Do individuals care about fairness in burden sharing for climate change mitigation? Evidence from a lab experiment Robert Gampfer ETH Zurich, Center for Comparative and International Studies and Institute for Environmental Decisions Haldeneggsteig 4 8092 Zurich Switzerland Phone: +41 44 632 4644 Fax: +41 44 632 1289 gampfer@ir.gess.ethz.ch Electronic Supplementary Material 1. Lab experiments vs. stated preference approaches, monetary incentives, external validity This revealed-preferences method complements a growing number of studies that use statedpreference approaches to examine individuals opinions on burden sharing, or on global climate politics in general. Large-scale surveys (e.g. World Bank 2010; Leiserowitz 2006) and survey experiments (e.g. Tingley and Tomz 2014; Bernauer and Gampfer 2013) have yielded important insights on individual preferences for diverse aspects of global climate governance; but their methods have the potential shortcoming that respondents rarely face real consequences from the choices they make in the survey. Responses might thus to a lesser degree motivated by self-interest than real-world choices (where, for example, support for ambitious climate policies could lead to higher energy taxes). Additionally, it might simply be difficult for respondents to gauge the consequences their choices would entail in reality, again rendering responses less meaningful. Monetary incentives in the lab make choices consequential in a rather direct and obvious manner, thus alleviating this potential validity problem. Lab experiments have however invited criticism themselves, of which two issues need to be mentioned here. First, since monetary incentives are usually not large, the stakes are small compared to the real-world problem of climate change. Furthermore, even if it is possible to lose a substantial part of their experimental endowment, participants can expect that losses will not extend to their actual private wealth, potentially leading them to make more risky choices. While there is still no conclusive evidence on this, numerous studies indicate that participants in lab experiments act relatively lossaverse regarding their endowment, and that the size of stakes has only limited influence (e.g. Carpenter et al. 2005; Franciosi et al. 1996). The second critique concerns external validity, as undergraduate students, the most common lab populations, are hardly a random sample from the general population. In defense, some researchers argue that laboratory experiments test theories about essential human interactive behavior in generic choice situations that are applicable to many concrete situations (Morton and Williams 2010: 341) - for example, international climate governance, a fundamental collective-risk social dilemma (Milinski et al. 2008: 2291). Still, caution is in order when generalizing insights from lab experiments. I do not claim that my results are representative e.g. for the population of a given country; I merely test average treatment effects of fairness criteria on the sample under study. 1
2. Technical details of the experiments Experiments were conducted in November 2012 and January 2013 at ETH s Decision Science Laboratory (www.descil.ethz.ch) using z-tree as software platform (Fischbacher 2007). Participants were recruited from the ORSEE Zurich pool (Greiner 2004), which contains predominantly students from Zurich universities. Individuals who had already participated in similar experiments were excluded from the sampling population to reduce the potential for super-game strategic behavior acquired through learning. Sample sizes were 234 for the first and 135 for the second experiment. To mitigate endowment effects and strategic behavior over the three rounds, I used a random-round payoff mechanism. At the end of the experiment, one of the three rounds was randomly selected for payout. Each player received what was left from his endowment in this round at an exchange rate of 1 Swiss Franc per ECU. This mechanism was common knowledge. Additionally, a 10 Franc show-up fee was paid. Mean payout in the first experiment was 29.2 Francs (standard deviation 6.75, range from 17.5 to 40); mean payout in the second experiment was 34.82 Francs (standard deviation 5, range from 21 to 43). Fig. S1 Above: decision structure of a round in the ultimatum game in experiments 1 and 2; payoffs shown next to each branch s end. Below: structure of a round in the preliminary stage of experiment 2 2
3. Sample introduction and rules of the game (from the capacity treatments) Welcome to the experiment! Please read this document carefully. Introduction In this experiment you can earn money. How much you earn will depend on your decisions during the experiment. Please do not talk to other participants or draw any attention to yourself. Anonymity is guaranteed throughout the experiment. The researchers cannot infer your identity from your decisions recorded during the experiment. Please read the following rules of the experiment carefully. Should you have any questions, now or during the experiment, please raise your hand and we will come to you. In the game we will simulate climate change negotiations. Scientists see global climate change as a serious environmental problem faced by mankind. The great majority of climate scientists expect the global average temperature to rise by 1 6 C by the year 2100. Virtually all scientists agree that the main cause for global warming is the emission of greenhouse gases, especially CO 2. CO 2 originates from burning fossil fuels like coal, oil, or natural gas in industrial processes, for energy production, and for powering cars and trucks. Each unit of CO 2 emitted has the same effect on the Earth s climate, regardless of where it has been emitted that means CO 2 is a global pollutant. If global warming continues unchecked, there are likely to be severe consequences. For example, melting ice caps at the poles will probably lead to sea level rises, and extreme weather events like droughts and heavy storms will become more common. Some of these effects could be catastrophic for nature and for societies and lead to heavy economic damage and human suffering. In order to keep global warming and the risks associated with it at a manageable level, mankind has to reduce greenhouse gas emissions. However, reducing such emissions is costly, for example because new industrial technologies have to be developed and economic growth could be reduced, at least in the short run. These costs have to be shared somehow among countries and individuals. In this experiment, you will negotiate with another participant how much each of you should pay for reducing global warming. Rules of play During the experiment we will use an artificial currency, called Experimental Currency Units (ECU). At the beginning of each round of the game, each of you receives an initial endowment of money. This endowment can be either 30 ECU or 15 ECU. Within groups of two players, you will have to negotiate how to share the cost of reducing greenhouse gas emissions. The total cost of reducing greenhouse gas emissions is 10 ECU. If you agree on a distribution of costs with the other player, you have successfully reached an agreement against global warming. The computer will then subtract the agreed share of the costs from each player s endowment. If you do not agree on a distribution of costs, there will be no agreement against global warming. In this case a climate catastrophe will occur with a probability of 50%, inflicting heavy economic damage. This means that if such a catastrophe occurs, each of you will lose half of your endowment. If you reach an agreement on the distribution of costs, there will be no climate catastrophe. We now describe how a round in the game will look like (there will be three rounds in total). In every group of two, one player will be assigned the role of proposer, and one player the role of responder. Every proposer and every responder receives either 15 or 30 ECU (the endowment ) at the beginning of each round. In the first screen, you will be shown which role you have been assigned, the size of your endowment and the other player s endowment, and the total cost of reducing greenhouse gas emissions. If you have read all the information, click Continue. The player who is the proposer will then be asked to enter how much of the total cost she/he offers to pay. This offer has to be a whole number between 0 and 10. If the proposer offers to pay an amount of x 3
ECU, this means that the responder would have to pay an amount of 10 x ECU. After the proposer has made the offer, the responder will be shown the suggested cost distribution and will be asked whether he wants to accept or reject this distribution. If the responder accepts, there will be no climate catastrophe and the computer will implement the agreement: the proposer s share of the cost is subtracted from the proposer s endowment, and the responder s share from the responder s endowment. If the responder rejects the suggested cost distribution, there will be no agreement and no costs have to be paid. However, with a probability of 50% a climate catastrophe will occur. In that case, each player loses half of their endowment (i.e. either 30/2=15 ECU or 15/2=7.5 ECU). Whether this catastrophe occurs is determined by the computer (it generates a random number between 0 and 1, and there will be a catastrophe when this number is smaller or equal than 0.5). At the end of each round, you will be shown the suggested cost distribution, the decision of the responder, whether there was an agreement, whether the climate catastrophe occurred, and how much of your endowment remains. You will play three rounds of this game. The rounds will be identical, except that your role (proposer or responder) and your endowment can change after each round. In each round you will start again with an endowment of either 30 ECU or 15 ECU. The outcome in one round does not depend on your decisions in a previous round. You will be matched with a different participant in each round. You will not know who the other player in your group is, nor will that player know your identity. The game will thus be completely anonymous. At the end of the experiment, you will be paid out your remaining endowment from one of the three rounds (in addition to your fixed show up fee). Which round will be paid out is unknown until the end of the experiment. The computer will then randomly select one of the three rounds. You will receive 1 Swiss Franc for each 1 ECU that is left in your endowment. After the third round, you will be asked to fill out a short questionnaire. There are no right or wrong answers, and your payment does not depend on your answers in the questionnaire. As for the whole experiment, researchers will not be able to infer your identity from the questionnaire. If you have any questions now or during the experiment please raise your hand. We will then come and assist you. 4
4. Sample screenshots from participants screens (High-capacity low-vulnerability proposer) Fig. S2 Information shown to responder at the beginning of each round. The screen shown to the proposer contains the same information and has the same layout. Fig. S3 Screen prompting the proposer to suggest a cost distribution (input in blue box). 5
Fig. S4 Screen showing responder the suggested cost distribution and asking for her decision whether to accept or not. Fig. S5 Information shown to proposer after each round. The screen shown to the responder contains the same information and has the same layout. 6
5. Potential learning effects during the experiment In the first experiment, means and distributions of offered proposer shares differ slightly between rounds in all treatment groups (Fig. A5), but these differences are mostly small in size and not statistically significant (the exception is between rounds 1 and 3 in the baseline group, where mean offers decrease by 1 ECU, p-value.0488). Potential learning effects from one round to the next are thus not apparent. Above all, patterns do not suggest any convergence. As regards the responder side, acceptance rates do not change significantly from one round to the next. Baseline LowCap HighCap Offered proposer share 0 5 10 0 5 10 LowCap+HighVul HighCap+LowVul 1 2 3 Graphs by Treatment 1 2 3 1 2 3 Round Fig. S6 Distribution of (offered) proposer shares by round in each treatment condition (first experiment). 6. Findings from post-experiment questionnaires Participants answered the questionnaires directly after the end of the game within the same software environment, before receiving their payout. It can thus not be ruled out that responses to a certain degree reflect ex-post rationalization of their behavior during the game. Participants were asked what they thought would have been the fairest offer a proposer could have made. Their responses largely reflected observed behavior in the experiments. In experiment 1, participants in the baseline group on average thought an offer of 5 would be fairest. In the poor and high proposer conditions, it was 3.6 and 6.75, respectively. In the poor, highly-vulnerable proposer group the average fairest offer was 4.4, and in the rich, little-vulnerable proposer group, it was 6.1. These responses are very consistent with the offers actually observed in the experiment. The differences between baseline and the capacity treatments are significant at the.1% level and that between low capacity and low capacity, high vulnerability at the 5% level. The difference between 7
high capacity and high capacity, low vulnerability treatments is significant only at the 10% level, corroborating the finding from the game itself that if poor, highly-vulnerable proposers make an offer, less asymmetric cost distributions are seen as fair than if proposers are rich and little-vulnerable. In the second experiment 48% of participants said a fair offer would be one based on the differences in contributions to climate risk of the two players. 14% said an equal share would be fair, 7% mentioned endowment as fairness criterion, and 8% said that a fair offer should equalize payoffs as much as possible (the remainder did not answer the question). Further questions asked participants to describe the motivation for their behavior in those cases where they were proposers and responders, respectively. Participants could list more than one motivation for their behavior (so percentages do not add up to 100). For their proposer role, in the baseline group of the first experiment, 28% of participants described motives of self-interest (e.g. maximizing payoff, protecting my endowment ), and 62% motives of fairness ( an equal share, a fair share etc.). In the capacity treatments, 27% cited self-interest and 67% fairness, either in general terms or related to endowment differences. In the capacity-plus-vulnerability treatments, 40% mentioned self-interest and 44% fairness. Only 15% in these two groups related fairness to vulnerability, but 17% listed the weaker bargaining position of the more vulnerable player as a motive for their decision. This corroborates the finding that including vulnerability in the player characteristics induces more selfinterested considerations. Regarding motives for their responder behavior, 28% in the baseline condition cited self-interest and 63% fairness. In the capacity treatments, self-interested motives were mentioned by 37% and fairness motives by 53%. The percentage of fairness motives was even smaller in the capacity-plusvulnerability treatments (47% vs. 35% self-interest), with only 7% explicitly mentioning vulnerability in connection with fairness. Overall, thus, responders who are only able to accept or reject a given offer seem to tend a bit more toward rational payoff maximization. In the second experiment, 31% of participants described self-interested motives for their behavior as proposers. 67% described fairness motives, and 60% explicitly mentioned historical responsibility (or the difference in it between proposer and responder). Regarding their behavior as responders, a little less mentioned fairness motives (56%, with 52% relating it to climate risk contributions, compared to 32% citing self-interest). Again, these stated motivations are rather consistent with the observed behavior in the second experiment. 8
Fig. S7 First experiment: treatment effects. Dots show difference in mean (offered) proposer shares between treatment and baseline conditions (i.e. size of treatment effect), whiskered lines depict t-tests 95% confidence intervals Fig. S8 Relationship between difference in historical responsibility (proposer s responsibility responder s responsibility) and proposer s offer. 9
Table S1 Parameters in the treatment conditions of the first experiment. Columns correspond to treatment conditions; rows to parameters. Proposer endowment Responder endowment Total cost of mitigation Proposer loss rate Responder loss rate Probability of catastrophe Baseline Low-capacity proposer High-capacity proposer Low-capacity high-vulnerab. proposer High-capacity low-vulnerab. proposer 30 15 30 15 30 30 30 15 30 15 10 10 10 10 10 0.5 0.5 0.5 0.66 0.33 0.5 0.5 0.5 0.33 0.66 50% 50% 50% 50% 50% Table S2 Responder acceptance, first experiment: permutation tests of Chi2 tests. Each treatment group is compared with the control group. c indicates the number of obtained Chi2 values that are higher than the actually observed Chi2 value (1000 permutations). Treatment c p-value SE(p) 95% confidence interval LowCap 123 0.1230 0.0104.1032769.1449722 HiCap 169 0.1690 0.0119.1462745.1936927 LowCap, HiVul 10 0.0100 0.0031.0048055.0183132 HiCap, LowVul 1 0.0010 0.0010.0000253.0055589 Table S3 Results of linear regression of proposer s offer on difference in responsibility, total climate risk, and endowment ratio (regression coefficients, standard errors in parentheses). Model P1 includes only constituent terms, model P2 includes an interaction between endowment ratio and difference in responsibility. Standard errors are clustered on proposer participant. Number of observations: 135. Offer Responsibility difference Total climate risk Endowment ratio (prop./resp.) End. ratio * responsibility diff. F-Stat. R 2 Model P1 16.1582*** (5.2921) -2.3661** (1.0923) -1.3568 (1.3554) -- 30.92.4386 17.6108*** -2.1406* Model P2 (6.0618) (1.0679) Significance levels: * <.1, ** <.05, *** <.01-3.4184 (4.6372) 4.5696 (9.6584) 25.07.4403 10
Table S4 Logistic regression of responder s decision on proposer s offer, responsibility difference, total climate risk, and endowment ratio (regression coefficients, standard errors in parentheses). Model R1 includes only constituent terms, models R2 and R3 add interaction effects. Number of observations: 135 Acceptance R1 R2 R3 Offered proposer share 1.3357*** (.2978) Responsibility difference -43.7046** (19.2184) Total climate risk 5.8534** (2.6642) Endowment ratio 8.7912* (4.9313) End. ratio * responsibility diff. Offer * responsibility diff. -- 1.3378*** (.2995) -44.9026* (25.7640) 5.7680** (2.6284) 9.9897 (14.9374) -2.3158 (24.6752) -- -- 1.3834*** (.3301) -40.9123* (24.0853) 6.4084** (2.8011) 7.5531 (14.4486) 8.1287 (27.9798) -2.3668 (2.8911) Wald Chi 2 24.29 24.31 28.71 Pseudo R 2.3039.3040.3114 Significance levels: * <.1, ** <.05, *** <.01 References Bernauer T, Gampfer R (2013) Effects of civil society involvement on popular legitimacy of global environmental governance. Global Environ Chang 23 (2):439-449 Carpenter J, Verhoogen E, Burks S (2005) The effect of stakes in distribution experiments. Econ Lett 86 (3):393-398 Fischbacher U (2007) z-tree: Zurich toolbox for ready-made economic experiments. Exp Econ 10 (2):171-178 Franciosi R, Kujal P, Michelitsch R, Smith V, Deng G (1996) Experimental tests of the endowment effect. J Econ Behav Organ 30 (2):213-226 Greiner B (2004) The Online Recruitment System ORSEE 2.0 - A Guide for the Organization of Experiments in Economics. Working Paper, University of Cologne Leiserowitz A (2006) Climate change risk perception and policy preferences: the role of affect, imagery, and values. Climatic Change 77 (1):45-72 Milinski M, Sommerfeld RD, Krambeck H-J, Reed FA, Marotzke J (2008) The collective-risk social dilemma and the prevention of simulated dangerous climate change. P Natl Acad Sci USA 105 (7):2291-2294 Morton RB, Williams KC (2010) Experimental Political Science and the Study of Causality. Cambridge University Press, New York Tingley D, Tomz M (2014) Conditional Cooperation and Climate Change. Comp Polit Stud, forthcoming World Bank (2010) Public attitudes toward climate change: findings from a multi-country poll. World Bank, Washington, D.C. 11