Wealth and Vulnerability to Climate Change: An Experimental Study on Burden Sharing among Heterogeneous Agents

Abstract

Agents in global climate negotiations differ with respect to their vulnerability to the negative consequences of climate change, but also their ability to contribute to its prevention. Due to this multidimensional heterogeneity, agents disagree about how the costs of emission reduction ought to be shared and, as a consequence, efficiency is low. This experiment varies the two dimensions separately in a controlled setting. The results show that in groups that succeed in reaching a predefined threshold, the rich and the more vulnerable, ceteris paribus, tend to carry a larger share of the burden. Surprisingly, groups are most likely to master the challenge when poverty coincides with high vulnerability. In this case the rich and less vulnerable abstain from interpreting fair burden sharing in a self-serving manner. Instead, they seem to acknowledge the double-disadvantaged position of the poor and more vulnerable and voluntarily carry a larger share of the burden.

Appendices

Appendix 1: Pledge Mechanism

Subjects in the experiment were asked at the end of round 3 and round 7 to make non-binding pledges about their contribution intentions for the remaining rounds. Comparing subjects’ pledges with their actual contributions in the respective rounds, I find that subjects in general respected their pledges. The two upper panels in Fig. 5 show that observations scatter closely around the diagonal line in successful groups. Wilcoxon signed rank tests find no significant differences between pledges and contributions in round 3 or round 7 (p = 0.707 and p = 0.667, respectively). In failing groups, depicted in the two lower panels of Fig. 5, contributions tended to be significantly lower than pledges in round 3 (Wilcoxon signed rank test, p = 0.000). While pledges in round 3 were not significantly different between successful and failing groups (Wilcoxon rank-sum test, p = 0.380), pledges in round 7 were significantly lower in failing groups (Wilcoxon rank-sum test, p = 0.000). There are no significant treatment differences.

Fig. 5

Contribution-pledge comparison for successful (top) and failing groups (bottom)

These findings suggest that pledges—if respected—can serve as a useful coordination mechanism allowing groups to reach the threshold. This is in line with Tavoni et al. (2011) who found that groups that are able to communicate through a simple pledge mechanism are significantly more successful in reaching the threshold than groups that have no means to communicate.

Appendix 2: Experimental Instructions

2.1 Sample Instructions for WLT-VUL [Translated from German]

Welcome to the experiment and thank you for your participation. Please do not talk to the other participants of the experiment.

During the experiment, you and the other participants will be asked to make decisions. Your payout depends on your own decisions and the decisions of the other participants. You are paid individually, in private and in cash after the experiment. During the experiment, you will not be using Euro, but tokens (game coins). They are converted after the experiment at the following exchange rate: 1 token = 0.25 Euro.

Please take sufficient time to read the explanations and to make your decisions. The experiment is completely anonymous. Neither you nor the other participants will be told the identity of the players, either during the experiment or after it.

If you have questions, please raise your hand. One of the lab assistants will come to you and answer your question. The experiment will take about 60 min.

2.2 Proceedings of the Experiment

At the beginning of the experiment, all participants are randomly divided into groups of 4 players. They stay in the same group throughout the experiment and play with the same people. All members of a group receive the same information and face the same decisions. All decisions in the experiment are anonymous. For this purpose, each player is randomly assigned a player ID (a letter between A and D). The player IDs remain the same throughout the experiment. Your own player ID will be displayed in the top right-hand corner of your screen throughout the experiment (see figure on page 2).

In each group there are two purple players and two orange players. At the beginning of the experiment you will learn your own type of player as well as the type of player of the other members in your group. You can recognize the player type by the color in which the player ID is written (see figure on page 2).

At the beginning of the experiment, all participants will receive an initial endowment of 40 tokens. Earnings during the experiment are added to this endowment while losses are deducted.

The experiment consists of 10 rounds. At the beginning of each round each player receives a budget (round budget). The exact amount of tokens a player receives as a round budget depends on the type of player. Purple players receive 5 tokens, orange players 3 tokens. From this round budget you can invest an amount of your choice in a joint project with the other group members. Purple players can invest between 0 and 5 tokens in the group project, orange players between 0 and 3 tokens. Any token that you do not invest in the group project will be added to your private wealth. In total, purple players possess round budgets of 50 tokens (10 round × 5 tokens per round = 50 tokens), while orange players possess a total of 30 tokens (10 round × 3 tokens per round = 30 tokens).

After all members of a group have made an investment decision, all members will receive the following information (see figure below):

• (1) Amount of tokens each group member has invested in the group project in the current round (member investment in round x)

• (2) Amount of tokens all group members together have invested in the group project in the current round (group investment in round x). This equals the sum of all investments in (1).

• (3) Amount of tokens that each group member has invested in the group project in all rounds played so far (total member investment in rounds 1—x).

• (4) Amount of tokens that all group members have jointly invested in the group project in all rounds played so far (total group investment so far). This equals the sum of all investments in (3).

All tokens from the round budget that you do not invest in the group project will be added to your initial endowment. During the experiment, you will be kept informed about the current amount of your private wealth.

• Private wealth = initial endowment + tokens from the round budgets you did not invest in the group project

2.3 Payment

Your payoff at the end of the experiment will depend on how many tokens your group has invested in the group project in total in all 10 rounds. There is a target value of 80 tokens. If this target value is reached, all members of the group will be paid the private wealth they accumulated up to the 10th round.

If the total investment in the group project at the end of round 10 is less than the target value of 80 tokens, each group member loses a certain amount of tokens. The lost tokens are deducted from the private wealth and the remaining wealth is paid out. The exact amount of tokens a player loses depends on the player type: Purple players lose 20 tokens, orange players lose 40 tokens.

2.4 Examples

• Example 1: You are an orange type and you have invested a total of 23 tokens in the group project. Your group has invested in total more than 80 tokens in the group project, i.e. the target value has been reached. Your payout is therefore 40 tokens (initial endowment) + 7 tokens (uninvested round budget [30-23]) = 47 tokens.

• Example 2: You are an orange type and you have invested a total of 23 tokens in the group project. Your group has invested in total less than 80 tokens in the group project, i.e. the target value was not reached. Your payout is therefore 40 tokens (initial endowment) + 7 tokens (uninvested round budget [30-23]) − 40 tokens (loss for orange types) = 7 tokens.

• Example 3: You are a purple type and you have invested a total of 23 tokens in the group project. Your group has invested in total more than 80 tokens in the group project, i.e. the target value has been reached. Your payout is therefore 40 tokens (initial endowment) + 27 tokens (uninvested round budget [50-23]) = 67 tokens.

• Example 4: You are a purple type and you have invested a total of 23 tokens in the group project. Your group has invested in total less than 80 tokens in the group project, i.e. the target value was not reached. Your payout is therefore 40 tokens (initial endowment) + 27 tokens (uninvested round budget [50-23]) − 20 tokens (loss for purple types) = 47 tokens.

• Payment (if target value reached) = private wealth

  Payment (if target value not reached) = private wealth – loss

2.5 Announcements

Twice during the experiment, at the end of round 3 and at the end of round 7, you will have the opportunity to tell the other group members how many tokens you are planning to invest in the remaining rounds in the group project. This announcement is not binding, i.e. you are not required to actually invest the announced amount. Your announcement and other members' announcements will be shown to all group members before the next round begins.

2.6 End of the Experiment

At the end of the experiment your payout will be converted at an exchange rate of 1 token = 0.25 Euro. Before you receive your payment, we will ask you to complete a questionnaire. You will receive an extra payment of 5 Euro for answering these questions. Please take your time to answer the questionnaire. The payment will be made only after all participants have completed the questionnaire.

After having read and understood these explanations and the summary on the next page, please press NEXT. Please raise your hand if you have questions.

2.7 Summary

The experiment is played in groups of 4 players (2 purple players, 2 orange players) and consists of 10 rounds.

	Purple player	orange player
Endowment	40 tokens	40 tokens
Round budget per round	5 tokens	3 tokens
Total round budget	50 tokens	30 tokens
Loss if target value not reached	20 tokens	40 tokens

• Investments: In each round you can invest tokens from the round budget in a group project. Tokens that you do not invest in the group project will be added to your private wealth.

• Private wealth: initial endowment + tokens from the round budgets you did not invest in the group project

• Target value:80 tokens

• Payment: Payment (if target value reached within 10 rounds) = Private wealth.

• Payment (if target value not reached within 10 rounds) = Private wealth–loss.

Appendix 3: Descriptive Statistics

Table 5 provides an overview of basic descriptive statistics for the variables sex, age, cognitive abilities, risk preferences, positive reciprocity, negative reciprocity and political attitude. Randomization across treatments with respect to these variables was successful (Wilcoxon rank sum tests, p > 0.1 for all treatments and variables).

Table 5 Descriptive statistics

Risk preferences were elicited by providing subjects with five interdependent hypothetical choices between a lottery and a safe option. I followed the more time-efficient “staircase” procedure suggested by Falk et al. (2016), which results in a measure for risk preferences on a scale from 0 to 31, with higher values indicating more risk-lovingness. Cognitive abilities were elicited using a test by Toplak et al. (2014) consisting of four questions, with more questions answered correctly indicating higher cognitive abilities. Inspired by Dohmen et al. (2009), the measures for positive and negative reciprocity are based on an individual’s agreement with a statement on a scale from 0 (‘does not apply at all’) to 10 (‘totally applies’). The statements for positive reciprocity cover willingness to return a favor, the one for negative reciprocity the willingness to take revenge for a serious wrong. To measure political attitude, subjects were asked to describe their political position on a scale from 1 (right) to 10 (left).

Appendix 4: Determinants of Individual Contribution Behavior

The OLS regression model depicted in Table 6 confirms the results presented in Sect. 5.4. In successful groups (model (1)) the coefficients for more vulnerable types in VUL, rich types in WLT and rich/less vulnerable types in WLT-VUL are positive and significant, indicating higher contributions of these types relative to subjects in BASE. The coefficients of the other types are negative and also significant. Subjects’ ex-ante preferences regarding their own type have a positive and significant effect on individual contributions.

Table 6 OLS regression—determinants of individual contribution behavior

In failing groups (model (3)) the contributions of all types deviate negatively from BASE (remember that in the case of failing groups contributions were the highest in BASE). Note also the sharp drop in the value of R² from 0.552 in model (1) to 0.149 in model (3), which emphasizes the strong differences in the contribution behavior of types, and thus their explanatory power in successful groups. Ex-ante preferences have no significant effect on individual contributions in failing groups.

Models (2) and (4) include several variables from the post-experimental survey that might affect individual contribution behavior: risk preferences, cognitive abilities, preferences for positive and negative reciprocity, sex and political attitude. There is, however, no evidence that any of these variables affects individual contributions in successful groups. This is also true for failing groups with the exception of positive reciprocity. An additional regression using first round contributions as dependent variable also finds no significant effect of these variables.

Appendix 5: Beliefs

This section complements the analysis provided in Sect. 5.5 with additional results on the role of subjects’ beliefs about group contribution and threshold achievement.

5.1 Beliefs About Group Contribution

From the literature on conditional cooperation we know that a large fraction of people makes their contributions contingent on the contributions of others or their expectations regarding others’ contributions (Fischbacher et al. 2001; Gächter 2006). The results of the panel model presented in Sect. 5.5 show that in this experiment subjects also contribute conditionally. In addition, there is a strong, positive and significant correlation between subjects’ beliefs about group contribution and individual contributions in the respective round (Spearman’s r = 0.505, p = 0.000).

Subjects’ beliefs about group contribution tend to be quite accurate, although there is a significant difference between beliefs and actual contributions (Wilcoxon signed rank test, p = 0.000). More precisely, beliefs were on average equal to 7.05 tokens per round (sd = 3.29) compared to an average actual group contribution of 7.49 tokens per round (sd = 3.01). Figure 6 shows how group contributions and the respective beliefs evolved over time in successful and failing groups. Note that beliefs in the first round were not significantly higher in successful groups (8.28 tokens) than in failing groups (7.93 tokens; Wilcoxon rank-sum test, p = 0.121). However, group contributions in the first round were significantly larger in successful groups (8.91 tokens) than in failing groups (6.89 tokens) (Wilcoxon rank-sum test, p = 0.000). Remember that groups needed an average contribution of 8 tokens per round to reach the threshold by round 10. While only 4 out of 33 successful groups (12.1%) contributed less than the required amount in the first round, this share was significantly larger in failing groups, where 5 out of 9 groups (55.5%) failed to contribute the required sum in the first round. This can be considered evidence for a signaling function of the first round of contributions since they demonstrate subjects’ willingness to contribute. Figure 6 also shows that groups did a good job of updating their beliefs. The correlation between group contribution in $t-1$ and subjects’ beliefs about group contributions in $t$ is strong and highly significant (Spearman’s r = 0.493, p = 0.000).

Fig. 6

Group contribution and beliefs about group contribution, by group success

5.2 Beliefs About Threshold Achievement

The results of the panel model presented in Sect. 5.5 show that subjects were hesitant in adjusting their beliefs about the likelihood of reaching the threshold unless it was definitely out of reach. In that case, subjects adjusted their beliefs and contributions immediately and drastically. The following complements this analysis.

Initially, subjects were in general very optimistic about the threshold being reached. On a scale from 1 (very unlikely) to 7 (very likely), subjects assessed the likelihood on average with 5.23 behind a veil of ignorance and with 5.19 in round 1 after being informed about their type. Notably, members of successful groups were significantly more optimistic in the first round about threshold achievement than members of failing groups (5.29 vs. 4.83; Wilcoxon rank-sum test, p = 0.018). Also, final treatment differences in success rates were already reflected in initial beliefs: Subjects were most optimistic about reaching the threshold in WLT-VUL (5.4 behind the veil of ignorance, 5.45 in round 1) and least optimistic in VUL (4.89 both behind the veil of ignorance and in round 1). The difference is weakly significant (Wilcoxon rank-sum tests, p = 0.016 behind the veil and p = 0.013 for round 1) for these two treatments and insignificant for all other treatment comparisons.

These results lend support to the hypothesis that threshold beliefs affect group success, even though subjects do not adjust their beliefs continually between rounds. Instead, initial beliefs seem to play a crucial role, in particular because they are correlated with first round contributions, which in turn have a signaling effect as has been shown in the first part of this section.