Announcement, Observation, and Honesty in the Voluntary Contributions Game

Transcription

1 Announcement, Observation, and Honesty in the Voluntary Contributions Game By Laurent Denant-Boemont, David Masclet and Charles Noussair January, 2005 Abstract In this paper, we study the effect of announcement and observation on voluntary public good provision. We find that requiring individuals to make a non-binding prior public announcement about their contribution level has no significant effect on average contributions. Making public each individual s contribution decision also has no significant impact on contribution levels. However, requiring announcements, in conjunction with making contribution decisions public, has a significantly positive effect on the average level of contributions. The treatments, in which announcements were elicited, permit the truthfulness of subjects announcements to be measured. We find that high contributors are more honest, the truthfulness of others is reciprocated with greater honesty, and announcements are more honest when contribution decisions are observable. Denant-Boemont: Université Rennes 1, Rennes, France. Masclet: Université Rennes 1, Rennes, France. Noussair: Department of Economics, Emory University, 1602 Fishburne Dr., Atlanta, GA , USA. We thank Gary Bornstein and participants at the June 2004 Economic Science Association Meetings in Amsterdam for constructive and helpful comments. We thank Elven Priour for programming and organization of the sessions. 1

2 1. Introduction Social dilemmas are situations in which a divergence exists between the outcomes that result when individual choices are based on self-interest and when they are based on group-interest. The prisoner s dilemma and the tragedy of the commons are well-known examples of social dilemmas. A large and active literature in experimental economics has investigated the behavior of individuals who face social dilemmas, and the factors that increase the extent of group-oriented behavior in such situations. The principal conclusion is that most individuals do not act only out of self-interest, nor do they act exclusively in the group interest. Rather, most individuals behavior tends to reflect both considerations. A number of factors, properties of both the environment and the rules of interaction, which encourage cooperation, have been identified (see Ledyard, 1995, for a survey). One commonly used context, in which the conflict between individual and group incentives is studied, is the Voluntary Contributions Mechanism. In this game, each individual member of a group receives an initial endowment of money. Each individual then has an opportunity to contribute any fraction of his endowment to a group account. The allocation decisions are simultaneous in that others choices are unknown at the time an individual makes his own decision. The total amount of money that all agents contribute to the group account is multiplied by a factor greater than 1 and then divided equally among all of the members of the group. Each individual has a dominant strategy to allocate zero to the group account, whereas the highest total group payoff is reached if all members contribute their entire endowment to the group account. The level of contribution can be interpreted as a measure of the extent that decisions are socially oriented. The value of the measure can be compared between treatments to identify factors that influence the level of cooperation. The experimental method allows the researcher to control the information individuals have about other group members regarding their actions and stated intentions. Most previous studies have considered environments where very little information about other participants decisions is available, prohibiting communication both before and after decisions have been made, and keeping individual contribution decisions private. However, many social dilemmas in the field do not occur in environments with this degree of anonymity, but rather are found in contexts in which some degree of contact between group members occurs before and/or after decisions are taken. In recognition of the potential importance of such contact, experimental economists have studied the effect of allowing individuals to acquire information about or from other group members, and have pursued the issue from several avenues. 2

3 A number of studies have explored the effect of different types of communication on outcomes in the Voluntary Contributions Mechanism context (see for example Isaac and Walker, 1988b) and have found that unstructured communication before contribution decisions are made serves to increase contribution levels. Furthermore, preplay communication changes the time profile of contribution levels from one that decreases (Isaac and Walker, 1988a), to one that increases, over time. Ostrom et al. (1994) and Muller and Vickers (2000) obtain similar results in experiments studying the social dilemma posed by the potential overuse of a common pool resource. Duffy and Feltovich (2001, 2003) study the effect of communication in a two-person prisoners dilemma (one of the players is permitted to transmit to the other a non-binding pre-play announcement consisting of an intended choice of action, before the actions are chosen), a social dilemma related to the voluntary contributions mechanism, and find that the existence of communication increases the incidence of cooperative play. Non-binding communication has a positive effect on contributions despite the fact that it does not alter the set of equilibrium outcomes. Another type of information that has been studied is public observability, on the part of other group members, of individual decisions. Sell and Wilson (1991) investigate the effect of public observability in a setting in which past contribution behavior of each member of the group is made public, and the authors observe an increase in contributions as a result. Duffy and Feltovich (2001, 2003) observe that displaying a history of each group member s contribution decisions increases the average amount contributed. Other studies have shown that associating agents with their contributions, when combined with an ability to sanction them, can create more cooperative behavior. For example, Yamagishi (1986), Fehr and Gaechter (2000a), Carpenter (2004), Anderson and Putterman (2004), Masclet et al. (2003) and others study the effect of allowing agents the ability to reduce other group members payoffs after observing their decisions, and they all find that this ability induces large increases in contribution rates. Masclet et al. (2003) also show that allowing agents to observe and to rate the decisions of each other group member, and to transmit the information to the agent being rated, increases contributions. Rege and Telle (2004) find that requiring agents to make their contributions while physically located in full view of the other group members increases contribution levels. Wilson and Sell (1997) are the only previous authors, to our knowledge, who consider the joint effect of both pre-play announcements and public observability after contribution decisions are made. They find that announcement plus observability leads to similar contribution levels as in a benchmark treatment in which neither instrument is available. However, they obtain other interesting results that are different in spirit from those of the studies discussed above. They find that if all players make a non-binding pre-play announcement of a contribution decision, contribution levels are lower than when such announcements are not possible. They also observe that if decisions are observable, there is a decrease in contributions compared to a benchmark 3

4 treatment in which actions are not observable (and no pre-play announcements could be made). Announcement plus Observability yields higher contributions than either Announcement alone or Observability alone, but no higher than in a baseline treatment where neither Announcement nor Observability is in effect. Their design allows honesty, the difference between an individual s announcement and his contribution, to be measured. This measure of honesty is in the spirit of the definition proposed by Khalil (2004), who writes honesty is understood here as telling the truth and fulfilling contractual obligations even when the benefit from cheating exceeds possible punishment. Wilson and Sell find that honesty is rare, with subjects announcements being systematically considerably higher than final contributions, regardless of whether or not the contribution decisions are observable. In this paper we report the results of an experiment in which we study the effect of both pre-play announcements and of post-play observability on contribution levels in the voluntary contributions mechanism. In a sense, we revisit the experiment of Wilson and Sell (1997). As is clear from the above discussion, the previous results on the effect of public observability and preplay announcements on contribution levels are mixed. The results of Wilson and Sell suggest that making decisions public or allowing pre-play communication has a negative effect on contribution levels, while the results of Duffy and Feltovich (who study a different but related social dilemma to the one considered here) suggest that each instrument has a positive effect. Furthermore, the positive effect of unstructured communication on contributions identified in previous studies suggests that announcement elicitation might have a positive effect, and that the increase in cooperation induced by public observability coupled with sanctions is consistent with observability alone having the effect of increasing contribution levels. In our study, we also consider, following Wilson and Sell (1997), the possibility of a positive synergy between the two instruments. There are three treatments in our experiment. The first is Announcement (A), in which each agent must make a public non-binding announcement consisting of his intended contribution level. The second is Observation (O), in which each individual s contribution decisions are revealed to all participants. The third is Announcement plus Observation (A+O), in which both the announcement of treatment A and the observability of treatment O are in effect. In addition to considering the effects of observability and announcements on contributions, we explore in some detail the determinants of announcements, contributions, and honesty. The treatments, as well as other procedural details of the experiment, are described in more detail in section 2. The results of the study are presented in section 3. Section 4 contains our concluding remarks. 4

5 2. The Experiment The experiment consisted of seven sessions. In each session, there were 30 periods of interaction, divided into three segments of 10 periods. All of the sessions were conducted at the Center for Research in Economics and Management (CREM), at the University Rennes I, Rennes, France. Between 12 and 20 subjects participated in each session. The subjects were recruited from undergraduate courses in business and economics at the university. None of the subjects had participated in an economic experiment previously. No subject participated in more than one session. On average, a session lasted 100 minutes including initial instruction and payment of subjects. The experiment was computerized using the Ztree program developed at the University of Zurich 1. Table 1 contains some summary information about each of the sessions. The first four columns indicate the session number, the number of subjects that took part in the session, the number of four-person groups in the session, and the treatment in effect. The fifth through seventh columns indicate the particular rules in effect in each of the three ten-period segments of the session. [Table 1: About Here] 2.1 Procedures Common to All Treatments A partner matching protocol was in effect for all sessions. The computer network separated the subjects taking part in a session into groups of size four. Group assignments remained fixed for the entire session and the members of a group interacted exclusively with their own group members for the entire session. Individuals received no information about the activity of any groups other than their own. There were 30 periods of play in each session, consisting of three ten-period segments. During each ten-period segment, subjects did not know whether or not the experiment would extend beyond the current segment. However, they knew the segment length and that each period within the segment would be identical. At the beginning of the experiment, the instructions were distributed and read to the subjects. There followed a quiz, consisting of four questions concerning the rules of the game and how earnings were determined, which all subjects were required to answer. The experimenter then announced and explained the correct responses. Subjects could indicate whether they had any questions about the process and the experimenter would answer them in private. In periods 1-10 and periods of each session, subjects played a voluntary contributions (VCM) game, described in the next paragraph, without announcements and public information 1 See Fischbacher (1999) for a description of the ztree computer program. 5

6 (except for session 7, in which the VCM without announcements and observation was played in periods 11-20). Activity in these periods proceeded as follows. At the beginning of each period, each agent was endowed with 20 Experimental Currency Units (ECU s), with each ECU convertible to Euros at 60 ECU = 1 Euro. Subjects simultaneously chose the portion of their endowment to contribute to the group account. They made this contribution decision by entering the number of ECU they wished to contribute in an appropriate field on their screens and clicking on another field to confirm their decision. Each ECU contributed to the group account yielded a payoff of 0.4 ECU to each of the four members of the group. Each ECU not contributed by the subject was credited to the subject s private account. 2 Therefore, the earnings, in ECU, of individual i in a period equaled where ci 4 E = 20 c i c k, (1) k = 1 is the contribution of player i. It is easily seen from (1) that individual i s earnings are maximized at c i = 0. Therefore, if the game is played once, there is a dominant strategy to contribute zero. If the game is finitely repeated, the only subgame perfect equilibrium of the game is for all players to contribute zero in each period Procedures Specific to Particular Treatments In periods of sessions 1 and 2, which constituted the Announcement (A) treatment, each period consisted of a two-stage game. In the first stage, subjects were required to announce a hypothetical contribution level. The announcement was non-binding. All group members announcements were simultaneous. There was no requirement that an individual s actual contribution level correspond in any way to his announcement. The announcement was described in the following terms (translated from the original French text). Before each period, each subject, including you, receives an endowment of 20 ECU. Each of the four members of your group, including you, decides separately on the amount of his endowment, in the range of 0 to 20 inclusive, which he announces that he will assign to the project. At the beginning of the second stage within each period from in the A treatment, subjects computer screens displayed the announcements of each of the other members of their group. The announcement of each individual was associated with a letter of the alphabet, A,..,D, which identified the person who made the announcement. In this treatment, individual contributions 2 The same parameters were used in the Fehr and Gaechter (2000a) study, as well as in several subsequent studies. At the group optimum, each member of the group contributes all 20 ECU, yielding a payoff of 32 ECU per person for the period. If every player follows his dominant strategy, of contributing zero, each player receives a payoff of 20 ECU. 6

7 were not displayed at any time, and thus other group members could not associate an individual s announcement with her contribution decision. Activity for the remainder of a period in the A treatment followed exactly the same rules as periods 1-10 and Participants simultaneously chose the portion of their endowment to contribute to the group account. Subjects were free to contribute less than, more than, or the exact amount specified in their stage 1 announcement. Specifically, the instructions indicated, You must choose your contribution to the project. This amount can be identical to, less than, or more than the amount that you have announced in the first stage. At the end of each period, each subject s computer displayed his own initial endowment, own contribution, the total group contribution, and own earnings. The earnings displayed consisted of those for the current period, as well as the accumulated earnings for the session up to and including the current period. After all subjects selected a field labeled [OK], the computer program continued to the next period. Periods of the Observation (O) treatment consisted of the same sequence of activity as in periods 1-10 and 21-30, except for the following difference. At the end of each period, each subject s computer screen displayed the contribution of each other group member, in addition to the individual s own endowment, contribution, earnings, and the total amount the group contributed in the period. Periods of sessions 5 and 6 of the Announcement plus Observation (A+O) sessions followed identical rules to periods of treatment A, except that under A+O, each subject was informed of the contribution level of each group member in the current period, as in the O treatment. Thus during periods of the A+O treatment, an individual s announcement in the current period could be associated with certainty to his contribution decision for the period. 3 In each treatment, the only subgame perfect equilibrium of the game, whether it is played once or finitely repeated, is for all players to always contribute zero to the public good. Any profile of announcements in the A and A+O treatments is compatible with a subgame perfect equilibrium. 3. Results Figure 1a displays the time series of total group contributions by period for each of the eight groups that participated in the Announcement plus Observation (A+O) treatment, in sessions 5 and 6, in which in periods announcements were elicited and decisions were made public. Figure 1b shows the data for session 7, in which announcement elicitation and observability were in effect in periods 1-10 and The vertical axis indicates the total number of tokens that the group contributed in the period, where the maximum possible total contribution is 80 (the total corresponding to a contribution of 20 for each agent). The horizontal axis indicates the period 3 In session 7, activity followed a different sequence. The activity described above for periods 1-10 and for sessions 1-6 characterized periods of session 7. The description above for periods in sessions 5 and 6 applied to periods 1-10 and of session 7. 7

8 number. Each time series represents the activity of one group. Table 2a presents the average contribution of each group in sessions 5 and 6, indicated as groups 1 8, as well as the data from session seven, indicated as groups The second column of the table indicates, for each of the eight groups numbered 1-8, the average individual contribution for periods 1-10 and The third column gives the same information for periods For groups 9 13, the table displays the data from periods in the second column and those from periods 1-10 and in the third column. For illustrative purposes, figure 2 compares the actual average contribution for sessions 5 and 6 of treatment A+O relative to hypothetical data without observation and announcement for periods The hypothetical benchmark data is an interpolation of the data in periods 1-10 and of sessions 5 and 6, constructed for each period t in the following manner. The average contribution of individuals in all groups participating in the two sessions (32 individuals) is calculated for period t 10, and the same calculation is performed for period t +10. The two resulting averages are averaged to yield an interpolated average level of contribution for period t 4. The data from the Announcement plus Observation treatment exhibit an increase in contributions in the periods in which Announcement and Observation are in effect, and a fall in contributions when they are no longer in effect. Table 2a indicates that for seven of eight groups participating in sessions 5 and 6, average contributions are higher when both Announcement and Observation are present than when they are both absent. The same is also true for four of the five groups in session seven. Figure 1a illustrates that greater heterogeneity between groups and higher contribution levels characterize periods of sessions 5 and 6, compared to the other periods of the same sessions. Some groups, in particular groups 2 and 5, exhibit very large increases in contributions in response to the change in the environment, while group 3 shows very little effect. Many groups exhibit a restart effect, a tendency for contributions to increase after a session is restarted (see Andreoni and Miller, 1993) in the first periods of new segments, periods 11 and 21. Figure 2 shows that in periods 11-20, average observed contributions are greater than in the hypothetical benchmark data, particularly in the last few periods of the segment. The principal finding that emerges from the Announcement plus Observation treatment is stated as Result 1. [Tables 2a 2c and Figures 1-2: About Here] 4 The time series of contributions that the interpolation generates exhibits two patterns typically associated with play of the VCM game: a decline in the level of contribution with repetition and the existence of a "restart effect (Andreoni and Miller, 1993), under which contributions in the period following the announcement of an unanticipated extension of play the game are on average greater than in the last period before the restart. 8

9 RESULT 1: The addition of both (a) prior non-binding announcements of contribution levels and (b) public observation of contribution decisions, increases the average contribution level. Support for Result 1: In sessions 5 and 6, contribution rates are higher in periods 11 20, when Announcement and Observation are available, than in the pooled data from periods 1-10 and for seven of eight groups. A nonparametric Wilcoxon matched pairs test shows that the difference in median contributions is significant at the p <.05 level. For the test, each group s activity over the entire session constitutes a paired observation. One element of the pair is the group s data from the period segment(s), in which Announcement and Observation are in effect (periods in sessions 5 and 6). The other element of the pair is the data from the periods in which they are absent. Thus we reject the hypothesis that Announcement plus Observation has no effect on contributions. In session seven, we also observe that contribution levels are higher under Announcement and Observation than in their absence for four of five groups. A nonparametric Wilcoxon matched pairs test shows that this difference in contributions is significant at the p <.1 level. The above result differs from that of Wilson and Sell (1997), who found that Observation and Announcement together did not increase contributions over the benchmark level. Wilson and Sell also obtained the striking results that Observation alone and Announcement alone decreased contribution levels. As we indicate in our support of result 2, we fail to replicate these effects. Rather, we find that observation alone has no significant effect on contributions, and the effect is positive in sign. The introduction of announcement alone also has a small and insignificant positive effect on contribution levels. Both effects are positive in sign, indicating that gathering more observations yielding similar outcomes would not yield a significantly negative effect on contributions, as Wilson and Sell have observed. These observations are summarized and supported in result 2. RESULT 2: Announcement alone does not change the average contribution level significantly. Similarly, Observation alone does not change the average contribution level significantly. Support for Result 2: Figures 3 and 4 display the time series of total contributions by period for each group in the A and the O treatments. Tables 2b and 2c display the average contribution levels for the A and O treatments in the same format as table 2a. For the A treatment, a nonparametric Wilcoxon matched pairs test of the hypothesis that median contributions are identical when announcements are required and when they are not, indicates that median contributions are not significantly different between periods 11 20, when announcements were required, and the pooled 9

10 data from periods 1-10 and when no announcements were required (p =0.1386). In the same manner, for the O treatment, median contributions were not significantly different between periods 11 20, when contributions were observed and the pooled data from periods 1-10 and when they were unobservable (p = ). [Figures 3 and 4: About Here] It appears that the elicitation of announcements has a negative effect on contributions for groups 2 and 3 only, in which contributions were already the lowest among all groups in periods It has a non-negative effect for all of the other groups. Thus it may be the case that eliciting announcements serves to lower contributions in groups that are particularly disposed to free riding and increases contributions otherwise, but on average does not yield a significant effect. Results 1 and 2 are consistent with an interaction effect between Announcement and Observation that serves to promote increased contributions. However, result 3 shows that any interaction effect is not significant. Requiring individuals to announce a priori their intended contribution levels has a positive but insignificant marginal effect when their contribution decisions are made public and also when they are not. Similarly, observation of individuals decisions also has a small and insignificant positive marginal effect in increasing contributions when the individuals are required to announce an intended contribution, as well as when they are not permitted to do so. However, the sum of the two small effects is strong enough to yield a significantly positive effect on contribution levels when both Announcement and Observation are in effect. RESULT 3: There is no evidence of an interaction effect between Observation and Announcement that increases contributions. Treatment A+O does not yield significantly higher median group contributions than treatments A or O. Support for Result 3: A Mann-Whitney rank sum test of the null hypothesis that there is no difference in median contribution between periods of treatment A+O (in sessions 5 and 6 only) and the same periods in treatment O, treating the average contribution of each group over the ten periods as an observation, yields a z-score of (p = 0.19). A similar test of the difference in median contribution level between the A+O and A treatments is also insignificant at conventional levels (z = 0.289, p = 0.772). Thus, the only significant difference we detect is an increase in contributions in the A+O treatment when both Announcement and Observation are introduced. The increase does not occur under A or O separately, indicating that neither has a significant direct effect. There is no significant 10

11 difference between A and A+O or between O and A+O, and the absence of these effects reveals a lack of a strong synergy between the two instruments 5. To study the decision-making behavior of individuals, we explore how their choices of contributions, announcements, and honesty levels, are updated from one period to the next. Our observations regarding contribution level are described as result 4 below. [Tables 3a 3c: About Here] RESULT 4: An individual s contribution in period t, in the A, O, and A+O treatments, is higher (a) the more he contributed in period t-1, and (b) the more the other members of the group contributed in period t-1. In the A and A+O treatments, an individual s contribution in period t is higher, (a) the higher his own announcement in period t, and (b) the higher the average announcement of others in period t. Support for Result 4: Table 3a contains the estimates from the following regression model for the A+O and A treatments: ( ) ( ) ( ) i i i i ( i c ) t = β0+ β1 c t 1 + β2 ct 1 + β3 at + β4 a t + β 5 t (2) For the O treatment, it also contains estimates of the equation: c i ( c t ) ( i i t = β0+ β1 1 + β2 t 1) c + β 5 t (3) The variables c t and i i a t denote person i s contribution and announcement respectively in period t, i c t 1 the average contribution of the members of the group other than i in period t 1, and i a t the average announcement of these other members of the group in period t. In all relevant equations, the estimates for β 1, β 2 β3 and β 4 are significantly positive. Thus, contributions exhibit some inertia in that individuals who make high contributions in one period are more likely than other agents to do so in the next period. Furthermore, high 5 It is also the case that A and O yield levels of cooperation that are not different from each other. A rank-sum test of the hypothesis that median contributions are equal in periods of the two treatments yields a z- score of.088, and a p value of

12 contributions on the part of other group members appear to be imitated or reciprocated with high contributions. Both own announcement and the average announcement of others in the current period positively affect the level of contribution. The positive relationship between own announcement and own contribution indicates that announcements are not pure noise, but informative, though imperfect, indicators of subsequent contributions. There is at least some degree of honesty in announcements in that on average an individual s contributions is higher, the higher his prior announcement. High announcements on the part of others in the current period are reciprocated with higher contributions, holding all else equal. This pattern shows, at least if the effect on behavior in future periods is ignored, that biasing one s announcement upward relative to one s subsequent contribution yields an advantage on average, because it does induce others to contribute more. We next explore possible influences on announcement decisions, and we observe the patterns described in result 5. RESULT 5: An individual s announcement in period t in the A+O and the A treatments is higher, (a) the higher the individual s announcement in period t-1, (b) the higher the announcements of other group members in period t-1, and (c) the more that i contributed in period t-1 (this last result is significant only in the A+O treatment). Support for Result 5: Table 3b contains the estimates for the equation: a i t = β 0 + β 1( i t 1 a )+ β 2 ( 1) i t a i + β 3 ( t 1) c + β 4 t (4) for the A and the A+O treatments. The signs of the coefficients β 1 β2 β3 are positive and significant (except for β3 in the A treatment). Announcements have similar properties as contributions. There appear to be individuals who make systematically higher announcements, as the β 1 term indicates that there is a correlation in individuals announcements from period to period. Furthermore, as the positive sign of β 2 reveals, because announcements are higher, the higher others announcements have been, there appears to be a reciprocal aspect to announcements. Announcements are higher for those who are high contributors, suggesting that honesty, the difference between announcements and contributions, may follow specific patterns. Two such patterns of honesty are summarized in result 6. 12

13 RESULT 6: An individual is more honest in period t, (a) the more honest he was period t-1, and (b) the more honest others were in period t-1. Table 3c contains the estimates from the following regression model: i i i i i i i t t β0 β1( t 1 t 1) β2( t 1 t 1) β3( t c a = + c a + c a + c 1)+ β 4 t (5) For both the A+O and the A treatments, β 1 and β 2 are positive and significant, while β 3 is insignificant. The regression supporting result 6 indicates that honesty is in part an endogenous phenomenon. The significance of β 2 suggests that honesty is subject to reciprocity, as are announcements and contributions. Honesty on the part of others is repaid with a higher level of honesty. However, the sign of β 2 indicates that those who were honest previously are more likely to be honest in the current period, all else equal. This suggests that there may also be an intrinsic component of honesty for individuals, which is in part independent of the behavior of others. Because player i is no more honest when others have contributed more unless they have also announced more, as the lack of significance of β 3 illustrates, the greater honesty associated with higher contributions of others is a reciprocation for the greater honesty of others rather than for the high contributions themselves. Figure 5 illustrates the fact that high contributors tend to be more honest than free riders. The figure shows the honesty level in relation to the contribution level. Each bar illustrates the average honesty level of those who contributed the quantity shown on the horizontal axis. For every average contribution level that is 14 or less in either the A or the A+O treatment, it is on average lower than the announcement. For any average contribution level greater than or equal to 16 in either treatment, the contribution exceeds the average announcement, indicating a positive correlation between own contribution and own honesty levels. [Figures 5-6: About Here] Figure 5 also shows that for 16 of the 21 possible contribution levels, the announcement was higher in A than for A+O. This suggests that honesty was greater when contributions were observable in A+O than when they were not in A. On average, individuals are more honest in the A+O treatment than in the A treatment. Figure 6 further illustrates the point. It shows the difference 13

14 between contribution and announcement over time in the two treatments. The vertical axis of the figure shows the total group contribution (the maximum possible level is 80). In each of the ten periods the difference between the average announcement and average contribution is greater in A than in A+O. Thus, individuals are on average more honest when their contributions can be observed. 4. Summary Our principal findings are the following. We find that the joint presence of pre-play announcements and post-play observation of decisions increases contribution levels significantly. However, observation or announcement alone does not induce a significant increase in the average contribution level. Individuals make higher contributions, the higher their own announcements, the higher the announcements of others, and the more they and others have contributed previously. Individuals announcements are higher, the higher the preceding announcements of other group members have been, and the more others have previously contributed. Honesty is greater, the higher one s own honestly level has been previously, and the higher the honesty level that others have shown previously. Cooperators tend to be more honest than free riders, and individuals are more honest when their contribution decisions can be observed than when they cannot. Like contributions, announcement and honesty levels are endogenous, and appear to reflect in part the reciprocation of the behavior of others. Our results do not replicate Sell and Wilson (1997) who report that Observation, coupled with Announcement, decreases contributions. Several factors might explain the discrepancies between their study and ours. There are some differences in experimental design. First, some parameters of the payoff functions (the initial endowment, the marginal per-capita return on contributions to the public good, and the number of members in a group) are not the same in the two studies, and it is possible that the parametric structure interacts with the announcement and observability requirements in a complicated manner. Second, in contrast to Sell and Wilson (1997), participants in our study were not given any information about other group members' individual contributions in past periods. It is possible that this information about past behavior may reduce contributions if it is made more obvious that past contributions of other group members are declining. Indeed, we do observe in our data that for groups in which contributions are already low, Announcement further reduces them. The increase in contributions resulting from the existence of pre-play announcements and post-play observation might arise because additional sources of emotional cost resulting from opportunistic behavior are created, and behavior adapts in an attempt to avoid or to reduce these costs. To help identify the sources of these costs, we distinguish, following Kandel and Lazear 14

15 (1992), between two types of emotional cost. The first is external in origin, and arises from the disapproval by other agents of one s perceived opportunistic behavior. We refer to this type of cost as shame. The second type, which arises internally within an individual after behaving opportunistically, is referred to as guilt. Shame can be thought of as arising from opportunistic actions that others can observe, while guilt originates from opportunistic actions that are unobservable to others. Gintis (2004, p. 63) has recently argued, from an evolutionary perspective, that The experience of shame, guilt and other visceral reactions plays a central role in sustaining cooperative relations. Several recent experimental studies have also supported the notion that guilt and shame influence decision making (see for example Bowles and Gintis (2002) or Charness and Dufwenberg (2001) for a discussion). Announcement elicitation and public observability may create additional potential sources of guilt and shame that do not exist in a baseline treatment with no announcements or observability of decisions. In a baseline treatment, contributions that are too low relative to the average may be a source of guilt, but the unobservability of contribution decisions eliminates any source of shame. Under the announcement treatment, the submission of a contribution that is lower than one s announcement is an additional possible source of guilt. Announcement also allows for a source of shame, the public association to an announcement that is viewed as too low, perhaps because it is below the group average. The observability treatment admits an additional source of shame, the association of an individual with a low contribution. In the A+O treatment, three sources of shame arise that are absent in the baseline treatment: (a) from making a low contribution, (b) from making a low announcement, (c) from a discrepancy between an individual s announcement and his contribution level. Thus the greater levels of contribution in the A+O treatment relative to baseline levels may reflect the tendency to avoid the cumulative costs of shame. Avoidance of shame is consistent with the greater level of honesty on the part of individuals when their announcements can be verified as in the A+O treatment, than when they cannot as in the A treatment. Thus, while Walker (2004) proposes the following definition, honesty is the pursuit and reinforcement of self-integrity and is therefore the pursuit of self-worth, suggesting that honesty is a consequence of internal forces, it appears that social pressure may also be important in encouraging honest behavior, at least in the setting that we study here. While dishonesty, announcements that exceed actual contributions, is widely observed, we find that the level of dishonesty is endogenous, depending on the history of activity earlier in the session. As Majeski and Fricks (1995) claim, "cooperation increases depending on whether promises are kept or not". This conditionality of honesty has been formalized by Alger and Renault (2002), who distinguish between unconditional honesty, which depends on an individual s ethics, and conditional honesty, which depends on the fairness of the contracts the individual has agreed to. In our experiment, honesty appears to have components of each type. 15

16 Our measurement of contributions, announcements, and honesty suggests that reciprocal behavior is widespread and multi-dimensional in human interaction. In a similar manner as has been previously documented for contributions (see for example Fehr and Gaechter, 2000b), we find that high announcements lead other group members to make higher announcements, and greater honesty leads other group members to be more honest. However, the absence of greater honesty in response to higher contributions suggests that the reciprocity may be specific rather than general in nature, perhaps reflecting a degree of specific tit-for-tat behavior, and not necessarily indicative of greater benevolence to those who have made higher contributions in the past. For example, in our data honesty is rewarded with honesty, but not necessarily with a general tendency to be more benevolent toward the honest person. However, we recognize that our experiment was not designed to isolate the determinants of honesty and a more focused investigation of the behaviors that increase or decrease honesty in human interaction in social dilemmas is an important potential area for research. References Alger, I. and Renault R Screening Ethics When Honest Agents Care About Fairness. Working Paper. Economics Department. Boston College. Anderson C. and L. Putterman Do Non-Strategic Sanctions Obey the Law of Demand? The Demand for Punishment in the Voluntary Contributions Mechanism. Working Paper. Univerity of Rhode Island. Andreoni J. and Miller J Rational Cooperation in the Finitely Repeated Prisoner s Dilemma: Experimental Evidence. Economic Journal. 103: Bowles, S. and H. Gintis Prosocial Emotions. Working Paper. Santa Fe Institute. Carpenter Punishing Free Riders: How Group Size Affects Mutual Monitoring and the Provision of Public Goods. Games and Economic Behavior, forthcoming. Charness G. and M. Dufwenberg. Promises and Partnership. Working Paper. University of California at Santa Barbara. Duffy, J., and Felkovitch N Words, Deeds and Lies. Working paper, University of Pittsburgh and University of Houston. Duffy, J., and Felkovitch N Do Actions Speak Louder than Words? An Experimental Comparison of Observation and Cheap Talk, Games and Economic Behavior, 39, Fehr, E., and S. Gaechter. 2000a. Cooperation and Punishment in Public Goods Experiments. American Economic Review, 90(4): Fehr E. and S. Gaechter. 2000b. Fairness and Retaliation: The Economics of Reciprocity. Journal of Economic Perspectives 14, Gintis, H The Genetic Side of Gene-Culture Coevolution: Internalization of Norms and Prosocial Emotions. Journal of Economic Behavior and Organization, 53, Isaac, M., and J. Walker. 1988a. Communication and Free-Riding Behavior: The Voluntary Contributions Mechanism. Economic Inquiry 26(4): b. Group Size Effects in Public Goods Provision: The Voluntary Contributions Mechanism. Quarterly Journal of Economics 103: Kandel. E and E. Lazear Peer pressure and partnership. Journal of Political Economy 100: Khalil, E. L What is Altruism? Journal of Economic Psychology

17 Ledyard, J Public Goods: A Survey of Experimental Research, in: J. Kagel and R. Roth, eds. Handbook Of Experimental Economics, Princeton University Press, Princeton, NJ. Majeski, S. J and S. Fricks Conflict and Cooperation in International Relations. Journal of Conflict Resolution 39: Ostrom, E., J. Walker, and R. Gardner Covenants With and Without a Sword: Self- Governance is Possible. The American Political Science Review 86(2): Rege, M., and K. Telle The Impact of Social Approval and Framing on Cooperation in Public Good Settings, Journal of Public Economics 88(7/8), Sell J. and Wilson. R Levels of Information and Contribution to Public Goods. Social Forces 70: Walker, C A Charitable View of Altruism: Commentary on What is altruism? by Elias Khalil. Journal of Economic Psychology Wilson R. and Sell J Liar, Liar : Reputation and Cheap Talk in Repeated Settings. Journal of Conflict Resolution.41: Yamagishi, T The provision of a sanctioning system as a public good. Journal of Personality and Social Psychology 51(1):

18 Appendix: Instructions, translated from the original French version used in the experiment <<<The instructions below were distributed and read before period 1>>> GENERAL INSTRUCTIONS You are participating in an economic experiment in which you can earn some money. Your earnings depend on your decisions and the decisions of other participants in the experiment. It is therefore important to read these instructions carefully. The instructions that have been given to you are your private information. You are not allowed to communicate with the other participants during the experiment. If you do, you will be excluded from the session and not be paid your earnings. Your earnings in the experiment will be in terms of ECU (Experimental Currency Units). Your earnings will be converted from ECU to Euros and paid in cash at the end of the experimental session on the following basis: Your final earnings in ECU are the total of your earnings during each of the periods that comprise this session. Your final earnings in ECU will be converted to Euros on the following basis: 60 ECUs are worth 1 Euro You will receive, in addition, a participation fee of 3. This experimental session is composed of a series of periods. In each period, the participants are divided in groups of 4. You will thus be grouped with three other people. For the entire session, you will be grouped with the same three other people. You will not find out the identity of these people at any time. Procedures for the next 10 periods The four subjects comprising a group has the opportunity to participate in a project, by contributing to a group total, which will then be divided among the four individuals. The group total depends on the amount each individual in the group contributes. Each period proceeds in the following manner: At the beginning of each period, each subject, including you, receives an endowment of 20 ECU Each of the 4 subjects, including you, decides on his own the amount of his endowment he assigns to the project. After you have chosen your contribution to the project, by indicating a number between 0 and 20 inclusive, you must click on the [OK] button. Once this action is performed, you may not change it for the remainder of the current period. Once all of the members of your group have made their decisions, your screen will indicate the total amount of ECU assigned to the project by all of the members of the 18

19 group together (including your contribution). The screen will indicate to you as well how much you have earned in the period. Your earnings are made up of two elements: - One element is the amount of your endowment you have kept for yourself (that is, 20 ECU minus your contribution to the project). - The other element is your income from the project: each ECU invested in the project earns 0.4 ECU to each member of the group, so that the income to you from the project is 0.4 times the total of the four individual contributions to the project from your group. - Therefore, your earnings for a period are calculated in the following manner: Your earnings = (20 your contribution to the project) times (total group contributions to the project) The earnings of each member of the group are calculated in the same manner, which means that each group member receives the same income from the total group contribution to the project. Each ECU of your endowment that you keep for yourself increases your earnings by 1 ECU. Each ECU that you assign to the project increases your earnings by 0.4 x 1 ECU =.4 ECU. The earnings of the other group members also increase by 0.4 ECU per person for each ECU you assign to the project. In the same manner, you receive earnings from each ECU that other members of your group assign to the project. For each ECU assigned to the project by another member of the group, you earn 0.4 x 1 ECU =.4 ECU. If you have any questions about what you have read, please raise your hand. We will respond individually to your questions right away. To be sure that you have understood the rules, please answer the following questions: 1) Suppose that each member of your group receives an endowment of 20 ECU. Nobody, including you, contributes any ECU to the project. What are your earnings? ECU What are the earnings of each other group member? ECU 2) Each member of the group receives an endowment of 20 ECU. You contribute 20 ECU to the project. Each of the three other members of the group contributes 20 ECU to the project. What are your earnings? ECU What are the earnings of each other group member? ECU 19

20 3) Each member of the group receives an endowment of 20 ECU. The other three members of the group have assigned a total of 30 ECU to the project What are your earnings if you contribute 0 ECU to the project? ECU What are your earnings if you contribute 15 ECU to the project? ECU 4) Each member of the group receives an endowment of 20 ECU. You contribute 8 ECU to the project. What are your earnings if the other members of the group contribute a total of 7 ECU to the project? ECU What are your earnings if the other members of the group contribute a total of 22 ECU to the project? ECU 20