Heterogeneity and cooperation in climate change mitigation: An experimental approach to common-pool resource games

Abstract

The collective action problem associated with CO₂ reduction presents a global challenge that requires cooperation among actors. Two factors are often argued to hinder successful cooperation in this context: endowment heterogeneity and the time delay between environmentally harmful actions and their consequences. This study examines the effects of endowment heterogeneity and delayed feedback on cooperation in common-pool resource management at the individual and group levels. Using a Common-Pool Resource (CPR) game in a laboratory experiment, we simulated CO₂ emission dynamics under various conditions of heterogeneity and time-varying feedback. The results highlight that cooperation in common-pool resource dilemmas is shaped through complex micro-macro interdependence. Endowment heterogeneity primarily affects the internal distribution of behavior within groups rather than aggregate outcomes. Low-endowment participants tend to emit less, while high-endowment participants tend to emit more, indicating that endowment heterogeneity affects emissions in opposite directions across group members, largely offsetting each other at the group level. This suggests that heterogeneity may redistribute contributions to mitigation efforts within groups. In contrast, delayed feedback primarily operates by influencing how individuals process and respond to social information. Feedback timing affects how participants react to prior group behavior and the informational conditions under which conditional cooperation is expressed. At the group level, this translates into a modest but detectable effect, suggesting that feedback structure can influence collective outcomes even when individual-level effects are limited. This implies that the determinants of cooperation in CPR settings cannot be fully understood by looking only at either the micro or the macro level.

Introduction

In recent decades, climate change has emerged as an increasingly urgent global issue. Human activities, particularly the emission of greenhouse gases like carbon dioxide (CO₂), are the primary drivers of climate change, resulting in severe consequences such as prolonged droughts, increased water scarcity, and devastating wildfires. These impacts extend to public health, food security, housing, safety, and employment (United Nations, nd). Within sociological research, the study of climate change has garnered significant attention due to its profound social implications for communities worldwide (Billi et al., 2019; Klinenberg et al., 2020; Lorenz, 2013).

To effectively mitigate the threat of climate change, cooperation among various actors across societies is required to achieve substantial reductions in CO₂ emissions (EEA, 2023). However, climate change mitigation presents a classic collective action problem, as it involves a social dilemma related to CO₂ emission reduction: while lowering global emissions benefits collective well-being, it often incurs immediate costs for individuals (Gillingham and Stock, 2018). This tension between individual and collective interests can lead to a ‘tragedy of the commons’ (Hardin, 1968), where individual actors are incentivized to overexploit a shared resource, resulting in its degradation and negative consequences for all. In the context of climate change, this means that although low or moderate emission levels would enhance collective well-being, it is often more profitable for individual actors to maintain or increase their own emissions and “free-ride” on the efforts of others.

Experimental research has approached the climate collective action problem using different paradigms that differ in their underlying incentive structure. Two central frameworks are common-pool resource (CPR) games (Ostrom, 1990) and the collective-risk social dilemma (CRSD) (Milinski et al., 2008). While CPR games model climate change as an appropriation problem, where individual emissions cumulatively deplete a shared carbon sink over time, the CRSD models it as a provision problem in which groups must contribute sufficient resources to reach a collective threshold and avoid a probabilistic climate catastrophe. Unlike CRSD designs, where risk is triggered by failing to reach a collective threshold, CPR games generate risk endogenously through cumulative over-extraction. Both paradigms capture key aspects of climate mitigation: CRSD experiments emphasize coordination under collective risk in threshold public good settings, whereas CPR games highlight cumulative emissions, dynamic feedback, and the gradual erosion of common resource through overuse. In this study, we focus on the resource depletion dynamics central to CPR games.

Contrary to standard predictions that self-interested actors will overexploit shared resources, research demonstrates that groups can sustain cooperation under certain institutional and structural conditions (Ostrom, 1990). A central theoretical question, therefore, is not whether cooperation is possible, but which structural conditions of a dilemma affect the stability of cooperative behavior. Two such structural features are particularly relevant in the context of climate change mitigation: endowment heterogeneity and delayed feedback on consequences of human action.

First, endowment heterogeneity reflects persistent global inequalities in income and emission capacities. From a theoretical perspective, heterogeneity can alter benefits and the distribution of costs. Actors with larger endowments face lower relative costs of overuse and may perceive mitigation burdens as disproportionately constraining, whereas actors with smaller endowments may view restraint as unfair given unequal historical contributions. Such inequalities shape actors’ responsibilities, vulnerabilities, and fairness perceptions, and may hinder cooperation in efforts to reduce emissions (Tavoni et al., 2011). Empirically, global emissions are highly concentrated: the wealthiest 10% are responsible for nearly 48% of global CO₂ emissions, while the poorest 50% contribute only 12% (Chancel, 2022). At the same time, those with fewer resources are more vulnerable to the adverse effects of climate change (EPA, 2021). In experimental terms, endowment heterogeneity translates into unequal emission capacities and unequal stakes in the collective outcome.

Second, delayed feedback on consequences of human action capture the temporal structure of climate change. Emissions today generate environmental damages only in the future, weakening feedback, learning, and perceived efficacy (Fennewald and Kievit-Kylar, 2014; Weber, 2006). This delay between action and consequence is built into the climate problem, because the effects of CO₂ build up slowly and often occur long after individual decisions are made. In CPR framework, such temporal delay reduces the salience of environmental deterioration, weakens learning, and increases incentives for short-term appropriation relative to long-term sustainability.

These two factors are theoretically central because they directly reshape incentive structure in CPR dilemmas: heterogeneity alters the distribution of costs and benefits across actors, while delay alters the temporal alignment between action and consequence. Unlike uncertainty, which affects beliefs about outcomes, both factors modify the underlying payoff structure itself. Understanding how they interact is therefore essential for explaining variation in cooperative behavior in climate-related common-pool settings.

In existing literature, endowment heterogeneity has been identified as one of the most significant determinants of cooperation in collective-action problem games (Suchon and Théroude, 2022). However, findings from previous studies in this domain remain contradictory. While some studies suggest that endowment heterogeneity hinders cooperation in resource management (Andersson and Agrawal, 2011), others did not find a significant impact of heterogeneity in CPR games (Momeni, 2021; van Klingeren, 2020). Along similar lines, experimental studies using the CRSD under resource heterogeneity have produced mixed results (Waichman et al., 2021). The role of delayed feedback remains debated as well, with some research indicating that delayed feedback undermines cooperation (Fennewald and Kievit-Kylar, 2014), while other studies fail to confirm this effect. Thus, a clear gap remains in understanding how these two factors shape cooperative behavior in climate change mitigation efforts.

Thus, investigating these dynamics in an experimental setting can provide valuable insights, offering not only a controlled environment but also precise observations of participants’ behavior and decision-making processes (Stoop et al., 2018) under varying levels of heterogeneity and delayed feedbacks. Such experiments allow researchers to disentangle the underlying mechanisms that affect cooperation in the context of climate change mitigation.

Common-pool resource (CPR) games are a valuable tool for studying the collective action problem inherent in climate change mitigation within an experimental framework (Stoop et al., 2018). In CPR games, participants withdraw units from a shared resource, which can be used sustainably as long as the rate of withdrawal does not exceed the resource’s natural regeneration rate (Ostrom, 1990). These games are well-suited for simulating collective CO₂ emission scenarios, as they reflect both the non-excludability of emission reductions (everyone benefits from lower emissions) and the rivalry involved (one person’s emissions can diminish the ‘pool’ available for others). Additionally, CPR games typically incorporate collective mechanisms that mirror real-world scenarios, such as the negative consequences of climate change, which can motivate individuals to cooperate in order to avoid detrimental outcomes.

This study therefore examines how endowment heterogeneity and delayed feedback shape cooperative behavior in repeated CPR games that simulate CO₂ reduction scenarios to address the following research questions: How does endowment heterogeneity and delayed feedback on consequences impact cooperation in repeated CPR games related to climate change mitigation?

To explore this question, we conducted an experiment using CPR games designed to simulate the collective action problem of CO₂ reduction. These games reflected unequal emissions resulting from income disparities and the consequences of depleting resources, mirroring the Earth’s capacity to absorb CO₂. The experiment consisted of two types of games: one in which players had equal endowments to invest in activities that generate profit but also cause CO₂ emissions (homogeneity), and another in which players had unequal endowments (heterogeneity). To test the effect of delayed feedback, half of the groups received delayed feedback on the environmental impacts of their actions, while the other half received immediate feedback.

Theoretical framework

In this section, we develop a theoretical framework grounded in the existing literature to explore how the social dynamics underlying climate change mitigation are shaped by endowment heterogeneity and delayed feedback regarding the consequences of collective behavior. We examine how these features characterize real-world climate dilemmas and influence cooperative behaviour. We further connect this framework to the literature on cooperation in collective action games by discussing the theoretical and empirical implications of endowment heterogeneity and delayed feedback for cooperative behavior in CPR settings.

Rather than providing a full formal game-theoretic analysis of the finitely repeated CPR game introduced below, we focus on the behavioral mechanisms that are most relevant for explaining cooperation in such environments. A purely equilibrium-based analysis would primarily generate predictions based on selfish and forward-looking incentives, implying strategic overexploitation of the resource over time. However, extensive experimental research has shown that cooperative behavior in CPR games is often more strongly shaped by fairness considerations, conditional cooperation, and participants’ expectations about the behavior of others. We therefore derive our hypotheses primarily from these behavioral mechanisms. This approach follows recent work by van Klingeren and Buskens (2024), who adopt a similar perspective and provide empirical support for its relevance. Because these behavioral mechanisms may operate in opposing directions, one of the treatments generates competing theoretical predictions.

Endowment heterogeneity and cooperation

The social framework surrounding climate change is characterized by a distinct form of heterogeneity, as CO₂ emissions are deeply intertwined with economic inequalities (e.g., Mi et al., 2020). Wealthier actors emit more CO₂, while poorer actors are more vulnerable to the consequences of climate change (Chancel, 2022; EPA, 2021).

Economic inequalities that influence both emissions and vulnerability to climate change can hinder cooperation among actors, obstructing efforts to reduce CO₂ emissions (Tavoni et al., 2011). Economic inequality has been identified as one of the most important factors influencing cooperation levels in collective-action problem games (Suchon and Théroude, 2022). In these games, economic inequality is typically modelled through unequal endowments among players, reflecting real-world economic disparities. In this research, these unequal endowments simulate the differing opportunities for CO₂ emissions among players, based on their income disparities.

In theory, there are two competing lines of thought regarding the impact of heterogeneity on cooperation in collective-action problem games, each predicting either a positive or negative impact of endowment heterogeneity on collective action. The first line argues that endowment heterogeneity can benefit collective action if those with the most economic power and interests take the lead (Olson, 1965). In this view, high-endowment individuals are expected to willingly contribute more to a shared goal, while low-endowment individuals can rely on the contributions of wealthier ones (Baland and Platteau, 1999). According to this theory, the success of CO₂ emission reduction efforts depends on the actions of the wealthy.

However, this idea—that resourceful actors act as catalysts for cooperation by bearing the cost of collective action—requires that these players exhibit a sense of fairness. For example, Fehr and Schmidt’s (1999) model of inequality aversion suggests that an individual’s utility is tied to the equality of payoffs among all participants. Individuals tend to dislike receiving a payoff that is significantly different from others. Based on this model, inequality aversion in unequal CPR games would predict that high-endowment players would consume proportionally less of the common-pool resource than low-endowment players to reduce inequality.

The second line of thought argues that economic inequality, in the form of unequal endowments, can hinder the development of cooperative behavior in CPR games. Indeed, most theoretical research on the effects of endowment heterogeneity on cooperation in CPR games suggests that inequality not only leads to disparate contributions and uneven rewards but also results in a wide variation in individual interests in resource management, which ultimately reduces the likelihood of cooperation (Adhikari and Lovett, 2006; van Klingeren, 2020). This diversification of interests due to endowment heterogeneity can be explained through Tajfel and Turner’s (1979) Social Identity Theory. According to this theory, individuals who perceive themselves as similar to others in their group are more likely to identify as in-group members. This sense of shared identity fosters a stronger orientation towards group interests and, consequently, increases cooperation. In contrast, in heterogeneous groups, individuals may perceive a weaker group identity, which can reduce their focus on collective interests and lead to lower levels of cooperation (Baland and Platteau, 1999).

Further, in the context of climate-related CPR games, endowment inequality translates into unequal CO₂ emission opportunities and unequal economic stakes in the shared resource. Individuals with higher endowments have greater extraction capacities and may experience the costs of restraint differently from those with lower endowments. This asymmetry can lead to diverging incentives regarding resource preservation. For instance, individuals with fewer resources may depend more strongly on the continued availability of the shared resource relative to their endowment, while wealthier individuals may be better able to absorb losses. Such differences in relative stakes and consequent divergence in interests may foster competitive rather than cooperative behavior.

Moreover, Reuben and Riedl (2013) argue that ‘normative disagreements’ can arise between unequal players, as high- and low-endowment individuals may hold different beliefs about what constitutes fair behavior. These conflicting norms and interests can make it challenging to achieve consensus (Hofmeyr et al., 2007), further increasing transaction costs and hindering cooperation. However, research on this issue is not conclusive. In contrast, Otten et al. (2020) found that normative disagreements do not necessarily impede cooperation.

These two argumentations offer competing hypotheses on the role of endowment heterogeneity in collective action, predicting either a positive or a negative effect on cooperation.

Empirical evidence on economic inequality in CPR settings is predominantly derived from case studies. For instance, Andersson and Agrawal’s (2011) meta-analysis of 228 qualitative case studies across three continents found that socioeconomic inequalities have a negative impact on conservation efforts when institutions are weak or absent. However, laboratory experiments that directly examine the effect of endowment heterogeneity on cooperation in CPR games are scarce, and the findings do not show a consistent pattern. For example, van Klingeren’s (2020) study, which explored the effects of endowment and sociocultural heterogeneity in combination with trust, found no effect of either type of heterogeneity on cooperation, though it did identify a negative effect of endowment heterogeneity on trust. Similarly, Momeni’s (2021) research on voluntary and mandatory provision of CPRs among heterogeneous users found no effect of heterogeneity on voluntary extraction choices within CPR games.

Empirical findings do not clearly support one or the other prediction of the lines of thought explained above. Therefore, drawing on both theoretical perspectives, we propose to test both competing hypotheses against each other:

H1a

Endowment heterogeneity will positively affect cooperation.

H1b

Endowment heterogeneity will negatively affect cooperation.

Given the competing theoretical arguments and the inconsistent empirical findings, our investigation of endowment heterogeneity is theory-informed but not derived from a single unified framework. The experimental approach, therefore, serves to explore which of these competing expectations receives empirical support in a controlled setting.

Delayed feedback and cooperation

In repeated CPR games, players’ behavior is often influenced by information about prior resource (over-)use, realized payoffs, and the observed actions of their co-players. As in many other cooperation dilemmas, such adaptive behavior based on past experiences and social interactions is closely related to the concept of conditional cooperation (Fischbacher et al., 2001; Fischbacher and Gächter, 2010). Conditional cooperators are willing to reduce their own resource use if others do so, but increase extraction when others free-ride. In a CPR game that simulates climate change mitigation, conditional cooperation implies that participants lower their CO₂ emissions when the majority of others did so previously.

In our experimental setting, individual emission decisions jointly determine the aggregate emission level, which in turn affects the profitability of emissions for all group members. Monetary payoffs are realized immediately in every round. However, in the delayed feedback treatment, information about the group’s aggregate emissions and the resulting profit rate is provided only after a time lag. The treatment therefore does not postpone material consequences in terms of payoffs; rather, it delays information about how collective behavior affects the shared resource environment.

To facilitate conditional cooperation, timely feedback is crucial for effective behavioral adjustment. When feedback is delayed, the link between actions and their collective outcomes becomes less salient. Participants receive less immediate information about whether current emission levels are sustainable or contribute to overexploitation. This weakens their ability to adapt behavior based on observed environmental responses.

Delayed feedback may further increase uncertainty about the current environmental state and its development, as well as the effectiveness of individual contributions. Compared to immediate information, postponed feedback makes it more difficult for participants to assess whether reduced emissions are sufficient to prevent resource depletion. Such uncertainty can undermine perceived behavioral efficacy, which is the belief that one’s actions meaningfully contribute to collective outcomes. Lower perceived efficacy has been shown to reduce cooperative engagement and may also serve as a convenient justification for non-cooperation (Kerr, 1992; van Lange and Rand, 2022).

Moreover, when participants lack timely information, expectations about others’ behavior become more speculative. If individuals are conditional cooperators, cooperation may either increase or decrease depending on whether they hold overly optimistic or pessimistic beliefs about their co-players’ emissions. This effect is likely to be particularly pronounced in early rounds, when participants have not yet accumulated sufficient experience to form stable expectations.

Finally, the delayed feedback manipulation relates to the psychological mechanism of temporal discounting (Ainslie, 1991). Temporal discounting refers to the tendency to assign lower weight to delayed outcomes relative to immediate ones. Although monetary payoffs in our experiment are realized immediately, postponing information about the environmental impact of emissions may create a subjective sense of temporal distance between actions and their broader consequences. When the connection between present emissions and cumulative environmental deterioration becomes less immediate and salient, participants may implicitly prioritize short-term profits over long-term collective sustainability. In this sense, delayed feedback may generate a psychological dynamic similar to discounting, even though material outcomes themselves are not temporally delayed.

The manipulation of delayed feedback captures an important aspect of climate change. In the real world, there is a substantial temporal gap between CO₂ emissions and many of their environmental and societal consequences. Even after emissions are reduced, global temperatures may continue to rise for an extended period (Fennewald and Kievit-Kylar, 2014). As Weber (2006: 103) notes, “The time-delayed, abstract, and often statistical nature of the risks of global warming does not evoke strong visceral reactions.” The gradual and indirect nature of climate impacts weakens the perceived connection between present actions and future environmental outcomes. By postponing information about aggregate emissions and profitability, our delayed-feedback treatment mirrors this weakened experiential link between individual behavior and collective environmental consequences.

While theoretical arguments suggest that delayed feedback should impair cooperation in CPR settings, empirical evidence remains scarce. Fennewald and Kievit-Kylar (2014) provide initial support, showing that delayed effects in climate-related CPR games increase players’ confusion and the likelihood of environmental degradation. Building on this reasoning, we hypothesize:

H2

Delayed feedback will increase the likelihood of overexploitation of the CPR compared to immediate feedback.

Finally, some of the mechanisms associated with endowment heterogeneity and delayed feedback may interact. For example, if participants disagree about the appropriate contribution levels under heterogeneous endowments, such normative disagreements may be more difficult to resolve when information about others’ behavior is delayed. Similarly, adjusting one’s behavior to expectations about the actions of others becomes more difficult under delayed feedback conditions. At the same time, the theoretical arguments regarding the effects of heterogeneity point in different directions. We therefore do not formulate a specific hypothesis concerning the interaction effect, but instead explore it empirically.

The experiment, methods and measurements

In the following sections, we first introduce the experimental setting. This includes describing the structures of the CPR game, the operationalization of emissions and consequences, and introducing the experimental conditions. As the CPR game is designed to simulate CO₂ emissions, we employ emission-related terminology throughout. Next, we provide an overview of the data collection process, covering participant recruitment, assignment to conditions, and data collection procedures. Additionally, we outline the measurements used in the analysis, encompassing both dependent and independent variables. Finally, we present the analytical strategy for testing our hypotheses, which includes descriptive analyses as well as the application of one-level and cross-level models.

Game design and variable operationalization

In this experiment, CPR games simulate an imaginary scenario for participants, aiming to reflect the real-world social framework of CO₂ emissions. Participants assume the role of Earth’s inhabitants, whose actions contribute to CO₂ emissions that affect the planet’s atmospheric CO₂ level. This CO₂ level serve as the common resource in the CPR game and must be collectively managed by the group. In the game, the emission actions of Earth’s inhabitants are translated into points that participants can decide to invest each round. The benefits gained from these emissions are represented by profit points awarded to participants based on their emission decisions. However, if participants collectively emit an excessive amount, the atmosphere cannot absorb the emissions, resulting in a subsequent rise in the atmospheric CO₂ level. This increase leads to negative consequences for all participants, as their profits relative to invested points decrease in the following round. Conversely, limiting emissions can yield positive outcomes; previously reduced profit rates resulting from excess emissions can be restored by collectively reducing emissions. Participants are able to observe these mechanisms and gain insights into the consequences of their actions through feedback on their own and their co-players’ emission behavior, as well as the subsequent impact on Earth’s atmosphere and profit rates.

Emissions and profit calculation

In each round, participants decide how much CO₂ to emit into the atmosphere by investing the points they receive at the beginning of that round. Each point invested corresponds to the emission of 1 ton of CO₂. Participants profit from their emissions based on the number of points used, multiplied by their group’s profit rate for that round, which is set at 1.4 in the first round of each game, yielding individual profits. The profit rate may vary in response to the collective emissions from previous rounds.

The total points that participants earn across all rounds are calculated as the sum of points used for emissions, multiplied by their group’s profit rate, plus the sum of unused points throughout all rounds:

$U_i={\sum }_{t=1}^T\left(e_{i,t}\mathbf{·}\hspace{0.17em}{r}_{g,t}+d_{i,t}\right)$

(1)

where

• $U_i$ is the total profit of participant $i$ in the game,

• $T$ is 15, the total number of rounds in the game,

• $e_{i,t}$ is the points used for emissions by participant $i$ in round $t$ ,

• $r_{g,t}$ is the profit rate for group $g$ , where i is a member of in round $t$ ,

• d _i,t is the points that have not been used for emissions by participant $i$ in round $t$ .

Emission endowment and decision stage

In each round of the game, the four participants can choose their CO₂ emissions. In the homogenous game, all participants have an endowment of 50 points to emit in each round $t$ . In the heterogeneous game, two participants have an endowment of 30 points, while the other two have 70 points. The emission decision (0 ≤ $e_{i,t}$ ≤ 30, 50 or 70) represents the emissions of participant $i$ at time $t$ .

R _t represents the atmospheric CO₂ level in round $t$ . In the next round, atmospheric CO₂ level, $R_{t+1}$ , is partly determined by adding the group`s collective emissions from the current round ${\sum }_{i=1}^4{e}_{it}$ to the current round’s atmospheric CO₂ level. Additionally, in each round, Earth’s atmosphere can absorb 100 tons of CO₂. Therefore, if all four participants collectively emit more than 100 tons of CO₂ in a round, the atmospheric CO₂ level in the next round R_t+1 will increase. However, if total emissions are less than 100 tons, the atmospheric CO₂ level can drop, but it will never drop below the sustainable threshold of 1000 tons: $R_{t+1}=\min \left(R_t+{\sum }_{i=1}^4{e}_{i,t}-100,1000\right)$ .

Additionally, the profit rate $r_g$ for the next round is influenced by the resulting CO₂ level in the atmosphere:

$r_{g,t+1}=\frac{1000}{R_{t+1\hspace{0.17em}}}\hspace{0.17em}\mathbf{·}1.4$

(2)

Equation (2) captures the assumption that investment returns decline as atmospheric CO₂ accumulates. The functional form translates higher resource depletion into lower future profit rates. While climate change may generate heterogeneous effects across sectors, creating both losses and new investment opportunities, we abstract from this heterogeneity and model the net effect of environmental degradation as reducing the profitability of productive investment. This simplification allows us to focus on how delayed feedback shapes strategic incentives in the common-pool setting.

Overemission

Participants are considered to have collectively overemitted if their combined emissions exceed 100 tons of CO₂. If participants collectively emit 100 tons in a round, this results in an average emission of 25 tons of CO₂ per person, representing the sustainable emission level for the group. The maximum amount of CO₂ that participants can emit without surpassing the sustainable amount is always 25—regardless of the existing atmospheric CO₂ level—due to the decreasing profit rate associated with rising CO₂ concentrations. Therefore, emitting more than 25 tons is unsustainable and considered uncooperative behavior.

As long as R _t is below 1400, participants have a short-term incentive to emit as much as possible since the profit rate remains above 1. This means that the returns from emissions exceed those from refraining from emissions. However, because the game extends over multiple rounds, collective overemissions are only more profitable than collective sustainable emissions for the first few rounds. For instance, with a group emission of 150 tons, collective profits are only higher than that of sustainable group emissions of 100 tons for the first 8 rounds. After round nine, collective profits fall below what it could have been if all participants had emitted sustainably. Conversely, if one participant overemits while others underemit, keeping the collective emissions at or below 100 tons, overemitting becomes more profitable for that participant. This phenomenon is referred to as free-riding. Moreover, when overall emissions are less than 100 tons, the atmospheric CO₂ level gradually regenerates to the sustainable threshold of 1000 tons, as long as they are not already at that threshold.

Feedback

After determining their emissions, participants receive feedback on their own and their co-players’ endowments, emissions, profits, the atmospheric CO₂ level, and the profit rate for the next round. Each game consists of a total of 15 rounds. Depending on whether participants are playing a game with immediate or delayed feedback, they receive this information either after each round of emission decisions or only after the 3rd, 6th, 9th, 12th and 15th round. Participants in the immediate feedback condition can track the consequences of their own and their co-players’ behavior more frequently. While profit rates for participants with delayed feedback can still change in every round, as previously explained, they are unable to track these changes continuously. Participants are randomly assigned to either feedback condition.

Data collection

A computerized laboratory experiment was designed and programmed using o-Tree (Chen and Wickens, 2016). The experiment was conducted from February to March 2024. Written consent was obtained from all participants before the start of each experimental session, and the data were anonymized prior to analysis.

Participants were recruited from the Online Recruitment System for Economic Experiments [ORSEE] (Greiner, 2015). The experiment comprised 10 sessions, involving a total of 192 participants. Half of the participants were randomly assigned to the delayed feedback condition, while the other half played CPR games with immediate feedback. In contrast to this between-subject condition assignment, the heterogeneity condition was assigned within participants; all participants played one CPR game with heterogeneous endowments and another game with homogenous ones. Half of the participants began with the homogeneous CPR game, while the other half started with the heterogeneous one to minimize potential biases arising from game order or learning effects.

At the beginning of the experimental session, participants were provided with written instructions in English and completed a preliminary test-round. They were then randomly assigned to groups of four to participate in the first CPR game, spanning 15 rounds. Subsequently, they were randomly reassigned into new groups of four to participate in a second CPR game, also spanning 15 rounds. Participants were not informed of the identities of the other three players in their groups nor the other experimental conditions. The experiment lasted approximately 75 min, during which participants played for real money (EUR) under an exchange rate of 110 points = 1 EUR (in the first two sessions, the exchange rate was 100 points = 1 EUR). This resulted in an average earning of €16.50 per participant.

At the end of the experiment, participants were asked to complete a survey. This survey included questions about their attitudes toward climate change, political orientation, views on economic inequality, and self-report items measuring their level of prosociality, developed by Caprara et al. (2005). Additionally, participants provided demographic information such as age, gender, nationality, field of study, prior experience with similar experiments, how many other participants in the session they knew, and their perception of the experiment.

Measurements

For the analysis, we used two variables to measure cooperation levels at both the group and individual levels. The experimental conditions serve as our main independent variables. Additionally, we included several control variables to account for potential confounding game characteristics, individual attitudes, and personal characteristics.

Dependent variables

Total CO₂ emissions

Total CO₂ emissions represent the cumulative collective emissions of the group in one game over all rounds, reflecting cooperation at the group level. Lower collective emissions indicate a higher level of cooperation, as higher emissions contribute to an increasingly unsustainable CO₂ level in the atmosphere. Specifically, group emissions of up to 100 tons can be considered cooperative, as they do not significantly elevate the atmospheric CO₂ level in the subsequent round.

Individual CO₂ emissions

Individual CO₂ emissions refer to each participant’s emission decisions in every round, reflecting cooperation at the individual level. Lower individual emissions indicate higher levels of cooperation. If each participant emits a maximum of 25 tons per round, the atmospheric CO₂ level in the next round will not increase. Therefore, individual emissions of up to 25 tons can be considered cooperative, as they do not diminish the resource size in the following round.

Independent variables

The independent variables in this study are derived from the experimental conditions. We constructed two treatment indicators and their interaction:

Heterogeneity

Coded as 0 for homogeneous endowments and 1 for heterogeneous endowments.

Delayed feedback

Coded as 0 for immediate feedback and 1 for delayed feedback.

Heterogeneity * feedback

Interaction term coded as the product of the two treatment indicators. Although we do not formulate a specific hypothesis regarding this interaction, we include it to exploratorily assess whether the effect of delayed feedback depends on the level of endowment heterogeneity. Methodologically, including the interaction term is important because omitting potentially relevant interactions may bias or distort the estimated main effects (see Muralidharan et al., 2025; Singh 2026: chapter 4).

Control variables

Several variables are included in our models to control for game characteristics as well as participants’ demographic characteristics, attitudes, and political orientations, all of which have been shown to significantly influence climate change awareness (Fraembs and Drobnič, 2024):

High endowment

Coded as 1 if participants in heterogeneous treatment groups were assigned a high endowment of 70 points. Coded as 0 if they were in homogeneous treatment groups (endowment of 50 points for each participant) or in heterogeneous treatment groups with a low endowment of 30 points.

Homogeneous first

Coded as 1 if participants started with the homogeneous condition and 0 otherwise. This variable helps to isolate the effects of starting conditions on behavior.

Second game

Coded as 1 for participants in their second game and 0 for those in their first game, to account for potential learning effects over the two games.

Group emission > 100

Coded as 1 if the group emitted more than 100 tons of CO₂ in the previous round, otherwise coded as 0. This variable considers the impact of past group behavior on individual emissions.

Player previous emission

This variable reflects the individual emission decision made by a participant in the previous round, controlling for the effect of past individual behavior on current emission decisions.

Feedback round

Coded as 1 for the rounds immediately following a feedback round—specifically, for participants in the immediate feedback conditions in all rounds, and for those in delayed feedback conditions after the 3rd, 6th, 9th, and 12th rounds of the game. Otherwise, it is coded as 0. This variable is intended to contribute to the examination of the feedback effect in delayed treatment groups.

Political right

Measured on a Likert Scale from 1 to 10 (1 = Extremely Left, 10 = Extremely Right).

Prosociality

Measured on a Likert Scale from 1 to 5, with 5 indicating high prosocial values. This variable was constructed by calculating the average score of 16 items that assessed participants’ prosocial values.

Belief in limiting emissions

Assessed how likely participants believe it is that large numbers of individuals will limit their CO₂ emissions in the face of climate change (0 = Not at all likely, 10 = Extremely likely).

Male

Coded as 1 for male participants and 0 otherwise.

Dutch

Coded as 1 for Dutch participants and 0 otherwise.

Unknown co-players

Coded as 1 for participants who did not know any of the other participants prior to the experimental session.

Experiment experience

Coded as 1 for participants who have previously participated in a similar experiment and 0 otherwise.

Understanding task

Measured on a Likert Scale from 1 to 5 to capture participants’ agreement on whether they understood their tasks during the experiment (1 = Completely disagree; 5 = Completely agree).

Gr. Em. > 100 * feedback round

An interaction between the variables Group Emission > 100 and Feedback Round. This interaction variable captures the effect of past behavior. It is coded as 1 in cases where, in the previous round, the participant’s group exceeded 100 tons of CO₂ emissions and received feedback about it, and 0 otherwise.

Delayed * feedback round

An interaction between the variables Delayed Feedback and Feedback Round.

This variable is coded as 1 for participants in the delayed feedback treatment groups if they received feedback in the previous round, and 0 otherwise.

Analytical strategy

To test our hypotheses, we performed descriptive analyses and ran regression models, including linear regression and cross-level multi-level models, to examine the effect of experimental conditions on CO₂ emissions at both the individual and group levels. First, the descriptive analysis involves examining total CO₂ emissions at the group level across all rounds under different experimental conditions. Next, we conducted OLS regression analysis on total CO₂ emissions at the group level to assess the impact of the conditions on group behavior. Given the limited overlap between groups and the limited number of sessions, we disregard potential interdependencies between groups arising from overlapping participant membership or co-occurrence within sessions.

To investigate emission behavior at the individual level, we used cross-level models on individual CO₂ emissions, where individual emission decisions in each round are modeled at the first level. Participants and groups are crossed at the higher level, since each group consists of four participants, and all participants take part in two groups. The models incorporate random effects at both the participant and group levels, addressing the crossed structure of the data: emission decisions are nested within individuals, who are themselves nested within two different groups. Moreover, we specified unstructured covariance for the random effects, which allows for flexible estimation without assuming a specific covariance structure (McNeish and Harring, 2020). This approach enabled us to capture potential variations in emission decisions that may arise due to both individual and group characteristics across rounds.

Results

Descriptive findings

Before testing the hypotheses, descriptive analyses were performed to explore the relationship between the dependent variables and the experimental conditions. Descriptive statistics for group emissions are presented in Table 1. Half of the participants played games with immediate feedback, while the other half played games with delayed effects (between-subject design). All participants played both a heterogeneous and a homogeneous game (within-subject design). Therefore, horizontal comparisons in Table 1 reflect within-participant differences, while vertical comparisons reflect between-participant differences.

Table 1.

Means and standard deviations of total group emissions in the 15 rounds over all conditions.

​	Heterogeneous endowments	Homogeneous endowments	Total
Immediate feedback	1672 (210) a N b = 24	1775 (236) N = 24	1724 (227) N = 48
Delayed feedback	1652 (187) N = 24	1654 (145) N = 24	1653 (165) N = 48
Total	1662 (197) N = 48	1715 (203) N = 48	1688 (201) N = 96

^aStandard deviations in parentheses.

^bN indicates number of groups.

Overall, the average total group emissions were lower under the heterogeneity treatment compared to the homogeneity treatment. This observation is in line with Hypothesis 1a rather than with Hypothesis 1b, suggesting that, on average, groups emit less CO₂ and therefore exhibit more cooperative behavior under heterogeneity. Also, average group emissions were lower in the delayed feedback condition, indicating that groups were more cooperative when feedback was delayed. This finding is not in line Hypothesis 2. However, the differences in average group emissions across the treatments are small.

Similar patterns were observed in the average total individual emissions over 15 rounds across all conditions (Table 2). Under both heterogeneity and delayed feedback, average individual and group emissions were lower compared to homogeneity and immediate feedback.

Table 2.

Means and standard deviations of individual emissions in the 15 rounds over all conditions.

​	Low-endowment a	High endowment b	Homogeneous endowment c	Total
Immediate feedback	358 (65 d ; 80% e ) n f = 720	478 (157; 46%) n = 720	444 (113; 59%) n = 1440	431 (125) n = 2880
Delayed feedback	349 (62; 78%) n = 720	476 (141; 45%) n = 720	419 (114; 56%) n = 1440	416 (120) n = 2880
Total	354 (63; 79%) n = 1440	477 (149; 45%) n = 1440	431 (114; 58%) n = 2880	​

^aLow endowment in the heterogeneous condition: endowment of 30 points per round, with a maximum possible total of 450 points for emissions across 15 rounds.

^bHigh endowment in the heterogeneous condition: endowment of 70 points per round, with a maximum possible total of 1050 points for emissions across 15 rounds.

^cEndowment in the homogeneous condition: endowment of 50 points per round, with a maximum possible total of 750 points for emissions across 15 rounds.

^dStandard deviations.

^ePercentage of mean emissions relative to the maximum possible points of emissions at the condition.

^fNumber of individual emission decisions.

Further analysis of the mean total individual emissions among high- and low-endowment players (Table 2) revealed that high-endowment players emitted, on average, 477 tons of CO₂, while low-endowment players emitted 354 tons, and players with homogeneous endowment 431 tons. This suggests that high-endowment players in heterogeneous treatment groups emitted considerably more in total than low-endowment players in heterogeneous treatment groups or players in homogeneous treatment groups.

However, when examining individual emissions relative to the possible maximum emissions for high-endowment, low-endowment, and homogeneous players (Table 2), we found that high-endowment players emitted only 45% of their endowment, on average, while homogeneous and low-endowment players emitted 58% and 79% of their endowment, respectively. Therefore, although high-endowment players emitted more in total, they emitted less relative to their wealth.

Figure 1(a) and (b) show the development of average group emissions across different experimental conditions. In line with the results in Table 1, these figures indicate that average group emissions were higher in most rounds under the homogenous treatments. Additionally, the figures reveal that in the first game, average group emissions were higher in the immediate feedback conditions, while in the second game, average group emissions were higher in the delayed feedback conditions. This suggest the possibility of a game order effect, which will be further controlled for in the multi-level analyses. Moreover, individual and group emissions tended to increase sharply towards the end of the second game, suggesting a “last-round effect” where players were more inclined to overemit CO₂, as there were no future consequences for their actions.

Figure 1.

(a): Average group emissions per condition in the first CPR game. (b): Average group emissions per condition in the second CPR game.

Building on the observed trends in average group emissions across separate experimental conditions, it was insightful to examine the frequency distributions of total group emissions across combined conditions (Figure 2). Notably, the largest proportion of groups that emitted less than 1500 tons of CO₂ in total, thus demonstrating sustainable behavior, were found in the heterogeneous treatment with delayed feedback. However, even with this treatment combination, some groups emitted unsustainably. In other groups, the largest proportion of groups emitted between 1500 and 1700 tons of CO₂. This was particularly striking in the heterogeneous treatment with immediate feedback, where 14 groups fell in this category. Moreover, the frequency distribution for this condition revealed an outlier: one group emitted between 2300 and 2500 tons of CO₂, the maximum total group emission observed across all combined conditions.

Figure 2.

Frequency distribution of total group emissions for all condition combinations.

Group emissions

The analysis of total group emissions across all rounds (Table 3) showed that heterogeneity did not have a statistically significant effect on group emissions (b = −102.75, p = 0.075), whereas delayed feedback had a statistically significant negative effect (b = −121.21, p = 0.036). Thus, heterogeneity did not lead to significant differences in overall emissions, while delayed feedback was associated with lower group emissions.

Table 3.

Total group emissions regressed on experimental conditions.

​	b (SE)	p
Heterogeneity	−102.75 (56.964)	0.075
Delayed feedback	−121.21 (56.964)	0.036
Delayed * heterogeneity	100.29 (80.559)	0.216
Constant	1775.21 (40.279)	0.000
Observations	​	96
Adj R-squared	​	0.034

Consequently, Hypotheses 1a and 1b are not supported. Hypothesis 2 is also not supported; contrary to expectations, delayed feedback reduced rather than increased group emissions. Although the interaction effect was not statistically significant, its direction suggests that the negative effect of delayed feedback on emissions was stronger under homogeneous than heterogeneous endowment conditions.

Since the analysis of total group emissions did not support our hypotheses, we next sought to deepen our understanding of the mechanisms underlying individual emission decisions. Using cross-level analyses, we examined individual emissions over the course of the game and explored the role of potential control variables in shaping individual-level behavior.

Individual emissions

The multi-level regression models for individual emission decisions, incorporating random intercepts for individual players and groups, are presented in Table 4. The empty model (Model 1) estimates the standard deviation of the group-level random effect to be virtually negligible (1.97e^-7), suggesting minimal variation between groups and relatively consistent behavior across different groups.

Table 4.

Cross-level models 1–6 on individual emission decisions with a random intercept for groups and players.

	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
	b (SE)	p	b (SE)	p	b (SE)	p	b (SE)	p	b (SE)	p	b (SE)	p
Game characteristics
Heterogeneity			−1.71 (0.468)	0.000	−6.14 (0.553)	0.000	−4.04 (0.814)	0.000	−4.06 (0.814)	0.000	−4.36 (0.848)	0.000
Delayed feedback			−1.64 (1.010)	0.104	−1.64 (0.973)	0.091	−1.23 (0.685)	0.074	−1.75 (0.716)	0.014	−1.62 (0.702)	0.021
Heterogeneity * delayed			1.30 (0.662)	0.050	1.30 (0.651)	0.047	0.94 (0.650)	0.148	0.91 (0.649)	0.160	0.91 (0.649)	0.160
High endowment					8.85 (0.614)	0.000	5.86 (0.583)	0.000	5.86 (0.582)	0.000	5.84 (0.581)	0.000
Homogeneous first							0.25 (1.208)	0.838	0.21 (1.209)	0.861	−0.41 (1.299)	0.752
Second game							2.13 (0.687)	0.002	2.12 (0.687)	0.002	1.80 (0.727)	0.013
Group emission > 100							0.35 (0.374)	0.351	0.93 (0.431)	0.031	0.95 (0.431)	0.028
Player prev. emission							0.35 (0.015)	0.000	0.35 (0.014)	0.000	0.35 (0.014)	0.000
Feedback prev. round									−0.22 (0.668)	0.744	−0.22 (0.668)	0.743
Attitudes
Political right											0.22 (0.174)	0.206
Prosociality											−1.33 (0.600)	0.026
Belief in limit emissions											−0.49 (0.230)	0.035
Individual characteristics
Male											−0.18 (0.657)	0.648
Dutch											0.02 (0.657)	0.975
Unknown co-players											0.95 (0.707)	0.179
Experimental experience											−0.18 (0.628)	0.775
Understanding task											0.20 (0.346)	0.555
Interaction effects
Gr. em. > 100 * feedback prev. round									−2.10 (0.739)	0.005	−2.10 (0.739)	0.005
Delayed * feedback prev. round									1.96 (0.718)	0.006	1.95 (0.718)	0.006
Constant	28.23 (.479)	0.000	29.59 (0.714)	0.000	29.59 (0.688)	0.000	15.93 (1.695)	0.000	16.04 (1.706)	0.000	20.89 (3.268)	0.000
SD cons. group	0.000 (0.000)		0.000 (0.000)		0.000 (0.000)		0.000 (.)		0.000 (.)		0.000 (0.000)
SD cons. player	6.221 (0.361)		6.202 (0.360)		5.940 (0.347)		3.520 (.)		3.526 (.)		3.358 (0.272)
N observations	5760		5760		5760		5376		5376		5376
N players	192		192		192		192		192		192
N groups	96		96		96		96		96		96
Log likelihood	−22,959.51		−22,851.64		−22,849.66		−21,058.95		−21,050.46		−21,043.95
Akaike inf. crit.	45,929.01		45,719.29		45,717.33		42,135.89		42,124.92		42,135.89
Bayesian inf. crit.	45,962.31		45,973.03		45,777.26		42,195.2		42,203.99		42,294.047

Model 2 examines the effects of heterogeneity, delayed feedback, and their interaction on individual emissions without accounting for individual-level variables. Heterogeneity has a significant negative effect (b = −1.71, p = 0.000) on individual emissions, indicating more cooperation in groups with heterogeneous endowments. This contradicts Hypothesis 1b, which predicts less cooperative behavior in heterogeneous groups, but supports Hypotheses 1a, which predicts greater cooperation in such groups. Delayed feedback does not show statistically significant effect on player emissions, suggesting that Hypothesis 2 is not supported. In addition, the interaction between heterogeneity and delayed feedback is positive and marginally significant (b = 1.30, p = 0.050), indicating that the cooperative effect of heterogeneity may be weaker under delayed feedback conditions. However, this interaction loses statistical significance once additional game characteristics are included in the models, suggesting that the moderating relationship is not robust.

Further, when controlling for high-endowment players in Model 3, the negative main effect of heterogeneity becomes even stronger (b = −6.14, p = 0.000), indicating that low-endowment players emitted less CO₂ compared to players in the homogeneous condition. In contrast, having a high endowment in the heterogenous condition shows a positive effect on emissions (b = 8.85, p = 0.000), meaning that high-endowment players in the heterogeneous condition emitted 8.85 tons more CO₂ than low-endowment players, and 2.71 tons more than players in the homogeneous condition. Thus, while high-endowment players exhibited less cooperative behavior, the lower emissions of low-endowment players compensated for this, leading to more favorable outcomes in heterogeneous groups.

In addition, Model 4 shows that players tended to emit more CO₂ in the second game compared to the first (b = 2.13, p < 0.002). This finding is consistent with earlier descriptive analyses suggesting a learning or adaptation process over time, potentially leading participants to overexploit the resource more strongly in the second game. The number of observations in the model decreased to 5376 after adding variables capturing previous behavior, since the first rounds of each game could not be included in these analyses.

In Model 5, the influence of feedback conditions on emissions becomes more evident. Players tended to emit more CO₂ (b = 0.93, p = 0.031) when their group emitted over the sustainable threshold of 100 tons of CO₂ in the previous round, suggesting an amplification effect driven by immediate feedback mechanisms. However, the interaction effect between emitting more than 100 tons in the previous round and the variable indicating whether players in games with delayed effects received feedback in the previous round (or in games with immediate feedback), is negative (b = −2.09, p = 0.005). This suggests that players in the delayed feedback conditions emitted less after learning about their co-players’ non-cooperative behavior, contrasting with players in the immediate feedback condition who tended to amplify non-cooperative behavior. This difference might be explained by the uncertainty in the delayed condition, leading to a possible “urgency-to-reduce” effect when high group emissions are observed. However, the positive interaction effect of being in the delayed condition and having received feedback in the previous round (b = 1.96, p = 0.006) shows that receiving feedback in the delayed treatment group leads to higher emissions, reducing cooperation.

When individual characteristics and attitudes were included in the full model (Model 6), the results indicate that both holding prosocial values (b = −1.33, p = 0.026) and believing that individuals in the real world would limit their CO₂ emissions to mitigate climate change (b = −0.49, p = 0.035) had negative effects on individual emissions, thereby promoting cooperation. Other controlled individual characteristics did not significantly influence individual emission decisions.

Discussion

The aim of this research is to study how endowment heterogeneity and delayed feedback affect the sustainable management of CO₂ emissions. Existing literature on cooperation in collective action problems presents two competing perspectives on the role of heterogeneity in cooperation. On the one hand, some research suggests that endowment heterogeneity can positively influence sustainable cooperation due to inequality aversion (Fehr and Schmidt, 1999) and the role of resourceful actors in catalyzing cooperation (Baland and Platteau, 1999; Olson, 1965). On the other hand, endowment heterogeneity can negatively impact cooperation due to disparities in contributions, rewards, and individual interests (Adhikari and Lovett, 2006; van Klingeren, 2020), as well as normative disagreements (Reuben and Riedl, 2013) and divergent priorities among individuals of different economic statuses (Hofmeyr et al., 2007). Additionally, literature suggests that delayed feedback on the consequences can undermine sustainable cooperation by introducing uncertainty, reducing perceived behavioral efficacy, and encouraging patterns consistent with temporal discounting (van Lange and Rand, 2022; Weber, 2006).

Our results showed that endowment heterogeneity negatively affects average individual emissions, thus positively influencing cooperation. Upon examining the effects of unequal endowments on individual emissions in more detail, we observed that lower emissions in heterogeneous conditions can be attributed to two components. First, individuals with lower endowments have fewer resources than those with higher endowments, and consequently emitted less overall. Second, while individuals with high endowments in heterogeneous groups tended to emit more, they still used a considerably smaller percentage of their resources compared to individuals in homogeneous groups. This suggests that cooperation in the context of climate change mitigation may not depend solely on the presence of endowment inequality. Rather, individuals’ relative emissions appear to be influenced by their overall budget, indicating that the dynamics of cooperation may be shaped by how both high-endowment and low-endowment participants manage their resources in response to collective challenges, rather than by inequality alone.

This observation aligns with the theoretical arguments of Adhikari and Lovett (2006) and van Klingeren (2020) regarding how inequality can lead to disparate contributions, uneven rewards, and significant variation in individual interests in resource management. However, contrary to their conclusions, our results suggest that these disparate contributions can actually enhance the preservation of the CPR. This finding supports Olson’s (1965) argument that heterogeneous groups may be more successful in managing resources. Resourceful actors are expected to act as catalysts for cooperation by bearing the larger share of the cost of collective action and contributing more toward a shared goal (Baland and Platteau, 1999).

Nevertheless, while this argument holds in our simplified CPR game, in which endowments were randomly assigned to participants, it might not directly apply to real-world contexts. In reality, resource allocation is often shaped by complex socioeconomic structures, power dynamics, and historical inequalities (Hurst et al., 2016), all of which can significantly influence cooperation dynamics. Also, factors such as the institutional frameworks (van Klingeren and Buskens, 2024), competition over resources, social norms, and varying levels of trust (Balliet and van Lange, 2013) may counteract the potentially positive role of resourceful actors in promoting cooperation.

It should also be noted that comparisons between our findings and evidence from CRSD experiments suggest that the effects of inequality on cooperation may depend on the specific structure of the dilemma. For example, Vicens et al. (2018) found that participants with fewer resources contributed significantly more to the public goods than wealthier participants, in some cases contributing twice as much. This differs from the pattern observed in our CPR setting and indicates that inequality may shape cooperation differently depending on whether the social dilemma involves collective provision or the appropriation of a common resource.

At the group level, we found no statistically significant effect of endowment heterogeneity on reducing emissions. One possible explanation is that unequal endowment allocation generated different behavioral responses within the same group. In heterogeneous group, individuals with low endowments tended to emit less. At the same time, individuals with higher endowments emitted more in absolute terms, even though they used a smaller proportion of their available resources than participants in homogeneous groups. When emissions were aggregated across all four group members, the lower emissions of low-endowment individuals were likely offset by the higher emissions of high-endowment individuals.

The different effects of endowment heterogeneity at the individual and group levels are therefore not necessarily contradictory. Rather, the findings suggest that heterogeneity may redistribute emissions within the group without substantially changing the collective outcome. In other words, in this simplified CPR game, the reduction in emissions among low-endowment participants was not strong enough to produce lower total group emissions. Although more individuals may have been willing to cooperate under heterogeneous conditions, the collective problem remained unresolved because emissions were disproportionately shaped by a relatively small number of high emitters.

Viewed more broadly in the context of climate change mitigation, these findings suggest that the success of collective action depends not only on how many actors are willing to cooperate, but also on which actors cooperate. Without the cooperation of high emitters, heterogeneity may generate visible individual restraint without producing meaningful collective gains. Moreover, it may create an unequal pattern of burden sharing in which those with few resources reduce emissions more strongly, while those with greater resources continue to drive the collective outcome.

Regarding the effects of delayed feedback on cooperation, results contradicted our predictions, as they showed no significant direct effect of delayed feedback on individual emissions. However, a more detailed exploration of the effects of different feedback types revealed how varying feedback frequencies influence individuals’ emission decisions differently. For instance, individuals in immediate feedback groups tended to respond to high previous group emissions by increasing their own emissions, while individuals in delayed feedback groups reduced their emissions after receiving feedback. This interaction between previous group emissions feedback frequency supports the theoretical expectation that cooperative behavior is supported by better opportunities for conditional cooperation (Fischbacher et al., 2001; Fischbacher and Gächter, 2010).

Contrary to the concerns raised by Weber (2006) and van Lange and Rand (2022), the effect of delayed feedback did not appear to impede cooperation through patterns consistent with temporal discounting or a lack of urgency. At the group level, delayed feedback even increased cooperation slightly. In this simplified CPR game, the uncertainty of acting without direct feedback, combined with the urgency-to-reduce effect that arises when learning about co-players’ behavior only after several rounds, seemed to prompt more cautious behavior or corrective actions in response to unsustainable group behavior—compared to the more reactive approach seen when feedback is provided continuously.

Overall, this study contributes to the research on cooperation in the context of climate change mitigation through an experimental approach. It theoretically elucidates how heterogeneity and delayed feedback can influence cooperation while empirically revealing the complex underlying mechanisms involved. Thus, it provides valuable insights into the nuanced nature of cooperation in climate-related CPR games. While this study adopts a more informal approach to theory, focusing on empirical insights and behavioral responses, its design is grounded in core game-theoretic concepts. For instance, the game-theoretic approach to social dilemmas—such as the ‘tragedy of the commons’ and the associated free-riding problem—forms the basis for understanding the strategic tensions present in CPR games. This study offers an applied perspective on these theoretical dynamics in the context of climate change mitigation, emphasizing the behavioral responses observed during the experiment. Future research could benefit from formally modeling these dynamics to better predict and understand deviations from theoretical equilibrium outcomes.

Moreover, it is essential to acknowledge some shortcomings of this study, highlighting potential areas for improvement in future research. First, a well-known criticism of laboratory experiments is their lack of generalizability. The generalizability of this experimental study is limited by the subject pool, which consists primarily of students. This self-selected group does not accurately reflect the composition of actors in real-world contexts. Nonetheless, regardless of the subject pool, such experimental studies have proven to be valuable tools for investigating relationships between behavior and various biological, economic, or social variables, even when the subject pool is non-representative (e.g., Anderies et al., 2011; Falk and Zehnder, 2013).

In addition, it can be argued that the total number of rounds in an experiment may impact participants’ behavior in a CPR game. For instance, in a longer game compared to our CPR game, the point at which uncooperative behavior becomes non-profitable would occur earlier relative to the total number of rounds. Consequently, participants would experience more rounds under the negative consequences of their overemissions in a longer game. In contrast, shorter games may allow participants to engage in uncooperative behavior for a relatively longer duration before facing negative consequences. Thus, the cost of unsustainable behavior differs based on the length of the game. Besides, it may take time for participants to understand the game and to develop a behavioral strategy. Furthermore, Kimbrough and Vostroknutov (2015) demonstrate that the regrowth factor is a significant determinant of the survival and sustainable use of resource in CPR experiments. Therefore, variations of our own experiment with different numbers of rounds and regrowth rates could provide interesting insights for future research.

Conclusion

This study explores the complex dynamics of cooperation in climate change mitigation through an experimental approach, specifically examining the impacts of endowment heterogeneity and delayed feedback within CPR games. Our findings challenge some prevailing assumptions about cooperation in collective action games while shedding light on the intricate mechanisms underlying cooperative behaviors.

We found that endowment heterogeneity tends to positively impact cooperation at the individual level. This is primarily due to low-endowment actors having fewer resources to emit in total, while high-endowment actors emit less relative to their resources and do not fully exploit their resources. This suggests that cooperation in climate change mitigation may not solely depend on endowment equality but rather on individuals’ resource management strategies and choices. However, conclusions regarding the wealthy bearing the costs of collective action by emitting less relative to their resources should be interpreted with caution. Our experimental setting cannot fully capture the real-world resource allocation mechanisms and the intertwined social contexts that affect cooperative behavior.

Furthermore, when exploring the effects of feedback types, we found that delayed feedback does not directly impact group or individual emissions, contrary to our predictions. However, we identified complex mechanisms underlying the relationship between delayed feedback and cooperation, with individual emission decisions influenced differently by feedback frequencies. In particular, the interplay between observed group behavior, feedback frequencies, and individual cooperation underscores the role of conditional cooperation in shaping cooperative behaviors. Notably, we observe that receiving information about others’ unsustainable behavior in a delayed manner leads to a more cautious behavior or corrective actions, resulting in higher cooperation than when feedback is provided continuously, which appears to reduce cooperation. Nonetheless, translating such findings to the real-world context of climate mitigation should be approached with caution, as our experimental setting, while informative, is a simplified version of real-world complexities.

Despite these limitations, our study contributes to the scarce research on heterogeneity and delayed feedback in CPR games, providing valuable insights into the underlying mechanisms and challenges within cooperative frameworks. Moving forward, future research should continue to explore the interplay of socioeconomic factors, feedback mechanisms, and cooperation dynamics in real-world settings. By bridging experimental findings with field studies, we can deepen our understanding of the factors and mechanisms influencing sustainable cooperation to collectively reduce CO₂ emissions and effectively address climate change.