Sanctions and international interaction improve cooperation to avert climate change

Abstract

Imposing sanctions on non-compliant parties to international agreements is advocated as a remedy for international cooperation failure. Nevertheless, sanctions are costly, and rational choice theory predicts their ineffectiveness in improving cooperation. We test sanctions effectiveness experimentally in international collective-risk social dilemmas simulating efforts to avoid catastrophic climate change. We involve individuals from countries where sanctions were shown to be effective (Germany) or ineffective (Russia) in increasing cooperation. Here, we show that, while this result still holds nationally, international interaction backed by sanctions is beneficial. Cooperation by low cooperator groups increases relative to national cooperation and converges to the levels of high cooperators. This result holds regardless of revealing other group members' nationality, suggesting that participants' specific attitudes or stereotypes over the other country were irrelevant. Groups interacting under sanctions contribute more to catastrophe prevention than what would maximize expected group payoffs. This behaviour signals a strong propensity for protection against collective risks.

1. Background

Humanity is faced with a wide range of threats involving the possibility of catastrophic collective losses. Such threats require international cooperation across widely different cultures [1]. While cooperation may be sustained by direct and indirect reciprocity [2] in small or culturally cohesive groups [3], cooperation in large groups of unrelated individuals is typically parochial; that is, it favours others perceived as belonging to one's own group at the expense of others perceived as belonging to other groups [4,5]. Since nationality is one of the strongest sources of parochial attachment [6], international cooperation seems to be at risk [7]. Some scholars and policymakers have proposed to introduce substantial and credible sanctions—trade sanctions in particular—for countries that do not comply with international agreements as a way to improve cooperation [8,9]. Sanctions could take the form of increased tariffs on imported goods from countries not complying with international agreements. Yet, applying sanctions may trigger a second-order cooperation problem [9,10]. Sanctioning is generally costly to the party applying sanctions, thus each party will prefer to free ride on others' sanctions. Nevertheless, individuals, just like countries [1], seem to favour sanctioning policies, even when this is costly to them [11–13].

Climate change is perhaps the most severe existential threat facing humanity. Currently, international cooperation falls critically short of the levels necessary to mitigate climate change [14]. Sanctions have been proposed as a possible remedy to the current stalemate. The ‘climate club' proposal hinges upon the idea that countries not complying with a climate agreement suffer a penalty in the form of increased tariffs from countries belonging to the club [15]. This approach seems promising. Even if the Paris Agreement is not a climate club as it does not allow for formal sanctioning, the number of climate provisions introduced in trade agreements is, in fact, increasing [16].

We designed an experiment to examine the effectiveness of sanctions for increasing international cooperation in interaction mimicking costs and incentives to prevent collective losses. Our experiment builds on the collective risk social dilemma [17] (CRSD; see electronic supplementary material, S7 for an abbreviation list). We modify the CRSD by introducing a sanctioning stage that reflects the characteristics of trade sanctions applied to international agreements. Controlled experimental evidence on sanctioning in inter-cultural contexts is rare [18] and lacking in international contexts. We involve participants from two countries—Germany and Russia—epitomising cultural areas where sanctions have been found to work or fail, respectively, as mechanisms to increase cooperation [19,20]. This approach puts the potential impact of sanctions as a method for underpinning international cooperation to a severe test.

Scientists classify climate change as having both a ‘gradual' and a ‘catastrophic' component. The former refers to incremental changes in underlying factors that continuously alter the climate, such as the progressive rise in sea levels. Catastrophic climate change refers to structural changes in ecosystems triggered by temperatures exceeding a ‘tipping point' and leading to irreversible change [21]. Examples are the collapse of the Amazon forest or the loss of ice sheets. The CRSD used in our experiments captures in a stylized way the potential gains and losses underlying catastrophic climate change [17]. Groups of individuals are faced with the possibility of losing part of their endowment if a random loss event occurs. To prevent such collective losses, individuals can contribute part of their monetary endowments to a collective fund that reduces the probability of the loss event occurring. Possible losses are large, thus simulating a major catastrophe in the offing. The consensus among scientists is that if temperatures increase less than 1.5°C from pre-industrial levels, no catastrophic loss will occur. We call this level the ‘certain safety threshold'. On the other hand, an increase by more than 5°C by 2100—which would occur in a ‘business-as-usual' scenario—will certainly trigger catastrophic climate change [21]. We call this the ‘certain unsafety threshold'. There is, however, uncertainty over which temperature level will actually trigger catastrophic climate change within the 1.5–5°C range [21,22]. We model uncertainty about the actual temperature level associated with this ‘catastrophe tipping point' using a uniform distribution over the interval bounded by the ‘certain safety' and the ‘certain unsafety' thresholds [22]. Collective loss is thus avoided with a probability proportional to the total amount of money that the group invests in the collective fund, relative to the amount of investment needed to achieve the certain safety threshold.

2. Experimental methods

Participants were involved in the CRSD at either the National or the International level, with sanctions being possible (S treatments) or not possible (NS treatments). Groups of six participants interacted in the CRSD, three of whom were university students in one city and three in another. In National treatments, participants were informed that the other city was located in the same country. In International treatments, one city was in Germany and one in Russia. The International treatments were conducted under two different settings. In the Open (O) treatments, German and Russian participants were informed that the other city was located either in Russia or in Germany. In the Blind (B-)treatments, participants were not made aware that participants from the other city were actually from another country [23]. Therefore, in both the National and the International Open treatments, participants were made aware that the other city was located either in the same country or in another country, although the exact location of the other city was never disclosed. Participants were citizens of their country of residence (see electronic supplementary material, S1.1 and S1.2 for participants' demographic and cultural characteristics). Behaviour in international interaction may be affected by prejudice and stereotypes about foreigners [17], by national pride, or by the desire to outperform the other group [24]. Our experimental design permits the comparison of outcomes between the case where such prejudices or inter-group motivations may affect behaviour (i.e. in the Open treatments) and the case where prejudice or inter-group motivations are limited by construction (i.e. in the Blind treatments). Ex-post questionnaire data confirm that our treatment manipulation worked because a large majority of participants in the Blind treatments believed that they were interacting with participants from the same country (electronic supplementary material, table S4). The outcomes of the Blind treatments can thus be attributed solely to the effect of participants' choices, reducing the relevance of beliefs and motivations relative to interaction with foreigners. The eight experimental treatments are summarized in table 1. The null hypotheses of equality of distribution across treatments of demographic and personal characteristics were not rejected in non-parametric tests within each country, except for an economics degree in Russia (electronic supplementary material, table S5). This entails that participants were, in general, not unevenly distributed across treatments with respect to such characteristics.

Table 1.

Summary of experimental design.

treatment name	within-country/ international	sanctions	nationality revealed	no. independent observations	no. participants
GER_NAT_NS	National Germany	no	yes	16	96
GER_NAT_S	National Germany	yes	yes	16	96
RUS_NAT_NS	National Russia	no	yes	16	96
RUS_NAT_S	National Russia	yes	yes	16	96
INT_Blind_NS	International	no	no	16	96
INT_Blind_S	International	yes	no	16	96
INT_Open_NS	International	no	yes	16	96
INT_Open_S	International	yes	yes	16	96
total				128	768

Note: GER, German participants; RUS, Russian participants; NS, No Sanctions; S, Sanctions; NAT, National interaction; INT, International interaction.

Participants interacted over 10 periods with the same partners in real-time via the Internet. Interactions were anonymous, but each group member could be identified by a number ranging from 1 to 6. Since participants knew that group members labelled from 1 to 3 were from one location while those labelled from 4 to 6 were from the other location, they could infer other group members' location from their numeric label. Each participant was endowed with 60 tokens in each period. Each token was worth 0.07 euros in Germany and 2.0 rubles in Russia. Such levels ensured equivalent purchasing power across countries. In the NS treatments, participants could contribute up to 50 tokens to a collective fund, the remaining 10 tokens being automatically added to their private accounts. If the sum of total contributions (C) to the collective fund matched or exceeded the certain safety threshold (T), there would be no loss to any player's private account. If, however, C < T at the end of the 10 periods, a loss of 75% to each player's private account would occur with probability $1-P$, where $P=\min \{(C/T);1\}$ and P is the probability of loss avoidance (PLA) (SM: electronic supplementary material, figure S4). P was the same for each group member. C = 0 is the certain unsafety threshold. Individuals' private accounts at the end of ten periods would equal the total endowment of 600 tokens minus total individual contributions to the collective fund. Participants earned either the full amount in the private accounts at the end of the ten rounds if no loss occurred, or else a quarter of this amount.

The CRSD in the S treatments took place in two stages. The first stage was identical to the NS treatments. In the second stage, each group member could use the remaining 10 tokens from their endowments to reduce other group members' private accounts in each of the 10 periods. Tokens spent on such sanctions were deducted from the private account. This sanctioning system had a number of characteristics in common with typical sanctions in international trade agreements. First, sanctions were observable [25] as tariff systems are known to all relevant parties. Second, the number of tokens deducted from a sanctioned participant's account increased more than proportionally in the number of tokens spent by other participants to sanction that participant. Similarly, the costs incurred by a country rise disproportionately as the number of sanctioning countries increases and as sanctions become more severe. The sanctioning cost structure is reported in electronic supplementary material, table S6. Final payoffs under the S treatments were equal to those under the NS treatments minus the sanctioning costs. At the end of each round of contributions, participants received information on each of the other group members' contributions in all the previous rounds, as well as the current PLA determined by total contributions. In S treatments, participants also received information on the sanctions assigned by a group member to any other group member.

We report details on the experiment protocol, measures adopted to ensure cross-country comparability, links to materials, and notes on determination of sample size, ethical approval and generalizability of results in electronic supplementary material, S4. Instructions and questionnaire are reported in electronic supplementary material, S5–S6.

3. Theoretical benchmarks

We use two theoretical benchmarks to analyse this interaction. The Nash equilibrium (NE) identifies the set of individual actions ensuring that each action is the best response to others' individual actions assuming that each agent maximizes their own monetary payoff. By contrast, the cooperative solution (CS) takes the perspective of the entire group and maximizes the total sum of expected monetary payoffs (see electronic supplementary material, S1.3 for the derivation of the two solutions).

For low levels of T, both the NE and the CS prescribe the avoidance of losses with certainty (figure 1). For intermediate levels of T, individual and collective interests diverge as the NE prescribes progressively lower contributions, while the CS prescribes full loss avoidance. For T = 2100, the threshold used in our experiment, the NE prescribes to contribute nothing—regardless of the individual's degree of risk aversion (Point A)—while the CS for risk-neutral agents prescribes a PLA of 69% (Point B). If agents are risk-averse, the CS prescribes a higher PLA (Point C) than for risk-neutral agents, which is typically lower than certain loss avoidance (Point D). At T = 2100, higher risk aversion does not affect the NE (Point A).

Figure 1.

Nash equilibria and cooperative solutions in the CRSD game for different levels of the certain safety threshold T (x-axis). The y-axis plots the group-level contribution for each solution. (Online version in colour.)

It can be observed that higher risk aversion reduces individual contributions in the NE and increases those in the CS for internal solutions. These results are due to the fact that individuals and groups balance differently the effect of contributions on the PLA, on the one hand, and on the share of the private account that is earned even if the loss event occurs, on the other. From an individual perspective, the effect of contributing more to the group account only slightly increases the PLA, while it for sure decreases the amount that is earned if the loss event occurs. Therefore, more risk-averse individuals will prefer to allocate more to the private account. From the group perspective, the effect of contribution on increasing the PLA is larger than in the individual case, because the CS takes into account that a token contributed to the group increases the PLA for everyone in the group. Therefore, in the CS with more risk-averse individuals, more resources will be allocated to increasing the PLA.

Overall, the interaction implemented in our experiment had the typical characteristics of a social dilemma [26] with individual interests maximized by no contribution to the collective account and group interests maximized by positive contributions. Both the CS and the NE predict that no sanctioning should occur. This is so either because the CS already prescribes the collectively optimal levels of contribution, or because sanctioning of others is a second-order cooperation problem and rational self-interested individuals should not sanction (in the case of the NE).

4. Hypotheses and research questions

Our first two hypotheses concern the national treatments and the effectiveness of sanctions. Since the CRSD in our setting has not been investigated neither in an international setting nor under sanctioning conditions, we ground our hypotheses on other types of cross-national or international cooperation experiments. Most experimental studies on cooperation show that contributions start at an intermediate level between the NE and the CS and tend to get closer to the NE, without actually reaching it, as interactions go by. Two studies [6,20] were concordant in finding that cooperation rates in Germany and Russia were no different from each other when sanctions were not available. Looking more generally at broad cultural areas, two studies found higher cooperation in countries from the ‘Protestant Europe' cultural group (to which Germany belongs [27]) than in the ‘Orthodox/ex-communist' cultural group (to which Russia belongs [6,19,27,28]), while another study found no difference between the two areas [20]. On the grounds of this evidence, we posit:

Hypothesis 1: Cooperation rates are not significantly different in Germany and Russia in National NS treatments.

The same studies [19,20] found that sanctions were effective in increasing cooperation in the ‘Protestant Europe' cultural group—and in Germany in particular. By contrast, in the ‘Orthodox/ex-communist' cultural group—and in Russia in particular—they were detrimental because of the widespread propensity to sanction high cooperators and to search for vengeance after having been sanctioned [19,20,28,29]. We therefore posit:

Hypothesis 2: Sanctions are effective in increasing cooperation in Germany but not in Russia. Consequently, cooperation is overall higher in Germany than in Russia in National S treatments.

As for international cooperation without sanctions, the parochial nature of human psychology [4,5] entails that cooperation should be lower in international than in national interaction, as national groups provide a strong source of attachment to individuals [30]. This is likely to be the case especially with repeated interaction because of the ‘bad apple' effect (i.e. the phenomenon whereby a few low cooperators in a group lead to a drastic reduction of willingness to cooperate with others [31]). Reduced cooperation in international interaction compared to national interaction has indeed been found [23,32]. However, other studies found no significant effect [33,34], with high-cooperators from one country possibly making up for the low cooperation rates by individuals from the other country [33]. Nevertheless, a large-scale study involving 42 countries found that ingroup bias between national and foreign groups was ubiquitous around the world [6]. On the grounds of this evidence, we posit:

Hypothesis 3: Cooperation is lower in International interaction than in National interaction in NS treatments.

As for international interaction with sanctions, we already noted that sanctions have opposite effects on cooperation in Protestant European countries vis-à-vis Orthodox/ex-communist countries [19,20,28,29]. It is therefore an open question whether sanctions will maintain the capacity to discipline low cooperators observed in Protestant European countries, whether they will trigger the retaliatory patterns observed in Orthodox/ex-communist countries, or whether such two effects will cancel each other out. The overall effects on cooperation are also unclear. Given the lack of experimental evidence on sanctions in an international setting, we leave our research question open to two mutually exclusive hypotheses:

Hypothesis 4a: Sanctions are effective in increasing cooperation in International interaction above cooperation in International NS treatments;

Hypothesis 4b: Sanctions fail to increase cooperation in International interaction above cooperation in International NS treatments.

5. Results

(a) . Cooperation is substantial and averages the cooperative solution

In contrast to the NE prediction, all groups achieved substantial levels of loss avoidance. We report the PLA achieved by each group in figure 2. Since the PLA is proportional to group members' total individual contributions, it is a measure of group-level cooperation. The PLA ranged from 12% to 100%, the grand mean being 70.1%, in line with the CS prediction for risk-neutral agents (figure 2). Eighty-six per cent of groups in the S treatments achieved a PLA higher than the CS, with seven groups achieving full loss avoidance. Overall, the PLA was 18% lower in NS treatments than S treatments (d = −1.03; p < 0.0001; n = 128; d is Cohen's d; all tests are two-sided Wilcoxon–Mann–Whitney (WMW) tests unless otherwise indicated) and only 34% of groups in NS treatments (as opposed to 86% in the S treatments) exceeded the PLA prescribed by the CS. Neither contributions nor sanctions differed significantly between the two locations within each country (electronic supplementary material, tables S7–S8). Therefore, we consider aggregate national observations only. Moreover, the PLA achieved in the International Open treatments was very close in size—and not statistically significantly different—from the PLA in the International Blind treatments, particularly so in the NS treatments (d = 0.04; p = 0.99; n = 32), but also in the S treatments (d = 0.41; p = 0.30; n = 32). We report results for the International Open treatments (O treatments) in the paper and those relative to the International Blind treatment in electronic supplementary material, S1.4, unless results between the two treatments differ.

Figure 2.

Probability of loss avoidance for each group and treatment.

(b) . Cooperation among Germans is higher than among Russians' in National treatments with sanctions

We first assess Hypotheses 1–2. In National treatments without sanctions, German groups did not achieve significantly higher PLA than Russian groups at conventional levels, although the effect size was medium (d = 0.71; p = 0.08; n = 32). Conversely, when sanctions were available, German groups did achieve significantly higher PLA than Russian groups in national interactions (d = 1.39; p = 0.0002; n = 32). The PLA was significantly higher in the S treatment than the NS treatments in German national interactions (d = 1.37; p = 0.0014; n = 32). The increase in PLA in Russian national interaction in the S treatment compared to the NS treatment was not large enough to achieve statistical significance at conventional levels, although it had medium effect size (d = 0.65; p = 0.072; n = 32). These results confirm Hypotheses 1–2 and are in line with previous comparative research [19,20,28,29].

(c) . Cooperation in International treatments without sanctions is not significantly different from cooperation in National treatments

On the basis of Hypothesis 3, we would expect lower cooperation in International treatments without sanctions than in National treatments. This was however not the case. Without sanctions, mean PLA in the International O treatment was 0.65, on a par with mean PLA in the German National treatment and higher than the mean PLA in the Russian National treatment (figure 2). Without sanctions, there was no statistically significant difference in PLA between the International O treatment and either the German National NS treatments (d = 0.03; p = 0.99; n = 32) or the Russian National NS treatment, at conventional levels, (d = 0.72; p = 0.055, n = 32), although in the latter case the effect size was medium.

This result may be due to German participants having increased contributions in International NS treatments to compensate for Russian participants' lower cooperation rates [33]. Alternatively, Russian participants may have increased their cooperation in International NS treatments compared to National treatments. The latter alternative is supported by the data. German participants' cooperation levels were virtually the same as Russian participants' in the International O treatment using two-tailed WMW matched-pairs sign rank tests (d = −0.18; signed-rank WMW: p = 0.67; figure 3a).

Figure 3.

Average cooperation rates by nationality and treatment. (a) No sanctions (NS) and (b) Sanctions (S). Mean contributions to the collective fund as a share of the certain safety threshold. Error bars are 95% confidence intervals with bootstrapped standard errors (10 000 repetitions).

We plot the evolution of average individual contributions per round broken down by nationality in electronic supplementary material, figure S5. It is noticeable that Russians' mean contributions in the International Open NS treatment started off at a level virtually identical to what was found in the Russian National treatment, but gradually caught up and matched up with Germans' mean contribution levels. However, a series of tests fail to reject the null hypothesis that Russians' contributions in the International Open NS treatment come from the same distribution as in the National NS treatment, except for periods 7 and 8 (electronic supplementary material, table S9 and S1.4.1). Overall, we cannot reject, at conventional levels, the hypothesis that Russian participants contributed differently in the International NS treatment and in the National NS treatment, although the effect size was medium (d = 0.57; p = 0.055, N = 32). Moreover, there was no significant difference between Germans' contributions in the International Open NS treatment and in the German National treatment neither in any interaction period (electronic supplementary material, table S9), nor across all periods (d = 0.06; p = 0.70, n = 32).

(d) . Cooperation in International treatments with sanctions is higher than without sanctions and, for Russians, higher than in National interaction

Hypothesis 4 leaves open whether sanctions are effective in International treatments. We find that the PLA was significantly higher in the International Open S treatment than in the Open NS treatment (d = 1.35; p = 0.0012; n = 32). Average PLA in International Open S treatments was similar in size and not significantly different to that achieved in National German S treatments (d = 0.40; p = 0.22; n = 32), but was significantly higher than average PLA in the Russian National S treatment (d = 1.18; p = 0.003; n = 32).

Again, this result may be due to Germans making up for Russians' lower cooperation in International treatments, or to Russians increasing their cooperation. As with the NS treatment, we find that German participants' cooperation levels were virtually the same as Russian participants' in the International Open treatment (d = 0.11; WMW signrank: p = 0.86; n = 16; figure 3b).

In the initial periods, contributions by Russian participants in International S treatments started below German participants' contributions but quickly caught up as interactions continued (electronic supplementary material, figure S5A–B). Non-parametric tests reveal that while Russian participants' contributions in the International Open S treatment were not significantly different from Russian participants' contributions in the National S treatment in periods 1 and 2, their contributions were significantly higher in International treatments than in National treatments in all subsequent periods (electronic supplementary material, table S9 and S1.4.1). Conversely, the hypothesis of equality of distributions for contributions in International and National treatments was never rejected for German participants in any period. We can thus conclude that in International treatments, Russian participants' contributions quickly increased in comparison with the National treatment and converged to German participants' contributions. Over the whole 10 periods, contributions by Russians were significantly higher in the International O treatment than in the National treatment with sanctions (d = 1.00; p = 0.011, n = 32). Conversely, there was no significant difference for Germans (d = −0.29; p = 0.45, n = 32). These results support Hypothesis 4a and indicate that international cooperation with sanctions was overall beneficial because Russian participants achieved higher PLA, while PLA remained stable for German participants.

We decompose the treatment effects of cooperation by Russian and German participants in electronic supplementary material, table S10 and figure S6, pooling the two International treatments. Introducing sanctions in National interactions in Russia increased cooperation by 13% in comparison with the Russian National NS treatment. Remarkably, the same increase was effected without sanctions by ‘internationalising' interaction (i.e. having Russians interact with Germans). While neither of these effects was statistically significant, introducing sanctions in an international context increased Russian participants' cooperation by 20% in comparison to either the Russian National S treatment or the International NS treatment. As we noted, both these increases were statistically significant. We can thus conclude that, while sanctions alone and internationalization alone brought about only marginal increases in cooperation by Russian participants, the combination of the two factors was necessary to significantly increase Russian participants' cooperation.

(e) . Little sanctioning suffices to spur cooperation

Next, we analyse the mechanisms that made sanctions effective in increasing cooperation. Only 7% of the available endowment was spent on sanctions and in 70% of cases no sanction was administered (electronic supplementary material, figure S7). Sanctions had a spike in the last period when no counter-sanctioning was possible anymore (electronic supplementary material, figure S10). This spike can only be accounted for as punishment for the previous or the present interactions, as it could not have any disciplinary function for the future.

Previous research found that sanctions are effective if people increase cooperation after having been sanctioned [20]. With an OLS econometric model controlling for period effects (see electronic supplementary material, S1.5 for specification details), we compute the impact on the contribution made in the next period of having a token deducted through sanctioning in the current period. On average, a token deducted by sanctioning raised cooperation by 0.42 tokens in the next period (p < 0.001; table 2, column 1), but the effect differed across treatments (table 2, columns 2–5). Sanctioning was more effective in the German National treatment than in the Russian National treatment (p = 0.005; table 2, column 3). Sanctioning in the International Blind treatment was as effective as sanctioning in the German National treatment (p = 0.76, table 2, column 4) and was significantly more effective than in the International O treatment. Hence, sanctions lost part of their effectiveness when nationality was revealed to participants than when it was concealed from them.

Table 2.

Econometric analysis of the impact of sanctions on cooperation.

	dependent variable: ${\mathrm{contribution}}_t\hspace{0.17em}-\hspace{0.17em}{\mathrm{contribution}}_{t-1}$
	(1)	(2)	(3)	(4)	(5)
treatments	estimated coefficients for sanction_losst−1		difference between RUS_NAT_S and other treatments' coefficients	difference between GER_NAT_S and other treatments' coefficients	difference between INT_B_S and INT_O_S
all treatments	0.42***
	[0.10]
	<0.001
RUS_NAT_S		0.36***
		[0.07]
		<0.001
GER_NAT_S		0.66***	−0.31***
		[0.09]	[0.11]
		<0.001	0.005
INT_B_S		0.72***	−0.36*	−0.052
		[0.14]	[0.16]	[0.17]
		<0.001	0.024	0.76
INT_O_S		0.29†	0.07	0.37*	0.43*
		[0.15]	[0.15]	[0.17]	[0.21]
		0.056	0.69	0.032	0.041
observations	3456	3456	3456	3456	3456
number of participants	384	384	384	384	384
R2_within	0.0887	0.105
R2_between	0.0436	0.0417
R2_overall	0.0748	0.0856

Note: We fitted an OLS estimator to a model having as dependent variable the variation in contribution between period t and t − 1, for t = 2, …, 10. The table reports the estimated coefficients for tokens lost due to sanctions by other group members in the previous period ($\mathrm{sanction}\mathrm{\_}{\mathrm{loss}}_{t-1})$ across all treatments (column 1) and for each treatment (column 2). Columns 3–5 report the results of t-tests over the null hypothesis that a certain treatment coefficient is different from the coefficient of another treatment. The full regression output and further specification details are reported in electronic supplementary material, table S11. See table 1 for treatment labels. ***p < 0.001; **p < 0.01; *p < 0.05; †p < 0.10.

In order to understand the reasons why sanction effectiveness differed across treatments, we decompose sanctioning into Prosocial and Antisocial sanctioning (electronic supplementary material, S1.4.3). We define Antisocial sanctions as instances in which an ego punished an alter who contributed no less than the group median, while Prosocial sanctioning is the residual category [25]. Antisocial sanctioning is puzzling because it targets individuals who are increasing social welfare in the group, but it has proved to be endemic in experiments with people from cultural areas classified as Orthodox/post-communist [20,28,29].

We analysed treatment differences between the International O treatment and the National treatments in the individual propensity to sanction through a Poisson regression controlling for period effects (see electronic supplementary material, S1.6 for the specification details). We controlled for the counterpart's contribution level, which, as expected, was a strongly significant predictor of sanctioning. The higher the contribution, the lower the probability of being sanctioned (b = −0.07, p < 0.001, n = 19 200; b is the coefficient estimated in the Poisson regression). It is notewothy that Germans spent significantly more on prosocial sanctioning than Russians in National treatments (b = 0.32, p = 0.016, n = 7210), and spent significantly less on antisocial sanctioning than Russians (b = −0.81, p = 0.002, n = 11 990). These results are at the bases of why sanctioning was effective in increasing cooperation in German National treatments but not in Russian National treatments. Germans significantly increased their prosocial sanctioning in the O treatment compared with the National treatment (Wald test on difference in coefficients—Wald henceforth: b = 0.34, p = 0.018, n = 7210), and so did Russians (Wald: b = 0.87, p < 0.001, n = 7210). As a result, we found no significant difference in prosocial sanctioning between Russians and Germans in the International O treatment (b = 0.21, p = 0.16, n = 7210). Interestingly, Germans increased their antisocial sanctioning in the International O treatment compared to the National treatments, while Russians decreased it, although in both cases the differences were at the margins of statistical significance (Wald: b = 0.53, p = 0.062, n = 11 990 for Germans; Wald: b = −0.47, p = 0.077, n = 11 990 for Russians). Overall, there were no significant differences between Russians and Germans in the International O treatment with respect to antisocial sanctioning (b = −0.20, p = 0.49, n = 11 990), nor in overall sanctioning (b = 0.12, p = 0.43, n = 19 200). Hence, Russians seem to have converged to the same patterns as Germans' sanctioning in the International O treatment.

In electronic supplementary material, S1.5, we show that these results are robust to demographic controls. We also analyse the effect of being sanctioned regardless of the amount of sanctioning (electronic supplementary material, table S12).

(f) . Additional results

We analyse demographic effects in electronic supplementary material, S1.5–1.7 and show that payoffs were significantly higher in NS treatments than S treatments in electronic supplementary material, S1.8.

6. Discussion

Many fear that as global-level cultural heterogeneity, complexity and institutional limitations make international cooperation even more difficult than the local or national variety [26,35,36], international cooperation will be unable to steer clear of a tragedy of the ‘global commons' [37,38]. Our results offer a glimmer of hope, indicating that the combination of sanctions and the internationalization of interaction brings about net positive effects. When sanctions were available, groups that are normally high cooperators in national interactions did not decrease cooperation internationally, while groups that are normally low cooperators nationally increased their cooperation internationally.

Our results indicate that individuals do not necessarily act parochially in social dilemmas where people cooperate to reduce collective risk. Theoretically, it may be argued that ingroup identity may be fostered by the common threat of losing part of the private account if the loss event occurs [39]. Such a shared fate may make the common goal of avoiding the loss particularly salient, thus prompting individuals to substitute collective goals for individual goals [40]. One may conjecture that a common ingroup identity is more easily created in a CRSD than in a standard public goods game [41]. The possibility of collective loss reduces the absolute differences in expected payoffs between high and low cooperators. Perhaps this aspect of the interaction also makes it possible to cement a stronger group identity than in linear public goods interactions.

It has to be noted that, with some rare exceptions [23,33], the available evidence on international interaction is limited to one-shot interactions [6,7,42]. The dynamic setting may have created additional motivations for cooperation. One theoretical account hinges upon the idea of quicker belief update by low cooperators than high cooperators. According to this account, Russian participants' initial beliefs on others' cooperation would be rooted in the cooperation rates observed in local and national environments and would thus be set at a relatively low level. After having observed higher-than-expected cooperation in the initial periods of interaction, though, Russian participants involved in international interactions would quickly revise their beliefs on their counterparts' cooperation upwards. Consistently with a motivational model of conditional reciprocity [19,43,44], adjusting beliefs upwards would then prompt Russian participants to be more cooperative in international than in national interactions. As for high cooperators, they may, contrary to the bad apple effect, behave and be perceived as role models [45], whose behaviour is imitated by low cooperators. Sanctions were necessary to achieve the result of significantly higher cooperation for low cooperators than in national interaction. A novel result of this study is that German high cooperators were as capable of disciplining low cooperators in international interactions as in national ones. This result further qualifies the characteristics of strong reciprocators in cooperation interaction [4–12].

It is an open question whether our results are specific to the German/Russian combination or could be generalized to other countries or other contexts. The similarity of results in Blind and Open treatments shows that it was the actual content of participants' actions rather than motivations linked to the specific nationalities involved, which determined the beneficial effects of international cooperation. This suggests that our results are not driven by awareness of the counterpart's nationality and may be generalized to other countries from the same cultural groups. It has to be noted that the effectiveness of sanctions was also found in a study between different ethnicities within Bosnia [18]. Moreover, the conflictual history between Germany and Russia suggests that international cooperation observed between these two populations may be a lower bound of what is the same in other countries from different cultural areas. On the other hand, while cross-cultural analysis of cooperation patterns shows a remarkable consistency of results within cultural groups [28], the variance of behaviour is higher both when sanctions are available compared to when they are not [28] and in international as opposed to national contexts [23]. Germany and Russia are at intermediate levels of cultural distance [46] and it is therefore unclear what may happen when cultural distance increases. These considerations suggest caution on the possibility of straightforward generalisation of our results.

Though our experiment reproduces in a stylized fashion various features of the consequences of climate change and trade sanctions on individual earnings, the problem of ‘scalability' is apparent in connection with the outcome of our experiment [47,48]. Nonetheless, at a more fundamental level, our experiment can be seen as revealing the willingness of the general population to abide by an agreement once an agreement has been reached [49], which is a fundamental feature of any international agreement [1]. In fact, individuals who cooperated in the experiment were also marginally more likely to conduct environmentally sustainable behaviour in real life, such as buying environmentally friendly goods, saving water, participating in ecological movements and recycling (electronic supplementary material, table S14 and S1.7).

Despite these limitations, our findings give rise to some policy recommendations. First, establishing international teams at several levels of government to seek solutions for collective risk social dilemmas seems a promising strategy. In spite of conspicuous cultural differences, our international groups were no less cooperative than national groups when a common threat loomed. Second, our results suggest that sanctions can be used in international interactions to discipline people who would otherwise not cooperate and that they can do this without risking a spiral of retaliation and counter-retaliation. This evidence supports the view that international sanctions can lead to significant and robust changes in standards of conduct and should be used more extensively in international agreements, particularly in climate agreements. A concrete policy recommendation would be to institutionalize sanctions, as is suggested through so-called climate clubs, by enforcing rules for members through internal sanctioning mechanisms, and implicitly sanctioning non-members through measures such as carbon border tax adjustments [50]. Our study has also shown a preference for remarkably high levels of collective loss avoidance, at rates exceeding those that would maximize payoffs (see electronic supplementary material, S1.8). Such preferences should be taken into account by policy-makers.

Ethics

Since our research could not provide any harm to participants and did not involve any medical treatment, the approval by an ethics committee or institutional review board was waived by our universities. We asked every participant to read an information sheet and sign an informed consent form before starting the research session. Participants were aware that they could leave the session at any point, but none did. Data are fully anonymized.

Data accessibility

The dataset generated and analysed during the current study and analyses codes are available from the project repository of the Open Science Foundation: https://osf.io/r3a2x/ and the Dryad Digital Repository: https://doi.org/10.5061/dryad.qv9s4mwgd [51]. Analysis has been conducted with Stata17.

Authors' contributions

G.G.: conceptualization, data curation, funding acquisition, investigation, methodology, project administration, writing–original draft, review and editing; A.B.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, writing—review and editing; H.H.-S: conceptualization, data curation, funding acquisition, investigation, methodology, project administration, writing—review and editing; T.R.: conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, writing—review and editing; M.V.R.: conceptualization, funding acquisition, investigation, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

The authors declare no competing interests.

Funding

The work was supported by the Center for Global Cooperation Research at the University of Duisburg-Essen, Christian-Albrechts-University of Kiel and by HSE within the framework of the Basic Research Program at the National Research University Higher School of Economics (HSE).