Empowering Neighbors versus Imposing Regulations: An Experimental Analysis of. Pollution Reduction Schemes*

Transcription

1 Empowering Neighbors versus Imposing Regulations: An Experimental Analysis of Pollution Reduction Schemes* Timothy N. Cason Department of Economics Purdue University and Lata Gangadharan Department of Economics Monash University June 2012 Abstract This paper presents an experimental study of two mechanisms that influence incentives to reduce ambient pollution levels. In the formal mechanism individuals face a penalty if the group generates total pollution that exceeds a specified target, whereas in the informal mechanism individuals can choose to incur costs to punish each other after observing their group members emissions. We examine the effectiveness of these mechanisms, in isolation and in combination. The results suggest that the formal targeting mechanism is significantly more effective than informal peer punishment in reducing pollution and increasing efficiency. Peer punishment however improves the performance of the formal mechanism. Keywords: Ambient Target, Non-point source pollution, Peer-punishment, Transfer Coefficients JEL Classification: C90, Q53, Q58 * We are grateful for valuable comments provided by Marco Casari, Lana Friesen, Marianne Lefebvre, Christian Vossler, and audiences at Indiana University, Victoria University (Wellington), the Universities of Massachusetts, Sydney, Western Australia, Oxford, Heidelberg, Nottingham, Birmingham, the Economic Science Association International Conference and the Australia-New Zealand Experimental Economics Workshop, and two anonymous referees. We retain responsibility for any errors. This research has been supported by a grant from the U.S. Environmental Protection Agency's National Center for Environmental Research (NCER) Science to Achieve Results (STAR) program. Although the research described in the article has been funded in part through EPA grant number R833672, it has not been subjected to any EPA review and therefore does not necessarily reflect the views of the Agency, and no official endorsement should be inferred. Part of this research was conducted while Cason was a visiting fellow with the Department of Economics, University of Melbourne. Justin Krieg and Kory Garner provided very capable research assistance. 1

2 1. Introduction Formal regulations have been widely used in recent years to control pollution, such as equipment and performance standards, emission taxes and tradable permit systems. Their costs increase as more pollutants are regulated or regulatory goals are strengthened, however, and formal mechanisms are often difficult to implement and enforce in regions with weak institutional structure, such as in many developing countries (Russell and Vaughan, 2003). In these cases informal mechanisms such as social sanctions and peer-punishment are often considered practical options to complement formal approaches and improve environmental performance (Ostrom, 1990). In this paper we use laboratory experiments to examine the relative effectiveness of formal and informal mechanisms, and their interaction, for reducing non-point source pollution. We find that the formal, group-tax targeting mechanism leads to a substantially larger and more reliable pollution reduction than informal peer sanctions alone. Informal sanctions nevertheless improve the performance of the formal mechanism. Controlling non-point source pollution is notoriously difficult because different individual sources contributions are often not observed directly by the regulator or can be identified only at prohibitively high costs. 1 Segerson (1988) suggested a formal, target-based regulatory mechanism that could address this information asymmetry and improve efficiency, building on the group moral hazard incentive mechanism introduced by Holmstrom (1982). The basic idea behind this incentive scheme is that all potential polluters can be penalized (or subsidized) if the ambient pollution level at a particular monitoring point exceeds (or is lower than) a certain target. This target is defined by the regulatory authority, for example based on a 1 Pollution that does not originate from a single source, or point, is called non-point source pollution. Examples of such pollution include soil erosion, pesticides run-off and nitrate leaching from farmland, traffic pollution in cities, or pollution due to livestock waste, deforestation and logging. In contrast, pollution originating from a single, identifiable source, such as a discharge pipe from a factory or sewage plant, is called point source pollution. 1

3 threshold beyond which ambient concentrations are perceived to increase the environmental risk to an unacceptable level. The liability of each source depends on the aggregate emissions from the entire group of polluters, not just its own level of emissions, since these emissions are unobservable to the regulator. Empirical research using economic experiments provides evidence that such targeting instruments can improve efficiency (Spraggon and Oxoby 2009, 2010; Suter, Vossler and Poe, 2009; Oxoby and Spraggon 2008; Vossler et al. 2006; Cochard, Willinger and Xepapadeas 2005; Alpizar, Requate and Schram 2004; Poe et al. 2004; Spraggon 2004, 2002). While most of these studies find that this moral hazard problem can be mitigated at the aggregate level under a wide range of conditions, including the number and heterogeneity of polluters, subject pool and communication opportunities, there is concern that this mechanism does not ensure compliance at the individual level. This creates inefficiencies due to differences in marginal abatement costs and can lead to adverse distributional impacts (Suter, Vossler and Poe, 2009; Giordana and Willinger, 2010). Both economic theory and experiments have therefore shown that ambient based instruments reduce group moral hazard and can be effective, so it is puzzling that these instruments are not frequently observed in the field. 2 Arguments against field implementation have centered on political feasibility and the fairness of penalizing non-polluters based on aggregate emissions. The reluctance to use these formal targeting schemes implies that it is important to assess the effectiveness of alternative informal mechanisms, and whether they could be combined with formal mechanisms to improve their performance and acceptability. The 2 Ribaudo and Caswell (1999) discuss several ambient based policies used by government authorities. For example, the Everglades Forever Act in the United States includes a provision to increase land taxes by the state of Florida if an aggregate objective of phosphorous reduction is not met in the watershed. Another policy in Florida requires dairy farmers to be taxed if water quality goals are not met in the Lake Okeechobee region (Karp, 2005). 2

4 primary goal of our study is to compare and interact this formal targeting mechanism with informal peer monitoring and punishment. Alternative non-regulatory, informal mechanisms that aim to reduce pollution and resource exploitation have been suggested in different contexts for several decades. For example, Ostrom s early work (1990) on overutilization of common pool resources emphasizes the role of informal institutions in governing the commons. Economic experiments suggest that informal mechanisms such as peer punishment that allow individuals to sanction socially bad behavior can help instill beneficial norms. Ostrom et al. (1992) and Fehr and Gaechter (2000) show that allowing individuals to sanction each other can decrease overexploitation of common pool resources and increase provision of public goods. These studies have inspired an extensive and influential literature exploring the robustness of peer punishment in promoting efficient cooperation (see Chaudhuri, 2011, for a recent review). Our experiment explores whether punishment can reduce non-point source pollution, and how it interacts with a more formal regulatory targeting mechanism. Pollution could be nonpoint from the regulator s perspective, who does not have the resources or does not find it costeffective to identify and monitor (small) individual sources; nearby individuals, however, could often observe emissions of their neighbors and impose sanctions even when regulators cannot (Romstad, 2003). Thus, informal peer monitoring can have an informational advantage over formal regulation. For example, a voluntary system of fines and sanctions is used in the Ankleshwar Industrial Estate, in the state of Gujarat in India. The estate has more than 400 chemical plants. The regulator may only be able to observe aggregate emissions from the borders of the estate, but firms located in the same industrial district have easier access in monitoring individual plant emissions than geographically remote environmental agencies. Sterner (2003, 3

5 chapter 25) analyses the monitoring efforts set up voluntarily by firms located in this estate and examines the effectiveness of several mechanisms used by estate members to coerce fellow members to clean up emissions - including provision of information, effluent fees and fines. Kathuria and Sterner (2006) provide a discussion of how the monitoring and enforcement activities of the estate are analogous to the management of a common pool resource. Similar schemes initiated by industry groups have been documented within the Responsible Care Program for Chemical Industries ( Members use peer pressure and other informal sanctions to improve performance indicators and verification procedures in the areas of environmental protection, health and safety (Millock and Salanie, 2005). Further to these examples that relate specifically to non-point source pollution, there are many examples of social sanctions and peer punishment used in the field to reduce overexploitation of common pool resources. Ostrom (1990) describes several common pool resource problems such as mountain grazing in Switzerland and extraction of timber from forests in Japan, which are resolved using peer-monitoring and costly punishment. Similarly Casari and Plott (2003) examine an ancient system of peer monitoring and peer sanctioning practiced in the Italian Alps for managing common pool resources. There are also examples where the threat of exclusion from common projects with high benefits (such as agricultural cooperatives, joint research to reduce pollution or increase output, and advertising) may serve as a punishment mechanism and deter deviations from cooperation (Seabright, 1993). Gaspart and Seki (2003) discuss the cooperative actions of Japanese fishermen who also engage in collective projects such as searching for good fishing spots, development of better fishing techniques, engaging in collective research and sharing expertise. As noted above, the nonpoint source pollution 4

6 problem has many features similar to the common pool resource problem (Kathruia and Sterner, 2006). Success of informal mechanisms could imply a different (and perhaps a less interventionist) role for government and this may be a promising approach to environmental protection, especially for governments struggling to achieve environmental objectives with limited budgets and staff. Moreover, informal mechanisms may mitigate some of the distributional problems associated with ambient pollution instruments. When individuals choose higher emission levels they impose a cost on the other members of the group, especially when this triggers a group penalty. This cost is offset for the offender by higher private payoffs but not for the other members of the group. In the absence of peer-punishment mechanisms, the best the other members of the group can do is chose a lower emission level and absorb the cost that is being inflicted on them by those who choose higher emissions. Providing opportunities for peerpunishment allows individuals to retaliate against this offender behavior. Hence it is reasonable to expect that punishment may mitigate the inequity inherent with these instruments. Both the formal and the informal mechanisms we consider have monetary implications. The formal mechanism charges all individuals a tax if they do not collectively meet the ambient target. The informal mechanism allows individuals to reduce the monetary payoffs of others by incurring a punishment cost. The difference between the two is that one has a centralized, regulatory origin and the other is decentralized, community- and peer-based. While previous studies have examined these kinds of mechanisms in isolation, very few have examined the comparative impact of formal and informal schemes. The interaction of these schemes is important to study because they may counteract each other. For example, enforcement of externally-imposed regulations can crowd out endogenously cooperative behavior if it 5

7 discourages the formation of social norms (Ostrom, 2000; Bowles and Polanía-Reyes, 2012). In a recent paper, Lopez et al. (2012) compare the effectiveness of informal schemes (probabilistic public disclosure) and regulatory enforcement (fines and probabilistic enforcement) in promoting efficient provision of a public good. They find that the threat of public disclosure of the individual contributions induces more cooperation than regulatory pressure. Hence the formal mechanism that can implement the social optimum as a Nash equilibrium (their High Penalty treatment) performs worse than an informal mechanism (their public disclosure Shame treatment). Although informal peer monitoring and punishment have been shown to effectively improve performance in related public goods provision experiments, our results show that formal group targeting regulation performs significantly better than peer punishment alone. Targeting using a group tax significantly lowers pollution and raises efficiency, either in the presence or absence of peer punishment opportunities. Punishment does reduce pollution when it occurs in conjunction with the formal targeting mechanism, and environmental damages are lowest when the targeting mechanism is supplemented with punishment. Punishment is less effective in the current study compared to the literature, however, which is based largely on the linear public goods/voluntary contribution mechanism. Subjects fail to punish as much or as frequently as in earlier studies, so they exert a weaker influence on their peers subsequent pollution choices. 2. Experimental Setting and Methodology 2.1 Economic Setting Our experiment is based on a non-point source pollution problem. We focus on firms that are heterogeneous. Heterogeneity across firms is introduced in a novel manner, by positioning firms 6

8 at different geographical proximity to a common monitoring or receptor point. Environmental quality and possible damage is measured at this receptor point and emissions from the firms located nearer to the receptor point are more damaging than emissions from other firms that are further away. More specifically, emissions from the nearer firms cause a larger rise in recorded concentrations than emissions from the firms further away. This is because the emissions from the more distant firms are more diluted by the time they arrive at the receptor point, while emissions from the nearer firms arrive as more concentrated. This kind of geographic heterogeneity is common in the field but it has not been studied systematically using data from controlled experiments. 3 Emissions can be controlled by the firms, but the concentrations at the receptor point are the target for policy. A transfer coefficient T k represents the constant amount the concentration at the receptor point will rise when a firm at location k emits one more unit of pollution. In each session subjects interact in groups and make decisions corresponding to the role of polluting firms. 4 Each firm i chooses an emissions level E i every period. This choice has an impact on their private payoffs and on the group payoffs. Private payoffs are given by a concave benefit function B(E i ) that increases at a decreasing rate up to a maximum that corresponds to the level of uncontrolled emissions. The experiment implemented the following quadratic functional form, similar to that used in Spraggon (2002) and much of the subsequent literature: 5 3 Spraggon (2004) shows that when firms have asymmetric payoff functions, so that some are larger capacity polluters, then significant inefficiencies and inequities exist at the individual level and suggests that the tax/subsidy targeting mechanism may not be optimal in such cases. Suter et al. (2009) compare the behavior of homogenous and heterogeneous firms and find more strategic actions by firms in the heterogeneous capacity treatments. 4 No actual pollution is released in the experiment, and as is common in experimental economics subjects make economically-motivated choices that were stripped of their environmental context to improve control (Cason and Raymond, 2011). Although subjects chose decision numbers in the neutral framing of the experiment instructions, we use pollution framing in the paper for clarity of exposition. 5 Spraggon (2002) also shows that experimental outcomes are not sensitive to the presence of uncertainty (i.e., the ambient level being observed with error), hence in this formulation we also abstract away from the problem of uncertainty. 7

9 E 2 B E i Thus, a source s benefits are increasing in emissions up to a maximum emission level of 125. The firms total payoff function (private plus group payoff) reflects the negative externality that their pollution choice imposes on the group. As noted above, environmental damage is represented by measured concentrations at the receptor point, which is a linear transformation of individual emissions through the (heterogeneous) transfer coefficients. Therefore, the group payoff depends on all of the emissions chosen by everyone and is the same for everyone in the group. The environmental impact of emissions from the firms nearer the receptor point is higher because the transfer coefficient is higher, hence the externality associated with the total emissions from this group is greater. The experiment employed two distinct firm locations, so k {n, f}, where we denote the M n near firms with index n and M f far firms with index f. The environmental damage ED suffered by each of the M = M n + M f firms is i M f M n 1 ED Tf Eg TnEh M g 1 h 1, and the aggregate total damage is M ED, which for simplicity just corresponds to the ambient pollution measured at the receptor. The damages are endogenous as in Cochard et al. (2005), hence the polluters themselves suffer from the externality they cause (for example, the pesticides and fertilizers used by farmers can pollute the water they drink). We require this framework as our aim is to compare the effectiveness of formal and informal mechanisms. Informal mechanisms such as peer punishment have been observed to be very effective in situations where the actions of individuals generate externalities affecting others in the group (as in a public good game). Without this externality agents have little incentive to punish their peers because their payoff is not affected by their peers decisions. 8

10 The experiment uses transfer coefficients of T f =0.24 for the far firms and T n =0.36 for the near firms, with 3 firms of each type (i.e., M f =M n =3 and M=6). The benefits from pollution-producing production B(E i ) less the environmental damages ED are the payoffs to each individual firm. The first-order condition for the emissions choice of a payoff-maximizing firm with transfer coefficient T k in this linear-quadratic parameterization is 0.004(125-E i )-(T k /6)=0. Therefore the unique Nash equilibrium is in dominant strategies, with emissions of 115 for the far firms and 110 for the near firms. In equilibrium the ambient pollution measured at the receptor point is 201.6, and total payoffs across the six firms are The socially optimal level of emissions can be found by solving the social planner s problem, which internalizes the negative externality generated by the pollution choice. In particular, the planner maximizes the difference between the aggregate benefits and the aggregate damages for all six firms. The optimal level of pollution is different for firms near and far from the receptor point. In our setting with three firms at each location, the planner s problem is to choose E f and E n to maximize Net Social Benefit = 3B(E f )+3B(E n )-[3T f E f +3T n E n ] For the parameters used in the experiment, the socially optimal emission level is 65 each for far firms and 35 each for near firms. These emission levels lead to ambient pollution of 84.6 and total earnings across the six firms of This is a substantial (81 percent) improvement over Nash equilibrium total earnings. As the private payoff function of firms is nonlinear, the Nash equilibrium and the social optimum are on the interior of the choice space. Segerson (1988) showed that this social optimum can be implemented through a group incentive instrument. The original mechanism considered both taxes and subsidies imposed on or granted to all members of the group for emissions that deviated from the target. The tax paid by 9

11 each individual firm is chosen equal to the marginal environmental damages of her emissions, so the mechanism leads the firm to internalize the damage externality in the same way as the social planner. For our linear-quadratic parameterization this mechanism with both taxes and subsidies implements the target as an equilibrium in dominant strategies. When the mechanism includes a subsidy for pollution below the target, however, this creates strong incentives for agents to collude, and collusion has been observed in previous experiments with subsidies (e.g., Poe et al., 2004). We therefore employ a version of the mechanism that only imposes a tax for pollution that exceeds the target level. For the ambient target D*, and total ambient pollution measured for the M firms M ED, the tax paid is M 1 ( M ED ) D * if ( M ED ) D * ( ED) M 0 if ( M ED) D* As we consider the version of this mechanism with only taxes and not subsidies, in the sequel we shall refer to it as the group tax mechanism. The individual firm s payoff function when facing this tax is B( Ei ) ED ( ED) if ( M ED) D* ( Ei ) B( Ei ) ED if ( M ED) D* For the quadratic benefit function used in the experiment, each firms marginal gain from increasing emissions is constant at 0.004(125-E i ), but for ambient pollution levels that exceed the target D* the tax increases the marginal cost from the private level of T k /M to the social cost of T k. It is straightforward to show that the unique Nash equilibrium implements the social optimum described above. Unlike the unregulated case or the mechanism that includes subsidies, however, this equilibrium is not in dominant strategies since firms have an incentive to increase emissions 10

12 if others lower their emissions when the ambient pollution falls below the target. Note also that while all firms pay the same tax whenever the pollution exceeds the target, regardless of their location, more distant firms emit more pollution in equilibrium because of their lower transfer coefficient and marginal environmental impact. 2.2 Treatments We collect data from four treatments using a 2 2 design. Our first treatment is one where there is no policy instrument (Baseline) and firms do not face an environmental target. They choose emissions given their private and group payoffs as described above. As noted above, for the parameters in the experiment the total damages in this treatment are units in equilibrium as firms both near and further away from the receptor point choose decision numbers that are much higher than what is socially optimal. Given the total earnings in the unregulated set up of and the social maximum of 109.2, the predicted efficiency of the unregulated setting is 60.45/109.2 = 55.4 percent. Data from this treatment reveal agents decisions in the absence of either formal or informal regulations. In the second treatment (Group Tax Only), all firms pay a tax penalty if the ambient pollution at the receptor point is above the target imposed by the regulator. The tax described above is a linear function of the difference between the ambient pollution measured and the target pollution. It is deducted from each firm s payoff if pollution is greater than the target, and the target is set equal to the social optimum of (The experiment instructions did not indicate that this target was the social optimum.) This treatment is designed to examine the impact of Segerson s (1988) formal regulatory targeting mechanism on pollution levels and efficiency, here for the setting with geographic heterogeneity and damage externality (the latter as explored in Cochard et al, 2005). In theory the targeting mechanism implements the socially efficient 11

13 outcome, and leads to total earnings (due to the environmental damages from the pollution of 84.6) of The group tax can substantially increase efficiency if it creates incentives as designed, potentially up to 100 percent. This requires firms to not only meet the aggregate damages target of 84.6 exactly, but also for all three far firms to emit 65 units of pollution each and for all three near firms to emit 35 units each. Any within-location variation necessarily reduces efficiency. In the third treatment (Punishment Only), agents make two decisions in each period. In the first stage, subjects choose their emissions, which can be observed by the other two subjects at their same location, at the end of that stage. (Subjects could also observe the total emissions generated at the other location but they could not identify the emissions of individual subjects in that location.) As mentioned in the introduction, peer monitoring can be feasible in many situations that are considered non-point pollution problems from the regulator s perspective because of limited resources or incentives to measure individual sources. To maintain comparability across treatments, the two nearby subjects emissions are always revealed to neighbors in all treatments, and in all treatments subjects were assigned to fixed locations throughout the session. Although subjects identity is kept anonymous, they are given labels that remain the same throughout the session. In the second stage of the Punishment treatment subjects are allowed to decrease the payoff of the other two members at their location by assigning deduction points. A punished group member can be assigned between 0 and 5 deduction points by each of the two nearby peers. Each assigned deduction point reduces the punished member s payoffs by 1.5 Experimental dollars and costs the punishing member 0.5 Experimental dollars. Our choice of punishment parameters is taken from Gaechter et al. (2008), who find that punishment improves 12

14 cooperation both in the short and long run for the linear voluntary contributions mechanism. Our subjects, like theirs, can assign a maximum of five deduction points and we also use the same fine to cost of punishment ratio (3 is to 1) that they use and which is widely used in the public goods punishment literature. 6 The largest payoff for a deviation from the social optimum (about 11 Experimental dollars for the near firms) is less than the maximum punishment that subjects could receive (15 Experimental dollars). To further enable comparisons with the literature on punishment, we employed a lightly edited version of the punishment instructions used in Gaechter et al. (2008) for this treatment and used similar terminology. We restrict subjects to assign deduction points to others at their same location for two reasons. First, we wanted to avoid exacerbating confounds from social preference considerations, which could arise if punishment was allowed across locations. Subjects closer to the receptor should emit less pollution and therefore earn lower payments in equilibrium. Subjects may therefore wish to punish firms in the other location due to envy (as those at the other location earn more money); due to inequality aversion (so as to equalize payoffs in both locations) or to instill social norms of emitting less pollution so as to reach the social optimum. The motives relating to envy or inequality aversion would be difficult to separate out from the latter reason to punish. As our main interest is in examining the effectiveness of peer punishment in reducing pollution, we chose to not allow across-location punishment. Second, given the nature of nonpoint source pollution, in which emissions from individual sources are difficult to observe by outsiders, actions that lead to a particular level of emissions may be easier to observe by the most closely neighboring sources (Romstad, 2003). Hence social sanctions are more feasible and effective in small groups that are located close to each other. For example, Blackman (2000) reports that the success of regulatory mechanisms in certain sectors in Mexico depends crucially 6 See, for example, Casari and Luini (2009, 2012) for an investigation of alternative punishment technologies. 13

15 on peer monitoring, which is most successful when facilitated by local groups or organizations that can effectively monitor each other. 7 This treatment is intended to examine if informal peer monitoring with opportunities for costly punishment lead to lower pollution levels and higher efficiency as compared to the Baseline, perhaps comparable to the Group Tax Only treatment. For the non-repeated version of this game, a profit maximizing agent will not choose to incur costly punishment, and knowing this the other agents will not alter their behavior due to the availability of punishment. However, while theoretical predictions in this treatment are the same as the baseline treatment, the literature on punishment suggests that behavior could be very different. For example, Fehr and Gaechter (2000, 2002) and Bowles and Gintis (2002), argue that altruistic punishment by homo reciprocans, i.e., humans who are willing to punish norm violators even when such punishment is costly to the punishers, may be the primary driving force behind sustaining cooperative norms in a variety of social settings. The threat of punishment helps discipline free riders and therefore leads to higher cooperation amongst group members. Hence we expect their decisions to be nearer the socially optimal level in this treatment. Punishment is also more frequent and effective in repeated (partner) matching across multiple decision periods, which we implement here to represent repeated interactions among local polluters. 8 Our last treatment (Punishment with Group Tax) combines the formal and informal mechanisms for pollution reduction. As in the Group Tax Only treatment, all subjects pay a penalty if ambient pollution exceeds the target. In addition, they also participate in the second 7 In Cason and Gangadharan (2012) we show that results for punishment effectiveness are similar when subjects can observe and punish all five other subjects rather than only their two neighbors. Carpenter (2007) also finds that performance is similar when subjects can punish half or all other group members, in a linear public goods game with 5 or 10 subjects per group. 8 Communication between subjects has also been shown to encourage coordination and cooperation in social dilemma games (Ledyard, 1995), even in challenging settings where subjects only observe others behavior with a substantial lag (Cason and Khan, 1999). To enhance comparability with the literature on punishment, we focus on peer punishment and not on communication as the social sanctioning mechanism. 14

16 stage where they can decide to assign deduction points to their neighbor group members exactly as in the Punishment Only treatment. Theoretical predictions are identical to the Group Tax Only treatment, and punishment is not needed to implement the social optimum as an equilibrium. This treatment is intended to determine whether combining the formal regulatory and the informal peer monitoring and punishment mechanisms helps agents achieve environmental targets. The design summarized in Table 1 indicates a larger number of independent groups in the group tax condition (14 per treatment) than in the no tax condition (8 per treatment). 9 Six subjects participated in each of the 44 groups, for a total of 264 human subjects. No subject participated in more than one session. 2.3 Procedures All sessions were conducted at the Vernon Smith Experimental Economics Laboratory at Purdue University, using Z-tree (Fischbacher, 2007). Subjects were drawn from the undergraduate student population, broadly recruited across the university by using ORSEE (Greiner, 2004). Although some had participated in other economics experiments, all were inexperienced in the sense that they had never participated in a similar experiment with common pool resource or nonpoint pollution characteristics and incentives. While subjects interacted anonymously in 6 person fixed groups, multiple groups under the same treatment conditions were conducted simultaneously in the laboratory, employing 12 to 24 subjects. Subjects were randomly assigned to different groups and locations upon arrival, and 9 This unbalanced design is due to two reasons. First, based on our initial data collection it was clear that the variance in outcomes across groups within the group tax condition was considerably greater than the variance within the no tax condition. This led us to an optimal sample arrangement with more observations in the high-variance condition (List et al., 2011). This greater variance in outcomes is also observed in the full dataset. For example, the standard deviation of damages across groups during late periods is 32 in the Group Tax condition but is only 13 without the tax. Second, we wanted to focus our data collection efforts on the two novel group tax treatments so as to provide some insights on as yet unexplored issues in the literature i.e., the impact of imposing a group tax on firms that are locationally dispersed and who may also face informal sanctions. 15

17 they remained in those positions throughout the experiment. Subjects made decisions in one or two stages as described above, depending on the treatment. They participated in 30 periods and this number of periods was common knowledge and announced in the instructions. Five periods were randomly chosen at the end of each session for payment. 10 To make the benefits of greater emissions transparent, subjects are given their private payoffs B(E i ) for a wide range of emission levels in a table. Subjects are also given a table which presents the group payoffs for different combinations of emissions from the two locations. In addition to these tables, we also designed a decision tool in the experiment software to help subjects understand more clearly how their choice of emissions would affect their total payoffs using a graphical user interface. Every period subjects were required to input at least one estimate regarding the emissions made by other subjects at the two locations. The decision tool graphs their potential total payoff based on that estimate for every possible emissions level and records and reports this on the screen while subjects explore the impact of their choices. They could submit multiple estimates regarding others emissions in each period to re-display new payoff functions, allowing them to perform searches over the strategy space before making a binding choice. 11 After all subjects submit their emission choice for the period, they are informed about the choices of the other two subjects at their same location, the total emissions for both groups, the ambient pollution measured at the receptor, and their total payoff for the period. In 10 Although in some periods subjects earned negative profits, in most periods most subjects earned positive profits so no subject was declared bankrupt within a session. If subjects had negative cumulative earnings after the five payment periods were drawn at the conclusion of the experiment, they received only their initial $5 starting balance plus a small payment from a lottery choice procedure employed to measure their risk attitudes. 11 See Healy (2006) for a discussion of advantages of this type of interactive decision tool. Requate and Waichman (2011) show that this kind of a profit calculator does not affect the degree of cooperative behavior as compared to standard payoff tables in a Cournot oligopoly setting. Our subjects used this tool frequently in initial periods. During the first round they submitted an average of 4.6 different estimates and 3.4 different estimates in the group tax and no tax conditions, respectively. After about 5 periods, however, subjects rarely entered more than the single estimate we required for each period. In the data analysis we do not analyze the specific numbers submitted for these estimates because they were non-incentivized and it is not possible to identify the difference between true expectations and simple exploration of the payoff function. 16

18 the treatments in which subjects were allowed to punish, they were informed about the total number of deduction points they received, the total number they assigned and the associated payoff reduction. Subjects were required to record this information on hardcopy record sheets at the end of each period. At the beginning of each experimental session an experimenter read the instructions aloud while subjects followed along on their own copy. As noted above, the instructions and computer decision screens used neutral terminology, as is the common practice in experimental economics. For example, the instructions simply refer to decision numbers, not emission levels that subjects chose. Also subjects received and assigned deduction points, not punishment points. The instructions for the Punishment with Group Tax treatment are shown in the Appendix. Subjects earned about US$27 on average. Including the instruction and payment distribution time, sessions usually lasted between 90 and 120 minutes. 3. Results We divide the results into three subsections. The first subsection compares pollution choices and environmental damages across treatments, and the second subsection compares realized efficiency of the formal and informal mechanisms. The third subsection examines punishment behavior. Except for the analysis of punishment, the treatment comparisons summarized below are based on statistics calculated for the later periods 16-30, after the initial learning phase during the first half of the sessions. All of the conclusions also hold when considering all 30 periods, but we omit those details to streamline the presentation. When examining punishment behavior and its impacts, we include the early periods because early peer punishment could influence later pollution choices. 17

19 3.1 Pollution and Environmental Damages We start with a comparison of the aggregate pollution levels across treatments. Table 2 reports the median pollution choice for firms in each location, separately for each individual group, ordered from lowest to highest. We use the median as a summary statistic of central tendency because it is robust for variability across groups and individuals, and extreme sample values and outliers. Using conservative nonparametric Mann-Whitney tests based on these independent medians, we can conclude that the group tax significantly lowers pollution choices in both the punishment and no punishment conditions, for both near and far firms (p-value<0.01 in all 4 comparisons). This formal group tax mechanism has a larger and more statistically significant influence on pollution choices compared to informal peer punishment. In the condition without the group tax, peer punishment does not have a statistically significant influence on pollution choices for either far (p-value=0.197) or near (p-value=0.392) firms. The availability of punishment significantly reduces pollution in the group tax condition for far firms (p-value=0.013), but for near firms the difference is not quite statistically significant (pvalue=0.128). Due to the different locationally-dispersed sources and heterogeneous transfer coefficients, equilibrium emission levels vary by location for both the group tax and no-tax treatments. The expected difference is only 5 units in the no-tax case (115 versus 110), whereas it is 30 units in the group tax case (65 versus 35). We find that the differences in median pollution between locations are not significant for the no-tax treatments for either punishment condition, but median pollution is lower for near firms than for far firms in the group tax condition (Wilcoxon signed-rank matched pairs test, n=14 pairs: p-value<0.05 for the 18

20 Punishment with Group Tax treatment; p-value<0.01 for the Group Tax Only treatment). 12 Parametric regressions of individual pollution choices, pooling across treatments and firm locations, provide more statistical power and strongly reject the null hypothesis of no treatment effects. For example, a tobit regression (n=3960) with subject-level clustered standard errors based on later periods indicates an average pollution choice of for the baseline case of far firms with no group tax or punishment. Highly significant dummy variables indicate that pollution choices are lower with the group tax (coefficient estimate -39.2), for near firms (-8.7), with punishment (-6.2) and for a punishment with group tax interaction (-9.9). Clearly the group tax has the strongest economically significant influence on individual pollution choices. 13 Figure 1 indicates that total environmental damage parallels the results presented for pollution choices by individual firms. Damages are not significantly different for the punishment and no punishment conditions in the no-tax treatment (Mann-Whitney p-value=0.248), and in both cases they are modestly (but significantly according to a Wilcoxon test, p-value<0.05) below the unregulated equilibrium level of Adding the group tax significantly reduces damages both for the no punishment and punishment conditions (p-value<0.01 for both cases). Thus, the group tax is effective in reducing pollution at the group (ambient pollution) as well as individual (emissions) level. Our factorial design permits an evaluation of the marginal impact of adding each factor (punishment or the group tax) independently, as well as their interaction, and this reveals that punishment does significantly reduce damages compared to the Group Tax Only 12 Table 2 indicates that average emissions are often significantly lower than the equilibrium predictions, except for the near firms under both treatments of the group tax condition where average emissions are significantly greater than the prediction of Subjects pollution choices are in general consistent with those of conditional co-operators in public goods games (e.g., Croson et al, 2005). In all treatments they choose pollution levels that are directly correlated with the average pollution level chosen by the others at their location in the previous period. Regressions for each treatment and location, which cluster standard errors at the subject level, indicate a significant (at p-value<0.01) influence of previous period group choices in all regressions except one (the Punishment with Group Tax treatment for near firms). Previous period average pollution at the other location has an inconsistent influence on current period pollution choices. 19

21 treatment, when the group tax is used (p-value=0.015). Total pollution is not significantly different from the target (and Nash equilibrium) of 84.6 when punishment is possible (Wilcoxon p-value=0.158), but it does exceed this target without punishment in the Group Tax Only treatment (p-value<0.01). Punishment also leads to a (marginally) significant increase in the rate that total pollution is below the target, from 17.6% without punishment to 35.7% with punishment (Mann-Whitney p-value=0.092). Although punishment has a weaker impact than the group tax, it significantly reduces individual and within-session variance in both the group-tax and no-tax conditions. Figure 2 displays the distribution of individual pollution choices in later periods (excluding the final period), separately for far firms (Panel A) and near firms (Panel B). The group tax clearly lowers pollution choices, consistent with the aggregate results presented above. Notice also that the availability of punishment substantially reduces high pollution choices, such as those at the maximum (125) in the no-tax condition and those above 100 in the group tax condition. Calculated by session for these late periods, average pollution standard deviations decline by 24 to 31 percent in the punishment compared to no-punishment condition. Substantial acrosssession variation exists, however, so these differences are not statistically significant. The only exception is for near firms in the no-tax condition where the difference is marginally significant (Mann-Whitney p-value=0.072). In the no-tax condition, the average within-session standard deviation of emissions is 11.4 without punishment but only 8.0 with punishment, a difference that is marginally significant (Mann-Whitney p-value=0.07). In the group tax condition, the average standard deviation of emissions is 11.9 without punishment but only 8.3 with punishment, which is also a marginally significant difference (Mann-Whitney p-value=0.06). Pooling across the tax and no-tax 20

22 conditions, the reduction in the standard deviation of emissions is highly significant (pvalue=0.01). Similarly, environmental damages are significantly less variable when punishment is possible for the no tax condition (p-value=0.046), marginally for the group tax condition (Mann-Whitney p-value=0.06), and when pooling across both tax conditions (p-value<0.01). This lower variance decreases the likelihood of extreme pollution levels and damages that could have significant environmental costs in some field applications Efficiency We define efficiency as the ratio of realized economic surplus over the maximum possible surplus (which is fixed at for our parameterization). Figure 3 indicates that median efficiency exceeds the 55.4 percent equilibrium benchmark in the no tax condition in 14 out of the 16 individual sessions. 15 This difference is significant in the Punishment Only treatment (Wilcoxon p-value=0.017) but it is only marginally significant in the Baseline treatment (Wilcoxon p-value=0.093). Efficiency increases as aggregate damages decrease, and as within-location pollution variance across firms decreases. Although efficiency is higher on average in the Punishment Only (64.4 percent) than Baseline (61.9 percent) treatment in part due to the lower variance in the punishment condition, this difference is small and is not statistically significant (Mann-Whitney p-value=0.834). The group tax significantly increases efficiency compared to the Baseline treatment benchmark, to an average of 86.8 percent in the Group Tax Only treatment and to an average of 90.0 percent in the Punishment with Group Tax treatment (p-values<0.01 in both cases). As in the no-tax condition, the marginal impact of punishment is 14 Punishment allows agents to retaliate against those who choose high pollution levels and impose a negative externality on the group. This suggests that punishment may be able to reduce inequity of the group tax arising from unequal pollution choices. Our data indicate that inequity (measured by the standard deviation of earnings across firms in each period, conditional on location) is always lower when punishment opportunities are available, but it is not statistically significantly lower than in the condition without punishment. 15 These figures always subtract the punishment costs and punishment penalties from earnings, since these represent deadweight losses. However taxes paid for failure to meet the target in the group tax mechanism do not reduce efficiency since they just represent a transfer from the firm to the regulator. 21

23 insignificant in the group tax condition (p-value=0.748). This occurs because the improvement in allocative efficiency due to lower and less dispersed pollution choices in the Punishment with Group Tax treatment is more than offset by the deadweight loss of the punishment costs and punishment penalties. 16 This negative impact of the deadweight loss from punishment is welldocumented in the peer punishment literature, even from its origins (Ostrom et al., 1992). Figure 4 displays the time series of efficiency, averaged across groups within each treatment, to highlight three features of the data. First, efficiency tends to rise over time for the group tax mechanism, especially when punishment is possible. Second, efficiency tends to fall across periods for the no-tax condition, and the time trend is stronger without punishment opportunities. Third, in the later periods efficiency tends to be higher in the conditions with punishment, both for the group tax and no-tax treatments, although as noted above these differences are not statistically significant. Note that efficiency drops in the final period in all treatments. For the punishment condition this end-period effect is due to a strong increase in punishment in the final period Punishment The treatment comparisons indicate that the formal group tax mechanism lowers pollution choices for near and far firms, lowers environmental damages, and raises efficiency. By contrast, the informal peer punishment mechanism has a more modest and inconsistent impact on performance. Punishment only decreases environmental damages when the group tax mechanism is also available, mainly through reductions in far firms pollution choices, although it does 16 We obtain identical results when employing an alternative Allocative Efficiency measure defined in Suter et al. (2008). This alternative metric subtracts the no regulation, status quo economic surplus (60.45 in our parameterization) from both the numerator and the denominator of the efficiency ratio, so it just lowers the average efficiency without changing the relative ranking across treatments or any of the statistical comparisons. 17 All of the conclusions drawn for the efficiency comparisons are robust to recalculation after excluding this final period. 22

24 reduce the variance of emissions and environmental damages. Punishment does not significantly raise efficiency in either the group tax or no-tax condition. Since imposing punishment is costly both for the punishers and the punished, efficiency levels are lower due to this deadweight loss. However, the failure of punishment opportunities to increase efficiency does not occur only due to substantial resources spent on socially-wasteful punishment, since punishment does not increase efficiency even if the calculations do not subtract any punishment costs exerted by punishers or suffered by the punished (Mann-Whitney p-values=0.401 for no-tax condition: comparing the Baseline with Punishment Only, for group tax condition: comparing the Group Tax Only with the Group Tax and Punishment treatment). Figure 5 indicates that average punishment points assigned by all six subjects per period varied considerably across sessions, ranging from about 1 to 8 punishment points. Figure 6 shows that punishment declined substantially over time in both treatments until the final period. The prominent increase in punishment during the final period, where it cannot affect future choices, suggests a strong emotional rather than strategic use of punishment (see Casari and Luini, 2012). The average punishment points also range over similar intervals for the group tax and no-tax treatments, and the use of punishment is not significantly different across treatments (Mann-Whitney p-value=0.562). Although Figure 5 suggests that efficiency tends to be higher in the sessions with less punishment in both treatments, greater punishment is significantly associated with lower efficiency only in the Group Tax treatment (correlation=-0.94; p- value<0.01). The data also do not indicate any differences in punishment rates between near and far firms (Wilcoxon p-values>0.5). 18 Following Fehr and Gaechter (2000), Figure 7 reports the average punishment points received as a function of the negative and positive deviations from others average pollution in 18 These conclusions are unaffected when dropping the final period with its pronounced increase in punishment. 23

25 the group. A subject who, for example, exceeded the average for the group by 8 to 14 units received on an average 2.45 punishment points in the group tax condition. The figure indicates that subjects received greater punishment when their pollution level exceeds the average pollution chosen by the other individuals at their location. 19 This is particularly true for large deviations, which led agents on average to receive up to an order of magnitude greater punishment. Consistent with the figure, Table 3 reports tobit regression models of the punishment received similar to those in the literature (e.g., Fehr and Gaechter (2000) Table 5), which confirm that positive deviations from average pollution levels result in significantly greater punishment. Our subjects do not exercise any systematic anti-social punishment, i.e., they do not sanction individuals who behave pro-socially and punish negative deviations, as has been observed in some previous studies especially outside of the United States and Western Europe (Herrmann et al., 2008). Table 4 indicates that individual subjects decisions on whether to punish, whom to punish and how much punishment to impose are motivated by relative pollution choices and how much group tax was paid in a given period. Subjects who punish choose lower pollution levels on average, and they apply punishment to peers whose pollution choice exceeds their own. The size of this increase is similar in the no-tax and tax conditions. The rightmost column with the group tax indicates that subjects punish even more strongly when they have to pay a larger group tax because of greater aggregate pollution. Finally, Figure 8 summarizes how subjects react to being punished. Modest amounts of punishment result in little change in pollution for the next period. When subjects receive 12 or 19 Most earlier research on punishment in social dilemmas has considered conditions in which choosing larger number (i.e., contributions to a public good) benefit others, so that negative deviations from the group average are punished. In our setting, larger numbers (pollution) harm others, and so positive deviations are punished. 24

26 more punishment points, however, they substantially reduce pollution. 20 Recall that 12 punishment points reduces the recipient s earnings for the period by 6 Experimental dollars, which is 60% of the mean equilibrium earnings in the No Tax condition and is 33% of the mean equilibrium earnings in the group tax condition. Regression estimates (that we do not present in the paper to save space) indicate that this relationship between punishment received in the previous period and the change in pollution emitted is highly statistically significant (coefficient of the lagged variable has a p-value <0.01 in both the regressions). 21 Although punishment clearly affects subjects subsequent choices, this figure and the previous Figure 7 suggest that punishment is used too infrequently and at levels too low to have a large impact on overall environmental performance. Subjects apparently must receive at least 12 punishment points before they systematically reduce pollution (on average), but Figure 7 indicates that less than 6 punishment points are received on average even when pollution exceeds the group mean by over 20. More specifically, for the 474 cases in which subjects chose pollution that exceeds their group mean by over 20 in the punishment condition, they received 0 punishment points 264 times (56%) and received 12 or more punishment points only 128 times (27%). Thus, although we implemented a 3-to-1 punishment effectiveness ratio that has been shown to be effective in the voluntary contributions literature, the large punishments required to substantially change pollution levels were infrequently assigned in this setting. This decreased the effectiveness of punishment. Our subjects also chose lower levels of punishment overall, since the average punishment points assigned per subject per period in our 30-period sessions (0.58) is at the low end of the range observed for average punishment in Gaechter et al. s (2008) 20 Median emissions of subjects who received 12 or more punishment points in the previous period were lower by 40% of the equilibrium level of emissions, whereas subjects who were not punished in the previous period reduced their emissions only by 3% of the equilibrium level. 21 The results are robust to the inclusion of demographic variables like gender and preferences for risk. 25

27 study. (Their subjects assigned per period punishment of 1.62 in 10-period sessions and 0.46 in 50-period sessions.) In the present study the low levels of punishment are insufficient to move subjects towards the socially optimal decision without the group tax. 4. Discussion Formal regulatory mechanisms have dominated environmental policy making in recent decades and less attention has been paid to non-regulatory and informal approaches that rely on social sanctions. Many (non-environmental) studies using experiments have documented increased cooperation and better social outcomes when subjects are allowed to sanction or punish others, even though theoretically profit maximizing individuals should not choose costly punishment. Most of this literature focuses on a homogenous framework with a linear transformation of contributions to the public good, where all individuals have similar incentives and similar payoff functions. 22 In most field applications, however, policy makers are faced with individuals who have heterogeneous and nonlinear effects on the environment. In this paper we examine a non-point source pollution problem where some agents have a more damaging environmental influence than others. Our findings can be summarized as follows. For this new setting with source heterogeneity, the formal targeting mechanism significantly increases efficiency and reduces environmental damages. With targeting, as predicted, firms nearer the receptor point choose lower emissions than far firms. The informal mechanism helps reduce the level and variability of emissions, but overall has a comparatively limited impact. In particular, peer punishment leads to 22 While individuals are homogeneous in terms of their incentives to cooperate, recent research has explored heterogeneity in the area of punishment. Nikiforakis et al. (2010) examine an asymmetric punishment institution in which subjects differ in terms of their punishment effectiveness. They find that in asymmetric conditions the levels of cooperation and efficiency are similar to the commonly-used symmetric setting. 26

28 lower damages only when it is combined with the formal mechanism, and it increases the frequency that damages stay below the formal target. Hence informal sanctions in this framework play a more complimentary role and do not have a consistent impact on their own. One of the reasons for this is that punishment is used too infrequently and at levels too low to strongly influence behavior. In our experiment we made numerous design choices to increase the likelihood that peer punishment would affect pollution and efficiency levels in this non-linear heterogeneous framework. First, we used the fine to cost of punishment ratio (3 is to 1) that has been shown to be very effective in improving cooperation in other studies. Second, we do not allow counter punishment as previous research finds that this option reduces the frequency and impact of punishment on cooperative behavior (Nikiforakis, 2008). Third, we allow for repeated interactions with fixed (partner) matching to allow for longer-term benefits of punishment. Fourth, to reduce the influence of across-location inequalities and social preferences on pollution and punishment decisions, we only allow subjects to observe and punish individuals from within their location group. In spite of these favorable design features, peer punishment in isolation is not very effective. This suggests that for informal mechanisms to have an impact they may need to be substantially stronger in a non-linear setting, compared to the homogeneous, linear conditions usually studied in the public goods literature. Stronger punishment, however, also implies higher costs and higher deadweight losses when used. Our findings therefore raise concerns about the efficacy and robustness of social norm enforcement alone in fostering cooperation in different contexts. 27

29 The formal targeting mechanism in our laboratory implementation has some advantages compared to the field, which should be kept in mind when drawing any policy implications. For example, we did not include any costs of imposing the tax, nor did we include litigation and enforcement costs. Nevertheless, the observed efficacy of the formal mechanism relative to informal social sanctioning suggests an important role for laws in focusing agents on particular social norms. When individuals are unsure of what the social norms are for example, in our framework they may be unsure about the appropriateness of particular emission levels in their location a regulatory mechanism can create a focal point by expressing social values that are codified into a legal system. 23 The specified target also identifies the social optimum for agents, although our experiment instructions did not explain this explicitly. Thus an appropriate norm for group cooperation that could justify informal punishment is clearer with targeting. The results highlight this important complementary role of social sanctions since environmental damages are lowest in the treatment where individuals face the combination of the formal and the informal mechanism. Moreover extreme pollution levels are less common when informal mechanisms are available, which could have important implications in the field. In the field, peer monitoring and peer punishment are successful often when there is an external agent who can regulate group members. Borrowers of micro credit, for example, monitor and punish their group members so as to successfully repay the group loan from the lender. In the Ankleshwar Industrial Estate example discussed earlier, one of the main reasons for imposing peer sanctions is to reduce attention received from the external regulator. The threat of an external agent can encourage the formation and enforcement of social norms within groups, and regulators can potentially use this to mitigate environmental problems. 23 Cooter (1998) argues that by expressing social values the law can tip a system into a new and perhaps better equilibrium. 28

30 References Alpizar, F., T. Requate, and A. Schram Collective Versus Random Fining: An Experimental Study on Controlling Ambient Pollution. Environmental and Resource Economics, 29(2): Blackman, A Informal Sector Pollution Control: What Policy Options do we have? World Development, 28(12). Bowles, S. and H. Gintis Homo Reciprocans. Nature, 415, Bowles, S. and S. Polanía-Reyes Economic Incentives and Social Preferences: Substitutes or Complements? Journal of Economic Literature, 50, Carpenter, J Punishing Free-Riders: How Group Size Affects Mutual Monitoring and the Provision of Public Goods. Games and Economic Behavior, 60 (1), Casari, M. and L. Luini Group Cooperation under Alternative Punishment Institutions: An Experiment. Journal of Economic Behavior and Organization, 71 (2): Casari, M. and L. Luini Peer Punishment in Teams: Expressive or Instrumental Choice? Experimental Economics, 15: Casari, M and C. Plott Decentralized Management of Common Property Resources: Experiments with a Centuries-Old Institution. Journal of Economic Behavior and Organization, 51(2): Cason, T. and F. U. Khan A Laboratory Study of Voluntary Public Goods Provision with Imperfect Monitoring and Communication. Journal of Development Economics, 58(2), Cason, T. and L. Raymond Framing Effects in an Emissions Trading Experiment with Voluntary Compliance. in R.M.Isaac and D. Norton (eds.), Research in Experimental Economics, vol. 14. (Bingley, UK: Emerald Group Publishing), Cason, T. and L. Gangadharan Communication and Peer Punishment in Nonlinear Public Goods Provision. Purdue and Monash Universities working paper. Chaudhuri, A Sustaining Cooperation in Laboratory Public Goods Experiments: A Selective Survey of the Literature. Experimental Economics 14, Cochard, F., M. Willinger, and A. Xepapadeas Efficiency of Non-Point Source Pollution Instruments: An Experimental Study. Environmental and Resource Economics 30(4): Cooter, R Expressive Law and Economics. Journal of Legal Studies, 27(S2),

31 Croson, R., E. Fatas and T. Neugebauer Conditional Cooperation in Two Public Goods Games: the Weakest Link and the VCM. Economics Letters, 87, Falk, A., Fehr, E., and Fischbacher, U Driving Forces Behind Informal Sanctions. Econometrica, 73 (6), Fischbacher, U z-tree: Zurich Toolbox for Ready-Made Economic Experiments. Experimental Economics 10(2), Fehr, E. and Gaechter, S Cooperation and Punishment in Public Goods Experiments. American Economic Review, 90 (4), Fehr, E. and Gaechter, S Altruistic Punishment in Humans. Nature, 415, Gaechter, S., E. Renner and M. Sefton The Long-Run Benefits of Punishment. Science, 5 December, Gaspart, F and E. Seki Cooperation, Status Seeking and Competitive Behavior: Theory and Evidence, Journal of Economic Behavior and Organization. 51, Giordana, G., and M. Willinger. Forthcoming. Regulatory Instruments for monitoring Ambient Pollution. in J. A. List and M.K. Price, eds., Handbook on Experimental Economics and the Environment. Northampton, MA: Edward Elgar Publishing. Greiner, B An Online Recruitment System for Economic Experiments. in K. Kremer and V. Macho, eds., Forschung und wissenschaftliches Rechnen Göttingen: Ges. Für Wiss. Datenverarbeitung Healy, P.J Learning Dynamics for Mechanism Design: An Experimental Comparison of Public Goods Mechanisms. Journal of Economic Theory, 129(1): Herrmann, B., C. Thoni and S. Gaechter Antisocial Punishment Across Societies. Science, 7 March, 319 (5868): Holmstrom, B Moral Hazard in Teams. Bell Journal of Economics, 13, Karp, L Nonpoint Source Pollution Taxes and Excessive Tax Burden, Environmental and Resource Economics Kathuria, V., and T. Sterner Monitoring and Enforcement: Is Two-Tier Regulation Robust? A Case Study of Ankleshwar, India. Ecological Economics, 57: Ledyard, J Public Goods: A Survey of Experimental Research. in Handbook of Experimental Economics, edited by J. Kagel and A. Roth, Princeton University Press. 30

32 List, J., S. Sadoff and M. Wagner So You Want to Run an Experiment, Now What? Some Simple Rules of Thumb for Optimal Experimental Design. Experimental Economics, 14, Lopez, M.C., J.C. Murphy, J.M. Spraggon, and J.K. Stranlund Comparing the Effectiveness of Regulation and Pro-Social Emotions to Enhance Cooperation: Experimental Evidence from Fishing Communities in Columbia. Economic Inquiry, 50: Masclet, D., C. Noussair, S. Tucker and M.-C. Villeval Monetary and Non-Monetary Punishment in the Voluntary Contributions Mechanism. American Economic Review, 93: Millock, K and F. Salanie Nonpoint Source Pollution When Polluters Might Cooperate. Topics in Economic Analysis and Policy, 5(1), Article 12: Nikiforakis, N. H. Normann and B.Wallace Asymmetric Enforcement of Cooperation in a Social Dilemma. Southern Economic Journal, 76: Nikiforakis, N Punishment and Counter-Punishment in Public Good Games: Can We Really Govern Ourselves?" Journal of Public Economics, 92: Ostrom, E Governing the Commons: The Evolution of Institutions for Collective Action. New York. Cambridge University Press. Ostrom, E Collective Action and the Evolution of Social Norms. Journal of Economic Perspectives, 14: Ostrom, E., Walker, J. and Gardner, R Covenants With and Without a Sword: Self Governance is Possible. American Political Science Review, 86, Oxoby, R. J., and J. M. Spraggon The Effects of Recommended Play on Compliance with Ambient Pollution Instruments. in T. L. Cherry, S. Kroll and J. R. Shogren, eds., Environmental Economics, Experimental Methods. New York: Routledge. Poe, G., W. D. Schulze, K. Segerson, J. Suter, and C. Vossler Exploring the Performance of Ambient-based Policy Instruments When Nonpoint Source Polluters Can Cooperate. American Journal of Agricultural Economics 86(5): Requate, T and I. Waichman A Profit Table or a Profit Calculator? A Note on the Design of Cournot Oligopoly Experiments Experimental Economics 14(1), Ribaudo, M.O., and M. Caswell Environmental Regulation in Agriculture and the Adoption of Environmental Technology. in F. Casey, A. Schmitz, S. Swinton and D. Zilberman, eds., Flexible Incentives for the Adoption of Environmental Technologies in Agriculture. Dordrecht: Kluwer Academic Publishers. 31

33 Romstad, E Team Approaches in Reducing Nonpoint Source Pollution. Ecological Economics, 47, Russell, C. S., and W. J. Vaughan The Choice of Pollution Control Policy Instruments in Developing Countries: Arguments, Evidence and Suggestions. in H. Folmer and T. Tietenberg, eds., The International Yearbook of Environmental and Resource Economics 2003/2004. Edward Elgar. Seabright, P Managing Local Commons: Theoretical Issues in Incentive Design. Journal of Economic Perspectives, 7, Segerson, K Uncertainty and Incentives for Nonpoint Pollution Control. Journal of Environmental Economics and Management 15(1): Spraggon, J Exogenous Targeting Instruments as a Solution to Group Moral Hazard. Journal of Public Economics. 84(2): Spraggon, J Exogenous Targeting Instruments with Heterogeneous Agents. Journal of Environmental Economics and Management 48(2): Spraggon, J. and R. J. Oxoby Game Theory for Playing Games: Sophistication in a Negative-Externality Experiment. Economic Inquiry 47(3): Spraggon, J. and R. J. Oxoby Ambient-Based Policy Instruments: The Role of Recommendations and Presentation. Agricultural and Resource Economics Review 39(2): Sterner, T Policy Instruments for Environmental and Natural Resource Management. Washington, DC: RFF Press. Suter. J. F., C.A. Vossler, G.L. Poe, and K. Segerson Experiments on Damage-Based Ambient Taxes for Nonpoint Source Polluters. American Journal of Agricultural Economics 90(1): Suter, J. F., C.A. Vossler, and G.L. Poe Ambient Based Pollution Mechanisms: A Comparison of Homogenous and Heterogeneous Groups of Emitters. Ecological Economics 68(6): Vossler, C.A., G.L.Poe, K. Segerson, and W.D.Schulze Communication and Incentive Mechanisms based on Group Performance: An Experimental Study of Non-point Pollution Control. Economic Inquiry 44(4):

34 Table 1: Experimental Design (264 Total Subjects) Formal Mechanism (Group Tax) No Yes Informal Mechanism (Peer Punishment) No 8 groups (48 subjects) 14 groups (84 subjects) Yes 8 groups (48 subjects) 14 groups (84 subjects) 33

35 Table 2: Median Pollution Choices by Firm Location and Treatment, Periods Far Firms Baseline (No Tax, No Punishment) Punishment Group Tax Only Only Individual Groups, average pollution per firm Average Across Groups Equilibrium Pollution Wilcoxon test p-value testing Ho: Average=Equilibrium Punishment with Group Tax Near Firms Baseline (No Tax, No Punishment) Punishment Group Tax Only Only Individual Groups, average pollution per firm Average Across Groups Equilibrium Pollution Wilcoxon test p-value testing Ho: Average=Equilibrium Punishment with Group Tax

36 Table 3: Tobit Models of Punishment Points Received Punishment Only Punishment with Group Tax Constant (5.92) ** (2.87) Others average emissions (0.06) 0.12** (0.04) Positive deviation from average emissions 0.18** (0.07) 0.33** (0.06) Negative (absolute) deviation from average emissions (0.04) (0.05) Number of Observations Notes: Clustered standard errors are robust to unspecified correlation within subjects and are shown in parentheses. ** denotes significantly different from zero at 1 percent level; * denotes significantly different from zero at 5 percent level (all two-tailed tests). Table 4: Tobit Models of Punishment Points Assigned Punishment Only Punishment with Group Tax Constant -4.44** (0.86) -6.45** (0.75) Amount of Group Tax Paid 0.04* (0.02) Positive deviation from i s own emissions 0.06** (0.02) 0.07** (0.01) Negative deviation from i s own emissions 0.01 (0.02) 0.02 (0.01) Number of Observations Notes: Clustered standard errors are robust to unspecified correlation within subjects and are shown in parentheses. ** denotes significantly different from zero at 1 percent level; * denotes significantly different from zero at 5 percent level (all two-tailed tests). 35

37 Environmental Damages Baseline (No Tax, No Punishment) Figure 1: Median Damages by Session, Periods Nash Equilibrium Punishment Only Group Tax Only Punishment with Group Tax 36

38 Figure 2, Panel A: Individual Emissions for Far Firms, Periods Frequency No Tax, No Punishment No Tax, Punishment Group Tax, No Punishment Group Tax, Punishment Pollution Choice Figure 2, Panel B: Individual Emissions for Near Firms, Periods No Tax, No Punishment No Tax, Punishment Frequency Group Tax, No Punishment Group Tax, Punishment Pollution Choice 37

39 1 0.8 Figure 3: Median Efficiency by Session, Periods Nash Equilibrium Baseline (No Tax, No Punishment) Punishment Only Group Tax Only Punishment with Group Tax 1 Figure 4: Time Series of Average Efficiency by Treatment Efficiency Baseline (No Tax, No Punishment) Group Tax Only Punishment Only Punishment with Group Tax Period 38

40 Session Efficiency Figure 5: Efficiency and Ave. Punishment Points Assigned by All 6 Subjects per Period, By Session (All Periods) Punishment with Group Tax Average Punishment Points Assigned Ave. Punishment Assigned per Period Figure 6: Time Series of Average Punishment Assigned by Treatment Punishment Only Punishment with Group Tax Period 39

41 Average Punishment Points Received Figure 7: Received Punishment Points for Deviations from Others' Ave. Pollution (All Periods) Punishment Only Punishment with Group Tax <-20 [-20,-14) [-14,-8) [-8,-2) [-2,2] (2,8] (8,14] (14,20] >20 Deviation from the Ave. Pollution of Other Group Members at Location Average Change in Pollution from t to t Figure 8: The Effect of Punishment on the Change in Pollution one Period to Next (All Periods) (Number of Observations shown next to bars) Punishment Only Punishment with Group Tax >15 Punishment Points Received in Period t

42 Appendix: Experiment Instructions INSTRUCTIONS FOR PART 1 YOUR DECISION In this part of the experiment you will be asked to make a series of choices in decision problems. How much you receive will depend partly on chance and partly on the choices you make. The decision problems are not designed to test you. What we want to know is what choices you would make in them. The only right answer is what you really would choose. For each line in the table in the next page, please state whether you prefer option A or option B. Notice that there are a total of 15 lines in the table but just one line will be randomly selected for payment. You ignore which line will be paid when you make your choices. Hence you should pay attention to the choice you make in every line. After you have completed all your choices a token will be randomly drawn out of a bingo cage containing tokens numbered from 1 to 15. The token number determines which line is going to be paid. Your earnings for the selected line depend on which option you chose: If you chose option A in that line, you will receive $1. If you chose option B in that line, you will receive either $3 or $0. To determine your earnings in the case you chose option B there will be second random draw. A token will be randomly drawn out of the bingo cage now containing twenty tokens numbered from 1 to 20. The token number is then compared with the numbers in the line selected (see the table). If the token number shows up in the left column you earn $3. If the token number shows up in the right column you earn $0. Any questions? A-1

43 Decision No. Option A 1 $1 $3 never Participant ID Option B $0 if 1,2,3,4,5,6,7,8,9,10,11,12,13, 14,15, 16,17,18,19,20 Please choose A or B 2 $1 $3 if 1 comes out of the bingo cage 3 $1 $3 if 1 or 2 comes out 4 $1 $3 if 1,2, or 3 5 $1 $3 if 1,2,3,4 6 $1 $3 if 1,2,3,4,5 7 $1 $3 if 1,2,3,4,5,6 8 $1 $3 if 1,2,3,4,5,6,7 9 $1 $3 if 1,2,3,4,5,6,7,8 $0 if 2,3,4,5,6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 3,4,5,6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 4,5,6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 5,6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 7,8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 8,9,10,11,12,13,14,15, 16,17,18,19,20 $0 if 9,10,11,12,13,14,15, 16,17,18,19,20 10 $1 $3 if 1,2,3,4,5,6,7,8,9 $0 if 10,11,12,13,14,15,16,17,18,19,20 11 $1 $3 if 1,2, 3,4,5,6,7,8,9,10 $0 if 11,12,13,14,15,16,17,18,19,20 12 $1 $3 if 1,2, 3,4,5,6,7,8,9,10,11 $0 if 12,13,14,15,16,17,18,19,20 13 $1 $3 if 1,2, 3,4,5,6,7,8,9,10,11,12 $0 if 13,14,15,16,17,18,19,20 14 $1 $3 if 1,2,3,4,5,6,7,8,9,10,11,12,13 $0 if 14,15,16,17,18,19,20 15 $1 $3 if 1,2,3,4,5,6,7,8,9,10 11,12,13,14 $0 if 15,16,17,18,19,20 A-2

44 INSTRUCTIONS FOR PART 2 This is an experiment in the economics of decision making. The instructions are simple and if you follow them carefully and make good decisions you will earn considerable money that will be paid to you privately in cash. All earnings on your computer screens are in Experimental Dollars. These Experimental Dollars will be converted to real Dollars at the end of the experiment, at a rate of Experimental Dollars = 1 real Dollar. At the beginning of the experiment everyone will begin with an initial balance of real Dollars. Today s session will be conducted using the computer network located here in this laboratory. This part of the session will last for 30 periods. Attached to these instructions you will find a sheet labeled Personal Record Sheet, which will help you keep track of your earnings based on the decisions you might make. You are not to reveal this information to anyone. It is your own private information. Part 2 You have been assigned to a group of six (yourself and five other) participants. This will be your group for the entire session (all 30 periods). In Stage 1 each period you (and the others in your group) will be asked to choose a number and to enter it into the computer. This is your Decision Number and it must be between 0 and 125. Your Total Payoff for Stage 1 each period is the sum of your Private Payoff and your Group Payoff: Total Payoff = Private Payoff + Group Payoff Your Private Payoff depends only on your own Decision Number. The formula describing your Private Payoff is Your Private Payoff = (125-D) 2, where D is your decision number. So that you do not have to worry about how to interpret this formula, Table 1 in these instructions shows the Private Payoff for a wide range of possible Decision Numbers. For example if your Decision Number is 60 Table 1 shows that your Private Payoff would be Experimental Dollars. Or if you had chosen 30 your Private Payoff would be Experimental Dollars. Notice that the higher your Decision Number the higher your Private Payoff. A-3

45 The Group Payoff depends on the total decision numbers chosen by everyone in the group and is the same for everyone in the group. In your 6-person group there are 3 BLUE individuals and 3 ORANGE individuals. You will be randomly assigned to one of these colors, and will remain this type of individual throughout this experiment. Denote the summed decision numbers for the 3 BLUE individuals as BLUE SUM, and the summed decision numbers for the 3 ORANGE individuals as ORANGE SUM. The computer will calculate the total weighted sum of the decision numbers chosen by all 6 individuals as TOTAL WEIGHTED SUM = [0.24(BLUE SUM)+0.36(ORANGE SUM)]. The computer will then compare this TOTAL WEIGHTED SUM number to If this amount is less than or equal to 84.6 then your Group Payoff will simply be negative one-sixth of this TOTAL WEIGHTED SUM amount: Your Group Payoff = (1/6)[0.24(BLUE SUM)+0.36(ORANGE SUM)] if [0.24(BLUE SUM)+0.36(ORANGE SUM)] If this TOTAL WEIGHTED SUM amount is greater than 84.6, then your Group Payoff is reduced further by (5/6) times the difference between 84.6 and the TOTAL WEIGHTED SUM amount. In this case, your Group Payoff is: Your Group Payoff = (1/6)[0.24(BLUE SUM)+0.36(ORANGE SUM)] {(5/6) [0.24(BLUE SUM)+0.36(ORANGE SUM)]-84.6} if [0.24(BLUE SUM)+0.36(ORANGE SUM)]>84.6. Notice that the impact of the Decision Numbers depends on the type of individual in the group. So that you do not need to worry about how to interpret these formulas, Table 2 in these instructions shows your Group Payoff for a wide range of combinations of BLUE and ORANGE SUM numbers. Notice that your Decision Number affects either the BLUE SUM or ORANGE SUM amounts, depending on your color. Suppose, for example, that you are a BLUE type and your Decision Number is 60 while the other 2 BLUE type individuals total decision numbers add to 105. Then the total BLUE SUM is =165. Suppose further that the 3 ORANGE type individuals total ORANGE SUM is 135. For this example the TOTAL WEIGHTED SUM is ( )+( )=88.2. Table 2 shows that your Group Payoff would be Experimental Dollars, which is the number found at the intersection of the BLUE SUM row labeled 165 and the ORANGE SUM column labeled 135. Notice that higher SUMS reduce your Group Payoff. A-4

46 Choosing your Decision Number Your task in Stage 1 each period is to choose your Decision Number. You may use the formulas or tables in these instructions to help you make your choice. You may also find it useful to use your computer to investigate how different choices that you and others in your group affect your Total Payoff for the period. When we start the computers to begin the experiment, your screen will look similar to the one shown in Figure 1 below. (It may look slightly different depending on whether you are a BLUE or ORANGE type of individual in the experiment.) You do not know what others in your group will choose for their decision numbers in the current period, but on the upper right part of this screen you can enter in guesses for the SUMS chosen by the other people in your group (not counting your own Decision Number). After you do this, click Enter Guess. The computer will require you to enter at least one guess at the start of each period. After you do this, your screen will look similar to Figure 2. The computer will draw a graph indicating your potential Total Payoff (including both your Private Payoff and your Group Payoff) for every possible Decision Number you might choose, based on your guess about the other group members choices. At this point you can enter a possible Decision Number for yourself. The screen will display your payoff on this graph with a red dot. You can enter in different Decision Numbers if you wish to see how your Total Payoff changes. Your different possible choices and guesses will be recorded in the upper left part of your computer screen. At this point you can go back to change your guess by clicking the Enter New Guess button on the lower right of the screen. This returns you to a screen similar to Figure 1, except that your previous guesses and possible Decision Numbers will be shown in the table on the top left. You can continue to enter many guesses and Decision Numbers to explore how they affect your Total Payoff. When you are ready to submit your actual Decision Number, click the Submit Final Decision Number button on the bottom of your screen. This will take you to a screen like Figure 3, to submit your final Decision Number for the period. The Second Stage each Period In the second stage you will see the Decision Numbers chosen for the two other individuals in your group who are the same color type as you, labeled with the names Circle and Square. These (anonymous) individuals will keep these name labels throughout the experiment A-5

47 today. Moreover, in this stage you can decide whether to decrease the payoff of these two other group members by assigning deduction points. These other two group members can also decrease your payoff if they wish to. This is apparent from the input screen at the second stage, shown in Figure 4. Your Decision Number is displayed in the first row, while the decision numbers of the other people with the same color type as you are shown in the lower two rows. You will have to decide how many deduction points to assign to each of these other two group members. You must enter a number for each of them. If you do not wish to change the payoff of a specific group member then you must enter 0. You can assign up to 5 points to each group member. You will incur costs from assigning deduction points. Every deduction point you assign costs you 0.50 Experimental Dollars. For example, if you assign 2 deduction points to one member, this costs you 1 Experimental Dollars; if, in addition, you assign 4 deduction points to the other member this costs you an additional 2 Experimental Dollars. In total for this example you will have assigned 6 points and your total costs therefore amount to 3 Experimental Dollars. After you have assigned points to each of the other two group members you can click the button Calculate (see Figure 4). On the screen you will then see the total costs of your assigned points. As long as you have not yet clicked the Submit Deduction Points button, you can still change your decision. To recalculate the costs after a change of your assigned points, simply press the Calculate button again. If you assign 0 deduction points to a particular group member (i.e., enter 0 ), you will not alter his or her payoff. However, if you assign one deduction point to a group member you will decrease the payoff of this group member by 1.50 Experimental Dollars. If you assign a group member 2 deduction points you will decrease the group member s payoff by 3 Experimental Dollars, and so on. Each deduction point that you assign to another group member will reduce his or her payoff by 1.50 Experimental Dollars. Similarly, each deduction point assigned to you by another group member will reduce your first stage payoff by 1.50 Experimental Dollars: Costs of received deduction points = 1.50 Sum of received deduction points. How much the payoff at the second stage is decreased depends on the sum of deduction points received. For instance, if somebody receives a total of 3 deduction points (from all other A-6

48 group members in this period), his or her payoff would be decreased by 4.50 Experimental Dollars. If somebody receives a total of 4 deduction points, his or her payoff is reduced by 6 Experimental Dollars. Your total payoff from the two stages is therefore calculated as follows: Total payoff at the end of the second stage = period payoff = = Payoff from the first stage 1.50 (sum of received deduction points) 0.50 (sum of deduction points you have assigned) After all participants have made their decision, your payoff from the period will be displayed on a screen such as the one shown in Figure 5. After you have viewed the payoff screen the period is over and the next period commences. Recall that 30 periods will be conducted in this part of the experiment. Recording Rules During every period you should write down the information shown on your results screen on your Personal Record Sheet. The screen where you choose deduction points shows the Decision Numbers you should record, and the final payoff screen shows your Deduction Points and payoffs. Be sure to record your total payoff for each period in the rightmost column. At the end of the experiment we will randomly select 5 periods from this part of the experiment for payment. You will sum up the earnings from these 5 periods and divide this amount by the conversion rate to determine your total earnings for this part in real Dollars. Summary You and the 5 others in your group chose a Decision Number each period Your Total Payoff before Deduction Points for each period is the sum of your Private Payoff and your Group Payoff You may use the Tables or your computer to determine how different choices affect your Total Payoff You may assign up to 5 Deduction Points to the two other individuals in your group who are the same color type as you. Each Point you assign costs you 0.50 Experimental Dollars. These two individuals can assign Deduction Points to you. Each Point assigned to you reduces your payoff by 1.50 Experimental Dollars You will interact with the same 5 other individuals for 30 decision periods in this part Results and payoffs should be recorded on your Record Sheet at the end of each period A-7

49 Figure 1: A-8

50 Figure 2: A-9

51 Figure 3: A-10

52 Figure 4: A-11

53 Figure 5: A-12