《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?.pdf

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 introduction 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

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 voluntary 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 decisions for the public good: (i) unconditional contributions and (ii) conditional contributions, i.e., what the subject would contribute 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 conditional 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 voluntary 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 distribution 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 contribution schedules of the conditional contribution elicitation, we see that conditional cooperators decrease their contributions 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 implementation 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 member 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

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 Beijing. 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 unconditional 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 contributions 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). Comparing across treatments, contributions with MCL = 7 are significantly higher at the 1% level than with MCL = 2 (twosided 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 (twosided 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 introduction of the minimum level? Table 3 provides detailed analyses of the effects of introducing an MCL. If behavior is unaffected 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

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 < 0.01; N = 22). Interestingly, low contributors seem to ‘‘overshoot” the MCL on average. When looking at those who contribute more than 2 tokens under MCL = 0, we observe averages of 10.76 when MCL = 0 and 9.10 when MCL = 2. Hence, on average, a highly significant crowding out effect occurs (two-sided Wilcoxon signed-ranks test, p < 0.01; N = 50). Taking the two groups together, the overall small change in average contributions from MCL = 0 (7.58) to MCL = 2 (7.69) can be explained by the offsetting effects of ‘‘overshooting” the MCL and the occurrence of crowding out for contributors who contribute more than 2 tokens even when MCL = 0. We conduct an analogous analysis for MCL = 7, compared with MCL = 0. The proportion of decision makers contributing at most 7 when MCL = 0 is 41.7%, while it is only 16.7% when MCL = 7. We can thus confirm the behavioral regularity of ‘‘overTable 3 Detailed descriptive statistics of unconditional contributions in the treatments. Treatment Contributions considered Proportion of subjects (%) Average contribution Sequences 1 and 2 MCL = 0 (contributions 0–2) 0, 1, 2 30.6 0.36 MCL = 0 (contributions 3–20) 3, 4, ..., 20 69.4 10.76 MCL = 2 (contributions 2) 2 29.2 2.00 MCL = 2 (contributions 3–20) 3, 4..., 20 70.8 10.04 MCL = 2 (those who contribute 0, 1, 2 in MCL = 0) ALL 30.6 4.50 MCL = 2 (those who contribute > 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 <0.01 Contributions in MCL = 2 of those who contribute > 2 in MCL = 0 are equal in MCL = 0 and MCL = 2 <0.01 Sequences 3 and 4 Contributions in MCL = 7 of those who contribute 0, 1, ..., 7 in MCL = 0 are equal to 7 <0.01 Contributions in MCL = 7 of those who contribute > 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

shooting” that we observed under MCL = 2. If all of our decision makers, contributing 0–7 tokens under MCL = 0, would contribute 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 (twosided Wilcoxon signed-ranks test comparing actual contribution levels and the projected level of 7, p < 0.01; N = 30). When looking at those who contribute more than 7 tokens without an MCL, we observe averages of 12.40 when MCL = 0 and 12.00 when MCL = 7. Our data therefore imply that, as for MCL = 2, there is crowding out for MCL = 7, yet it is relatively weaker and far from being significant (two-sided Wilcoxon signed-ranks test, p = 0.60; N = 42). The significant overall increase in contribution levels from MCL = 0 (8.44 tokens) to MCL = 7 (10.92 tokens) is a combination of the mechanical increase of contributions to the MCL, of ‘‘overshooting” of decisions makers who are forced to increase their contribution levels under MCL = 7 compared to MCL = 0, and of almost no crowding out effects of MCL = 7 on those theoretically not bound by the MCL. Table 4 shows the transition matrix in the two treatments in order to give information on the distribution of changes when the MCL is introduced. Not surprisingly, the vast majority of decision makers either adjust exactly to the MCL if they contribute less without an MCL (i.e., they contribute exactly 2 and 7 if they contribute less than 2 and 7 without the relevant MCL) or stick with higher contributions when they already contributed more than the MCL before its introduction. If we take such behavior as a measure of stability across different institutional setups, 76.4% are classified as stable when an MCL = 2 is introduced, and 66.7% behave in a stable way when an MCL = 7 is implemented. A much more rigorous definition of stability is one that does not allow for changes at all across institutional environments, except for forced ones (increasing one’s contribution to the MCL if one contributed below this level without an MCL). For instance, if one contributes 1 token under MCL = 0, then the only ‘‘stable” choice of the same person under MCL = 2 is 2 tokens; if one contributes 5 tokens under MCL = 0, then the only ‘‘stable” choice of the same person under MCL = 2 is also 5 tokens. Taking such a strict definition, 40.3% of the decision makers submit stable decisions across MCL = 0 and MCL = 2, and 39. 0% of the decisions makers do the same across MCL = 0 and MCL = 7. 2.2.2. Conditional contributions Does the introduction of MCLs change the distribution of contributor types, i.e., does the distribution of free riders and conditional cooperators change? Fig. 1 presents the average results from the contribution table. As expected, on average, the contributions increase as the average contributions of others increase, regardless of the MCL. Interestingly, however, while the slope in MCL = 2 is smaller than in MCL = 0 and, hence, the MCL does not increase average conditional contributions throughout the whole interval of others’ average contributions, the slope in MCL = 7 is approximately similar to the one in MCL = 0. As a consequence, conditional contributions are higher in MCL = 7 than in the other two treatments, also in the range clearly above the lower censoring limit of 7. In order to study this effect further, we ran regressions to explain contriTable 4 Transition probabilities in the treatments. MCL = 0 MCL = 2 Contribution > 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