The joy of ruling: an experimental investigation on collective giving

Data on charity giving are taken from Giving USA 2016 report.

Warr (1982) and Roberts (1984) are precursors of this approach. The provision of a public good through voluntary contributions may generate utility by a joy-of-giving, independent of any concern for the interest of others. In these models, for example Bergstrom et al. (1986) and Andreoni (1989, 1990), the motivations underlying individual voluntary donations typically combine pure altruism (linked to the recipients’ well-being) and warm-glow (or impure altruism, associated to the joy of donating).

Provided that tax evasion is not allowed. Voting over taxes when tax evasion is a possibility has been recently considered by Traxler (2009)

These mechanisms resemble some production functions in team production settings that date back to Hirshleifer (1983): the weakest link mechanism, where the team output is given by the minimum effort, the best shot mechanism, where the team output is determined by the largest effort, and the linear mechanism in which the team output is the average effort. For a comprehensive experimental analysis of the performance of these production functions in team production settings see Croson et al. (2015). They find that contributions in the weakest-link are smaller than in the linear and the best shot mechanism. Despite similar qualitative results, there are notable differences between the team production setting and our collective decision mechanism: in the former, “votes” are costly—meaning that group members individually bear the cost of their own effort—and therefore team members obtain different levels of material payoffs whereas in our setting, all players get the same material payoff, e.g., the same combination of private–public good provision but enjoy different utility levels because of the existence of heterogeneous preferences over the provision of the public good.

In a Bayesian framework, each player is characterized by a type—defined by their social preferences—and the beliefs held about the types of other players.

Andreoni (1989, 1990) extends the framework by assuming that the gift to the public good also enters in the utility function. None of the conclusions we arrive at in this paper depends on the existence of this warm-glow associated to giving. To keep things simple, we stick to the standard analysis in Bergstrom et al. (1986).

Note that for the super-dictatorial mechanism, there is no strategic interaction among players, because the procedure by which a message is chosen is independent from the messages that others send to the mechanism. This is why the optimal message is the solution to the unipersonal decision problem in (1). The super-dictatorial decision \(g_i^{\mathrm{SD}}\) exists as long as the utility function complies with the standard quasi-concavity assumption.

This is in sharp contrast to the results in the voluntary provision game. As the group grows in size, the equilibrium gift tends to zero under pure altruism, whereas zero convergence is not obtained under pure warm glow (and no altruism, see Ribar and Wilhelm 2002).

As the number of subjects attending every session was not a round number, the perception that subjects were participating in different group sizes was toughened, and credible.

Participants only knew their group size when entering the first stage. Subjects were also informed that different group sizes were predefined from a natural base (\(N=1\)) to an arbitrary and reasonable ceiling (\(N=10\)).

This implies that dictators (\(N=1\)) made three decisions under three equivalent rules. The reason for that was to get a baseline to compare.

They had to predict the average reported type of the other participants in their group in the AVG mechanism, the smallest reported type of the other participants in their group in the MIN mechanism and the largest reported type of the other participants in their group in the MAX mechanism.

Note that this prediction exercise is insubstantial for \(N=1\). They were however requested to predict their own vote to make procedures and payoffs homogeneous.

Recall that no information feedback was provided until the end of the experiment; as was explained in the previous subsection.

A translated version of the questionnaire is also available upon request from the authors. 91 out of 96 subjects passed the quiz on the first attempt. The remaining five did it in the second attempt with no additional explanations.

Our charity effect is not as strong as observed by Eckel and Grossman (1996), whose percentage of donations went up to 30%, although it is slightly larger than that observed by Hoffman et al. (1994) which maintained the recipient’s anonymity (9%).

The average success rate is 25.20%, with the highest score in the MIN mechanism (52.44%) and the lowest in the MAX mechanism (8.54%).

A correct prediction was rewarded with €2.50, and one euro was deducted for every ten percentage points/€1 difference in Q1 and Q2, respectively.

We do not have a good rationale for this difference. Note that the two experiments have very different structures and subjects made the same decision in very different framings. While participants in the individual condition of Experiment I obtained their earnings almost exclusively from their individual decisions, participants in Experiment II knew their first decision would be used to compute their final earnings with only a relatively small probability.

Control variables are Order, Female, Age, Economics and Trust (see the coefficients of these control variables in Table 1A in the appendix of Electronic Supplementary Material).

The econometric results in Table 7 are robust to different specifications. If we use the expected donation Q the results hold with very minor changes (Type (D1) for the MAX mechanism is less marginally significant). As adding a quadratic term of Type does not improve the models’ goodness of fit (and the quadratic term is insignificant in all cases but the MAX mechanism), we report the simpler model in Table 7, and include it in the analysis of the super-dictatorial decision D5 in Table 8 (see the discussion below).

Figure 3 is a standard whisker and box graph. The box contains the 25–75% quartiles, the bar corresponds to the median, and whiskers include the adjacent values in each condition.