Are conditional cooperators willing to forgo efficiency gains? Evidence from a public goods experiment

Abstract

We use a two-person public goods experiment to investigate how much agents value conditional cooperation when symmetric positive contributions entail efficiency losses. Asymmetric marginal per capita returns allow only the high-productivity player to increase group payoffs when contributing positive amounts. Asymmetric contributions, however, yield unequal individual payoffs. To assess a priori cooperative preferences, we measure individual ‘value-orientations’ by means of the decomposed game technique. We find that contributions remain negligible throughout the experiment, suggesting that people are not willing to contribute positive amounts if this may lead to damage efficiency.

Keywords:

• public goods experiments
• conditional cooperation
• fairness
• efficiency
• social values

Notes

1. Note, however, that in these studies efficiency concerns (i.e., being willing to sacrifice a portion of own payoff to increase social welfare) do not always require putting oneself at a relative disadvantage.

2. In dictator dilemma games, the recipient receives more than the dictator donates.

3. Conformity implies that others' behavior serves ‘as a guide to what is socially or morally appropriate’ (Bardsley & Sausgruber, 2005, p. 665).

4. Parks (1994) illustrates the predictive abilities of value orientations in public goods games.

5. Previous linear public goods experiments allowing for heterogeneity in the marginal per capita return include Brandts and Schram (2001), Fisher et al. (1995), and Goeree et al. (2002).

6. A translation of the German instructions for both experiments is available upon request from the authors.

7. All statistical tests in this part of the analysis rely on averages over periods and players for each pair as an independent unit of observation.

8. Following Offerman et al. (1996, p. 827), random players are those subjects whose consistency measures are below 33%. The overall index of consistency in the experiment is about 93%.

9. We employ a robust estimation of the variance to control for potential heteroskedasticity. Furthermore, the fact that the dependent variable is taken in differences provides a control for potential correlation of the errors at the individual and group level.