1 Introduction

Individual actions—referred to as voluntary climate action (VCA)—are needed to reduce climate change’s harmful effects (Goeschl et al. 2020). VCA takes different forms on an individual level; however, one key unifying aspect of VCAs is that they necessitate incur a cost to the individual to provide a benefit to the environment, a general public good that is largely consumed in the future (Fischer et al. 2004; Diederich and Goeschl 2014; Hauser et al. 2014; Lohse and Waichman 2020). Examples of VCAs include investing in energy-saving technology (e.g., solar panels), switching to CO2-friendly purchasing habits (e.g., buying less red meat), or even engaging in small, everyday behaviors, such as spending less time in the shower (Wynes and Nicholas 2017). In our study, we are interested in VCA that has a long-lasting positive effect on the environment (Layton and Levine 2003; Steinke and Trautmann 2021): we focus on CO2 offsetting, using a foresting program that plants climate-efficient trees, as such programs have become increasingly widespread and available as means for individuals to help reduce their “carbon footprint” (Kollmuss et al. 2010).

While past research has examined contextual changes (“nudges”) to motivate VCAs (Thaler and Sunstein 2008; Hauser et al. 2018) (see Appendix C for related literature), we propose a novel perspective on how to solve VCA dilemmas by leveraging the intergenerational aspect of VCA. Extant research has focused on public goods within the same generation (Fehr and Gächter 2000; Milinski et al. 2006; Rand et al. 2009) or cooperation between different generations (Charness and Villeval 2009), whereas the literature on intergenerational goods, where future generations cannot reciprocate the actions of the acting current generation and the incentives to cooperate with the future are low, is still in its early stages (Andre et al. 2021; Fischer et al. 2004; Sutter et al. 2013; Hauser et al. 2014; Kamijo et al. 2017; Ponte et al. 2017; Shahrier et al. 2017; Dengler et al. 2018).

However, this does not imply that there exists no link to future generations: people (i.e. parents) who have children are genetically related to the next generation and have an incentive and responsibility to care for their offspring’s wellbeing. Parents make up a large fraction of the population (Eurostat 2017) and, as they are in their adult life stage, they are among the largest contributors to CO2 emissions through their consumption, production, and work (Zagheni 2011). Thus, getting parents to engage in any kind of VCA is likely to result in economically meaningful changes.While parents “in isolation” have not been found to be more willing to give to VCA than other groups (Diederich and Goeschl 2014), we argue that, when their own child observes their VCA decision, the personal genetic link to the future makes them more likely to engage in VCA, as the intergenerational benefits of VCA are more salient to them.

Specifically, we predict that parents will be especially likely to engage in VCA when observed by their offspring relative to other observers. While past work has shown the importance of observers to motivate costly cooperative behaviors (Yoeli et al. 2013; Hauser et al. 2016), a parents’ offspring is critical here because it lets them recall their genetic link to the future. Therefore parents, who have their children’s wellbeing at heart, are reminded of the benefits of investing in the future when their genetic beneficiaries are present (Smith 1977; Nowak 2006).

2 Methods

2.1 Voluntary Climate Action and Study Context

We carried out a novel lab-in-the-field experiment with 368 participants in Innsbruck, Austria (for details regarding sample size, pre-registration, and power calculation, see Appendix D). The experiment included an incentive-compatible survey programmed in oTree (Chen et al. 2016), and data were collected with tablets (see Appendix E). Participation took no longer than 20 min, and participants were randomly assigned to a treatment when they were handed a tablet. Using a recruitment stand in public spaces, we recruited parents who were accompanied by at least one of their own children aged between 7 and 14 years. At all times during the experiment, only one parent (the DM) and one of the parent’s own children (who is an observer in one condition and not involved in the experiment in the other conditions) were allowed to participate. In conditions where the child was not an observer, s/he was asked to wait outside the study booth and participate in various games and activities (supervised by research assistants). In addition, for our conditions with observers who are not related to the participant, we employed confederate adults and confederate children who were introduced to the participant as “helpers from the community” to act as observers.

The VCA in our study was carefully designed based on the extant literature. It is worth noting that VCAs are generally undersupplied, and previous attempts to get participants to invest in VCA have generally not resulted in large effects; knowing this, we, therefore, aimed to design an attractive VCA for the participants in our study by incorporating some of the “best practices” from past literature. For example, past work has found that the general public prefers investing in VCA with local mitigation goals (Torres et al. 2015). Thus, in our setting, the VCA to offset CO2 takes the form of a local foresting program, for which we collaborated with the forestry office Innsbruck (“Amt für Wald und Natur” of the city of Innsbruck). We chose a foresting program for forest restoration because such programs are among the best climate change solutions available today (Bastin et al. 2019). Participants were asked to choose between keeping money for themselves or spending that money on planting trees. All trees that participants decided to plant will be planted in 2020 and 2021 on the “Nordkette” and “Patscherkofel” mountain ranges close to Innsbruck, ensuring that the mitigation strategy is truly local. Moreover, this particular area has a high suitability for the VCA, as it has a high net plant productivity with the potential for forest restoration (Bastin et al. 2019).

Following Goeschl et al. (2020), subjects received a short and neutral description of the foresting program. In particular, they were informed that the foresting program has the following characteristics: (a) The trees would only be planted if participants in our study actually chose to spend their money on planting a tree. This ensured that the participants’ decision was incentive-compatible and truly contributed to reducing CO2 in the environment. (b) The trees were selected to lead to a climate-friendly mixed forest, including climate-efficient species of different fir trees or deciduous trees. These tree types would usually not be planted as frequently due to their cost. (c) Each tree has an expected minimum age of 120 years (estimate provided by the forestry office Innsbruck). This means that each tree our participants planted lasts at least the equivalent of four average (human) generations (following the Cambridge dictionary definition of a “generation”). (d) The trees would be monitored and controlled annually to ensure they are healthy, and they would be listed in the governmental forest database “Walddatenbank” to ensure a “paper trail” of the planting exists. (e) The trees would be planted in a forest that is certified with an internationally recognized “Program for the Endorsement of Forest Certification” (PEFC) certificate, ensuring environmental sustainability. All these characteristics ensured the maximally possible credibility of our CO2 offsetting program. More information on the foresting program can be found in Appendix F.

Moreover, as part of the instructions, subjects were given information about greenhouse gas emissions and trees’ role for CO2 reductions before deciding on the VCA. Since the general population has relatively little prior knowledge about VCAs (Diederich and Goeschl 2014), we ensured that all participants first gained a basic understanding of the VCA in this study and correctly answered several comprehension questions. Whereas MacKerron et al. (2009), Löschel et al. (2013), and Goeschl et al. (2020) provided information as text on the screen, our study participants watched a short video. The video informed participants about the public goods character of CO2 reductions by explaining how planting trees removes CO2 from the atmosphere and mitigates global climate change. In particular, the video highlighted that reducing CO2 has an impact not only on current generations but also on future generations.

2.2 Experimental Conditions

We implemented four conditions in a between-subjects design, varying observability and the type of observer (see Table 1). In all conditions, a parent received a windfall endowment of 69€ and was asked to decide how much of that money to keep for themselves and how much to invest into the VCA (i.e., planting trees). Using their endowment, participants could purchase between 0 and 46 trees, with each tree costing 1.50€ (the average cost of planting a tree in the foresting program). Any money not invested in planting trees was paid to participants in cash at the end of the experiment.

Table 1 Experimental conditions, varying who observes the participant

When making their decision, participants had detailed information on how much CO2 emissions a tree would offset every year (0.015 tons) and that the total possible amount of 46 trees would offset 10% of the average CO2 emissions of a person living in Austria (OECD 2018). Based on these numbers, the price to offset one ton of CO2 emissions per year with our VCA is fixed at 100€/t (see Diederich and Goeschl 2014 who argue that this price reflects “an economically meaningful maximum abatement cost for one ton of CO2 emissions”, see also Tol 1999, 2009, 2010), which is higher than the EUA (EU ETS) Future Price for one ton of over 60€/t (as of September 2021). We also collected data on basic demographics (e.g., gender, age, education, etc.) and included a short survey at the end of the experiment (see Appendix E).

In our baseline NoObserver condition, the DM decided in private without being observed by anyone. In the StrangerAdult condition, the DM was observed by another adult who is a hired actor (confederate) to act as the observer and who is unrelated to the DM (see detailed information about the observability procedure in Appendix E). This condition is similar to the standard procedure used in observability experiments in the lab, where a DM is observed by another adult, which helps establish a “general observability” effect. In the StrangerChild condition, the observer was an actor who is a child between 7 and 14 years old and unrelated to the DM, which helps identify whether the VCA can be encouraged by having an observer from the future (beneficiary) generation. Finally, in the OwnChild condition, the observer was the DM’s child, enabling us to study whether the DM’s own child uniquely affects the DM’s VCA behavior.

All observers were provided the same information as the DMs, including information about greenhouse gas emissions and the trees’ role for CO2 reductions. They also read the description of the local CO2 mitigation program. All this information is common knowledge to the DM and the observer, as both DM and observer read the instructions simultaneously during the experiment. Detailed information on the experimental design can be found in Appendix E.

2.3 Experimental Sample

Our experiment involved 368 parents, 92 in each of the four treatment conditions. Data were collected starting at the end of 2019 until early 2020 in three different locations in the city of Innsbruck. In Appendix Table 10, we provide background details on our participants based on the post-experimental questionnaire. In Appendix Table 11, descriptive statistics are further broken down by treatment, showing that randomization worked: the randomly assigned participants are comparable across a number of relevant characteristics.

Across all treatments, 67% of our participants are female (248 out of 368), and the average age is 42 years. Participants have, on average, 2.06 children, and the vast majority (96%) are currently employed. With regards to education, 86% received a high school diploma (by completing an exam called “Matura”), which provides general access to higher education and labor market qualifications. Out of those with a high school diploma, half (50%) have a university degree. The majority is married or in a registered relationship (66%), and there is approximately an equal split between those living in the city (49%) versus those living in rural areas. Our recruited sample is largely representative of the general population of Innsbruck (Austria), where our trial took place.

Following Goeschl et al. (2020), we included a survey question asking participants how risk-seeking they viewed themselves (based on Falk et al. 2018). The mean reported value was 5.35 on a scale from 0 (not risk-seeking at all) to 10 (fully risk-seeking) and did not differ between treatments (Kruskal–Wallis test, p = 0.255). Additionally, we asked participants how patient they believe they are as a proxy of their time preferences. The average reported score was 5.92, measured on a scale from 0 (not at all patient) to 10 (fully patient). We did not find any treatment differences for the patience measure (Kruskal–Wallis test, p = 0.397).

In three out of four treatments, the participant was observed, for which we provide summary statistics of the different observer characteristics. Stranger adult observers (who were hired by the experimenters as confederates) in the StrangerAdult condition were on average 39.89 years old and observing children 11.33 years (StrangerChild: 12.23 years; OwnChild: 10.43 years). The majority (55%) of observers was female, and the number of female observers did not differ across treatments (Fisher’s exact p = 0.231). In Appendix Table 15, we summarize the gender matches of participants and observers for the treatments with observers. Because both the participant sample and the observers were made up of more women (F) than men (M), we have 99 FF matches, 88 FM, 54 MF, and 35 MM matches. Gender matches were balanced across treatment conditions (χ2, p = 0.497).

2.4 Econometric Specifications

Our analytical strategy is twofold: First, we estimate the treatment effects on VCA using ordinary least squares (OLS) regressions, shown in columns (1) and (2) in all our regression tables. Second, we employ Tobit regressions, which can be found in columns (3) and (4) in the regression tables, to estimate treatment effects, taking into account that the dependent variable is the number of trees planted (i.e., VCA), which is bounded by 0 trees on the lower end (if the participant keeps the entire endowment for him/herself) and by 46 trees on the upper end (if the participant invests the entire endowment into the VCA). For both models, we use the following specifications for columns (1) and (3), which shows the main effects of the independent variables (treatment dummies) without any control variables:

$$ VCA_{i} = \beta_{0} + \beta_{1} StrangerAdult_{i} + \beta_{2} StrangerChild_{i} + \beta_{3} OwnChild_{i} + \varepsilon_{i} $$

where i = 1, …, n indicates participant i, VCA is a continuous variable (ranging from 0 to 46) measuring the number of trees a participant decided to plant, and the StrangerAdult, StrangerChild, and OwnChild dummies are 1 in the respective treatments and 0 otherwise, \({\upvarepsilon }_{{\text{i}}}\) measures unobserved scalar random variables (errors).

We also report in columns (2) and (4) the same specification with several control variables:

$$ \begin{aligned} VCA_{i} ~ = & ~\,\beta _{0} + ~\beta _{1} StrangerAdult_{i} + ~\beta _{2} StrangerChild_{i} + ~\beta _{3} OwnChild_{i} + \beta _{4} Location_{i} \\ & + \beta _{5} Age_{i} + \beta _{6} Female_{i} + \beta _{7} NrKids_{i} + \beta _{8} Risk_{i} + \beta _{9} Patience_{i} \\ & + \beta _{{10}} HighSchoolDipl_{i} + \beta _{{11}} Employed_{i} + \beta _{{12}} Rural + \varepsilon _{i} \\ \end{aligned} $$

where Location is a categorical variable controlling for the three study locations Rathausgalerien, Sillpark, and Herbstmesse, Age is a continuous variable and Female a dummy variable for the participant’s age and gender, NrKids is a continuous variable capturing the participant’s number of kids, Risk and Patience are self-reported scale measures (scale range from 0 to 10), High School Dipl. is a dummy variable which is 1 if the participant completed secondary education (“Matura”), Employed is a dummy variable which is 1 if the participant is currently employed and Rural is a dummy variable which is 1 if the participant lives outside a city; all other variables are as defined in Eq. (1).

3 Results

3.1 Descriptive Results

For the 368 participants in our study (see Appendix Tables 7, 8 and 9 for descriptive statistics), we find a remarkable willingness to engage in VCA: across all conditions, all participants combined chose to plant a total amount of 13,988 trees (our outcome measure, labeled “VCA” throughout; out of a maximum possible 16,928 trees across all participants). On average, participants invested 82.63% of their 69€ endowment into the VCA, with 66.58% of participants choosing to invest their entire endowment into planting all possible 46 trees. Note that this extent of VCA is substantially higher than in previous studies, suggesting the VCA on offer in our study was unusually attractive to our participants. The average VCA does not differ by the participant’s gender (female participants: 37.79 trees vs. male: 38.47 trees; WMW, p = 0.724).

Following the literature (see Diederich and Goeschl 2014), we begin by examining which variables are predictive of VCA across conditions (see Appendix Table 13). Age is a significant predictor of VCA (coeff = 2.22, p = 0.042), whereas gender (coeff = 0.48, p = 0.757) and the participant’s number of kids are not significant (coeff = 0.92, p = 0.207). These results are all consistent with several past findings (Löschel et al. 2013; Diederich and Goeschl 2014 but see also Andre et al. 2021). A high school diploma (“High School Dipl.”) is associated with higher VCA (coeff = 10.68, p < 0.001), in line with Diederich and Goeschl (2014). Employment is also positively associated (coeff = 11.51, p = 0.001), as one might expect that being employed implies greater disposable income (Löschel et al. 2013). Meanwhile, neither risk nor patience preferences are significantly associated with VCA (Risk: coeff = 0.17, p = 0.553; Patience: coeff = − 0.22, p = 0.381). Living in a rural area or not does also have no significant effect (coeff = − 1.37, p = 0.328). Lastly, we find some variation by study location, which we discuss in more detail in Appendix A.

3.2 Treatment Effects

Turning to our conditions, we first summarize the raw VCA values (see Fig. 1). We observe the lowest VCA in NoObserver (mean = 37.12, 25th percentile = 35.00 and 75th percentile = 46.00) and StrangerAdult (mean = 37.09, 25th percentile = 32.00, 75th percentile = 46.00). VCA is slightly higher in StrangerChild (mean = 38.24, 25th percentile = 34.00, 75th percentile = 46.00) and it is highest in OwnChild (mean = 39.60, 25th percentile = 43.00, 75th percentile = 46.00).

Fig. 1
figure 1

VCA: Number of trees planted by treatment condition (N = 368 subjects). Each box plot shows the average VCA of participants in each treatment. Box plots show the mean (indicated by black X signs), the 25th and 75th percentiles, Tukey whiskers (median ± 1.5 times the interquartile range), and individual data points. Larger dots indicate a higher number of participants who invested in the corresponding number of trees

We next examine the effect of our treatments econometrically. As Table 2 shows, the largest coefficient relative to the baseline NoObserver is the OwnChild treatment. Without control variables, the OwnChild coefficient is positive but not significant (OLS: coeff = 2.48, p = 0.236; Tobit: coeff = 9.44, p = 0.149), whereas with control variables, the OwnChild treatment leads to significantly larger VCA (OLS: coeff = 3.69, p = 0.064; Tobit: coeff = 11.88, p = 0.050). Neither the coefficient on StrangerChild nor StrangerAdult is significant with or without control variables. Thus, in line with our descriptive and graphical results (Fig. 1), our econometric results (sometimes significantly but always directionally) suggest that OwnChild leads to the highest VCA, relative to the NoObserver baseline condition.

Table 2 Regression results for the entire sample

These results suggest directionally that parents may be affected by their own children’s presence when making the VCA decision, but not with other observers present. To isolate the potential mechanisms at work, we explore three explanations using pre-registered non-parametric tests. First, we investigate to what extent the genetic link in particular matters, holding constant the “observer’s generation”. While VCA is higher, as predicted, in OwnChild (39.60 trees planted) than in StrangerChild (38.24 trees planted), this difference is not significant (WMW, p = 0.419).

Second, we test whether a representative of the future generation as an observer has a larger impact on VCA than an adult observer. We pool VCA across the two treatments, in which a child is the observer (OwnChild and StrangerChild) and compare it with VCA in StrangerAdult. Again, as expected, the average VCA is higher (38.92 trees planted) when being observed by a child, but not significantly different from the average VCA (37.09 trees planted) when being observed by an adult (WMW, p = 0.471).

Finally, we investigate a general observability effect, comparing VCA in NoObserver with the average VCA from across the three treatments with observers (OwnChild, StrangerChild, and StrangerAdult). Even though VCA is higher, the difference between the pooled observer conditions (38.31 trees planted) and the NoObserver condition (37.12 trees planted) is also not significant (WMW, p = 0.328).

3.3 Treatment Effects by Education

We found education to be an important determinant of the willingness to invest in VCA, in line with prior work (Diederich and Goeschl 2014). Thus, we examine our treatment effects in two sub-analyses for participants with versus without high school diploma (see Appendix Table 14 for descriptive statistics by educational background). Average VCA by treatment and high school diploma groups are graphically summarized in Fig. 2. First, we observe a substantial main effect of having a high school diploma, pooled across treatments, consistent with prior research (Diederich and Goeschl 2014): whereas participants with secondary education invested in planting 39.37 trees on average (25th percentile = 40.00, 75th percentile = 46.00), participants without secondary education invested at a significantly lower rate of 27.61 trees (25th percentile = 10.00, 75th percentile = 46.00; WMW, p < 0.001).

Fig. 2
figure 2

VCA: Number of trees planted, by condition and education (N = 363 subjects). Each set of four box plots shows the average VCA of participants for each education level. In each education subplot, the order of conditions is as follows: NoObserver, StrangerAdult, StrangerChild, and OwnChild. Box plots show the mean (indicated by black X signs), the 25th and 75th percentiles, Tukey whiskers (median ± 1.5 times the interquartile range), and individual data points. Larger dots indicate a higher number of participants with the corresponding number of trees

Participants with a high school diploma constitute the majority of our sample (312 of 368 participants, or 86%). Focusing on these participants first, we observe consistent and sizeable effects of the OwnChild treatment: across all specifications, parents who are observed by their own child are significantly more likely to invest in VCA (see OwnChild coefficient all columns in Table 3). We do not find any evidence that being observed by a stranger adult or stranger child leads to higher VCA. Consistent with our results across the entire sample, these findings suggest that the OwnChild condition leads to the highest VCA for more educated parents.

Table 3 Regression results for parents with a high school diploma

In fact, we also find evidence that parents invest uniquely more when being observed by their own child than being observed by a stranger child (WMW, p = 0.031). This result supports that, for more educated parents, the genetic link between a parent and their own child uniquely matters for VCA, even when holding constant that the observer is a representative of a future generation.

Turning to participants without a high school diploma (N = 51), we find that the treatment effects look qualitatively different. Specifically, the average VCA is low in the NoObserver condition (20.13) and, remarkably, also in the OwnChild condition (23.94). The highest mean VCA is observed in the StrangerChild condition (35.73), which is significantly different from the NoObserver condition without covariates and with covariates using Tobit (see columns 1, 3, and 4 in Table 4) but not significant with covariates using OLS (in column 2, p = 0.104). The StrangerAdult condition (29.44) falls in the middle.

Table 4 Regression results for parents without a high school diploma

3.4 Treatment Effects by Climate Change Perception

Following the experimental intervention and VCA task, participants completed a short survey which included a questionnaire based on Howe et al. (2015). In this questionnaire, participants were asked about their attitudes towards climate change, allowing us to test how our treatment effects vary by differing climate change perceptions. To reduce variance from any single question item, we constructed an index—which we refer to as climate change perception index—based on the following three questions from Howe et al. (2015): First, “What do you think: is global warming happening?” with three possible answers: “Yes”, “No” and “I don’t know”. Second, “Which of the following statements do you most agree with? Global warming …” with answers: “…is mainly caused by human activities.”, “…is mainly caused by human activities and natural changes.”, “…is mainly caused by natural changes in the environment.“, and”…does not take place.“. Third, “Which statement is closest to your opinion?” with answers: “Most scientists believe that global warming takes place.”, “There is a lot of disagreement among scientists about whether global warming occurs.”, “Most scientists do not believe that global warming takes place.” and “I don’t know.”

The higher the climate change perception index, the more the participant is convinced by climate change and the human contribution to it: specifically, if a participant chose “Yes” to question 1, “…is mainly caused by human activities.” to question 2 and “Most scientists believe that global warming takes place.” to question 3, the index is equal to 3. If they chose two out of these three answers, the index is equal to 2. If they chose only one out of the three, the index is equal to 1; and 0 if none of those answers was chosen. Based on this climate change perception index, 41% of all participants show some skepticism towards climate change (i.e. received an index lower than 3). We observe that VCA is significantly lower for those classified as skeptical (climate change perception index < 3) than those who believe in climate change (climate change perception index = 3) (35.68 vs. 39.56, respectively; WMW, p = 0.023).

We conducted subgroup analyses for those who received an index equal to 3 (i.e. convinced that climate change is happening, mainly caused by humans and that most scientists agree) versus lower than 3 (i.e. skeptical of climate change). As Table 5 shows, interesting but somewhat opposing findings emerge: participants with a climate change perception index below 3 are not more likely to give when their own child is observing them, but perhaps more so when a genetically unrelated child or another adult is watching them. In contrast, participants with a climate change perception index of 3 are only affected by the presence of their own child but not affected by other children or adults. All these effects are directionally the same but smaller in magnitude and less likely to be significant without control variables (columns 1 and 3) but become statistically significant when fixed effects and control variables are included in the regressions (columns 2 and 4).

Table 5 Regression results depending on the climate index

Interestingly, participants with a climate change perception index below 3 show a similar pattern to those without a high-school diploma, while participants with a climate change perception index of 3 behave similarly to participants with a high-school diploma. Indeed, we observe a positive correlation between higher education and the climate index (Spearman's rho = 0.233, p < 0.001), which is also consistent with past literature (Stevenson et al. 2014; Lee et al. 2015).Footnote 1

4 Discussion

In an intergenerational public good (for example, planting trees that offset CO2 emissions), the beneficiaries (future generations) are not the same as the decision-makers (current generation). Parents, who have a genetic link to the future through their children, would be particularly likely to invest in VCA. Indeed, we find a high willingness to invest in VCA, with over 80% of all parents investing in the VCA to plant trees. This is more than the usual VCA contributions found in the literature: Bruns et al. (2018) report that participants spent 35% of a default amount of money on VCA, while Diederich and Goeschl (2014) find that only 16% of subjects chose the emission reduction instead of a cash amount. It is possible that our participants were more willing to invest in VCA because we designed the VCA based on some of the “best practices” from the VCA literature (see the Methods section). However, in addition to offsetting CO2 emissions, it is possible that planting more trees in the surrounding area of Innsbruck (where the experiment took place) may also provide recreational value to many of our participants (see, e.g., Pittel and Rübbelke 2008; Baranzini et al. 2018). Or that planting trees may be a form of CO2 reducing activity that is perceived as more transparent, trustworthy, and less abstract than investments in other technologies to reduce CO2 emissions (Schwirplies et al. 2019). Our findings, in combination with the literature, suggest that there are already many lessons to be learned from the prior literature to make VCAs attractive.

We proposed that VCA would be heightened when a parent is being observed by their own offspring. The parent’s own child would serve multiple purposes, most importantly as a reminder of the fact that a (genetic) link connects the parent (decision-maker) to their own child (future beneficiary). Across our entire sample, we find some evidence that parents give more when their children are watching their VCA decision but, importantly, education plays a key role: as Diederich and Goeschl (2014) note, participants with a high school diploma usually exhibit greater willingness to engage in VCA, and, in our setting, our treatment effects are substantially larger in the subsample of participants with a high school diploma. While it is unclear why educated parents respond more to the treatment, it may be that their educational background means that they are more aware of the environmental impact of their decision (as reflected in their higher climate change perception index) or because better education is usually correlated with more disposable income, which means that sacrificing the experimental endowment as an investment into the future would affect their finances today less than those of less-educated (and potentially lower-earning) parents. Future research is needed to fully explore these effects.

Similarly, our study points to the importance of beliefs and perceptions about climate change in understanding the treatment effects. Using questions first used in a well-known study by Howe et al. (2015), we show that those who believe climate change is happening, is mostly caused by human activities and that scientists are in agreement that climate change is particularly affected by their own child’s presence when deciding to invest in VCA. On the other hand, those who are skeptical of climate change are, surprisingly, mostly affected by other children (to whom they are not related) or other adults. This suggests that social pressure to invest into climate-mitigating strategies may be better applied to this skeptical group by members, not of their “in-group” or close family. It is possible that climate change-skeptical parents think that their children agree with them, would judge them less harshly for giving to this cause, or believe that they could convince them of their merits of such a decision, which they may not be able to do to unrelated observers.

Our study makes several contributions to the literature. First, we contribute to the literature on VCA. Previous studies have investigated both the personal characteristics that determined engagement with VCA (Diederich and Goeschl 2014) and contextual cues—often in the form of nudges—that can lead to more VCA (Araña and León 2013; Böhm et al. 2020; Carattini and Blasch 2020). In this paper, we focus instead on a novel context that we use as an intervention—the role of the genetic link across generations. Since VCAs are intergenerational by nature, we argue that VCA interventions can benefit from considering the intergenerational structure of families and we demonstrate that parents are indeed more willing to invest in future public goods when they are observed by their children, not just other adults or children.

Second, we contribute to the burgeoning literature on intergenerational public goods. Previous research has studied resource replenishment rates (Fischer et al. 2004), institutions (Hauser et al. 2014), and peer punishment (Lohse and Waichman 2020). However, a previously neglected aspect of intergenerational public goods is relatedness (Nowak 2006): decision-makers may not be present to reap the benefits of their actions in the future, but their own descendants could benefit. As a result, genetic offspring should be considered in other interventions to increase contributions to intergenerational public goods.

Furthermore, our study also speaks to the standard public goods game: although observability is a widely studied intervention in economics (Hoffman et al. 1996), our study suggests that the type of observers matters: while adults are typically recruited for studies using observability, we show that variation in observers can yield differing results. In our setting, adult observers did not affect VCA, either in the main analysis or the subgroup analyses. Our findings document the importance of choosing an observer that manipulates the theoretical construct in question.

Finally, we contribute to the literature on child-parent interactions. Most work has investigated one direction of this causal relationship—how parents influence their children—such as, for example, the extent to which parents’ sharing behavior in the dictator game influences their child’s subsequent dictator game behavior (Ben-Ner et al. 2017). Similarly, prior fieldwork has found that preferences are shaped by one’s parents’ behavior in childhood and persist into adulthood (Fernández et al. 2004). Here, we reverse the causal direction of this relationship, finding that parents’ behavior can be shaped by their children. Our intervention is relatively minimal and only involves observation by children, leaving open the possibility that children’s actual influence on their parents is much larger in reality.

Our findings have implications for policy-makers and offer new research questions for scholars across a variety of domains. We focused on VCA, specifically planting trees, for which we found a notably high willingness to invest into the future. Based on our fixed price for reducing one ton of CO2 (by design, fixed at 100€ per ton) and the average CO2 reduction for each tree (0.015 tons per year), we can estimate that our participants were willing to offset 0.570 tons of CO2 annually.Footnote 2 Extrapolating these annual reductions into the future, our participants, on average, invested into offsetting 68.42 tons of CO2 over the expected lifetime of these trees of about 120 years. We believe these estimates represent meaningful ecological and economic trade-offs, which should afford policy-makers some hope that citizens are willing to incur costs today to help the future. Indeed, Steinke and Trautmann (2021) find equally high willingness among their respondents who are willing to incur a cost today to help the future and even the far future (from which they themselves and even their own kin no longer benefit from).

However, parents make many more important decisions in daily life that have consequences, if not always for future generations, at least for years and decades to come that also shape the lives of the next generation. Consider, for instance, voting: in many countries, adults are not allowed to take their children into the voting booth. Would parties that emphasize long-term investments in education and environmental protection receive a greater voting share if parents had to choose under the watchful eyes of their own children? While this is an open empirical question, one could imagine that voting systems may take such considerations into account (Kamijo et al. 2017). Ultimately, this perspective can be extended further, including to seemingly mundane activities, such as shopping for groceries (e.g., buying meat or vegetarian alternatives), or choosing whether to take the bike to work or on the school run: children may be a powerful, yet underappreciated way to shape their parent’s behavior.