Abstract
Bastian et al. (2014) found that sharing a painful experience promoted later intergroup cooperation. In Bastian et al.’s second experiment, 62 participants were assigned to groups of two to six people each. They performed either two painful or two painless tasks and then played an economic game. The present study consisted of two replications of the experiment: The first was a nonpreregistered pilot study (N = 153 students from the Czech Republic), and the second was a preregistered direct replication (N = 158 students from Slovakia). Important deviations from the original procedure were that (a) gender homogeneity of the small groups was balanced across the conditions and (b) the number of participants in each small group was fixed at three. No relevant effect of shared pain on cooperation emerged. The findings indicate that the true effect of shared pain on cooperation obtained in the original study may have been an overestimate or that the effect is not generally valid across various contexts.
Past studies have shown that cooperation and cohesion between members of a group can be strengthened if its members carry out an activity concurrently, thus gaining the feeling of a shared experience (Sullivan et al., 2015; Wiltermuth & Heath, 2009). Similarly, the conviction that our subjective experience is viewed and experienced in the same way by another person contributes to greater feelings of connection and cohesion with that person (Pinel et al., 2006). Shared traumatic experiences from a war zone, including the death of friends, may lead to stronger bonds and feelings of cohesion between the surviving members of the unit (Elder & Clipp, 1988). Research further suggests that experience with injury in the highly performance-oriented world of professional ballet also helps foster a greater level of social support between members of the ensemble (Turner & Wainwright, 2003). Italian children who experienced extreme distress during two earthquakes displayed a high degree of prosocial behavior and intense feelings of closeness to one another (Vezzali et al., 2016).
Whether emotional or physical, the intense experiences described above all result in some form of shared pain. The importance of shared pain as “social glue” has long been a subject of interest in sociology and social anthropology, particularly in connection to initiation and religious rituals. These often contain elements in which painful experiences are an integral part (Whitehouse, 1996). One important meaning of these rituals relates to the possible increase in trust, cooperation, and internal group cohesion (Power, 2018; Whitehouse & Lanman, 2014). Painful initiation or religious rituals strengthen participants’ prosocial behavior and level of identification with the group (Xygalatas et al., 2013).
Bastian et al. (2014) carried out three experiments to study the influence of shared pain. In the second of the three experiments, they found support for the hypothesis that shared pain promotes cooperation between the group members (a moderate to large effect was identified, d = 0.72, 95% confidence interval [CI] = [0.20, 1.24]). A sample of 62 Australian university students was divided into small groups (2–6 members each). In the experimental condition, each group carried out two physical tasks designed to elicit unpleasant, painful feelings. The participants in the control condition carried out similar tasks without any pain being elicited. The researchers then measured the level of cooperation in an economic game.
The participants who initially carried out the painful tasks subsequently made decisions that enabled others to get a higher payoff from the economic game, relative to decisions made by the participants in the control condition. Thus, the authors concluded that experience of shared pain promotes cooperation between members of a group. The effect of shared pain on cooperation was replicated by Bastian et al. (2014) in their third experiment, which had a slightly modified design: Instead of physical tasks in the first phase, the participants ate hot chili peppers. They found a medium-sized effect, but the CI ranged from a strong effect to almost zero, d = 0.53, 95% CI = [0.00, 1.06].
We believed that another external replication of the effect of shared pain on cooperation was needed for several reasons. First, the effect found by Bastian et al. (2014) in their Experiment 2 was nonsignificant when analyses controlled for group-level variance. We did a secondary multilevel analysis using the data set from the original experiment. The intraclass correlation coefficient (ICC) was .57, which shows that the level of cooperation was affected by membership of a particular group. In the multilevel model, the effect of shared pain on cooperation was not significant, b = 0.86, 95% CI = [−0.05, 1.77], t(17.79) = 2, p = .061.
Second, the effect of shared pain found by Bastian et al. (2014) was nonsignificant or significant with a p value close to .05 when analyses controlled for the nonequivalence between the experimental and control groups. When analyses controlled for the influence of group size on the level of cooperation, the probability of Type I error (p = .042) rose markedly, and the effect decreased. Similarly, in the subsequent replication with hot chili peppers (Bastian et al., 2014, Experiment 3), the effect of shared pain became nonsignificant when analyses controlled for the nonequivalence in gender (p = .054) and group size (p = .050).
Statement of Relevance
Several psychological, anthropological, and sociological studies have shown that people cooperate more when they have an intense shared experience. One Australian psychological study suggested that a shared painful experience may promote cooperation as well and provided empirical evidence that shared pain can act as “social glue.” However, that study provided evidence from only one specific country. We replicated the study twice, once on a Czech sample and then on a Slovak sample, using a similar experimental procedure as in the Australian study. Neither of our replications found any significant effect of shared painful experience on subsequent cooperation in the economic game. Our results show that the effect of shared pain on cooperation is probably not as strong and valid across cultures and contexts as the authors of the original study assumed. This finding calls into question the claim that a short painful experience is sufficient to promote cooperation in work teams or other groups.
Third, although the difference in the number of men and women between the experimental and control condition was controlled for, the influence of the ratio of men and women in the individual groups who played the economic game was not controlled for. The gender homogeneity of the group is one of the important factors influencing the amount of competition and cooperation (Lee et al., 2016). The number of participants in both conditions was low (ns = 34 and 28), as was the number of groups (approximately nine in each condition). It would therefore be easy to arrive at nonequivalence between the experimental and control conditions, which could have affected the results.
Fourth, to our knowledge, the experiment was not replicated by other researchers before we planned our replication. Bastian et al. (2014) generalized the effect. However, this generalization was based only on analyses that were conducted on a sample of Australian university students.
Because of the aforementioned points, we believed that the data from the original study (Bastian et al., 2014) did not provide strong support either for or against the effect of shared pain and that the effect presented by Bastian et al. could not be considered generally valid without further replication. Therefore, we prepared a preregistered direct replication of Bastian et al.’s second experiment, which was designed to test the hypothesis that shared pain promotes cooperation between group members.
After approval of the Stage 1 plan for the present study, we discovered that Qi et al. (2020) had independently performed two replications of the second experiment of Bastian et al. (2014) on Chinese samples. Neither experiment was preregistered, and neither controlled for the eventual effects of group size and male-to-female ratio or for group-level variance. The first experiment (N = 75) found a moderately positive effect of shared pain on cooperation, d = 0.50, 95% CI = [0.03, 0.95]. The second experiment had a 2 × 2 between-subjects design (N = 117) in which half of the participants completed the physical tasks individually (n = 57; the experience was not shared) and the other half completed them in a group (n = 60; the experience was shared). The authors did not find an effect of unshared pain on cooperation, but they found a very strong effect of pain on cooperation in the shared condition, d = 1.01, 95% CI = [0.45, 1.52]. We were not able to consider these replications when we prepared our experiments. However, we will consider them in the meta-analysis at the end of the article and in the Discussion section.
General Method
Preregistered procedure
The present research consisted of a pilot study (Study 1) and a preregistered study (Study 2). Both studies closely replicated the procedure from the second experiment of Bastian et al. (2014). All materials, including a record of the sample procedure with an English transcript, are available on OSF as a part of the preregistration (https://osf.io/4xsz2/).
Participants were assigned to groups of three people, each of which was then randomly assigned to either the experimental (pain) or the control (no-pain) condition. Shared pain was manipulated using two physical tasks, which were presented as tasks designed to assess physical acuity. The participants in both conditions performed both tasks concurrently with other members of the group. Before the start of each task, the experimenter emphasized that the tasks should not be viewed as a competition. It was supposed that this would ensure that the participants perceived their experience during the physical tasks as shared.
The participants in the experimental condition performed a task based on the cold-pressor test (Walsh et al., 1989), in which they were asked to submerge their hands in ice-cold water (< 3 °C) and then move pellets from the bottom of the container into a small jar inside the container. They were directed to continue this task for as long as possible. In the control (no-pain) condition, the procedure for the initial task was the same, but the participants carried it out in water at room temperature (24 °C) for a period of 90 s. The administrator then counted how many pellets had been moved to the jar so participants would feel that there was a purpose to the task. In the second part, participants’ task in the experimental condition was to maintain an upright wall squat with their backs straight and knees bent at 90° for as long as possible. Participants in the control condition had to balance on one leg with their eyes closed for 60 s, and they could switch legs to prevent the activity from becoming unpleasant.
After finishing the two tasks, participants completed two questionnaires—the Positive and Negative Affect Schedule (PANAS; Watson et al., 1988) and the Appraisal of Life Events Scale (ALES; Ferguson et al., 1999). The PANAS questionnaire measured currently experienced positive and negative emotions. The ALES questionnaire evaluated the extent to which the participants perceived the initial physical tasks as a threat and a challenge. The variables measured by PANAS and ALES were relevant neither for the main hypothesis nor for the manipulation check. Therefore, we considered excluding them. However, both questionnaires filled the time between the two parts of the experiment and influenced what the participants were thinking about between the manipulation with the independent variable and the measurement of the dependent variable. Because we wanted to do a close replication, we decided to keep the two questionnaires in the procedure, following the original experiment.
After participants had completed the questionnaires, we presented the economic game, which was identical to the one used in the original experiment, in order to assess their level of cooperation. Before the start of the game, the administrator read out the instructions, which were also printed on paper for each participant. The game was based on an economic-game paradigm (Hirshleifer, 1983). It had six rounds. In each round, the participant chose a number from 1 to 7 and wrote it down privately on paper. The higher the lowest number in the group was, the higher the payoff for the group members. However, when various numbers were taken from within the group, the participants who had chosen lower numbers had a higher payoff than the participants who had selected higher numbers. Therefore, choosing Number 1 represented the least cooperative option because it gave the participant a medium-sized payoff but also minimized the potential payoff for the other members of the group. The participant received the highest payoff by choosing 7, but only if all other players had also picked 7. This choice maximized the win for the whole group but also represented the greatest risk for the individual player. If another player chose 1, the player who chose 7 received the smallest payoff. Choosing Number 7, therefore, represented the most cooperative option. At the end of each round, the experimenter announced the lowest selected number to the whole group and then, individually and in private, informed each participant what their payoff was for the given round.
Before the start of the game, the experimenter clarified that participants would be paid at the end of the experiment for only one round, which would be randomly selected by the toss of a die. Participants were also asked to keep their choices private throughout the game. Following the original experiment, we estimated the level of cooperation as the average choice of the participant in all six rounds of the game. A higher value indicated a higher level of cooperation.
When the economic game was finished, participants completed a questionnaire measuring the level of pain and the unpleasantness of the sensations (Price et al., 1983) they experienced when carrying out the initial physical tasks. They were also asked whether they perceived the experience as having been shared. These questionnaires served as a manipulation check. To be able to describe the sample, we also asked the participants to answer sociodemographic questions. To control for possible bias, we used three open-ended questions to ask them about the perceived objectives of the research experiment, about their relationship to other participants in their group, and about whether they felt they understood the rules of the economic game. Afterward, we debriefed them. The procedure was reviewed and approved by the ad hoc ethical committee of the Faculty of Social Studies at Masaryk University.
Differences between the preregistered procedure and the procedure of the original experiment
The above-described preregistered procedure differed slightly from the original procedure (Bastian et al., 2014). The following six differences were based on our critical appraisal of the original procedure and the experience of the pilot study.
First, we presented the game as a final part of the experiment that would enable us to determine how much the participants would be paid for taking part in the research.
Second, in the original experiment, the groups that shared the physical tasks were composed of two to six people (M = 3.54). According to Bastian et al. (2014), the effect of shared pain should not be contingent on the group size. However, the number of people in the group negatively influenced the level of cooperation. In the original experiment, the groups in the control condition were larger than those in the experimental condition. This difference partially explained the effect of shared pain because the participants in the smaller groups cooperated slightly more than the participants in the larger groups. Therefore, we fixed the number of people in one group at three to avoid possible bias.
Third, on the basis of previous findings concerning gender differences and the tendency to trust other people (Feingold, 1994) and of differences in competitiveness and cooperation between men and women in homogenous and heterogeneous gender groups (Lee et al., 2016), we matched the groups so there would be a similar number with the same gender ratio for the experimental and control conditions. For example, when a group of two women and one man was randomly assigned to the pain condition, the next group with the same gender ratio was automatically assigned to the no-pain condition.
Fourth, before the start of the economic game, we gave participants two specific examples that showed them how the number they selected would influence their payoff. We also showed them two sample rounds to make sure the participants understood the instructions. (For example, the experimenter asked the participants to imagine that they had selected Number 6 and that the lowest number selected in the group was 2. She then asked them to indicate their hypothetical payoff in the payoff table for a given round.) This addition to the procedure emerged from the pre-pilot test, in which some of the participants did not understand the rules and did not recognize the impact of the selected number on the payoff for the group members. These participants selected numbers at random. Not understanding the rules and randomly choosing numbers could have resulted in an invalid measurement of the level of cooperation.
Fifth, perceived pain in the physical tasks was measured in the original experiment as a manipulation check. However, this was not a complete manipulation check because the independent variable was not pain but shared pain. From our interviews with participants in the pilot study, we found that some participants in the experimental condition might perceive the initial tasks as a competition and, therefore, might not perceive the experienced pain as shared. To control for this possibility, we added two items that were not included in the original experiment so we could assess participants’ tendency to view the initial tasks as a competition. The wording of these items was, “To what extent did you strive to perform better than the other members of your group during the initial physical tasks?” and “To what extent did you perceive the initial physical tasks as a competition?” Participants responded on a 7-point scale.
Sixth, we adjusted the payoff table (https://osf.io/uwbd4/) for the economic game to give values in the local currency.
Study 1: The Pilot
Method
Deviations from the preregistered procedure
The procedure of the pilot study slightly differed from the above-described procedure, which was preregistered for Study 2. We had completed the pilot study first, and than we modified the procedure for Study 2 using the experience from the pilot study and considering the comments of reviewers who evaluated the proposal for the replication study.
First, in the pilot study, the participants selected their numbers on laptops (not on paper). Their payoff and the lowest number in each round were shown privately on their screens. Second, we did not ask all the participants in the experimental condition whether they perceived the pain as shared or whether they perceived the physical tasks as a competition; we added these questions to the procedure of this study after we had interviewed the participants in the pilot study. Third, the average payoff was Kcˇ 41 (Czech). The payoff table was calculated by taking the values from the original payoff table, converting them at an exchange rate of $1 (Australian) to Kcˇ 16, and multiplying them by a coefficient of 0.6, which corresponded to the difference in price level between Australia and the Czech Republic. Fourth, before the start of each physical task, the experimenter said nothing about whether the tasks should be viewed as a competition. And finally, participants were instructed not to communicate at all during the economic game.
Participants
The sample consisted of 153 university students from one Czech university (101 women, 52 men; age: M = 21.29 years, SD = 2.10), who were recruited indirectly via a leaflet campaign or via social media or recruited directly in the faculty building. We included all participants in the analyses because none of them met the exclusion criteria (a preexisting friendship with any of the group members, excessive feelings of pain in the control condition or insufficient feelings of pain in the experimental condition, not understanding the rules of the economic game, or insight into the main hypothesis). The participants were placed in three-member groups that experienced either pain (n = 75, 66.7% women, 32% gender-homogeneous groups) or no pain (n = 78, 65.4% women, 30.8% gender-homogenous groups).
Results
Preliminary analyses
The pain manipulation was successful. According to the self-assessment questionnaire, the participants in the pain condition, M = 6.04, SD = 1.92, felt significantly more intense pain than participants in the no-pain condition, M = 1.54, SD = 0.98, t(151) = 18.38, p < .001, d = 2.95, 95% CI = [2.49, 3.41]. For the participants in the experimental condition, the tasks were more unpleasant, M = 5.84, SD = 1.64, than for the participants in the control condition, M = 2.01, SD = 1.30, t(151) = 15.98, p < .001, d = 2.59, 95% CI = [2.16, 3.02].
To test the effect of shared pain on cooperation, we first performed a one-way analysis of variance (ANOVA), as in the original experiment by Bastian et al. (2014). The difference between the average value of cooperation in the pain condition, M = 3.87, SD = 1.60, and the no-pain condition, M = 4.02, SD = 1.72, was very small and nonsignificant, F(1, 151) = 0.34, p = .561, d = −0.09, 95% CI = [−0.41, 0.22].
Following the small-telescopes approach (Simonsohn, 2015; for details on this approach, see Study 2), we also used two one-sided t tests (Lakens, 2017) to test whether |d| < d33%, where d33% is the effect size that could be found in the original experiment with 33% power. We estimated that d33% would equal 0.39 (using G*Power; Version 3.1; Faul et al., 2009). Both one-sided t tests were significant—upper: t(151) = 3.00, p = .002; lower: t(151) = −1.83, p = .035 (analyses were run in jamovi [Version 1.6; The jamovi Project, 2021] and TOSTER [Version 0.3.3; Lakens, 2017]), which means that the analyses failed to provide support for the existence of a relevant positive effect of shared pain on cooperation.
Hypothesis testing: multilevel model
In regard to the violation of the assumption of independence, because the participants were clustered in small groups, we ran a multilevel linear model to test the hypothesis. The individual level was represented by the individual participants, and the group level consisted of the three-member groups. The ICC was .55, which indicates that the individual choices of the participants in the economic game were strongly influenced by their membership in the small group. Adding condition (pain vs. no pain) into the multilevel model did not significantly increase the fit of the model, χ2(1) = 0.16, p = .689. The effect of shared pain on the level of cooperation was negative and nonsignificant, b = −0.16, 95% CI = [−0.94, 0.62], t(51) = −0.41, p = .687. The results showed that the shared pain experience did not lead to a greater level of cooperation in the group. Therefore, we did not find support for the hypothesis that shared pain promotes cooperation between group members.
There was, however, a significant difference between the intercepts for the three-member groups, which suggests that the linear mixed model provides a more adequate analysis for this type of data. The results of the pilot study stressed the need for a replication study because we were unable to replicate the effect reported in the original experiment when using a sample nearly 2.5 times larger. The pilot study also enabled us to test the procedure and modify it on the basis of the experiences of some of the participants, which is why we added the questions on perceived competition and shared pain to the final design of the preregistered study.
Study 2: Preregistered Direct Replication
Method
After completing the pilot study, we submitted a proposal for a Preregistered Direct Replication, which was approved by Psychological Science. After modifications described above, the design and analysis plan for the replication were preregistered on OSF (https://osf.io/4xsz2/). One of the reviewers was the author of the original study (Bastian et al., 2014), whose comments allowed us to bring the procedure even closer to that of the original experiment.
Because the preregistered replication took place in Slovakia, we adjusted the payoff table for the economic game to give values in euros (see https://osf.io/uwbd4/). We used the data from the pilot study to reveal the average result of the economic game, and we set the payoff for this result to €7.70 (equivalent to $9.27 U.S. or $12.09 Australian), because this should be the average payoff for people who participate in economic experiments at our institution.
We collected the data in fall 2020 between the first and second wave of the COVID-19 pandemic. To prevent the potential transmission of the virus, we modified the procedure to ensure the safety of administrators and participants. Therefore, all administrators and participants wore face masks during the experiment.
We estimated the sample size using the small-telescopes approach (Simonsohn, 2015). According to Simonsohn, if the true effect is zero, the replication study needs 2.5 times as many observations as the original experiment to have approximately 80% test power to reject d33%, where d33% is the effect size that could be found in the original experiment with 33% test power. Therefore, the estimated minimum sample size was 156 participants (i.e., 78 in each condition).
A simulation in GLIMMPSE software (Kreidler et al., 2013) showed that a sample size of 156 is also sufficient to find support for a main effect (d) of 0.38 and higher with 80% power (α = .05) in a multilevel analysis. Taking into account these analyses and the possible reduction of the sample by 10% because of preset exclusion criteria (see below), we planned to recruit 174 participants.
We recruited 174 university students in Slovakia by means of social media and online school groups. We divided them into 58 three-member groups in which they completed the procedure. Before the analyses, we excluded 16 participants according to the preset criteria that were described in the preregistration protocol. We excluded one participant because he said he did not understand the rules of the economic game. We also excluded four groups of participants (i.e., 12 participants) because of preexisting friendships between the group members and one group of participants (i.e., 3 participants) because of a spoiled procedure. The manipulation of the independent variable was apparently unsuccessful in this group, given that one participant reported intense pain (> 3 SD above the mean in the control condition) despite being in the no-pain condition. This group also contained the only participant who was able to surmise the hypothesis that we were testing in the experiment. No participants were excluded because of the final criterion of sharing the chosen number with other group members during the economic game. The final sample consisted of 158 people, 78 of whom were assigned to the pain condition (67.9% women; age: M = 22.85 years, SD = 2.97) and 80 to the no-pain condition (63.8% women; age: M = 22.76 years, SD = 2.59).
Results
Preliminary analyses
Table 1 shows the descriptive statistics for the sample. The manipulation of the independent variable was successful, as indicated by the participants in the experimental condition reporting significantly more intense pain than participants in the control condition, t(101.46) = 17.96, p < .001, d = 2.83, 95% CI = [2.28, 3.37]. Participants perceived the physical tasks as more unpleasant in the experimental condition than in the control condition, t(117.73) = 16.79, p < .001, d = 2.65, 95% CI = [2.13, 3.17]. Participants in the experimental and control conditions did not differ either in striving for better performance, t(156) = 0.51, p = .613, d = 0.08, 95% CI = [–0.23, 0.39], or in perceiving the physical tasks as a competition, t(156) = 0.46, p = .644, d = 0.07, 95% CI = [−0.24, 0.39].
Descriptive Statistics for the Sample in Study 2
Note: N = 158 (n = 78 in the no-pain condition and n = 80 in the pain condition). CI = confidence interval.
To compare our results with those of the original experiment, we used a one-way ANOVA. We did not find a significant difference between the experimental and control groups, F(1, 156) = 0.18, p = .676, and observed only a negligible effect of shared pain, d = 0.07, 95% CI = [−0.24, 0.38].
Following the small-telescopes approach (Simonsohn, 2015), we also tested for equivalence of means (Lakens, 2017). We used two one-sided t tests to determine whether the effect of shared pain is smaller than a d33% of 0.39. 1 Both one-sided t tests showed a significant result—upper: t(155) = 2.03, p = .022; lower: t(155) = −2.87, p = .002 (analyses were run in jamovi [Version 1.1; The jamovi Project, 2019] and TOSTER [Version 0.3.3; Lakens, 2017] ). Therefore, the analysis failed to provide support for the existence of a relevant positive effect of shared pain on cooperation.
Hypothesis testing: multilevel model
Following the preregistration, we tested the hypothesis about the effect of shared pain on cooperation using a linear mixed model with condition as a factor, presence in a group of three participants as a cluster, and level of cooperation as a dependent variable (α = .05). The ICC was .56, which shows that more than half of the variance in the level of cooperation can be explained by membership in the three-member groups. Adding a predictor (i.e., condition) into the model did not significantly increase the fit of the model, Δχ2(1) = 0.07, p = .785. The effect of shared pain on cooperation was very small and nonsignificant (see Table 2). Therefore, the analysis did not provide support for the hypothesis that shared pain promotes cooperation between group members.
Results of Linear Mixed Models Predicting Cooperation From Condition (Study 2)
Note: For both models, the variance of residuals was 1.42 (SE = 0.19), p < .01, and the variance of intercepts was 1.80 (SE = 0.45), p < .01. Model 1: −2 × log likelihood = 586.81. Model 2: −2 × log likelihood = 586.73.
Nonpreregistered analyses
In the pilot study, the interviews with participants revealed that the effect of pain might be attenuated by the perception that the experimental tasks were some kind of competition. If participants perceived the tasks as a competition, it could reduce their perception of the pain being shared. To control for this effect, we added the perception of the task as a competition in the model as a covariate, and we also included its interaction with condition. Neither adding the perception of task as a competition, Δχ2(2) < 0.01, p = .964, nor adding the interaction, Δχ2(3) = 0.13, p = .722, improved the fit of the multilevel model. Both effects were very small and nonsignificant (see Table 3). Although the perception of the tasks as a competition varied across the sample, it was similar in both conditions (see Table 1) and did not affect the effect of pain on further cooperation.
Results of Linear Mixed Models Predicting Cooperation From Condition, Controlling for Perception of the Task as a Competition (Study 2)
Note: For both models, the variance of residuals was 1.42 (SE = 0.19), p < .01; the variance of intercepts was 1.79 (SE = 0.45), p < .01, for Model 3 and 1.78 (SE = 0.45), p < .01, for Model 4. Model 3: −2 × log likelihood = 586.73; Model 4: −2 × log likelihood = 586.60.
Meta-Analysis
As a part of our supplementary analyses, we conducted a meta-analysis to compute the average effect of shared pain on cooperation in the available studies. In addition to our studies and the study by Bastian et al. (2014), we also included two experiments by Qi et al. (2020).
We conducted a meta-analysis with restricted maximum likelihood estimation using jamovi (Version 1.6; The jamovi Project, 2021) and the R package metafor (Viechtbauer, 2010). When we included the five studies that used the cold-pressor task to manipulate shared pain, the average effect differed significantly from zero, z = 1.97, p = .049, Hedges’s g = .39, 95% CI = [.00, .77] (see Fig. 1), and the results indicated that the true effect could be negligible to moderate. Because all five studies had used a similar design and the same procedure to measure cooperation, we can also report that the raw mean difference in cooperation between the pain and the no-pain condition was 0.53, 95% CI = [0.03, 1.03], on a scale ranging from 1 to 7. When we also included Experiment 3 of Bastian et al. (2014), in which shared pain was manipulated by having participants eat chili peppers, the result was similar, z = 2.41, p = .016, Hedges’s g = .40, 95% CI = [.07, .73] (see Fig. 2).

Forest plot showing the mean effect size from the five studies that used the cold-pressor task. Results are also shown for the random-effects (RE) model. The size of the squares is proportional to the weights used in the meta-analysis. The effects reported here differ slightly from the effects presented in the original studies because they were computed from the published means and standard deviations rounded to two decimal places. Error bars represent 95% confidence intervals (CIs).

Forest plot showing the mean effect size from the six studies that focused on shared pain and cooperation. Results are also shown for the random-effects (RE) model. The size of the squares is proportional to the weights used in the meta-analysis. The effects reported here differ slightly from the effects presented in the original studies because they were computed from the published means and standard deviations rounded to two decimal places. Error bars represent 95% confidence intervals (CIs).
General Discussion
Our studies failed to replicate the findings of Bastian et al. (2014), which suggested that shared pain increases cooperation among group members. A nonpreregistered pilot study and a preregistered replication, both with a sample size that was 2.5 times larger than that of the original experiment, found an effect that was close to zero and nonsignificant. There are three possible explanations for not finding a relevant effect in our replication study: methodological problems with the replication study, an overestimation of the true effect size in the original experiment, and differences between the studies that moderated the effect (Wicherts, 2018, as cited in Van der Cruyssen et al., 2020, p. 465).
The first explanation may be applicable to Study 1 (the pilot study), in which our experimenter did not emphasize that the physical tasks should not be viewed as a competition and in which the participants were instructed not to communicate at all during the economic game. These eventualities could have reduced the perception of pain as being shared and limited the opportunities for collaboration during the economic game. However, the procedure in Study 2 (preregistered direct replication) was reviewed by the author of the original study, so the experimenter’s instructions did not significantly differ from those given in the original experiment, and the result was similar to that of Study 1. In Study 2, the participants used face masks to protect themselves from the spread of the COVID-19 virus. It is possible that not perceiving the facial expressions of other group members during the painful tasks may have suppressed the experience of shared pain. Although we are not aware that such an effect exists, we cannot rule it out.
With regard to the second explanation (i.e., that the true effect size obtained in the original study may have been an overestimate), we consider this a plausible explanation. The analysis plan was not preregistered in the original experiment, and the sample size was small. The CI for the main effect was wide, and the lower bound indicated the possibility of a marginal effect. When we considered the multilevel data structure in the original experiment, the effect of shared pain became nonsignificant. There were three other replications that successfully replicated the effect of shared pain on cooperation, but all of them had rather small samples, none of them were preregistered, and two of them (Bastian et al., 2014, Experiment 3, and Qi et al., 2020, Experiment 1) presented barely significant effects. Furthermore, in one of them (Bastian et al., 2014, Experiment 3), the effect became nonsignificant when models controlled for group size, gender inequivalence, or the multilevel structure of the data. The other two replications (Qi et al., 2020) did not control for group size or the gender homogeneity of the groups, and they also did not consider the multilevel structure of the data. As our supplementary meta-analyses suggested, it is possible that the true effect of shared pain on cooperation is much lower than the effect presented in the original experiment and that the original experiment found such a large effect partly because of chance and partly because of the inequality of the experimental groups that favored cooperation in the pain condition.
With regard to the third explanation (i.e., that the effect of shared pain on cooperation varies across the different conditions and contexts), the most notable difference between the original experiment and our replication study was the cultural context. People in individualistic cultures may tend to act more competitively or interpret certain situations as more of a competition than people in collectivistic cultures do (Leibbrandt et al., 2013; Nukić & Braje, 2017). This may have affected whether the participants perceived the physical tasks as a shared painful experience or an individual painful experience, which in turn could have weakened the observed effect. When comparing the countries on the dimension of individualism using Hofstede’s (2021) model, we found that Australia’s scores were notably higher than those of Slovakia and the Czech Republic. Considering the higher tendency of individualistic countries to interpret situations as more of a competition, we can deduce that higher values of competitiveness would be observed in Australian culture. Therefore, it is reasonable to believe that if the effect of shared pain on cooperation exists, it would have been observed in Slovakia. However, the average level of cooperation was similar in the Australian and Slovak samples (and also in the Czech sample) and therefore seems not to have influenced the results. Nevertheless, there might be other cultural effects that we are not aware of that might weaken the effect of shared pain in the Central European context or strengthen it in Australia and China.
Although we failed to replicate the effect of shared pain on cooperation, we cannot conclude that the effect does not exist. Given all the available published studies, including the new nonpreregistered replication with Chinese participants (Qi et al., 2020), it seems that there might be a weak to moderate effect that is unstable across various contexts. Further preregistered replications are needed to focus on the possible moderators of the effect of shared pain and to provide data for meta-analysis that would be less subject to publication bias than our supplementary meta-analysis was.
We believe that the main limitation of the original experiment and our replications lies in the ecological validity. We assume that the theoretical background provides support for a relevant effect of negative shared experience on later prosocial behavior, especially in the case of physically or emotionally intense and challenging experiences, such as natural disasters (Vezzali et al., 2016), war trauma and loss (Elder & Clipp, 1988), and physical injuries (Turner & Wainwright, 2003). However, the above-mentioned experiences represent significant and character-shaping events in a person’s life, which cannot be compared with the cold-pressor task or squats. The fact that we found that the effect of pain induced by the cold-pressor task and squats might not exist or could be significantly smaller than in the original experiment does not mean that shared pain does not promote cooperation at all. It is possible that the effect of shared pain is more evident when the shared painful experience is more intense than in standardized experimental tasks.
Footnotes
Acknowledgements
We thank Lucia Kupkovicova, Alexandra Surinova, Valeriia Yasenko, and Margarita Timonteyeva for their help with data collection.
Transparency
Action Editor: Patricia J. Bauer
Editor: Patricia J. Bauer
Author Contributions
J. Prochazka, K. Parilakova, and P. Rudolf contributed equally to the study and share first authorship of this article. P. Rudolf, J. Prochazka, V. Bruk, R. Jungwirthova, K. Parilakova, and S. Fejtova developed the study concept. P. Rudolf, V. Bruk, and R. Jungwirthova participated in Study 1 as part of their student project. K. Parilakova participated in Study 2 as part of her bachelor’s thesis. Data for the pilot study were collected by P. Rudolf, V. Bruk, R. Jungwirthova, K. Parilakova, and S. Fejtova. P. Rudolf, V. Bruk, and R. Jungwirthova analyzed and interpreted the Study 1 data under the supervision of J. Prochazka and M. Vaculik. P. Rudolf and J. Prochazka prepared the preregistration, and R. Masaryk and M. Vaculik provided critical revisions. The data collection for Study 2 was coordinated by K. Parilakova, J. Prochazka, and R. Masaryk. K. Parilakova and J. Prochazka analyzed and interpreted the data from Study 2 and drafted the final manuscript, and P. Rudolf, M. Vaculik, and R. Masaryk provided critical revisions. All authors approved the final version of the manuscript for submission.
