This paper aims to test the significance of the gender effect in the “Vote-with-the-Wallet” game (VWG). This game is a multiperson version of the prisoner's dilemma (PD), with a binary choice that we use to describe the dilemma typically faced by consumers confronted with more or less socially and/or environmentally sustainable consumption options.¹

In our experiment, players (consumers) choose between a standard product that costs less and a product that costs more and is advertised as “responsible.” The purchase of the responsible product creates a monetary externality for all other players, functioning as a public good and making cooperation difficult (i.e., achieving a high share of responsible choices). In this context, we are interested in studying the effect of gender on the dynamics of these choices, which aligns with evidence that consumers need help identifying sustainable options and that gender norms can influence their perception (Thøgersen, 2021; Bloodhart and Swim, 2020).

Focusing on gender is particularly relevant because differences are strongly associated with variations in social preferences, ethical sensitivities, and responses to contextual cues. Women, on average, tend to exhibit stronger other-regarding preferences, greater aversion to inequality, and more prosocial behavior in consumption contexts (Croson and Gneezy, 2009; Arnocky and Stroink, 2010; Zelezny et al., 2000; Bloodhart and Swim, 2020). Previous experimental evidence shows that gender differences often emerge in risk attitudes (women tend to be more risk-averse; Costa, 1994), responses to incentives and competition (women are generally less attracted or responsive), and, to a lesser extent, in altruistic and cooperative behavior (Croson and Gneezy, 2009). While meta-analyses suggest that overall cooperation rates are practically indistinguishable across genders globally (Spadaro et al., 2022), women tend to display greater sensitivity to contextual cues, stronger other-regarding preferences, and a higher likelihood of conditional cooperation (Croson and Gneezy, 2009; Furtner et al., 2021). These traits are particularly relevant for ethical consumption dilemmas such as the VWG, where cooperation relies on recognizing and acting upon socially desirable options without external enforcement (Bloodhart and Swim, 2020; Zelezny et al., 2000).

Studying gender-based behavioral patterns in controlled experimental settings enriches the behavioral economics of gender and provides insight for designing effective interventions and policies that promote sustainable and ethical consumption (Croson and Gneezy, 2009). These tendencies are especially relevant in consumption choices with moral or public-good dimensions, such as environmental sustainability or ethical sourcing. Marketing and policy strategies often target “ethical consumption” by appealing to values commonly associated with femininity, such as care, responsibility, and fairness (Bloodhart and Swim, 2020).

The VWG experiment is particularly suitable for gender analysis because it models a clear moral trade-off in a controlled setting, allowing intrinsic prosocial tendencies to emerge. The absence of social cues in the baseline isolates behavioral differences without external influence. Its repeated, binary structure with a gender-balanced sample enhances internal validity and reveals stable patterns in cooperative behavior, making it an effective tool to study gendered responses in ethical consumption contexts.

The treatments considered beyond the baseline systematically vary institutional cues that may interact with gendered preferences. The redistribution mechanism introduces fairness and punishment, potentially amplifying women's aversion to inequality. The legality frame activates normative expectations and moral signaling, to which women may be more responsive due to social-role internalization. The conformity treatment tests sensitivity to peer behavior, aligning with evidence that women exhibit greater social norm compliance. These manipulations help uncover conditional patterns of cooperation that would remain hidden in a neutral setting.

In this study, we make an original contribution to the debate on gender differences in social preference experiments. In the baseline treatment without redistribution women exhibit a higher propensity to cooperate. Unlike studies suggesting women are “conditional” cooperators (Furtner et al., 2021), our experimental design—lacking group information in the baseline indicates that women start from a higher baseline level of cooperation even without external cues.

We analyze data collected by Becchetti et al. (2024), which includes multiple VWG treatments with and without redistribution, a “legality frame,” and a conformity treatment. The companion study investigated an ex-post redistribution mechanism, the impact of a legality frame for the responsible product, and the effect of a conformity information design. The main findings of the complementary study are:

The remainder of the paper is structured as follows: Section 2 details the experimental design, the formulation of the VWG, and the general hypotheses of the study. Section 3 presents the initial empirical findings from the static tests, followed by an in-depth econometric analysis and a Bayesian multilevel probit model to assess the robustness of the gender effect. Section 4 discusses our results, contextualizing them within the existing literature and addressing the study's implications and limitations. Section 5 provides the main conclusions and final remarks.

The theoretical benchmark for this experiment is the VWG, a multiperson version of the PD. The PD is one of the most studied games in the social sciences as it formalizes the conflict between what is socially efficient and what is individually optimal, a theme that underlies many interesting interactions, economic and otherwise (Embrey et al., 2017).

The n players choose between the “responsible” product A, which costs more but produces a positive (environmental, social, legal) externality, and the conventional product B, less costly but without externality.

Following the original formulation in Becchetti and Salustri (2015), the game is defined as

and the ith player payoff function

with U being the player's utility function, S being the chosen strategy, and x the number of players buying product A excluding player i. The crucial parameters of the game are the price differential between the two products (γ ∈ [0, + ∞)), the other-regarding preference component (α ∈ [0, + ∞)) satisfied by the purchase of the “responsible” product,², and the positive externality (β ∈ [0, + ∞)) accruing to the utility of every player (irrespective of her/his product choice) in proportion to the share of players buying the responsible product A.

The game G_n always has a unique (Nash, 1950) equilibrium (B, B) when 1nβ+α<γ, and (A, A) otherwise, and we are in the PD area for intermediate values of γ where 1nβ+α<γ<β+α. This condition implies a large PD region as the number of players increases. The implication is that, in consumer markets where “responsible” products involve a public goods dilemma similar to the VWG, this PD-type problem may be highly relevant. However, this inference is based on data from a specific sample of Italian students, highlighting the need to investigate its generalizability to other markets and cultures.³

The design of the experiment is presented in Table 1, which details the number of sessions and the different treatments.

In each experimental session, a group of 10 players chooses between products A and B over 20 rounds. Product A costs 10 Experimental Currency Units (ECUs), while product B costs 5 ECUs. Each player buying product A generates 3 ECUs for every player in the given round (2 ECUs=1 euro). In each round players are given an endowment of 20 ECUs. The sequence of actions in each round is as follows: i) the ith player formulates her/his expectation on the number of players choosing product A; ii) chooses the product to buy; iii) is informed about the number of players choosing product A; iv) is asked about her/his satisfaction about the game, her/his behavior and the behavior of the other players in the game with three different questions on a 0–10 scale. Half of the rounds (the first or the last 10, depending on the session considered (see Table 1 for details), present a redistribution phase in which at the end of each round players choosing product B know that they have to transfer 1 ECU in a pool that will be divided into equal parts among players choosing product A at the end of each round.⁴

This mechanism simulates a redistributive fiscal policy intended to encourage responsible (environmental, social, legal) consumption choices. The above-described baseline game has been played with two variations.

The experiment was run at the University of Cagliari, Italy, with 180 volunteers (with an exact gender balance in each session) in November 2015. The z-Tree platform (Fischbacher, 2007) was used to program the experiment. Given the payoffs described, the experiment gives the following values to the crucial VWG model parameters: n = 10, β= 30, γ=5, α=0. Based on these values, if we assume that the player has no other-regarding preferences (i.e., α=0) the unique (inefficient) Nash Equilibrium (NE) of the multiplayer game in non-redistribution treatments is obtained when all players chose B, since 1nβ+α<γ<β+α (i.e., 3 < 5 < 30). However, in redistribution treatments, buying product B yields a lower payoff when there is only one cooperator and the same payoff as buying product A when there are two cooperators. The redistribution mechanism renders it more convenient to buy product B as the number of “responsible” choices gets larger. However, it is still always better (equilibrium of the game) to keep buying product B.

To test the gender effect, we formulate the following general hypothesis:

where S(f)_{t, s}[S(m)_{t, s}] is the share of cooperative choices (buy product A) selected by female [male] participants in round t of treatment s. The six considered treatments are those shown in Table 1. Each treatment is considered in full and separately when it occurs in the first 10 or the last 10 rounds (i.e., baseline before in sessions 1–3 and baseline after in sessions 4–6). Under the null, the share of cooperating females is not significantly different from that of cooperating males.

We estimate the following logit specification for baseline “before” treatments in sessions 1–3, where the dependent variable (VWGChoice) is a 0/1 dummy taking value 1 if the player chooses product A contributing to the production of the public good, Male is a (0/1) dummy for the male gender, Age is the player's age, DIncome is five income dummies picking up different income brackets and Round is the experiment round testing for dynamic effects in our game:

To estimate the statistical power of the fixed-effects logit model under different effect sizes for the gender variable, we conducted a Monte Carlo simulation that replicates the structure of our panel dataset. Specifically, we examined the ability of the model to detect small, medium, and large effects by simulating datasets with hypothetical true coefficients of β = 0.2, 0.5, and 0.8, respectively. As shown in Figure 2 the results indicate that the current sample provides limited power to detect small effects (24.8%) but sufficient power for medium (83.2%) and large effects (97.9%).

Additionally, we evaluated the empirical power to detect the specific effect size estimated from the real data (β^=-0.839). We generated 1,000 simulated datasets using this coefficient as the true value and re-estimated the model in each case, recording the proportion of statistically significant results at the 5% level. This analysis yielded a power estimate of 98.7%, suggesting that the study design is highly capable of detecting effects of the magnitude observed.

After controlling for round and age, we find that the gender effect in the reference treatment remains significant. It is also observed that being male significantly reduces the probability of choosing product A by ~ 0.145 percentage points (Table 3). To verify the robustness of our econometric findings, we test whether gender significantly affects satisfaction with the game in the “before” baseline treatment. The selected ordered probit estimated specification is

where SatGame_{i, t, s} is the satisfaction of the ith player about her/his behavior in round t of treatment s, VWGChoice × Male is the interaction between the choice of product A and male gender, and AvgGroup × VWGChoice is the share of players choosing good A in the given round. Other regressors are as in Equation 4.

Our findings show that β₂ is negative and significant, indicating a negative interaction between choosing VWGChoice and being male. In other words, men are significantly less satisfied than women when choosing the more expensive product that contributes to the public good. Specifically, choosing VWGChoice = 1 reduces the probability of men reporting the highest level of satisfaction (level 10) by ~ 0.144 percentage points.

None of the other treatments significant in static hypothesis testing are robust to the econometric checks of this section (results omitted for reasons of space).

Figure 3 presents the marginal effects on the probability of VWGChoice being equal to 1. It can be seen that being male significantly reduces this probability by ~ 0.145 percentage points compared to being female. Starting in period 3, and particularly in periods 5, 6, and 10, there is a statistically significant decrease in the probability of VWGChoice being equal to 1 compared to the base period. The most pronounced effect is observed in period 6. Periods 4, 7, 8, and 9 also show a negative trend, albeit with marginal significance. Conversely, the age variable does not have a statistically significant effect on the probability of VWGChoice.

To assess the robustness of the effect of gender on the probability of making a given choice, we estimated a series of xtlogit models with different combinations of controls. Figure 4 shows the specification curve corresponding to the coefficient of the male variable, together with its 95th percentile confidence intervals. As can be seen, the estimated coefficient is consistent in direction (negative) and reaches statistical significance in the most saturated models. This suggests that the gender effect is robust to reasonable model specifications.

Figure 5 shows the marginal effects of a discrete change in VWGChoice from 0 to 1 on the probability of observing each of the 10 satisfaction levels. It is observed that VWGChoice = 1 is associated with a statistically significant increase in the probability of being in the lowest satisfaction categories (1, 2, 4, 5, and 6). Conversely, it is associated with a statistically significant decrease in the probability of being in the highest satisfaction categories (8, 9, and 10). The effects in categories 3 and 7 are not statistically significant. This suggests that VWGChoice = 1 shifts the satisfaction distribution toward lower value categories.

Our results reveal that in our experimental study with Italian students, women cooperate significantly more than men in the baseline treatment of the “Vote with the Wallet Game” (VWG), a finding that contributes to the ongoing debate regarding gender differences in social preferences. Unlike studies suggesting that women are “conditional” cooperators (Furtner et al., 2021), our experimental design—which lacks group information in the baseline—suggests that, without external cues, women start from a higher baseline level of cooperation. This may reflect internalized prosocial preferences linked to gender norms that associate femininity with caregiving and collective responsibility (Xiao et al., 2019; Bloodhart and Swim, 2020).

This result contrasts with meta-analyses that do not identify significant differences between genders (Spadaro et al., 2022), underscoring the crucial role of the specific experimental context. The VWG is not only an abstract social dilemma; it simulates a consumption decision with explicit moral implications. In our study, the choice of the “responsible” product (A), especially in a cultural context like Italy's, which includes the “Legality Rating” system, could activate culturally rooted gender identities. In contexts like ours, where relative gender equality might be assumed (Falk and Hermle, 2018), women are able to express their prosocial preferences with fewer restrictions from traditional norms, which could explain the differences observed in our baseline treatment. Thus, we challenge the notion that gender gaps in cooperation arise solely from external factors (Croson and Gneezy, 2009), proposing that in morally charged contexts, women emerge as key agents in promoting the voluntary provision of public goods.

We are also aware that specific characteristics of our game can crucially affect outcomes. We model the VWG with positive externality. A different pay-off structure based on a negative externality from the conventional product could have made the gender effect stronger due to the existing evidence on higher women's loss aversion (Wang et al., 2017; Dawson, 2023).

Along the same lines, we also acknowledge that the decision to act cooperatively can be affected by the weight of the positive externality on the total players' payoffs. Changes in the payoff-structure of the vote with the wallet game can be fruitful directions for future research to test whether our main findings are robust or change according to them.

An additional finding is how the redistribution of resources reduces gender disparities. The asymmetry between men and women is mitigated by penalizing free riding - similar to policies that internalize social costs (Thøgersen, 2021). This does not imply an innate female altruism, but rather distinct logics of cooperation: men seem to prioritize individual efficiency in the face of ambiguous incentives (Burnham, 2018), while women seek to avoid moral costs tied to irresponsible consumption, even without a guarantee of reciprocity. This behavior aligns with their greater aversion to violating social norms perceived as fair (Bloodhart and Swim, 2020).

Finally, the legality frame enhances cooperation but does not eliminate gender differences. This indicates that institutional frameworks influence prosocial preferences but do not nullify them. By labeling product A as “legal,” the decision shifts from the individual moral domain to the institutional realm, thereby reducing the weight of gender identity. This outcome aligns with studies linking institutional legitimacy (such as ethical certifications) to promoting prosocial behaviors and mitigating gender gaps (Bolderdijk et al., 2018).

While our findings point to a significant gender difference in cooperative behavior in the baseline condition, we emphasize that this analysis is post-hoc and exploratory. The sample consists exclusively of university students in Italy, which limits external validity and the generalizability of our results. Cultural context, socioeconomic background, and demographic homogeneity may all shape observed behaviors. As such, our results should be interpreted as suggestive rather than definitive, and they call for replication in more diverse and representative samples.

The original (Becchetti et al., 2024) experiment was designed to test the impact of corporate legality ratings on consumer choices and their willingness to pay for legality, testing whether consumers are willing to pay for this public good when properly informed about it or when a redistribution mechanism is introduced that makes the sustainable choice more convenient.

In this paper, we focus on the existence of gender differences in the VWG. The hypothesis of such an effect is grounded on a large literature that stresses a generalized higher sensitivity of women to contextual elements, such as frames (for a review, see Croson and Gneezy, 2009) and also on previous results with the VWG (Becchetti et al., 2018b).

Our findings document that women are inherently more cooperative - specifically, they are significantly more likely to choose the cooperative option in the baseline treatment, thereby deviating from the Nash equilibrium prediction of universal defection. This departure from the individually rational but socially inefficient equilibrium moves outcomes closer to the socially optimal scenario in which all players choose the responsible product and collectively benefit. This result may reflect differences in internalized prosocial preferences, as observed in studies on altruism and gender roles (Xiao et al., 2019), and it may be amplified in contexts characterized by greater gender equality (Falk and Hermle, 2018). This finding adds new evidence to the debate on gender differences in prosocial behavior with special reference to the VWG.

Our results seem to strengthen the position of those who consider women to be naturally more cooperative. From the policy-maker standpoint, these conclusions are good news since the same policy (legality frame and/or redistribution) does not have a differential impact depending on the gender of the consumer.

As a final caveat it is important to remark that the gender-based analyses emerged in the course of exploratory analysis rather than from a pre-registered hypothesis. This, along with the use of a student sample in a specific cultural setting, limits the broader applicability of our conclusions. Future studies with more heterogeneous populations and pre-registered designs would be essential to validate and extend these insights.