from remote unobserved exchanges influence investments to a far greater degree when recordkeeping is not possible.

Panel B Correlations of period 10 investment with past within-dyad ROI

This graph indicates that stronger positive correlations between period 10 investments and past within-dyad return on investment (ROI) are observed when recordkeeping is possible. The more positive correlations for the complex recordkeeping setting after a lag of one period suggest that a deeper history of past within dyad exchange influences investment when recordkeeping is possible.

21

Panel C Correlation of period 10 investment with average ROI received by investor

from other trustees

This graph indicates more positive correlations between period 10 investments and past average ROI received by the investor from other trustees after lag 4 when recordkeeping is not possible in a complex exchange condition. The more positive correlations for the no recordkeeping setting suggest that investors have more limited recall of the source of returns as the lag between investment and returns increases when recordkeeping is not possible.

22