1 Introduction

We analyze the cooperative behavior of individuals in a Public Good game, with a focus on how the relationships among different generations of same-family members (e.g., family ties among youth, parent, and grandparent) are related to their cooperation. The cooperation of individuals has been widely analyzed in different fields. The analysis of cooperation by experimental science has a reference milestone in Nowak and May (1992), who developed the Prisoner’s Dilemma (PD) to prove the survival of the cooperative agent in a network, and since then different versions of the PD have been used to analyze cooperative behaviors. Public Good games, in which individuals secretly choose how many of their private tokens to put into a public pot, have also been used to analyze the cooperative behavior of individuals, finding that individuals contribute more to the Public Good than expected, with observed cooperation declining over time (see, for a survey, Chaudhuri 2011).

Among the mechanisms that may help to explain cooperation among individuals, Becker (1962, 1974) established the generosity criterion guiding cooperation or transfer of resources from donor to recipient, with this approach contrasting with alternative hypotheses (Cox 1987), under which donors expect some sort of reciprocity from the receivers in a cooperative context. Furthermore, family relationships (or family ties) may also help to explain cooperation among individuals, so that those who share genes have a cooperative behavior motivated by generosity. This reasoning has been mathematically formalized through Hamilton’s rule (Hamilton 1964), according to which cooperation between two individuals will take place if the benefit obtained, corrected by the kinship relationship between these individuals, exceeds the cost of giving or helping. This concept has been tried and tested empirically in the literature (West et al. 2011; Oli 2003; Peters et al. 2004; Waibel et al. 2011).

Prior research has specifically analyzed how cooperative behavior changes with age and/or across generations; for example, experimental research has shown that younger children are less altruistic (Peters et al. 2004; Fehr et al. 2008; Fehr et al. 2011). Charness and Villeval (2009) and Gutiérrez-Roig et al. (2014) have included subjects of different ages who were involved in the same experimental set-up, to test cooperativeness differences across generations, with their main findings being that the elderly are more cooperative. In both studies, individuals of different ages play within the same generation, and while the repeated PD has been used extensively in experiments with adults (Blake et al. 2015), PD experiments with children remain rare, with some few exceptions that include Blake et al. (2015) and Molina et al. (2013).

Prior research has also analyzed cooperation in families.Footnote 1 A trust game played by couples and strangers is shown in Görges (2015), finding that women are more likely to cooperate (e.g., give up their income autonomy and perform the unpaid task) when playing with a partner rather than with an unfamiliar man. Beblo and Beninger (2017) test in-couple income pooling in Germany, and find that income pooling, and thus cooperation, is more common among couples in which the spouses’ socio-economic characteristics are more similar. Cochard et al. (2017) test for a willingness to cooperate and share income by men and women who are either in couple with each other or complete strangers, finding that lack of preferences for joint income maximization, having children, and being married lead to higher defection rates in the social dilemma. Dauphin et al. (2017) test the collective rationality (e.g., consumption efficiency) within households in Burkina Faso, using monogamous and polygamous households. They find that while the collective rationality within monogamous households is not rejected, it is rejected in polygamous households. Consumption efficiency is based on Pareto efficiency, where the individuals within the household cooperate, and thus this evidence serves as evidence that households are not as cooperative as previously hypothesized.

The issues of family, age/generation, and cooperation have all been studied by Peters et al. (2004), Reynolds (2015) and Porter and Adams (2016). Peters et al. (2004) place parents and children in a laboratory to participate in a Public Good game, where subjects are endowed with an initial income and must choose how to allocate that income between private and group accounts. The study finds that parents may behave more altruistically than do their children in a family setting, and that parents and children contribute more to a Public Good when grouped with family members than when grouped with strangers. Reynolds (2015) applies two games to test for Pareto efficiency and bargaining in a sample of adolescent women who live with their mothers in Brazil, and the results confirm a cooperative relationship. Porter and Adams (2016) study the motivations behind transfers between two generations (adult children and their parents) in an experimental setting, with results indicating that a greater proclivity for giving appears when parents, rather than strangers, are the recipients of transfers while playing a series of dictator games.

Against this background, we contribute to the literature on the factors affecting the cooperative behavior of individuals, by focusing on the relationship between family ties, age/generation, and cooperation. In particular, we focus on how family ties are related to cooperation, and also on whether older generations cooperate more than younger ones. To that end, in an experimental setting, three generations of individuals from the same family (youth, parent, and grandparent) play a repeated Public Good game in three different treatments: one in which three members of the same family play each other (family), a second with the youth and two non-family members, but preserving the previous generational structure (intergenerational), and a third in which three randomly-selected individuals play each other (random).

Several conclusions can be drawn from our experiment. We find that the level of cooperation of the youths is significantly lower than that of the parents and the grandparents, consistent with prior literature showing that young people are less cooperative (Peters et al. 2004; Charness and Villeval 2009; Gutiérrez-Roig et al. 2014). Furthermore, in comparison with a setting where players are randomly assigned to a group, family ties have a positive relationship to the contributions to the Public Good, with this effect being greater for the youths than for the parents and the grandparents. Also, youths and parents contribute more to the Public Good when the group has a generational structure, while grandparents are not affected in this behavior. This evidence indicates that the behavior of parents and grandparents is not greatly affected by the treatment, while youths contribute significantly less when not playing with their own family members. Finally, the differences in the contribution by youths, on the one hand, and parents and grandparents, on the other, are larger when playing with randomly-assigned players.

The results shown are interesting, and represent an important contribution to the literature. Despite that we cannot know the mechanisms behind the differential behavior of youths in comparison to parents and grandparents, we can hypothesize several possible explanations: an age effect (i.e., older individuals may have more income, increasing their level of generosity), a cohort effect (i.e., individuals become less generous towards strangers) or a selection effect (i.e., those who become parents are more generous). Furthermore, family ties may have a positive effect on the cooperation of individuals, which may serve as a mechanism to stabilize cooperation in different settings. The evidence presented here opens interesting lines of research.

The rest of the paper is organized as follows. Section 2 describes the inter-generational Public Good game. Section 3 presents the lab-in-the-field results. Section 4 shows the empirical strategy and the estimation of results, and Section 5 sets out our main conclusions.

2 Experimental design and implementation

2.1 Design

Our design involves observing individuals participating in a repeated Public Good game, where each volunteer participates in three different 3-player games, corresponding to three different treatments: (a) one in which three members of the same family (youth, parent, and grandparent) play each other (family treatment, F); (b) a second with three non-related members (strangers) but preserving the generational structure with one youth, a parent, and a grandparent (intergenerational treatment, IG); and (c) a third in which three randomly-selected players are assigned and play each other (random treatment, R).Footnote 2 The number of rounds (decision periods) is set at 10 for each treatment, although this is not known a priori by the players. To account for the possibility of order effects, volunteers are assigned to 2 different groups, by arrival order, and the first group (n = 34 families) play treatments in the order F-IG-R, while the second group (n = 21 families) play treatments in the opposite order, R-IG-F. Arrival order is completely random, as all families are called at the same time.

As argued by Peters et al. (2004), to ensure that behaviors when playing with strangers and with their families are comparable, each participant’s group in the strangers’ treatment must have the same composition as the participant’s family treatment, and thus the only difference between treatments F and IG is that members of the same group are not from the same family as in treatment IG. Furthermore, treatment R eliminates the effects related to matching so that it constitutes a null-model in which there are neither kinship nor (inter)generational influences. This allows us to compare non-random treatments F and IG, in which there can be family and intergenerational effects, with the null-model treatment R that provides a baseline. This comparison provides a method to evaluate respectively the influence of kinship and generation on cooperative behavior. Although there is no consensus on the merit of random treatments in Public Good game experiments (Andreoni and Croson 2008), many experimental studies of cooperation show that non-random treatments exhibit a much higher level of cooperation than randomized treatments (see Duffy and Ochs 2009).

In each round of each treatment (3 treatments × 10 rounds = 30 rounds), participants are endowed with 10 monetary units (ECUs), and their decision problem is how to allocate this amount between a private account and a group account. For each round, the sum of the contributions made by all 3 players is calculated, and the total contribution is multiplied by a factor of 1.5 and then shared equally by the 3 players. The obtained payoff in each round is the sum of this share plus the ECUs held back, i.e., not invested in the common fund. In each round, players are informed about how many units the other two players in the same group have contributed to the Public Good in the previous round (except in the first round of each treatment).

2.2 Recruitment

Fifty-five families participated in the experiment, accounting for 165 volunteers: 55 volunteers aged between 17 and 19 years old, plus one of their parents and one of their grandparents. For the recruitment, different channels were used to advertise the experiment, including advertisements in the principal newspapers of the region (Heraldo de Aragón, El periódico de Aragón), ads in the city council, the University of Zaragoza, and regional government web pages, and posters sent to local high schools. The Ibercivis Foundation was in charge of the diffusion of the experiment, with the aim of recruiting individuals from the general population of the city of Zaragoza (Spain). The selection process consisted of attracting sets of three volunteers of different generations of the same family, by filling-in an on-line form. Applications were open to any set of three relatives meeting the above conditions, regardless of their social status or other demographic variables. Thus, the entire population of Zaragoza was open to participation in the experiment, not only young students (which is important in our framework, since students may have participated in economics experiments before, or have had some training in economics). Furthermore, the type of game was only known by the volunteers after arriving at the lab, and the aim of the experiment was never revealed to the volunteers. Unfortunately, we do not have information on whether or not the young volunteers were students, but we have reason to believe that the lower contribution rates among youths relative to parents and grandparents (as shown above) is not due to previous experience with Public Good games and/or their studies in economics.

One third (33%) of the players are male, and the average age is 48.09 years. By group of players, 35, 38, and 27% of youths, parents, and grandparents are male, with an average age of 17.44, 50.62, and 76.22, respectively. Furthermore, 38% of the players were assigned to the second session. As a fundamental ethics statement, the anonymity of all participants in the experiments reported was always preserved (in accordance with the Spanish Law on Personal Data Protection) by assigning them, randomly, a username that would identify them in the system. No association was ever made between their real names and the results. As is standard in socio-economic experiments, no ethical concerns are involved, other than preserving the anonymity of participants.

2.3 Implementation

We ran 3 treatments in each of the 2 sessions. All sessions were conducted using a local area network of the Institute for Biocomputation and Physics of Complex Systems at the University of Zaragoza, where a total of 102 computers were available for the experiment. The experiments were performed on a web-based platform, and results were gathered in a database for further analysis. Individuals could not interact during the experiment, and non-personal data only was collected. Before the first treatment of each session, detailed instructions were read aloud and shown on the screens, that participants had to read and accept before beginning the experiment. After reading the instructions, participants had one game to get acquainted with the computer program.Footnote 3 At the end of the experiment, each player received the sum of the payoffs corresponding to all rounds of the three treatments in which he/she participated, converted into Euros with a factor set by the research team to adjust to the available budget (1,720€). Typical earnings ranged from 8 to 12 euros, including a 5-euro attendance fee per person.Footnote 4 The typical duration of a session was around 20 min.

3 Lab-in-the field results

Figure 1 shows, by round number and generation, the mean contribution when individuals play in the different treatments. We also show confidence intervals of the contributions at the 95% confidence level. Panel A shows results when individuals play in the F treatment. We do not observe between-group differences in contributions to the Public Good when individuals play in the F treatment, as confidence intervals intersect in most rounds. Panel B shows the average contributions when individuals play in the IG treatment. In this case, we do not find statistically significant differences (e.g., confidence intervals do intersect) in the contributions parents and grandparents make during this treatment, and the same applies when parents and grandparents play in the R treatment (Panel C). But youths make relatively lower contributions to the Public Good in comparison to parents/grandparents when they play in both the IG and F treatments. In the IG treatment, youths make lower contributions in comparison to grandparents in rounds 4, 6, 8 and 9, and in comparison to parents in rounds 4 and 5. Furthermore, in the R treatment, youths make lower contributions in comparison to grandparents and parents in rounds 4, 5, 6, 8, 9 and 10. Thus, while we do not find inter-generational differences in contributions during the F treatments, youths make lower contributions to the Public Good in both the IG and R treatment.Footnote 5

Fig. 1
figure 1

Contributions to the Public Good, by treatment. Notes: Data come from the experiment “Colabora con la ciencia en Familia” (http://www.ibercivis.es/projects/colabora-con-la-ciencia-en-familia/). “Contribution in monetary units” are calculated per round, averaged over all the participants for the same generation. Confidence intervals calculated as \(\overline X {\mathrm{ \pm }}1.96 \ast SE\), where \(\overline X\) represents the mean value and SE represents the standard error

Full size image

Figure 2 shows, for each generation and round number, the mean contribution of players according to the treatment. Panel A shows the average contribution of youths in the three treatments, Panel B the average contribution of parents in the three treatments, and Panel C the average contribution of grandparents in the three treatments. We also show confidence intervals of the contributions at the 95% confidence level. We observe that, for the three generations, the contribution made by the players in the F treatment is comparatively higher than the contribution made in the IG and R treatments, for all rounds. In the case of youths, the confidence intervals of the F treatment do not intersect with the confidence intervals of the IG and R treatments in rounds 2 to 5 and 7 to 10, in rounds 7 and 10 for parents, and in round 5 for grandparents. Thus, we can assume that the contributions to the Public Good made during the F treatment are higher than in the IG and R treatments, especially in the case of youths, where differences are larger, and in higher numbers of rounds.Footnote 6

Fig. 2
figure 2

Contributions to the Public Good, by generations. Notes: Data come from the experiment “Colabora con la ciencia en Familia” (http://www.ibercivis.es/projects/colabora-con-la-ciencia-en-familia/). “Contribution in monetary units” are calculated per round, averaged over all the participants for the same generation. Confidence intervals calculated as \(\overline X {\mathrm{ \pm }}1.96 \ast SE\), where \(\overline X\) represents the mean value and SE represents the standard error

Full size image

Several conclusions can be drawn from this analysis. First, we observe that parents and grandparents contribute more to the group account (Fig. 1), although differences across generations disappear when the individuals play in the F treatment (Fig. 1). Youths tend to cooperate less than the other generations when they play in the IG and R treatments. We also observe that the level of cooperation is higher when individuals play in the F treatment for the three generations than when playing in the IG and R treatments, a difference that is especially large for the group of youths (Fig. 2).Footnote 7 We interpret this result as evidence that family ties are positively related to cooperation. Third, from Fig. 2, we observe that in the absence of family ties, inter-generational interactions do not change cooperative behavior, as the level of cooperation in the IG and R treatments is similar for the three generations.

When we analyze the intra-personal (within) variation in contributions by treatment, we observe that the variation in contributions is lower in the F treatment, in comparison with the IG and R treatments. For youths, parents, and grandparents, the variation in contributions during the F treatment is 2.14, 1.55 and 2.03, respectively. For the IG treatment, the variations are 2.32, 1.78 and 2.06, respectively, and for the R treatment the variations are 2.27, 1.69 and 2.06, respectively. Thus, this variation in contributions is lower in the F treatment, which may indicate that family ties serve as a mechanism to stabilize the cooperation of individuals. Parents and grandparents appear to react less to the game setting, as their intra-personal variation is lower in all the treatments. The intra-personal variations of youths, considering the three treatments together, is 2.75, much higher than the variations of parents (2.21) and grandparents (2.30), indicating that youths vary their behavior more, in comparison to parents and grandparents.

When we focus on the evolution of the contributions made by the players over rounds, prior research on voluntary contribution games has found that cooperation (contributions) decreases over time (Andreoni 1988; Isaac and Walker 1988; Palfrey and Prisbrey 1996; Fischbacher et al. 2001;Andreoni and Croson 2008; Muller et al. 2008; Fischbacher and Gätcher 2010). One possible explanation for this behavior is that strategies depend on the history of play and therefore may cause players to change their actions over the course of the game. Another is that participants slowly begin to better understand the game and refine their strategies accordingly.

To determine whether cooperation decreases over time in our experiment, we first compute the correlation between contributions and the number of rounds, for each generation and treatment. For the group of youths, the correlation between the contribution and round number is −0.09 for the R treatment, while for the F and IG treatments the correlations are not statistically significant at the 95% confidence level. For the group of parents, the correlation between the contribution and round number is −0.10 for the IG treatment, while for the F and R treatments the correlations are not statistically significant at the 95% confidence level. For the group of grandparents, there are no statistically significant correlations at standard levels in any of the treatments. Thus, it seems that the level of cooperation remains relatively constant over time.

One specific mechanism that appears to explain the cooperation of individuals is that of conditional cooperation (Fischbacher et al. 2001; Fischbacher and Gächter 2010). According to this mechanism, many individuals’ propensity to cooperate is dependent on others’ cooperative behavior, and is considered a consequence of certain fairness preferences, such as altruism, commitment, reciprocity (Croson 2007) or warm-glow (Andreoni 1990; Crumpler and Grossman 2008). Thus, the (moody) “conditional cooperation” assumption implies that individual cooperation is a function of their own history of play and the average cooperation level of their neighbors (Gracia-Lázaro et al. 2012; Grujić et al. 2012). To test this hypothesis, we analyze whether players react to the contributions they observed in the previous round, by assuming that players have a one-step memory. In doing so, we examine the variation of contributions (i.e., the change in individual contributions) as a function of the difference between the own and partners’ average past contributions.

Figure 3 plots the change in player’s contributions (y-axis), as a function of the difference between the own contribution and other players’ contributions in the previous round (x-axis).The change in contribution is measured as the difference between the player’s contribution and the contribution in the previous round, with a positive value indicating an increase in the contribution in comparison to the previous round. The difference between own and other players’ contributions in the previous round considers both the own and other players’ average contributions, with negative values indicating that, on average, other players contributed more to the Public Good in the previous round than the player under consideration.Footnote 8 By generation and treatment, we average for all the cases with the same difference (e.g., −1) between own and other player’s contributions, the change in the contribution.Footnote 9 We observe in all cases a very strong positive dependence of the contribution increment on the difference between own and partners’ last action, for all the generations and treatments, as the linear fits have positive slopes. Thus, we observe that the contribution to the common good is strongly conditioned by both the player’s previous action and the last observed contributions.

Fig. 3
figure 3

Change in contribution in relation to other player’s previous contribution. Notes: Data come from the experiment “Colabora con la ciencia en Familia” (http://www.ibercivis.es/projects/colabora-con-la-ciencia-en-familia/). Contributions depend on own and partners’ last action. Own contribution increment (y-axis) as a function of the difference between averaged partners’ contribution and own contribution in the last round

Full size image

4 Empirical strategy and results

We estimate models that take into account the unobserved heterogeneity of individuals, since there may be certain unobserved factors at the individual level that are correlated with the level of contribution, which would bias the results. For instance, past personal experience, the mood on the day of the experiment, or personal attitudes towards justice, equity, and confidence, all may condition the decisions of individuals in our experiment. Thus, we estimate a random-effects (RE) linear model to control for the unobserved heterogeneity of individuals, using the following equation:Footnote 10

$$C_{ij}{\mathrm{ = }}\alpha _i{\mathrm{ + }}\beta _1F_i{\mathrm{ + }}\beta _2IG_i{\mathrm{ + }}\beta _3Parent_i{\mathrm{ + }}\beta _4Grandparent_i{\mathrm{ + }}\delta X_i{\mathrm{ + }}\varepsilon _{ij}$$
(1)

where Cij represents the decision (contribution) by individual “i” in round “j”, and αi represents the individual effect. The dependent variable is a variable that measures the amount given by individual “i” in round “j” to the common fund. We include two dummy variables to indicate whether player “i” is playing in the F treatment or the IG treatment, and thus the reference group is players in the R treatment. According to our hypothesis (that family ties are positively related to the level of contributions), we should obtain that β1 > 0. We include two dummy variables to indicate whether the player belongs to the group of grandparents or parents (ref.: youths). The vector Xi includes the gender (1 = male, 0 = female) of participant “i”.Footnote 11 We also control for the number of the round to account for possible learning effects, and we include a dummy variable to control for whether the player belongs to the second group (1) or not (0).Footnote 12εij is a random variable (standard error) that represents unmeasured factors, capturing all the factors for which we do not have information, and that may affect those participant decisions, and we assume that the error term follows a normal distribution, εij~N(0,σ2).

Figure 2 shows that the increase in cooperation when playing in the F treatment, in comparison with playing in the R treatment, is greater for youths than for parents and grandparents. Thus, it could be that there are inter-generational differences in how treatments affect individuals. Thus, we need to reveal whether different treatments affect the generations differently, and to that end we estimate Equation (2) as follows:

$$\begin{array}{ccccc}\\ C_{ij} = \hskip-6pt & \alpha _i{\mathrm{ + }}\beta _1F_{ij}{\mathrm{ + }}\beta _2F_{ij} \ast Grandparent_i{\mathrm{ + }}\beta _3F_{ij} \ast Parent_i \hfill \\ & {\mathrm{ + }} \beta _4IG_{ij}{\mathrm{ + }}\beta _5IG_{ij} \ast Grandparent_i{\mathrm{ + }}\beta _6IG_{ij} \ast Grandparent_i \\ & + \beta _7Parent_i{\mathrm{ + }}\beta _8Grandparent_i{\mathrm{ + }}\delta X_i{\mathrm{ + }}\varepsilon _{ij}\hfill \\ \end{array}$$
(2)

where Cij represents the decision (contribution) by individual “i” in round “j”, and αi represents the individual effect. We include two dummy variables to indicate whether player “i” is playing in the F treatment or the IG treatment, and we include their interactions with two dummy variables to indicate whether the player belongs to the group of grandparents or parents (ref.: youths). We also include the variables to indicate whether the player belongs to the group of grandparents or parents (ref.: youths). Thus, the reference group is youths in the R treatment. The vector Xi represents the rest of the factors included in Equation (1), and εij represents the error term in the equation.

Column (1) of Table 1 shows the results of estimating Equation (1) for the sample of players. We observe a positive and statistically significant coefficient for the variable representing the F treatment. Playing in the F treatment is related to an increase of the average contribution by 1.72 monetary units, in comparison with the R treatment. When we focus on the IG treatment, in comparison with the R treatment, we observe that the coefficient of the dummy variable is not statistically significant. Thus, we find that, in comparison with the R and IG treatment, playing with family members increases the contribution of players to the Public Good, indicating that family ties are positively related to the contribution of the players. Furthermore, the coefficients for the dummy variables controlling for whether the player belongs to the group of grandparents or parents are positive and statistically significant at the 99% level. Given that the reference group are youths in the R treatment, we observe that grandparents contribute, on average, 1.15 more monetary units, and parents 0.90 more monetary units, when they play in the R treatment, in comparison to the contributions of youths to the Public Good in the same treatment. This is consistent with Fig. 1, as it shows that youths contribute less to the Public Good in comparison with parents and grandparents, especially in the IG and R treatments.

Table 1 RE results for contributions
Full size table

Although prior evidence has shown that the cooperative behavior of males and females is different (Thomas 1990; Lundberg et al. 1997; Molina et al. 2013; Cochard et al. 2017), the coefficient for males is not statistically significant in our sample.Footnote 13 Furthermore, we find that the coefficient controlling for whether individuals were assigned to the second group (i.e., Second group in experiment) is positive and statistically significant at the 99% level, indicating that the contributions of the second group are comparatively higher than in the first group, independent of the treatment, and such differences cannot be attributed to the difference in the order of treatments. This difference may be due to a session effect, since, perhaps, individuals could create ties before the experiment, given their higher waiting time, or it is possible that the students knew each other before participating in this specific session. However, the positive relationship between family ties and cooperation is present after controlling for this session (i.e., order of treatments) effect, which indicates that this result holds net of this session effect.

Column (2) of Table 1 shows the results of estimating Equation (2) on the contribution to the Public Good. Considering that the reference group is youths in the R treatment, our results show that playing in the F treatment is related to an increase in the contribution to the Public Good of 2.53 monetary units by youths, 1.68 (2.53–0.85) monetary units by grandparents, and 0.95 (2.53–1.58) monetary units by parents.Footnote 14 Thus, consistent with previous results, we observe that, in comparison with the R treatment, playing with family members increases the contribution to the Public Good. Furthermore, there are differential effects of the F treatment by generation, as the greater increase is present in the group of youths, while lower increases are present in the group of parents and grandparents.Footnote 15 Thus, playing with family members increases the contribution of players more in the case of youths.

Moreover, playing in the IG treatment is, in comparison with playing in the R treatment, related to a statistically significant increase in the contribution to the Public Good of 0.34 monetary units by youths and parents, and with a non-statistically significant decrease of 0.08 (0.34–0.42) monetary units by grandparents. These results indicate that the relationship between the contribution to the Public Good and the IG treatment is mixed, as it depends on the generation under analysis. Finally, when playing in the R treatment, grandparents and parents contribute 1.58 and 1.47 more monetary units to the Public Good, which is consistent with Fig. 1.

5 Conclusions

Prior research has analyzed cooperation within families (Reynolds 2015; Görges 2015; Beblo and Beninger 2017), but these studies focus on same-generation or two-generation cooperation, with little or no experimental evidence provided on the level of cooperation among three generations. To bridge this gap, we report a repeated Public Good game with three generations (youths, parents, and grandparents) with the participation of 165 volunteers (55 individuals aged between 17 and 19, one of their parents, and one of their grandparents).

All the age groups cooperate more when playing with relatives, with this trend being more evident for the youths and the parents than for the grandparents. Thus, family ties appear to have a positive relationship to contributions to the Public Good. We find that the level of cooperation of the youths is significantly lower than that of the parents and the grandparents, consistent with the prior literature showing that young people are less cooperative (Peters et al. 2004; Charness and Villeval 2009; Gutiérrez-Roig et al. 2014).

One limitation of the paper is that, in the current context, it is not clear whether the findings for the youths reported here imply a behavior driven by altruistic motives, or driven by the fear of criticism by parents and grandparents later on. We simply do not know whether the youths contribute more within the family because they recognize that the pooling of resources benefits them (the selfish motive), or they fear being shamed by their parents after the fact (psychic costs), or they act purely from generosity due to kinship. The absence of anonymity in the family treatment may be conditioning our results, which does not allow us to extract any causal claim. The experiment did not allow us to disentangle the motives behind this specific behavior, and a promising line of research remains open.

Our results indicate that the level of cooperation of the youths is significantly lower than that of the parents and the grandparents. This may well be driven by an income effect (Chowdhury and Jeon 2014), but the absence of income information in the experiment does not allow us to disentangle the motives behind this specific behavior, despite prior evidence showing the differential behavior of older generations in terms of cooperation (Gutierrez-Roig et al. 2014). Other channels that could explain our results are a cohort effect (i.e., individuals become less generous towards strangers) or a selection effect (i.e., individuals who become parents are more generous). We leave this issue also for future research, where experiments similar to the one presented in this paper, and that collect information on personal or household income, will be crucial in discerning the reasons behind generational differences in cooperation.