《实验经济学》课程教学资源（参考文献）Is there a hidden cost of imposing a minimum contribution level for public good contributions?

Journal of Economic Psychology 56(2016)74-84 Contents lists available at ScienceDirect Economic Journal of Economic Psychology Psychology journal homepage: www.elsevier.com/locate/joep ELSEVIER Is there a hidden cost of imposing a minimum contribution level Cross Mark for public good contributions? Martin G. Kochera.b.c, Peter Martinsson b.d, Emil Perssonb, Xianghong Wange.* Department of Economics, University of Munich, Germany Department of Economics, University of Gothenburg, Sweden School of Economics and Finance, Queensland University of Technology, Australia "Department of Economics, Linkoping University, Sweden School of Economics, Renmin University of China, China ARTIC LE INFO ABSTRAC T Article history: We examine the effects of either exogenously imposing or endogenously letting subjects Received 26 November 2014 choose whether to impose minimum contribution levels (MCLs) in a linear public goods Received in revised form 19 May 2016 experiment using the strategy method. Our results on contribution levels to the public Accepted 30 May 2016 goods are fairly independent of how MCLs are imposed. We find that the main effect of Available online 1 June 2016 an MCL on unconditional contributions is that it increases low contribution levels to the MCL imposed, while the effect of those contributing more than the MCL before its introduc- JEL classification: tion depends on the size of the MCL. Unexpectedly, there is much more crowding out for a C91 D03 low MCL than for a relatively high MCL. However, the distribution of contribution types is D64 stable across different MCLs. 2016 Elsevier B.V. All rights reserved. Keywords: Cooperation China Experiment Minimum level Public good 1. Introduction One option available to policy makers for increasing contributions to public goods is to introduce minimum contribution levels (MCLs). For example, some municipalities introduce minimum levels regarding the sorting of waste, while others introduce driving restrictions in order to contribute to the public good of clean air or of low emission levels. Other examples are requests of minimum monetary contributions in disaster relief or in keeping the sidewalks in front of one's house free of snow and ice in winter; in the former case there are such practices in China after the Wenchuan earthquake, and in the latter case, there are certain minimum requirements established by law in Germany and Austria. Existing studies have analyzed the effects of an MCL, often framed as a tax, on voluntary contributions using public goods experiments (e.g., Andreoni 1993; Chan, Godby, Mestelman, & Muller, 2002; Eckel, Grossman, & Johnston, 2005; Gronberg, Luccasen, Turocy, & Van Huyck, 2012; Sutter and Weck-Hanneman, 2004). The general finding is that a minimum level does not completely crowd out voluntary contributions by reducing intrinsic motivations to contribute to the public good (for an in-depth discussion on * Corresponding author at: School of Economics, Renmin University of China, #59 Zhong Guan Cun Ave., Beijing 100872, China. E-mail address: shwang06@ruc.edu.cn (X. Wang). http:/dx.doi.org/10.1016/j.joep.2016.05.007 0167-4870/ 2016 Elsevier B.V. All rights reserved

Is there a hidden cost of imposing a minimum contribution level for public good contributions? Martin G. Kocher a,b,c , Peter Martinsson b,d , Emil Persson b , Xianghong Wang e,⇑ aDepartment of Economics, University of Munich, Germany bDepartment of Economics, University of Gothenburg, Sweden c School of Economics and Finance, Queensland University of Technology, Australia dDepartment of Economics, Linköping University, Sweden e School of Economics, Renmin University of China, China article info Article history: Received 26 November 2014 Received in revised form 19 May 2016 Accepted 30 May 2016 Available online 1 June 2016 JEL classification: C91 D03 D64 Keywords: Cooperation China Experiment Minimum level Public good abstract We examine the effects of either exogenously imposing or endogenously letting subjects choose whether to impose minimum contribution levels (MCLs) in a linear public goods experiment using the strategy method. Our results on contribution levels to the public goods are fairly independent of how MCLs are imposed. We find that the main effect of an MCL on unconditional contributions is that it increases low contribution levels to the MCL imposed, while the effect of those contributing more than the MCL before its introduc￾tion depends on the size of the MCL. Unexpectedly, there is much more crowding out for a low MCL than for a relatively high MCL. However, the distribution of contribution types is stable across different MCLs.
2016 Elsevier B.V. All rights reserved. 1. Introduction One option available to policy makers for increasing contributions to public goods is to introduce minimum contribution levels (MCLs). For example, some municipalities introduce minimum levels regarding the sorting of waste, while others introduce driving restrictions in order to contribute to the public good of clean air or of low emission levels. Other examples are requests of minimum monetary contributions in disaster relief or in keeping the sidewalks in front of one’s house free of snow and ice in winter; in the former case there are such practices in China after the Wenchuan earthquake, and in the latter case, there are certain minimum requirements established by law in Germany and Austria. Existing studies have analyzed the effects of an MCL, often framed as a tax, on voluntary contributions using public goods experiments (e.g., Andreoni, 1993; Chan, Godby, Mestelman, & Muller, 2002; Eckel, Grossman, & Johnston, 2005; Gronberg, Luccasen, Turocy, & Van Huyck, 2012; Sutter and Weck-Hanneman, 2004). The general finding is that a minimum level does not completely crowd out voluntary contributions by reducing intrinsic motivations to contribute to the public good (for an in-depth discussion on http://dx.doi.org/10.1016/j.joep.2016.05.007 0167-4870/
2016 Elsevier B.V. All rights reserved. ⇑ Corresponding author at: School of Economics, Renmin University of China, #59 Zhong Guan Cun Ave., Beijing 100872, China. E-mail address: shwang06@ruc.edu.cn (X. Wang). Journal of Economic Psychology 56 (2016) 74–84 Contents lists available at ScienceDirect Journal of Economic Psychology journal homepage: www.elsevier.com/locate/joep

M.G.Kocher et aL/Joural of Ecomomic Psychology56(016)74-84 75 motives,see.e.g.Benabou&).In the context of principal-agent games where the principal can set a minimum effort level that whi in a follow-up studyzem MCL this paper is toprovide amore detaile fhow toa publicgood are affected by ar subject would con their three-fourthon of oher ential side en two types of players in a social dilemma (accounting fo Kelley St ent MCL in ord er to asses t impact on vo Iunaltected by the free riders and conditional cooperators change? remain largely the same as without an mcl despite the forced increase in contributions for very low contributors this is who have becn willing more than the also affected negatively ver,it seer s tha without an M ersho oking at the c NmNo in Nmo de the the aking into account the m and 7,res n Feld 2006) ed to th decision makers slightly downwards.because the bulk of the distribution of contributions is above the MCL without any of the MC 色c心mplcton of oureuts.to s.In fac and the results of our first experiment n Section3 we with endogenou MCLs the results and udes the paper 2.Experiment 1-Exogenously imposed MCLs 2.1.Design of Experiment 1 Our experime tal setup builds on the design by Fischbacher et al.(2001).It is a one-shot linear public goods experimen ehadto deide how u toallocate triateddevl for this suggestio

motives, see, e.g., Bénabou & Tirole, 2006). In the context of principal-agent games where the principal can set a minimum effort level that has to be exerted by the agent, Falk and Kosfeld (2006) find that MCLs have a negative overall effect on vol￾untary contributions, while in a follow-up study Ziegelmeyer, Schmelz, and Ploner (2012) do not generally find a similar effect. The objective of this paper is to provide a more detailed account of how contributions to a public good are affected by an MCL. To this end, we augment an incentivized version of eliciting individual cooperative preferences based on the strategy method (Fischbacher & Gächter, 2010; Fischbacher, Gächter, & Fehr, 2001; Herrmann & Thöni, 2009; Kocher, Cherry, Kroll, Netzer, & Sutter, 2008; Martinsson, Pham-Khanh, & Villegas-Palacio, 2013; Martinsson, Villegas, & Wollbrant, 2015) with two different MCLs. In the design introduced by Fischbacher et al. (2001), individuals make two types of contribution deci￾sions for the public good: (i) unconditional contributions and (ii) conditional contributions, i.e., what the subject would con￾tribute to the public good given different average contribution levels by the other group members. The setup eliminates the strategic uncertainty that is inherent to social dilemma games, and thus lets us focus on the incentive effects of MCLs and their potential side effects. It allows for distinguishing between two types of players in a social dilemma (accounting for about three-fourths of the population): free riders, who contribute nothing regardless of the contributions of others, and con￾ditional cooperators, who are willing to increase their own contribution if they know that others will do so as well (see also Kelley & Stahelski, 1970; Keser & van Winden, 2000). In our laboratory experiment, we implement two different MCLs – a low and a high (2 out of 20 and 7 out of 20 tokens, respectively) – in order to assess their potentially distinct impact on vol￾untary contributions. We are primarily interested in three effects of MCLs on the level of conditional cooperation: (i) Does the MCL affect those who are actually bound by the minimum level in addition to the pure effect of introducing the MCL? For example, do subjects who contribute fewer or equal to 2 tokens without an MCL contribute exactly 2 tokens with an MCL = 2? (ii) Is the distribu￾tion of contributions among those contributing more than the imposed minimum level unaffected by the introduction of the minimum level? (iii) Does the introduction of MCLs change the distribution of contributor types, i.e., does the distribution of free riders and conditional cooperators change? Our first experiment implements exogenous MCLs. The data show that decision makers are affected by MCLs and that the effects on contributions under the low MCL and the high MCL are different. In the case of a low MCL, average contributions remain largely the same as without an MCL, despite the forced increase in contributions for very low contributors. This is explained by a decrease in average contributions by those who have been willing to contribute more than the minimum requirement without the MCL already. Hence, there is crowding out. Under the higher MCL, conditional contributions among those contributing more than the minimum level already before the implementation of the MCL are also affected negatively. However, it seems that those who contribute less than the MCL without an MCL ‘‘overshoot” the MCL. Looking at the con￾tribution schedules of the conditional contribution elicitation, we see that conditional cooperators decrease their contribu￾tions for MCL = 2 (a smaller slope) compared to MCL = 0, whereas the contribution function for MCL = 7 looks almost identical to the one for MCL = 0, taking into account the lower-limit censoring at 2 and 7, respectively. Since the efficiency of institutional mechanisms may depend on how they are implemented (e.g., Tyran & Feld, 2006), we run a second experiment where MCLs are implemented endogenously. More precisely, we elicit preferences over the imple￾mentation of the MCL by letting group members choose whether to implement an MCL or not. One randomly selected group member’s choice then decides whether the MCL is implemented or not.1 Interestingly, we find less ‘‘overshooting” but also less crowding out compared to exogenously implemented MCLs. One explanation for the different impact of the two MCL regimes in both experiments is that MCLs could serve as signals, anchors, or reference points for intrinsically cooperative decision makers. While a low MCL drags the contribution of some decision makers slightly downwards, because the bulk of the distribution of contributions is above the MCL without any restriction, it could be the opposite for higher MCLs. Note however that any potential reference point effects are not as straightforward as they seem: We do not observe the mass of distribution of contributions moving exactly to the levels of the MCLs. In fact, the shifts are more gradual. As a practical implication of our results, it seems important not to introduce too low MCLs in the field in order to avoid potentially adverse reference point effects. The remainder of the paper is organized as follows. Section 2 presents the design and the results of our first experiment. In Section 3 we focus on our second experiment with endogenous MCLs. Section 4 discusses the results and concludes the paper. 2. Experiment 1 – Exogenously imposed MCLs 2.1. Design of Experiment 1 Our experimental setup builds on the design by Fischbacher et al. (2001). It is a one-shot linear public goods experiment, using a variant of the strategy method. In the experiment, subjects were randomly matched into groups of four. Each mem￾ber received an endowment of 20 tokens and had to decide how much to allocate to a private and a public good, respectively. The payoff function for subject i is given by 1 We thank an anonymous referee for this suggestion. M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84 75

M.G.Kocher et al.Journal of Economic Psychology 56(2016)74-84 %=20-G+0.69 (1) mma betweer edto two decuone frt n odtionud combonto the how mucl (rounded to integers).In order to make each choice in the experiment incentiv be yoff-relvhuoof the d public goods game s ne th 2097 exp wer 2.2.Results of experiment i （SeWat9aishbctwecnesusforuncondtionalcontntbutions(ection22）andiesultsforcontdional I contributions r order effects are p sent in our data.for each minimum level.we test whether uncondi tional contri ions are the s e in Part I orders in the follo analyses.The unconditional contributions with MCL=0 are well in line with comparable studies that usea smone-shot eal.(200)for instance,report averages for unconditionlcori n ab2.we summarize to the public good.s expected,average contributions are highe vs.092),and this in e is signin cant at the 1 level compared with are significantly higher at thelevel than with MCL(two e effects of mcis on the level of conditional co on (i)Do e who are actu ually bound by the minimum level inaddition to the pure effect nong thos e tha d by the intr f the of fected by the(of).the sub ts contributing 2(7)in treatment MCL-2(MCL-7) 2.the average would ver,the average contribution is uess the ondtionalcontbutionofteoterthregoupmcmbcrsouadedtoiatcgersl

pi ¼ 20 ci þ 0:6 X4 j¼1 cj; ð1Þ where ci denotes the contribution of subject i to the public good. In the experiment, each token was exchanged for 0.7 yuan.2 Assuming that subjects are selfish and rational, the dominant strategy for any marginal per capita return below one is to free ride, i.e., to contribute nothing to the public good. However, since the social return from any contribution to the public good was 2.4 tokens, everybody is better off if all group members contributed. Hence, the participants faced a social dilemma between privately optimal and socially optimal behavior. In the Fischbacher et al. (2001) design, subjects are asked to make two decisions: first an unconditional contribution to the public good and then a conditional contribution. The unconditional contribution is an integer number of tokens in the permissible range. Our three treatments (see Table 1) allow for ci 2 ½0; 20 (as in Fischbacher et al., 2001), ci 2 ½2; 20 (low MCL treatment), and ci 2 ½7; 20 (high MCL treatment). For the conditional contributions, each subject stated how much she would contribute to the public good for any possible average contribution of the three other players in her group (rounded to integers). In order to make each choice in the experiment incentive-compatible, both the unconditional and the conditional contribution must be potentially payoff-relevant. Thus, one of the four subjects was randomly selected, and her conditional contribution, corresponding to the average of the other three members’ unconditional contributions, was relevant as the contribution to the public account. Individual earnings can then be calculated according to Eq. (1). 3 Essentially, the mechanism transforms the simultaneous public goods game into a sequential variant. We implemented a mixture of a within-subject and a between-subject design with full control for potential order effects. The design gives four combinations (sequences) of two one-shot public goods experiments (here denoted Part I and Part II), as summarized in Table 1, given that we always wanted to keep the comparison with the treatment MCL = 0. There was no feedback between Part I and Part II. After conducting Part I, our subjects were randomly matched into groups with new members in Part II, i.e., we implemented a perfect stranger matching, and this procedure was common knowledge from the beginning of the experiment. Both parts were payoff-relevant. The experiment was run with context-free instructions (see Appendix). It was computerized using z-tree (Fischbacher, 2007) and conducted at Renmin University of China in Bei￾jing. Average earnings amounted to 43.90 yuan.4 2.2. Results of experiment 1 We distinguish between results for unconditional contributions (Section 2.2.1) and results for conditional contributions (Section 2.2.2). 2.2.1. Unconditional contributions We begin by testing whether order effects are present in our data. For each minimum level, we test whether uncondi￾tional contributions are the same in Part I and Part II. Since we cannot reject the hypothesis of no order effects in any of the combinations at the 5% significance level based on a two-sided Mann–Whitney U-test, we pool the data over the different orders in the following analyses. The unconditional contributions with MCL = 0 are well in line with comparable studies that use a similar one-shot design in other countries. Kocher et al. (2008), for instance, report averages for unconditional contri￾butions of 8.11 in the U.S.A, 7.53 in Austria, and 7.22 in Japan. In Table 2, we summarize unconditional contributions to the public good. As expected, average contributions are higher when a minimum level of 7 is introduced (8.44 vs. 10.92), and this increase is significant at the 1% level compared with MCL = 0, using a within-subject comparison (two-sided Wilcoxon signed-ranks test, p < 0.01; N = 72). There is no significant change in the levels of contributions when a minimum level of 2 is introduced compared to MCL = 0 (7.58 vs. 7.69). Com￾paring across treatments, contributions with MCL = 7 are significantly higher at the 1% level than with MCL = 2 (two￾sided Mann–Whitney U-test, p < 0.01; N = 144). It is important to note that the baseline level of cooperation under MCL = 0 is not significantly different between the two treatments, i.e. Sequences 1 and 2 versus Sequences 3 and 4 (two￾sided Mann–Whitney U-test, p = 0.32; N = 144). We now look at two questions related to the effects of MCLs on the level of conditional cooperation: (i) Does the MCL affect those who are actually bound by the minimum level in addition to the pure effect of introducing the MCL? (ii) Is the distribution of contributions among those contributing more than the imposed minimum level unaffected by the intro￾duction of the minimum level? Table 3 provides detailed analyses of the effects of introducing an MCL. If behavior is unaf￾fected by the introduction of an MCL of 2 (of 7), the proportion of subjects contributing 2 (7) in treatment MCL = 2 (MCL = 7) is expected to be equal to the proportion contributing 0, 1, and 2 (0, 1, 2, 3, 4, 5, 6, and 7) in treatment MCL = 0. We start our analysis with the low MCL, i.e., MCL = 2. In the case of MCL = 2, 29.2% of decision makers contribute 2 tokens to the public good, compared with 30.6% contributing 0, 1, and 2 under MCL = 0. If all decision makers that are bound by the MCL would contribute exactly 2 tokens in MCL = 2, the average would be 2, obviously. However, the average contribution is 2 At the time of the Experiment 1 USD = 6.45 yuan. 3 Moreover, subjects in the experiment were asked to guess the average unconditional contribution of the other three group members (rounded to integers). The subjects were monetarily rewarded depending on the accuracy of their guesses as in Gächter and Renner (2010). 4 For comparison, a lunch in the student restaurant cost 10 yuan at the time of the experiment. 76 M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84

MC.Kocher et aL/Joural of Economic Psychology 56(2016)74-84 77 Table 1 ry of the experimental design Part I Part I quence 4 Note:MCL=minimum contribution level. Table 2 o(standard) men MCL=0 MCL-2 MCL-7 tencand 358副 769(5.81】 10.92(3.83) Null hypothesis P.values 0.32 W-144 <0.01 a27aa2wa-2a- ributions in the treatments sidcred Average contribution ac 42 20 838 3.420 MLb 99 48 8024867 87 89.…20 ML-hm Mc) 83 88 Null =0 and MCL=2 88l Note:We se two-sided signed-ranks tests 4.50 tokens under MCL-2.This differen tribute more than 2 tokens even when MCL0

4.50 tokens under MCL = 2. This difference between the ‘‘projected” contribution of 2 and actual average contributions is highly significant (two-sided Wilcoxon signed-ranks test comparing actual contribution levels and the projected level of 2, p 2 in MCL = 0) ALL 69.4 9.10 Sequences 3 and 4 MCL = 0 (contributions 0–7) 0, 1, 2, 4, 5, 6, 7 41.7 2.90 MCL = 0 (contributions 8–20) 8, 9, ..., 20 58.3 12.40 MCL = 7 (contributions 7) 7 16.7 7.00 MCL = 7 (contributions 8–20) 8, 9, ..., 20 83.3 11.70 MCL = 7(those who contribute 0, 1, ..., 7 in MCL = 0) ALL 41.7 9.40 MCL = 7 (those who contribute > 7 in MCL = 0) ALL 58.3 12.00 Null hypothesis P-values Sequences 1 and 2 Contributions in MCL = 2 of those who contribute 0, 1, 2 in MCL = 0 are equal to 2 2 in MCL = 0 are equal in MCL = 0 and MCL = 2 7 in MCL = 0 are equal in MCL = 0 and MCL = 7 0.60 Note: We use two-sided Wilcoxon signed-ranks tests. Table 2 Average unconditional contributions (standard deviations in parentheses). Treatment MCL = 0 MCL = 2 MCL = 7 Sequences 1 and 2 7.58 (6.53) 7.69 (5.81) Sequences 3 and 4 8.44 (6.02) 10.92 (3.83) Null hypothesis P-values Contributions in MCL = 0 (Sequences 1 and 2) = MCL = 0 (Sequences 3 and 4) (N = 144) 0.32 Contributions in MCL = 2 = MCL = 7 (N = 144) <0.01 Note: We use Mann–Whitney U-tests. Sequences 1 and 2: MCL = 0–2 and MCL = 2– 0; Sequences 3 and 4: MCL = 0–7 and MCL = 7–0. Table 1 Summary of the experimental design. Part I Part II No. of subjects Sequence 1 MCL = 0 MCL = 2 36 Sequence 2 MCL = 2 MCL = 0 36 Sequence 3 MCL = 0 MCL = 7 36 Sequence 4 MCL = 7 MCL = 0 36 Note: MCL = minimum contribution level. M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84 77

M.G.Kocher et al./Journal of Economic Psychology 56(2016)74-84 shooting"that w ed under MCl=2 if all of our decisio under MCL=0 would cor ctu the proi and actu d tverage co bution 0dngathOewgComnDutemOehan7okem without an MCL we observe a ages of 12.40 when MCL-0 and 12.00 CL=7 beshows in the two treatments inorder to give the changes LI the tthe ons when they than the MCL efore its introduction. we tak cros MCL-0,then the only =0 and M 以epCe8 Whale the sohoeMC des ot incr ease average conditional contribu ers average contrib ons,th slope in M 7is approximately similar to theon range clearly above the lower of.In order to study this effect further.we ran regressions to expain contri ion probabilities in the treatments MCL-0 MCL-2 Contribution>2()Contribution=2( ≥61or2 58 1 M1.7 Contribution >7()Contribution-7 ( 234567891012131415161718192 一PCCMCL0-MCL2-M7

shooting” that we observed under MCL = 2. If all of our decision makers, contributing 0–7 tokens under MCL = 0, would con￾tribute exactly 7 tokens when MCL = 7, the average contribution should be 7; however, their actual average contribution is 9.40 tokens under MCL = 7. The difference between the projected and actual average contributions is highly significant (two￾sided Wilcoxon signed-ranks test comparing actual contribution levels and the projected level of 7, p 2 (%) Contribution = 2 (%) Contribution > 2 58.3 11.1 Contribution = 0, 1 or 2 12.5 18.1 MCL = 7 Contribution > 7 (%) Contribution = 7 (%) Contribution > 7 54.2 4.2 Contribution = 0, 1, 2, 3, 4, 5, 6 or 7 29.1 12.5 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 PCC MCL=0 MCL=2 MCL=7 Fig. 1. Average conditional contribution (others’ average contributions on the horizontal axis and own average conditional contribution on the vertical axis). Note: PCC = perfect conditional cooperation. 78 M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84

MC.Kocher et aL/Joural of Economic Psychology 56(2016)74-84 79 MCL= ndes☒ p-shaped(☒ M=0 4 00 MCL=7 onditional cooperators ( Free riders ( Hump-shaped ( Others ( MCL-0 fHiticnlcoperatois 444 83 14 funtion of othars'au contributions clusterin on the individual level in order to account for multiple entries for the individual.The the visual impression from are available on request hump-shaped,and others Here are the definitions that we use:If a subject's conditional contribution in oution of the other group members,the subject is fied as a condition e) easing the in rank coneia ve,their ibutions decrease based ona Crsed&asihcaionasheoneuseduptotheiniecionpomt e M n the stability ofthe classifica for an assessment of temporal stability of contribution types,but not across different institutional environments)The table e for in 0n0 not or 2010:c do001co0perato05na5 2009:Kocher et al.2008:Martinsson et al 2013:Martinsson et al 2015)and switches from one type to another are rare. 3.Experiment 2-Endogen usly imposed MCLs 3.1.Design of Experiment 2 The setup follows Experiment 1 closely.As in Experiment 1.subjects were randomly matched into groups of four.and iect and a ber mbre of (sequences)of two one-shot public noted Part as alre eady s nmarized in table Th or MCL-2 in Sequences 1 and 2.and the preference for either MCL-0or MCL-7 in Sequences 3 and 4.Once everyone had

butions as a function of others’ average contributions, clustering on the individual level in order to account for multiple entries for the individual. The results confirm the visual impression from Fig. 1 and are available on request. Based on the conditional contribution schedules, we can classify subjects into different types of contributors. We follow the convention (e.g., Fischbacher et al., 2001) and define four general types of subjects: conditional cooperators, free riders, hump-shaped contributors, and others. Here are the definitions that we use: If a subject’s conditional contribution increases weakly monotonically with the average contribution of the other group members, the subject is classified as a conditional cooperator. A subject is also classified as a conditional cooperator if the relationship between own and others’ average contri￾butions is positive and significant at the 1% significance level based on the Spearman rank correlation coefficient. Hump￾shaped contributors are subjects who show weakly monotonically increasing (or increasing with a Spearman rank correlation coefficient at the 1% significance level) contributions up to some, non-maximal level of others’ contributions; above that level, their conditional contributions decrease based on a reversed classification as the one used up to the inflection point. A free rider is a subject who contributes zero for all levels of the other group members’ contributions under MCL = 0, 2 tokens under MCL = 2, and 7 tokens under MCL = 7. Finally, those who cannot be categorized into any of the above categories are referred to as others. Tables 5 and 6 report results on the stability of the classification across different MCLs by providing a transition matrix with regard to types. Overall, the distributions of contribution types are quite stable across MCLs (see Ruigrok, Volk, & Thöni, 2012, for an assessment of temporal stability of contribution types, but not across different institutional environments). The tables show, by looking at the diagonal, that a large proportion of subjects do display the same type of behavior under different insti￾tutional mechanisms. The few changes go in all directions and seem to be more or less random. Some conditional cooperators become free riders; particularly under MCL = 7, these are decision makers who have an increasing contribution schedule under MCL = 0 with a maximum contribution of 7 tokens or less. Overall the distribution of conditional cooperators and free riders are not out of the ordinary (see for instance, Fischbacher & Gächter, 2010; Fischbacher et al., 2001; Herrmann & Thöni, 2009; Kocher et al., 2008; Martinsson et al., 2013; Martinsson et al., 2015) and switches from one type to another are rare. 3. Experiment 2 – Endogenously imposed MCLs 3.1. Design of Experiment 2 The setup follows Experiment 1 closely. As in Experiment 1, subjects were randomly matched into groups of four, and they made contribution decisions following the Fischbacher et al. (2001) design. Again, we implemented a mixture of a within-subject and a between-subject design with full control for potential order effects. The design gives four combinations (sequences) of two one-shot public goods experiments (denoted Part I and Part II), as already summarized in Table 1. The only new feature is the endogenous determination of the MCL: the experiment began with a ‘‘voting” stage where subjects indicated whether they wanted Part I or Part II to be payoff-relevant. This choice indicates the preference for either MCL = 0 or MCL = 2 in Sequences 1 and 2, and the preference for either MCL = 0 or MCL = 7 in Sequences 3 and 4. Once everyone had casted their vote, Experiment 2 proceeded as Experiment 1. It was common knowledge that, in the end, one participant in each group would be randomly selected and her choice would be implemented (thus determining which part of the exper￾iment would be payoff-relevant). This means that contribution decisions in both parts of the experiment were made in a sit￾uation where the MCL is endogenously imposed by one randomly selected group member. Table 5 Distribution of contributor types at MCL = 0 and MCL = 2 (at individual level). MCL = 2 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 33.3 4.2 4.2 4.2 Free riders 6.9 27.8 5.6 0.0 Hump-shaped 0.0 2.8 0.0 2.8 Others 1.4 0.0 2.8 4.2 Table 6 Distribution of contributor types at MCL = 0 and MCL = 7 (at individual level). MCL = 7 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 44.4 8.3 1.4 0.0 Free riders 1.4 16.7 0.0 0.0 Hump-shaped 9.7 2.8 2.8 1.4 Others 1.4 0.0 4.2 5.6 M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84 79

M.G.Kocher et al./Journal of Economic Psychology 56(2016)74-84 The ment was run by using context-free instructions(see Appendix)that followed the instructions nExpe xplain the choice.It was again computerized usingz-tree(Fischbacher.2007)and con me experime enmin Univ nt (24 i rsity of China in Beijing as the fir experment.A tota subject 3.2.Results of Experiment 2 In order to facilitate comparisons of the two experiments.we proceed with our analysis of Experiment 2 in the same way as for Experiment 1.We first present results for unconditional contributions(Section 3.2.1)and then results for conditional Umcomditional coibutiont unconditional contributions(standard deviations in parentheses)for Exper MC -0 MCL-7 Sewuencea and 6涮 683549 10.17(4.46 Null hypothesis P-values 0.77 001 Treatment Contributions considered Proportion of subjects (% Average contribution s 1 and 2 842 20 沿 859 MCL(ontibutions20) 3.420 6 28 MCL(Choe cnmbueM 絀 8287 9 Mobui) 8.9.-20 68 20g a 6 Null hypothesis P-value ces 1 and 2 89 Note:We use rwo-sided Wilxon signed-ranks tests

The experiment was run by using context-free instructions (see Appendix) that followed the instructions in Experiment 1, except for the necessary changes to explain the choice. It was again computerized using z-tree (Fischbacher, 2007) and con￾ducted in the same experimental lab at Renmin University of China in Beijing as the first experiment. A total of 96 subjects participated in the experiment (24 in each sequence). Since only one part was payoff-relevant in Experiment 2 by the design of the endogenous mechanism, we doubled the exchange rate compared to Experiment 1 to make the monetary incentives comparable in the two experiments. Average earnings amounted to 44.0 yuan. 3.2. Results of Experiment 2 In order to facilitate comparisons of the two experiments, we proceed with our analysis of Experiment 2 in the same way as for Experiment 1. We first present results for unconditional contributions (Section 3.2.1) and then results for conditional contributions (Section 3.2.2). Finally, we show the results on subjects’ choices for the implementation of the MCLs and also how they relate to unconditional contributions and the distribution of contribution types (Section 3.2.3). 3.2.1. Unconditional contributions Since there are again no significant order effects, we pool the data over the different orders. Unconditional contributions are summarized in Table 7. One can see a significant increase in contributions following the introduction of a minimum level of 7 tokens (6.27 vs. 10.17) (two-sided Wilcoxon signed-ranks test, p 2 in MCL = 0) ALL 66.7 9.13 Sequences 3 and 4 MCL = 0 (contributions 0–7) 0, 1, 2, 4, 5, 6, 7 60.4 2.59 MCL = 0 (contributions 8–20) 8, 9, ..., 20 39.6 11.89 MCL = 7 (contributions 7) 7 43.8 7.00 MCL = 7 (contributions 8–20) 8, 9, ..., 20 56.2 12.63 MCL = 7(those who contribute 0, 1, ..., 7 in MCL = 0) ALL 60.4 8.21 MCL = 7 (those who contribute > 7 in MCL = 0) ALL 39.6 13.16 Null hypothesis P-values Sequences 1 and 2 Contributions in MCL = 2 of those who contribute 0, 1, 2 in MCL = 0 are equal to 2 0.16 Contributions in MCL = 2 of those who contribute > 2 in MCL = 0 are equal in MCL = 0 and MCL = 2 0.37 Sequences 3 and 4 Contributions in MCL = 7 of those who contribute 0, 1, ..., 7 in MCL = 0 are equal to 7 7 in MCL = 0 are equal in MCL = 0 and MCL = 7 0.21 Note: We use two-sided Wilcoxon signed-ranks tests. 80 M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84

M.G.Kocher et aL/Joural of Ecomomic Psychology56(016)74-84 81 tributions followinga minimum level of 2 tokens.but it is only significant at the 10-level(6.58 vs.683)(two-sided Wil- ohelaodaokiIeonPnO0giNin48,QaamtntYety,overalunconditoalcontmibutioasinEpeniment2arevenysimilai the o n MO ted' observe signif ing at most tokens in MCL significant overshootingamong this of subiects.since the average cor ed con buto fo of 7 (two-s ed Wiloxon signed-rank MCL-7,p=0.21:N=19 32.cantionalantnbuions are very similar to the ones in Experiment1. ults from the contribution table.By and large.the slopes for MCL-0.MCL-2.and MCL-7 ferent MCLs.The proportions shown on the diagonal indicate those who were classified as the same contribution type in the different MCLs 3.2.3.Preferences for MCLs and contribution behavior MCL=0 MCL-7 Contmbuto.r 0 1234567891011121314151617181920

tributions following a minimum level of 2 tokens, but it is only significant at the 10%-level (6.58 vs. 6.83) (two-sided Wil￾coxon signed-ranks test, p = 0.09; N = 48). Qualitatively, overall unconditional contributions in Experiment 2 are very similar to the unconditional contributions in Experiment 1. We provide a more detailed analysis in Table 8. In the case of the low MCL of 2 tokens, 35.4% of our subjects contribute 2 tokens to the public good, compared with 33.3% contributing 0, 1, and 2 under MCL = 0. The average contribution in MCL = 2 for this group of subjects is 2.25, and it is not significantly different from the ‘‘projected” contribution of exactly 2 (two-sided Wilcoxon signed-ranks test, p = 0.16; N = 16). For subjects who contribute more than 2 tokens under MCL = 0, the average con￾tribution decreases from 9.59 in MCL = 0 to 9.13 in MCL = 2, but the change is not significant on conventional levels (two-sided Wilcoxon signed-ranks test, p = 0.37; N = 32). Thus, in the case of a low MCL in our endogenous treatments, neither do we observe significant ‘‘overshooting” among low contributors, nor significant crowding out among high contributors. In case of the high MCL of 7 tokens, 43.8% of subjects contribute 7 tokens to the public good, compared to 60.4% contribut￾ing at most 7 tokens in MCL = 0. We observe significant overshooting among this group of subjects, since the average con￾tribution of 8.21 in MCL = 7 is significantly different from the projected contribution of 7 (two-sided Wilcoxon signed-ranks test, p 2 (%) Contribution = 2 (%) Contribution > 2 60.4 6.3 Contribution = 0, 1 or 2 4.2 29.2 MCL = 7 Contribution > 7 (%) Contribution = 7 (%) Contribution > 7 37.5 2.1 Contribution = 0, 1, 2, 3, 4, 5, 6 or 7 18.8 41.7 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 PCC MCL=0 MCL=2 MCL=7 Fig. 2. Average conditional contribution in Experiment 2 (others’ average contributions on the horizontal axis and own average conditional contribution on the vertical axis). Note: PCC = perfect conditional cooperation. M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84 81

82 M.G.Kocher et al./Journal of Economic Psychology 56(2016)74-84 r MCL=0 in Sea ces 3 and 4 on av s are higher for subiects w the av erage of 941 fo Man-Whitney: the distr as free riders in MCL.with the opposite preference are classifed as free riders nt 2 (at individual level) MCL-2 MCL-0 00 typesMCd MCL7rindivdu lev MCL=7 Conditionalcooperaiors☒] Free riders ( Hump-shaped ( Others(图 MCL-0 Treatmeot Preference MCL-0 MCL-2 MCL-7 M:8思离 MC9（835 Note:We use Mann-Whitney U-tests.Sequences 1 and 2:MCL-0-2 and MCL-2-0:Sequences 3 and 4:MCL-0-7 and MCL-7-0. MCL=2 Conditional cooperators( Free riders(图 Hump-shaped ( 0thes(图 MCL-0 iinlcopcratons 00 MCL-2 Hump--shaped( Others(离) MCL-0 eidelcopeao

MCL = 7 over MCL = 0 in Sequences 3 and 4. On average, unconditional contributions are higher for subjects who prefer non￾zero MCLs. For example, in MCL = 2, the average contribution is 7.78 tokens for subjects with a preference for MCL = 2, and this is significantly larger than the average of 4.94 tokens for subjects who prefer MCL = 0 (two-sided Mann–Whitney U-test, p = 0.03; N = 48). Similarly, the average contribution for subjects with a preference for MCL = 7 is 3.05 tokens larger than the average contribution from subjects with a preference for MCL = 0 (two-sided Mann–Whitney U-test, p = 0.05; N = 48). Tables 13–16 report results on the distribution of contribution types. By and large, one can see that the proportion of free riders is smaller with those who have a preference for non-zero MCLs. For example, while 43.7% of subjects who prefer MCL = 0 are classified as free riders in MCL = 2, only 21.9% of subjects with the opposite preference are classified as free riders Table 10 Distribution of contributor types at MCL = 0 and MCL = 2 in Experiment 2 (at individual level). MCL = 2 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 62.5 4.2 0.0 0.0 Free riders 0.0 25.0 0.0 0.0 Hump-shaped 2.1 0.0 4.2 0.0 Others 2.1 0.0 0.0 0.0 Table 11 Distribution of contributor types at MCL = 0 and MCL = 7 in Experiment 2 (at individual level). MCL = 7 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 56.3 4.2 0.0 0.0 Free riders 2.1 25.0 0.0 0.0 Hump-shaped 0.0 0.0 8.3 2.1 Others 0.0 0.0 0.0 2.1 Table 12 Average unconditional contributions separated by preference for MCL. Treatment Preference MCL = 0 MCL = 2 MCL = 7 Sequences 1 and 2 MCL = 0 (33.3%) 4.19 (5.52) 4.94 (4.82) Sequences 1 and 2 MCL = 2 (66.6%) 7.78 (5.84) 7.78 (5.63) P-values 0.01 0.03 Sequences 3 and 4 MCL = 0 (12.5%) 3.50 (4.28) 7.50 (1.22) Sequences 3 and 4 MCL = 7 (87.5%) 6.67 (5.85) 10.55 (4.63) P-values 0.19 0.05 Note: We use Mann–Whitney U-tests. Sequences 1 and 2: MCL = 0–2 and MCL = 2–0; Sequences 3 and 4: MCL = 0–7 and MCL = 7–0. Table 13 Distribution of contributor types at MCL = 0 and MCL = 2 (at individual level) for subjects who prefer MCL = 0 (N = 16). MCL = 2 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 50.0 0.0 0.0 0.0 Free riders 0.0 43.7 0.0 0.0 Hump-shaped 0.0 0.0 0.0 0.0 Others 6.3 0.0 0.0 0.0 Table 14 Distribution of contributor types at MCL = 0 and MCL = 2 (at individual level) for subjects who prefer MCL = 2 (N = 32). MCL = 2 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 68.8 6.3 0.0 0.0 Free riders 0.0 15.6 0.0 0.0 Hump-shaped 3.1 0.0 6.3 0.0 Others 0.0 0.0 0.0 0.0 82 M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84

M.G.Kocher et aL/Joural of Ecomomic Psychology56(016)74-84 83 MCL=7 ndes☒ -shaped MCL=0 fMCand MCL7atndnvdual eve)orrrMN) Free riders( Hump--shaped( ohes(图 MCL-0 -shaped in the same situation(MCL-2).The pattern is similar but less pronounced for MCL-7.Taken together.the results on uncon ditional contributions and contribution types seem to indicate that subjects who prefer non-zero MCLs are more cooperative. 4.Discussion and conclusion than the minimum level have to increase their.If this is the only behavioral change.the proportion betore it is introduced is the same as the irewopotentialbethavioraleiertsotintoduCing9mnlmLmIeefheaioemeatonedPpopotonisnotSbieand ()the distributions of contributions among subjects contributing more than the imposed minimum level differ.when tions under e lo MCL anc he high MCL are not n rily the same.Our nrs st experme act that t the positive shift for low con of th effect take over.Looking at the full schedules of the conditional.wesee that respectively. ately very weak.We have no voting.no group discussion,and no o other form of group interaction that would give the MCL t to s ing out voluntary c is that the endogenous MCL sends a positive signal to the gro p members rather than one of mistrust.However.on ehas to b careful in directly comparing Experiment 1and Experiment 2.Experiment 2 was added later on.followi iments the subject pool might have changed in dimensions that are difficult to observe(e.g..cohort effects)

in the same situation (MCL = 2). The pattern is similar but less pronounced for MCL = 7. Taken together, the results on uncon￾ditional contributions and contribution types seem to indicate that subjects who prefer non-zero MCLs are more cooperative. 4. Discussion and conclusion We study different MCLs experimentally in a linear public goods game, where we elicit both unconditional and condi￾tional contributions. An obvious direct effect of introducing a minimum level is that all subjects intending to contribute less than the minimum level have to increase their contributions. If this is the only behavioral change, the proportion contribut￾ing at most the introduced minimum contribution level before it is introduced is the same as the proportion contributing exactly the minimum level after it is introduced, resulting in an overall positive effect on contributions. However, there are two potential behavioral effects of introducing a minimum level: (i) the aforementioned proportion is not stable and (ii) the distributions of contributions among subjects contributing more than the imposed minimum level differ, when the minimum level requirement is imposed. Our experimental results indicate that decision makers are affected by MCLs, but, interestingly, the effects on contribu￾tions under the low MCL and the high MCL are not necessarily the same. Our first experiment investigates the effect of exogenous MCLs. In the case of a low MCL, average contributions remain largely the same as without an MCL, despite the forced increase in contributions for very low contributors. This is explained by the fact that the positive shift for low con￾tributors – indeed, on average, ‘‘overshooting” the MCL – is offset by a crowding out effect of the MCL for relatively high con￾tributors. The latter effect is also present in our high MCL environment, but it is much less pronounced, making the former effect take over. Looking at the full contribution schedules of the conditional contribution elicitation, we see that conditional cooperators decrease their contributions under MCL = 2 (a smaller slope) compared with MCL = 0, whereas the contribution function for MCL = 7 looks almost identical to the one for MCL = 0, taking into account the lower-limit censoring at 2 and 7, respectively. In a second experiment, we investigate the effects of endogenous MCLs. The second experiment can also be seen as a robustness check for the results in Experiment 1, since the implementation of the common decision for the MCL is deliber￾ately very weak. We have no voting, no group discussion, and no other form of group interaction that would give the MCL a stronger normative connotation. Nonetheless, in comparison with exogenous MCLs, the endogenous implementation of an MCL could send a signal of distrust to group members, similar to the hidden cost of control in Falk and Kosfeld (2006), crowd￾ing out voluntary cooperation. Comparing the effect of MCLs across the two experiments, we find that there is less ‘‘over￾shooting” but also less crowding out when MCLs are endogenously imposed by one randomly selected individual in each group. Overall, the endogenously implemented MCLs work marginally better than the exogenously implemented ones, but given the size of the difference, one has to be careful not to over-interpret the result. In any case, the endogenous imple￾mentation seems to induce a more stable behavior (less ‘‘overshooting” and crowding-out). If stability is an objective, an endogenous implementation should be preferred. Taking a look at the contribution schedules, MCL = 2 leads to a steeper slope in Experiment 2 than in Experiment 1; for MCL = 7, we observe the reverse. However, the overall picture (positive slope) looks similar. When one wants to interpret the differences between the two experiments cautiously, one could argue that the endogenous MCL sends a positive signal to the group members rather than one of mistrust. However, one has to be careful in directly comparing results from Experiment 1 and Experiment 2. Experiment 2 was added later on, following the request of a referee. Given that some time has passed between the two experiments, it is unclear whether we can assume full randomization of participants across the two experiments. Within the experiments, this is not an issue, but across the exper￾iments the subject pool might have changed in dimensions that are difficult to observe (e.g., cohort effects). Table 15 Distribution of contributor types at MCL = 0 and MCL = 7 (at individual level) for subjects who prefer MCL = 0 (N = 6). MCL = 7 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 33.3 0.0 0.0 0.0 Free riders 16.7 33.3 0.0 0.0 Hump-shaped 0.0 0.0 16.7 0.0 Others 0.0 0.0 0.0 0.0 Table 16 Distribution of contributor types at MCL = 0 and MCL = 7 (at individual level) for subjects who prefer MCL = 7 (N = 42). MCL = 7 Conditional cooperators (%) Free riders (%) Hump-shaped (%) Others (%) MCL = 0 Conditional cooperators 59.5 4.8 0.0 0.0 Free riders 0.0 23.8 0.0 0.0 Hump-shaped 0.0 0.0 7.1 2.4 Others 0.0 0.0 0.0 2.4 M.G. Kocher et al. / Journal of Economic Psychology 56 (2016) 74–84 83