Conformity and reciprocity in public good provision

Journal of Economic Psychology 26 (2005) 664–681 www.elsevier.com/locate/joep

Conformity and reciprocity in public good provision q Nicholas Bardsley b

a,*

, Rupert Sausgruber

b

a CeDEx, Nottingham School of Economics, University Park, Nottingham NG7 2RD, UK Department of Public Economics, University of Innsbruck, Universita¨tsstr. 15, 6020 Innsbruck, Austria

Received 26 February 2004; received in revised form 3 December 2004; accepted 25 February 2005 Available online 3 May 2005

Abstract People contribute more to experimental public goods the more others contribute, a tendency called ‘‘crowding-in.’’ We propose a novel experimental design to distinguish two possible causes of crowding-in: reciprocity, the usual explanation, and conformity, a neglected alternative. Subjects are given the opportunity to react to contributions of a payoff-irrelevant group, in addition to their own group. We find evidence of conformity, accounting for roughly 1/3 of crowding-in.
2005 Elsevier B.V. All rights reserved. JEL classification: C72; C92; H41 PsycINFO classification: 3000 Keywords: Conformity; Reciprocity; Public good experiment

q The authors would like to acknowledge valuable comments from Frans van Winden, Simon Gaechter, Jonathon Baron and two anonymous referees. All errors are the responsibility of the authors. The experiments were conducted whilst Bardsley was affiliated with CREED, University of Amsterdam, and financed by the EU TMR project FMRX-CT98-0238 (‘‘ENDEAR’’) and by the Austrian National Bank (Jubilaeumsfonds) under Project no. 9134. * Corresponding author. E-mail addresses: [email protected] (N. Bardsley), [email protected] (R. Sausgruber).

0167-4870/$ - see front matter
2005 Elsevier B.V. All rights reserved. doi:10.1016/j.joep.2005.02.001

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

665

1. Introduction People tend to contribute more to a public good the more the others contribute. This pattern, known as crowding-in, is a replicated result in public good experiments (Bardsley, 2000; Croson, 1998; Fischbacher, Ga¨chter, & Fehr, 2001; Keser & van Winden, 2000; Weimann, 1994). The prevalence of crowding-in has encouraged theories of reciprocity which have recently received much attention (see Dufwenberg & Kirchsteiger, 2004; Falk & Fischbacher, 1998; Rabin, 1993; Sugden, 1984).1 An alternative explanation of crowding-in is conformity, which has in contrast been largely ignored, both by economic theorists and experimentalists. Conformity is an effect such that ‘‘we view a behaviour as correct in a given situation to the degree that we see others performing it’’ (Cialdini, 1993, p. 95). In this interpretation, conformity is a social phenomenon (for a review, see Moscovici, 1985), and differs sharply from economic concepts of rational imitation and information cascades, sometimes called ‘‘conformity’’ or ‘‘social influence’’ in the economics literature (Becker, 1991; Bernheim, 1994; Bikhchandani, Hirshleifer, & Welch, 1998; Selten & Ostmann, 2000). The economic concepts involve different ways in which people learn from othersÕ behaviour how to further their self-interest. In contrast, normative conformity involves perceiving othersÕ behaviour as a guide to what is socially or morally appropriate. It therefore predicts that people may conform independently of the material consequences of doing so. Reciprocity and conformity are both important potential causes of ‘‘endogenous social interaction’’ effects – processes whereby agents who share membership of a particular group come to exhibit similar behaviour (Manski, 1993, 2000). ‘‘Endogenous’’ here means coming about through the behaviour of the agents, rather than through their prior characteristics. An oft-cited example is the correlation of criminal activities within a neighbourhood, after controlling for the usual determinants of crime (Glaeser, Sacerdote, & Scheinkman, 1996; Ludwig, Duncan, & Hirschfield, 2001). Accordingly, a given individual is more likely to become a criminal in an unruly than in a law-abiding neighbourhood. Reciprocity might explain this observation, if people commit crimes in response to crimes suffered personally. Alternatively, people may just view a certain crime as more socially or morally acceptable the more common it is in their neighbourhood (Kahan, 1997). Recently, Falk, Fischbacher, and Ga¨chter (2003b) have developed an experimental design to empirically isolate endogenous social interaction effects. Exploiting the advantages of experimental control, this study avoids notorious difficulties in identifying endogenous social interaction effects with field data. In their design, every subject is simultaneously a member of two economically identical public good groups. Members of these groups contribute to both public goods separately. Random composition of groups controls for correlated effects, which would arise if

1 Reciprocity means responding in kind to the way one is treated by others. In economistsÕ reciprocity models agents return material benefits and harms received (Falk & Fehr, 2002).

666

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

subjects share similar individual characteristics. Therefore, the only reason for a subject to contribute different amounts to each group is that subjects respond to othersÕ contributions. The authors find strong evidence for endogenous social interaction, that is, a subjectÕs propensity to contribute to a public good varies substantially with the behaviour of the group. In this paper, we propose a novel design to discern reciprocity from conformity to explain such endogenous effects of social interaction. Our design also employs two groups. In group A, subjects observe only their own groupÕs behaviour. In group B, they also observe the behaviour of group A. In this set-up, our interest is to see whether contributions of group B correlate with those of group A. In contrast to Falk et al. (2003b), our agents only belong to one public good group. This allows us to distinguish causes of inter-subject responsiveness, since last-moving agents with self-interested or reciprocal preferences cannot learn anything relevant from group A. In this environment, the only reason for such subjects in group B to react to the decisions of group A is conformity. Our broad strategy is the same as that of Cason and Mui (1998) in an experiment on ‘‘social influence’’. In their study subjects make two consecutive choices in a dictator game. Before the second choice each subject receives information about another subjectÕs first choice. The authors find no significant difference between this treatment and a control providing irrelevant information only. A weak, indirect effect is found only in that decisions are more self-regarding in the irrelevant information treatment. In a similar experiment by Brandts and Fata´s (2001), subjects in a first round participate in a two-person public good game. Subjects are then informed about the average contribution of a set of subjects in this round. In a second round, subjects choose contingent on the other playerÕs possible actions. The authors test whether the subjectsÕ choices in the second round correlate with the information about average behaviour from the first round. The authors find no effects and conclude that models of social behaviour need not incorporate conformity. We believe that such a conclusion is premature. In the Cason and Mui (1998) study, very little social information was imparted; the decision of just one other individual was revealed. The design of Brandts and Fata´s (2001) provides no direct test for conformity. Rather it tests whether subjectsÕ responsiveness to others is influenced by aggregate information about what others did. This is a separate hypothesis, independent of conformity. Finally, our work is related to that of Carpenter (2004), who also attempts to explore conformity in a public good experiment. The study tests whether conformity exerts any dynamic effects in repeated interaction. Repeated interaction renders it difficult, however, to differentiate conformity and reciprocity. Even if subjects are randomly matched every round into group of strangers, as in Carpenter (2004), the behaviour of others in the previous period contains information about expected contributions in the current period. In this case subjects may copy the behaviour of others because of reciprocity and not because of conformity. To avoid this difficulty our study follows a different approach and tests for conformity in a one-shot public good game.

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

667

Our study reveals two main results. First, contributions correlate with those from a completely separate group. Since this holds for cases in which subjects cannot deduce anything about their own groupÕs prospective behaviour, we interpret this as evidence for conformity. Second, regarding the importance of conformity compared to reciprocity, we find that conformity explains roughly 1/3 of the total crowding-in effect. The remainder of the paper is organised as follows. Section 2 describes the experimental design in full detail. Section 3 reports our results. Section 4 provides some discussion and Section 5 concludes.

2. Experimental design Our experiment is a variant of a typical public good experiment. Subjects are organised into groups. Each subject i is endowed with 10 tokens. A token is worth €0.70. Subjects decide how to divide their 10 tokens between a public and a private account. ci denotes iÕs contributions to the public account. The payoff pi is a linear function of tokens in the private account and the sum of contributions to the public account: pi ¼ 10
ci þ a

N X

cj .

ð1Þ

j¼1

We set a = 0.5, and the group size is N = 6. Because a < 1 and aN > 1, the individually payoff-maximising strategy is to contribute nothing, although positive contributions increase group earnings. We know from previous experiments that many people contribute and that contributions exhibit a strong positive correlation in this game (see the references in the opening paragraph). To test for conformity we enable subjects prior to their own choice to observe the contributions of another group. Call A and B two groups consisting of NA = 6 and NB = 6 members, respectively. Group A subjects observe only their own groupÕs behaviour. Group B subjects also observe the behaviour of another group, that of group A. Our interest is to see whether contributions of group B correlate with those of the observed group A, even though a group B memberÕs payoff depends only on the behaviour of group B members. Conformity, if any, exists in both groups. However, in group A its effects are not separable from reciprocity. Therefore, although we need to have group A subjects participating in our experiment, our design is a pure within-subject design to analyse whether group B subjects conform with group A behaviour. By this procedure we hold fixed all potential individual characteristics which may affect a subjectÕs behaviour. Assume that group B subjectsÕ contributions in this set-up correlate indeed with those of another group A. To attribute this behaviour to conformity, we need to exclude that subjects in group B make inferences about their own groupÕs behaviour from that of group A. If learning of this kind was possible, contributions in group B could correlate with those in group A regardless of conformity. For subjects with reciprocal preferences wish to contribute more when they expect members of their

668

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

own group to contribute more. Information about what group A contributes could influence expectations about group BÕs contributions, and so change group BÕs reactions to each other. We take two measures to exclude this possibility. First, our experiment is oneshot; subjects interact only once. This assures that subjects cannot infer anything from othersÕ behaviour that has been experienced in previous decision rounds. Second, we employ a sequential procedure in four stages: Stage 1: In group A, NA
1 subjects move first and simultaneously choose their contributions to the group A public good. Stage 2: In group A, the NAth subject moves last. This subject sees a vector of (NA
1) contributions on behalf of first movers in group A and then chooses. Stage 3: In group B, subjects see a vector of NA contributions on behalf of group A. Based on this information NB
1 subjects contribute to the group B public good. Stage 4: In group B, the NBth subject moves last. When deciding, this subject sees both the contributions of NB
1 first movers in group B and all the contributions of group A. Notice the following about this procedure. First, and most importantly, last movers in stage 4 have full information about the behaviour of group B, so nothing can be learnt from the behaviour of group A about BÕs prospective actions. Our test of conformity exploits this fact and only concerns behaviour in group B in stage 4: if the behaviour of last movers in these groups correlates with contributions in group A, this will establish the relevance of conformity (see Section 3). Second, we choose as group size NA = NB = 6, which is larger than in many other public good experiments. We do so because a larger group is presumably more representative if subjects view the behaviour of others as a normative guide for their own (Cialdini, 1993). 2.1. The conditional information lottery (CIL) design In real-choice experiments, to collect enough data in the relevant contingencies would be a costly task. Experimental economists often work around this problem using the ‘‘strategy method’’ (Selten, 1967). Under this method, subjects submit a choice at each possible decision node that can be reached in the course of a game. Actual play is then computed from the strategies submitted, and payments generated accordingly. While the strategy method seems useful to test for crowding-in due to reciprocity (see for example, Fischbacher et al., 2001), its use appears less appropriate in an attempt to isolate conformity. One reason is that subjects under the strategy mode receive no information about the distribution of othersÕ choices. Presumably, for conformity to arise it is important that people can observe what normal behaviour is. Therefore, to test for conformity we regard it as essential that people are enabled

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

669

to respond to specific behaviour, as opposed to abstract considerations of what others could possibly do.2 Also, conformity may be a spontaneous, reactive behaviour, whilst the strategy method elicits deliberate strategies. Instead, we employ the Conditional Information Lottery (CIL) procedure (Bardsley, 2000). This mimics the Random Lottery (RL) method, which is widespread in individual choice experiments. In the RL-method, subjects face a number of decision tasks in which a choice has to be made between different lotteries. At the end of the experiment one of these tasks is randomly selected and the lottery the subject has chosen in this task is played out for real. Because subjects only afterwards learn which task is paid, they always have an incentive to choose their preferred lottery. In the CIL-method, likewise, subjects face a number of decision tasks. Each one dramatises the play leading to a node of the game tree. For instance, in one task subjects might see high contributions from others to a public good before making a decision. In another, subjects might move first before being shown low contributions from others. Decisions in these tasks are not paid out and ‘‘otherÕs contributions’’ are deliberately chosen by the experimenter. Hidden amongst these is one bona fide game, in which actual decisions are shown and subjects interact for real. The decisions taken in that task alone determine payoffs. As in the random lottery design, subjects do not know ex-ante which task is real and, thus, have an incentive to treat each one as if it is the game.3 In total our subjects face 12 tasks in which othersÕ contributions are simulated, plus one real game. Table 1 shows how these tasks look in our experiment. In this table, the first 6 columns, labelled X
i, show the vector of othersÕ contributions in a subjectÕs own group, whether the subject is in A or B. The remaining 6 columns, labelled Y, show the vector of contributions in the other group, shown only to group B. To illustrate how this table reads consider task 1. Here, ‘‘–’’ at the 6th position of the subjectÕs own group indicates that all subjects move last in the sequence. Therefore, in stage 1, every subject in a group A waits until five (fictional) first movers of the own group have ostensibly completed their choices. In stage 2, every subject in a group A sees these (fictional) contributions X
i, which are 2-0-2-1-0 in this case, and makes his or her own choice. In stage 3, every subject in a group B sees (fictional) contributions Y of the other group. In task 1 these contributions are 0-0-3-0-1-0. In this stage, again, every subject in a group B waits until five (fictional) first movers of their own group have made their choices. In stage 4, every subject in a group B sees the (fictional) contributions X
i of the own group, which are 2-0-2-1-0 in this case, and makes his or her own choice.

2

In experimental economics there is a debate whether it matters to run an experiment ‘‘hot’’, that is, in the real choice mode, or ‘‘cold’’ in the strategy mode. Evidence on this issue is inconclusive. For instance, Brandts and Charness (2000) find no difference between hot and cold decisions. In contrast, Weber, Camerer, and Knez (2004) find that the decision mode matters. 3 Notice that the CIL does not involve deception as subjects are fully informed about the procedure. See the instructions in the appendix, which included a demonstration to aid comprehension. Nor does it involve ‘‘playing games with computers’’; there is only one payoff-relevant task and that is with people.

670

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

Table 1 Tasks used in the experiment Task a

The subjectÕs own group: X
i

The other group: Y

1 (LL) 2a (LH) 3a (HL) 4a (HH) 5 6 7 8 9 10 11 12

2 1 9 10 5 10 – – – – – –

Real game

Actual behaviour shown for both groups

0 2 10 9 10 0 0 0 0 1 6 0

2 0 8 8 9 9 0 10 9 0 6 10

1 0 9 8 5 10 0 1 10 3 3 0

0 1 9 10 9 9 0 5 10 5 0 7

– – – – – – 0 1 5 5 0 5

0 10 0 9 0 0 0 1 0 0 4 3

0 10 0 10 0 0 0 0 10 9 5 5

3 7 2 8 3 3 10 4 0 0 10 0

0 9 1 7 0 0 9 0 0 0 7 0

1 9 0 10 3 4 8 0 2 7 10 5

0 8 0 9 2 2 6 7 0 0 0 5

Numbers show fictional contributions, ‘‘–’’ indicates a subjectÕs position of contribution in the sequence. a Tasks 1–4 are our test conditions for conformity. Tasks 5–12 serve to conceal the real game.

Tasks 2 to 6 are run in the same way. In these tasks, all subjects actually decide last. In tasks 7 to 12, all subjects actually choose first. Take task 7 as an example: in stage 1 of this task every subject in a group A contributes to the public good before being shown NA
1 contributions (the four zeroes in the table plus his or her own decision). These are followed by an ostensible last contribution in stage 2. In stage 3, every subject in a group B sees the (fictional) contributions Y of group A, which are 0-0-10-9-8-6 in this case. The subject then chooses, before being shown an ostensible last contribution in stage 4. In the real game, only real contributions are shown, in the order they actually occur. This means that NA
1 subjects in a group A contribute first in stage 1, followed by the contribution of only one (the NAth) subject in a group A in stage 2. Then, NB
1 subjects in a group B contribute first in stage 3, followed by the real contribution of only one (the NBth) subject in a group B in stage 4.4 Including the real game every subject makes 13 contribution choices. Tasks 1 to 12 are presented to every subject in random order; that is, subjects go through these tasks in an independent random order. The position of the real game, though, is necessarily the same for all subjects in a session. Across sessions the position of the real game is also randomly determined. In order to render the CIL incentive-compatible it is necessary that subjects attach a positive probability to each taskÕs being real. We took several precautions to ensure this. First, to make the fictional contributions appear as realistic as possible, data from a previous public good experiment were used to generate them.5 Second, our tests of conformity concentrate on four test conditions, which are implemented in tasks 1 to 4 (see next section). To implement these test conditions whilst concealing 4 5

Whether a subject moves first or last in the sequence in the real game is randomly determined. Data was obtained from Bardsley (2000), from the simultaneous play, one-shot treatment.

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

671

the real game, we added eight further fictional tasks (5 to 12), six of which involve contributing in first position. We also took care that subjects could not detect the real game by watching or listening when the experiment was running. Subjects were seated non-adjacently in the computer laboratory, at terminals separated by partitions, such that they could not see anyone elseÕs screen. Moreover, whenever it was not someoneÕs turn to make a contribution, the subject had to enter a letter randomly generated by the program, disguising the timing of subjectsÕ decisions. 2.2. Experimental hypothesis Our test for conformity is based on tasks 1 to 4. These tasks systematically vary contributions of others in both groups. We will use the notation X
iY to represent a task, where each component can take the value Low or High, depending on whether low or high contributions are shown. X
iY = LL is task 1, depicting low contributions in a subjectÕs own group and the other group. In task 2 subjects see low contributions from their own group and high ones from the other group. Hence, we call this task LH. Similarly, tasks 3 and 4 are named HL and HH respectively, because these tasks leave constant a subjectÕs own groupÕs contributions at high but vary the other groupÕs contributions from low (HL) to high (HH) (see Table 1). Table 2 arranges tasks 1 to 4 in a 2 · 2 matrix which manipulates both groupsÕ contributions. Call cX
iY the average contribution of subjects in group B in task X
iY = LL, LH, HL, HH. If conformity exists, it will surface in two comparisons. First, contributions of subjects in group B in task LH will be higher than in task LL, that is, cLH > cLL . Second, contributions of subjects in group B in task HH will be higher than in task HL, that is, cHH > cHL . Both comparisons hold constant the level of contributions X
i within a subjectÕs own group, but change the level of contributions of the other group from low to high. Summing over both effects, we test for conformity by a single hypothesis: Conformity Hypothesis: If conformity exists, the sum of average contributions of subjects in group B over tasks LH and HH is larger than over tasks LL and HL, that is, ðcLH þ cHH Þ > ðcLL þ cHL Þ. 78 group A subjects and 78 group B subjects participated in 13 sessions (6 group A and 6 group B subjects per session). Since we concentrate on group B subjects, this gives us 78 paired observations to test our conformity hypothesis. All sessions were conducted in the experimental laboratory of CREED at the University of AmsterTable 2 Illustration of treatment conditions within-subjects of group B in tasks 1–4 Contributions X
i of B subjectsÕ own group Low High

Contributions Y of the other group Low

High

LL (task 1) HL (task 3)

LH (task 2) HH (task 4)

672

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

dam, in January 2002. Subjects on average earned €13.70 in approximately one hour. Subjects went through detailed instructions on their computer screens (see Appendix A). They experienced three practice rounds before the actual experiment started. In addition, a summary containing the vital elements of the procedures was read aloud to subjects by the experimenter. We checked subjectsÕ understanding of how many situations they would be shown (a total of 13) and how many of these would be real and so relevant for payoffs (only one). We explicitly pointed out that nothing could be learnt from one situation to another, because of the CIL set-up. 3. Results In the real game (inserted randomly into the sequence of 12 tasks) subjects on average contributed 47.7 percent of their endowment to the public good. This replicates findings from previous studies exploring behaviour in one-shot public goods environments (Ledyard, 1995). In addition we observe the usual crowding-in within groups (see Result 1 below). We take this as an indication that subjects understood the procedures of the experiment and that our implementation of the CIL was successful. Fig. 1 summarises the contribution behaviour in the four test tasks (for summary statistics regarding all tasks see Appendix B). The columns show the average contributions of subjects in group B. Prior to their own choice these subjects see the fictional contributions X
i of their own group plus the fictional contributions Y of the other group. Total crowding-in: We first show that there is crowding-in. Fig. 1 reveals that contributions of other members of a subjectÕs group exert a strong impact on their personal contribution: between tasks LL and HL average contributions shift from 2.45 tokens, when the subjectÕs groupÕs contributions are low, to 3.97 tokens, when those contributions are high. Similarly, between tasks LH and HH average contributions 6 4.83

Mean contribution

5 3.97

4 3

2.45

2.77

2 1 0

LL

LH

HL

HH

tasks Fig. 1. Mean contributions in group B in conformity-testing tasks.

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

673

move from 2.77 to 4.83 tokens. Summing over both effects, the total crowding-in effect is ðcHL þ cHH Þ
ðcLL þ cLH Þ ¼ 3.58 tokens. This effect is strongly significant (p = 0.000, Wilcoxon signed-rank test based on 78 observations, one-sided).6 We state this as our first result: Result 1. There are strong effects of crowding-in. SubjectsÕ contributions correlate positively with contributions of others. Conformity: Result 1 replicates what we know from previous studies. The main question we pose in this paper is, however, to what extent result 1 can be explained by conformity. From Fig. 1 we see that high contributions of the other group seem to raise contributions in task LH relative to task LL and that low contributions of the other group reduce contributions in task HL relative to task HH. We now evaluate our conformity hypothesis as stated in Section 2: the sum of average contributions over tasks LH and HH is ðcLH þ cHH Þ ¼ 7.60; that over tasks LL and HL is ðcLL þ cHL Þ ¼ 6.42. The difference between these numbers, ðcLH þ cHH Þ
ðcLL þ cHL Þ ¼ 1.18, is significant at the 5% significance level (p = 0.049), i.e., consistent with our conformity hypothesis we find ðcLH þ cHH Þ > ðcLL þ cHL Þ. Conformity predicts an effect when one systematically varies the contributions of the other group. In contrast, conformity does not predict an effect when the contributions of the other group overall remain the same. The sum of contributions over tasks LL and HH should therefore be the same as that over tasks LH and HL. The sum of average contributions over tasks LL and HH is ðcLL þ cHH Þ ¼ 7.28, and that over tasks LH and HL is ðcLH þ cHL Þ ¼ 6.74. This difference is not significant (p = 0.452), consistent with conformity. Remember that our test is based on the behaviour of last movers in the sequence, that is, subjects have full information about the behaviour of others in both groups. Moreover, subjects interact only once and know this. Therefore, belief-based reciprocity cannot explain our findings. We conclude that the observed responsiveness towards a payoff-irrelevant group cannot be explained by reciprocity. We therefore state the following result: Result 2. SubjectsÕ contributions to the public good correlate positively with contributions in a payoff-irrelevant group; conformity exists. Separating the crowding-in effect into conformity and reciprocity: Result 2 shows that conformity, distinct from reciprocity, is another cause of crowding-in in public good experiments. We now consider the relative magnitudes of these forces.7 To quantify them let us assume that reciprocity and conformity are additive. In the previous paragraphs we have quantified the total crowding-in effect as ðcHH þ cHL Þ
ðcLH þ cLL Þ ¼ 3.58 and the total conformity effect as ðcLH þ cHH Þ
ðcLL þ cHL Þ ¼ 6 Unless otherwise stated, we use the same test throughout this section. For all our findings this test reveals qualitatively the same results as the parametric matched-pairs t-test. 7 We are grateful for helpful comments by Jonathan Baron and an anonymous referee on how to estimate the size of the conformity effect.

674

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

1.18 tokens. Hence, the proportion of crowding-in due to conformity is roughly one third (1.18/3.58 = 0.33). The remaining two thirds of crowding-in are due to reciprocity.8 We summarise this finding by our final result: Result 3. Assuming that reciprocity and conformity are additive, reciprocity accounts for roughly 2/3 and conformity for 1/3 of crowding-in.

4. Discussion The evidence for conformity reported in this paper is implausibly characterised as inequality aversion, intention-based reciprocity, or rational imitation. Inequality aversion is a motive according to which individuals dislike differences between their own payoffs and those of others. There is some experimental evidence of inequality aversion (see Camerer, 2003, for a survey). To evaluate its scope for our design, though, consider for examples tasks LH and LL. A subject in task LH sees another group which is cooperative. Because observed contributions of a subjectÕs own group are low, payoffs of the other group will exceed those of his or her own group in this task. Reducing oneÕs contribution compared to task LL raises oneÕs own payoff and is the only way a subject can bring about a reduction of inequality between oneÕs own payoff and that of the other group. This is the opposite of what we find.9 Models of inequality aversion follow standard economics practice and define the utility of an action only in terms of its payoff consequences. However, this strict consequentialism is rather implausible. Harris and Joyce (1980), for example, find evidence that people care about equality of inputs to group effort tasks, not just outcomes. If, however, one were to assert that subjectsÕ perceptions of fair inputs are a function of othersÕ contributions, regardless of group membership, this seems equivalent to a normative conformity concept. It seems more natural to interpret a concern for input-equality as reciprocity based, though. Indeed this is explicit in SugdenÕs (1984) reciprocity model, according to which, very roughly, people believe that if enough members contribute (or ‘‘input’’) at a certain level to a public good, others are obliged to contribute equally.10 8 Implicitly, this calculation assumes that the conformity effect towards the other group is the same as towards the subjectÕs own group. In our design subjects see 6 contributions of the other group and only 5 contributions of their group. Normalizing the conformity effect with the number of (other) subjects in each group, would impose an alternative assumption, for which the proportion of total crowding-in due to conformity reduces to (5/6)*1.18/3.58 = 0.28. 9 Similarly, in task HL and HH, a subject sees another group which is less cooperative than their own. In this situation, a reduction of their own contribution, as predicted by conformity, would increase the individualÕs payoff and expand the inequality between that payoff and the other groupÕs payoffs. Therefore, inequality aversion cannot explain our findings. 10 One could, alternatively, assert that subjects perceive it to be unfair to the other group to make lower contributions. This would provide a distinct explanation. However, it would be a peculiar sentiment. For the other group would have no cause to complain because of the separation of public good technologies. We know of no evidence of such sentiments.

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

675

Other reciprocity models depart from consequentialism in emphasizing that agents seek to reward kind intentions or punish hostile ones (Falk, Fehr, & Fischbacher, 2003a; Rabin, 1993). Regarding our design, the information about what another group did might, conceivably, shift a subjectÕs perception of what constitutes (un)kind behaviour of the own group. For instance, if subjects reward their own group for being more cooperative than the other group, contributions will be higher in task HL than in task HH. Likewise, if subjects punish their own group for being less cooperative than the other group, contributions shall be lower in task LL than in task LH.11 Notice that these predictions again run against our hypothesis of conformity though. It does not appear, therefore, that non-consequentialist versions of reciprocity or inequality aversion provide competing explanations for our result. Finally, consider models of rational imitation and information cascades. Regarding the former, learning the selfishly rational strategy in our game amounts to learning to contribute zero. So this could not explain any positive impact of the other groupÕs contributions. Also we do not provide feedback about othersÕ payoffs, so this kind of learning is hindered by design. Information cascades require imperfect information (for a discussion, see Bikhchandani et al., 1998). In our test conditions, group B subjects have full information about the contributions of all other subjects in their group. Hence, there is no imperfect information problem. 5. Conclusion This paper has distinguished two accounts of crowding-in: reciprocity and conformity. Our main result is that conformity exists and explains roughly one third of the total crowding-in of voluntary contributions. We regard these results as important, despite the apparent weakness of conformity compared to reciprocity. For, firstly, the scope of the two effects outside the laboratory differs. Conformity has potentially broader scope than reciprocity, in that reciprocity obtains only between agents whose actions affect each otherÕs welfare. For instance, there appears to be crowding-in in voluntary contributions to charities. Reciprocity often provides a problematic explanation since, for example, overseas famine relief does not involve returning benefits obtained from other donors. Whilst altruism, alternatively, implies crowding-out (see, for example, Andreoni, 1998). This is because altruistic utility is some convex combination of own and othersÕ utility, and therefore predicts that someone donates less as others donate more. Conformity provides an intuitive explanation for crowding-in effects in charitable giving, because contributions of others may serve as a signal for appropriate behaviour. This is conjectured, for instance, by List and Lucking-Reiley (2002) to explain a seed-money effect in charitable fundraising.12 11

We have also tested this argument statistically. We find no evidence that subjects punish (reward) their own group when it performs worse (better) than the other group. 12 Related econometric studies are Paque´ (1986), Abrams and Schmitz (1984), Steinberg (1985, 1991), Khanna, Posnett, and Sandler (1995) and Connolly (1997). Whilst theory implies complete crowding-out, it is estimated to be either minimal (the first three studies) or negative – that is, crowding-in (the others).

676

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

Secondly, evolutionary game theorists have recently introduced conformity to stabilise punishment-sustained cooperation. Only a weak conformity effect, such as we find, is necessary for this (see, for example, Henrich & Boyd, 2001). Finally, the effects of conformity might ÔsnowballÕ when interaction is repeated, as Carpenter (2004) claims. In this study, we did not allow for such effects to evolve, since isolation of conformity required a one-shot game protocol.

Appendix A. Instructions Welcome. In this experiment, which investigates decision making, you will be placed in a number of situations. Only one of these will be real, the others will be fictional. An example of this procedure follows below. Example: The experimenter, through this program, is about to make you an offer. The offer is one of three possibilities. One but only one of these is REAL. You have to decide whether to accept or reject each of the three possible offers, before you learn what the offer was. What happens then depends on how you responded to the real offer. 1. You will be paid 5 euros if you do a maths test and score above 60% AND take part in this experiment. Make your choice: accept/reject. 2. You will be paid 4 euros if you fill out a questionnaire about your diet AND take part in this experiment. Make your choice: accept/reject. 3. You have been offered 7 euros. You do NOT have to do anything in return [other than participate in this experiment. Make your choice: accept/reject. The offer was in fact number 3. The 7 euros you have just accepted are your resources to be used in the experiment. They have been given as 10 tokens worth 0.7 euros each. You can walk away with more or less than this depending on how you and other subjects behave in the real situation, but you cannot leave with less than 3.5 euros. Below you are given more detailed instructions. Scroll through these and press the button at the end to indicate that you are ready. The 13 situations you will confront are all set in the following context. There are two sets of six people, A, B, C, D, E, F (persons A–F) and G, H, I, J, K, L (persons G–L). Persons A–F are linked to each other but not to G–L. Similarly, G–L are linked to each other but not to A–F. You are in set G–L. Each person has 10 tokens [worth 0.7 euros each]. In each situation, each person must decide how to use their ten tokens. Each person will leave the experiment with a monetary reward. The size of this depends on what everyone, with whom they are linked, themselves included, does with their tokens in the real situation. There are two possible uses for each token: it can be taken by you personally or put into an account: account #2. To begin with there are no tokens in this account. Only people G–L can put money into it; persons A–F have an entirely separate account, account #1. The money in account #2 will be multiplied by 3 and split equally

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

677

between agents G–L. The money in account #1 will be multiplied by 3 and split equally between agents A–F. This means that each token taken by you leads to a reward to yourself of 0.7 euros but does not benefit anyone else, whilst each token put into account #2 results in a payment of approximately 0.35 euros to everyone in G–L including you, there being 6 people in this set: (0.7 · 3)/6 = 0.35 euros. You decide how much of the 7 euros goes into account #2 by entering a number of 0.7 euro tokens [0–10] to go in. Any token you do not put into account #2 goes to you; not putting it into account #2 is the same as taking it directly. A table showing the amount of cash you will be paid based on how many tokens there are in account #2 has been placed by your computer for ease of reference. A.1. Instructions for group A In each situation, you will be given the following details: (i) numbers representing othersÕ decisions; (ii) whether or not the decisions of A–F are shown to G–L before they make their decisions. As in the example of the offer, only in one case does any of this correspond to anything real. In this real situation, both (i) and (ii) are genuine information, so, for example, the numbers will show the actual choices made by persons A–F. In the others, (i) and (ii) describe purely fictional situations, so nothing really happens, nothing you decide will actually be shown to anyone else and the numbers are made up by the experimenter. Note that because there is only one real situation it is not possible to learn about other peopleÕs behaviour as the experiment progresses. Regarding (i), each situation shows five of the six people deciding at the same time what to do with their tokens. Then the sixth person makes their decision. So, if you are one of the five people in first position, you will decide without seeing othersÕ choices, but you will be shown othersÕ decisions after you have made yours. If you are placed in last position, you will decide after seeing the five other decisions from persons A–F. The letters A–F do NOT correspond to the order; any person may be placed in either position. Regarding (ii), it is possible, in the real situation only, that the decisions made by A–F including yours, will be shown to agents G–L before they make their decisions. After each situation has been played out, you will be shown how much money you will be given if that is the real one. We would like you to treat each situation as if it is real and the only situation. Note that, for all you know, each one could be the real one, in which case ALL information you are given about it is accurate, and only the real one has any effect on the outcome. (Remember that for one of the offers in the example, all the information turned out to be accurate but the other offers were purely fictional. The experiment works in the same way.)

678

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

The section below shows you the display you will see during the experiment.

This shows that 23 tokens were put into account #1 by persons A–F. When it is not your turn to make a decision you will be asked to enter a letter. You will now face three practice situations to get used to the program. These are fictional situations and therefore none of them are paid. A.2. Instructions for group B In each situation, you will be given the following details: (i) numbers representing decisions of persons A–F; (ii) numbers representing decisions of other persons in G–L. As in the example of the offer, only in one case does any of this correspond to anything real. In this real situation, both (i) and (ii) are genuine information, so the numbers will show the actual choices made by persons A–F and the actual choices made by persons G–L. In the others, i and ii describe purely fictional situations, so nothing really happens, nothing you decide will actually be shown to anyone else and the numbers are made up by the experimenter. Note that because there is only one real situation it is not possible to learn about other peopleÕs behaviour as the experiment progresses.

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

679

Each situation shows five of the six people in A–F deciding at the same time what to do with their tokens. Then the sixth person makes their decision. Then, decisions from A–F are shown to G–L. Then the same process happens with persons G–L; 5 of the 6 decide first, followed by the last person. So, if you are one of the five people in first position, you will decide after seeing decisions from A–F but without seeing othersÕ choices from G–L. In this case you will be shown othersÕ decisions from G–L after you have made yours. If you are placed in last position, you will decide after seeing the five other decisions from persons G–L. Decisions by persons G–L are NOT shown to persons A–F. The letters G–L do NOT correspond to the order; any person may be placed in either position. After each situation has been played out, you will be shown how much money you will be given if that is the real one. We would like you to treat each situation as if it is real and the only situation. Note that, for all you know, each one could be the real one, in which case ALL information you are given about it is accurate, and only the real one has any effect on the outcome. (Remember that for one of the offers in the example, all the information turned out to be accurate but the other offers were purely fictional. The experiment works in the same way.) The section below shows you the display you will see during the experiment.

This shows that 17 tokens were put into account #1 by persons A–F. 24 Tokens were put into account #2 by persons G–L. When it is not your turn to make a decision you will be asked to enter a letter.

680

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

You will now face three practice situations to get used to the program. These are fictional situations and therefore none of them are paid.

Appendix B. Summary statistics by tasks N = 156 (78 subjects in group A, 78 subjects in group B). Task

Real 1

2

3

4

5

6

7

8

9

10

11

12

Subjects in group A Mean 4.83 2.14 1.67 4.49 4.29 4.03 3.87 5.51 4.74 5.09 5.03 5.00 4.88 Std 3.50 3.38 3.12 4.20 4.28 3.80 4.14 3.43 3.61 3.60 3.72 3.56 3.77 1a

2a

3a

4a

5

6

7

8

9

10

11

12

Subjects in group B Mean 4.71 2.45 2.77 3.97 4.83 4.14 4.69 4.64 4.37 4.62 4.71 5.26 4.62 Std 4.02 3.73 3.86 4.35 4.47 4.25 4.30 4.14 3.81 4.00 4.00 3.95 3.93 a Test conditions for conformity. References Abrams, B., & Schmitz, M. (1984). The crowding out effect of government transfers on private charitable contributions: Cross sectional evidence. National Tax Journal, 37, 563–568. Andreoni, J. (1998). Toward a theory of charitable giving. Journal of Political Economy, 106(6), 1186–1213. Bardsley, N. (2000). Control without deception: Individual behaviour in free-riding experiments revisited. Experimental Economics, 3, 215–240. Becker, G. (1991). A note on restaurant pricing and other examples of social influences on price. Journal of Political Economy, 99, 1109–1116. Bernheim, B. D. (1994). A theory of conformity. Journal of Political Economy, 109(5), 841–877. Bikhchandani, S., Hirshleifer, D., & Welch, I. (1998). Learning from the behaviour of others: Conformity, fads, and informational cascades. Journal of Economic Perspectives, 12, 151–170. Brandts, J., & Charness, G. (2000). Hot and cold decisions and reciprocity in experiments with sequential games. Experimental Economics, 2, 227–238. Brandts, J., & Fata´s, E. (2001). Social information and social influence in an experimental dilemma game. LINEES Working Paper 29/00. University of Vale`ncia. Camerer, C. F. (2003). Behavioral game theory – experiments in strategic interaction. Princeton: Princeton University Press. Carpenter, J. (2004). When in Rome: Conformity and the provision of public goods. Journal of SocioEconomics, 33(4), 395–408. Cason, T. N., & Mui, V. L. (1998). Social influence in the sequential dictator game. Journal of Mathematical Psychology, 42, 248–265. Cialdini, R. (1993). Influence: Science and practice. New York: Harper Collins. Connolly, L. (1997). Does external funding of academic research crowd out institutional support? Journal of Public Economics, 64, 389–406. Croson, R. T. A. (1998). Theories of altruism and reciprocity: Evidence from linear public goods games. Working Paper. The Wharton School. University of Pennsylvania. Dufwenberg, M., & Kirchsteiger, G. (2004). A theory of sequential reciprocity. Games and Economic Behavior, 47, 268–298.

N. Bardsley, R. Sausgruber / Journal of Economic Psychology 26 (2005) 664–681

681

Falk, A., & Fehr, E. (2002). Psychological foundations of incentives. European Economic Review, 46, 687–724. Falk, A., Fehr, E., & Fischbacher, U. (2003a). On the nature of fair behavior. Economic Inquiry, 41, 20–26. Falk, A., Fischbacher, U., & Ga¨chter, S. (2003b). Living in two neighborhoods: Social interactions in the lab. IEW Working Paper No. 150. Institute of Empirical Research in Economics. University of Zurich. Falk, A., & Fischbacher, U. (1998). A theory of reciprocity. IEW Working Paper No. 6. Institute of Empirical Research in Economics. University of Zurich. Fischbacher, U., Ga¨chter, S., & Fehr, E. (2001). Are people conditionally cooperative? Evidence from a public goods experiment. Economics Letters, 71, 397–404. Glaeser, E. L., Sacerdote, B., & Scheinkman, J. A. (1996). Crime and social interactions. Quarterly Journal of Economics, 111, 507–548. Harris, R. J., & Joyce, M. A. (1980). What is fair? It depends on how you phrase the question. Journal of Personality and Social Psychology, 38, 165–179. Henrich, J., & Boyd, R. (2001). Why people punish defectors: Weak conformist transmission can stabilise costly enforcement of norms in cooperative dilemmas. Journal of Theoretical Biology, 208, 79–89. Kahan, D. M. (1997). Social influence, social meaning, and deterrence. Virginia Law Review, 83(2), 349–395. Keser, C., & van Winden, F. (2000). Crowding-in and voluntary contributions to public goods. Scandinavian Journal of Economics, 102(1), 23–39. Khanna, J., Posnett, J., & Sandler, T. (1995). Charity donations in the UK: New evidence based on panel data. Journal of Public Economics, 56, 257–272. Ledyard, J. (1995). Public goods: A survey of experimental research. In J. Kagel & A. Roth (Eds.), The Handbook of Experimental Economics (pp. 111–194). Princeton: Princeton University Press. List, J. A., & Lucking-Reiley, D. (2002). The effects of seed money and refunds on charitable giving: Experimental evidence from a university capital campaign. Journal of Political Economy, 110(1), 215–233. Ludwig, J., Duncan, G. J., & Hirschfield, P. (2001). Urban poverty and juvenile crime: Evidence from a randomized housing-mobility experiment. Quarterly Journal of Economics, 116, 655–679. Manski, C. F. (1993). Identification of endogenous social effects: The reflection problem. Review of Economic Studies, 60, 531–542. Manski, C. F. (2000). Economic analysis of social interaction. Journal of Economic Perspectives, 14(3), 115–136. Moscovici, S. (1985). Social influence and conformity. In L. Gardner & E. Aronson (Eds.), The handbook of social psychology (pp. 347–412). New York: Random House. Paque´, K.-H. (1986). The efficiency of tax incentives to private charitable giving: Some econometric evidence for the Federal Republic of Germany. Weltwirtschaftliches Archiv, 122, 690–712. Rabin, M. (1993). Incorporating fairness into game theory and economics. American Economic Review, 83, 1281–1302. Selten, R. (1967). Die Strategiemethode zur Erforschung des eingeschra¨nkt rationalen Verhaltens im Rahmen eines Oligopolexperimentes. In H. Sauermann (Ed.), Beitra¨ge zur experimentellen Wirtschaftsforschung (pp. 136–168). Tu¨bingen: J.C.B. Mohr (Paul Siebeck). Selten, R., & Ostmann, A. (2000). Imitation equilibrium. University of Bonn Discussion Paper 16/2000. Steinberg, R. (1985). Empirical relations between government spending and charitable donations. Journal of Voluntary Action Research, 14, 54–64. Steinberg, R. (1991). Does government spending crowd out donations? Interpreting the evidence. Annals of Public and Cooperative Economics, 62, 591–617. Sugden, R. (1984). Reciprocity: The supply of public goods through voluntary contributions. The Economic Journal, 94(376), 772–787. Weber, R. A., Camerer, C. F., & Knez, M. (2004). Timing and virtual observability in ultimatum bargaining and ‘‘weak link’’ coordination games. Experimental Economics, 7, 25–48. Weimann, J. (1994). Individual behavior in a free riding experiment. Journal of Public Economics, 54, 185–200.