Gender differences in trust and reciprocity in repeated gift exchange games

Abstract

We explore gender differences in trust and reciprocity using the Berg et al. (1995) trust game and the Fehr et al. (1997) gift exchange game. Our study differs from previous ones in that subjects play the stage game repeatedly with random re-matching of players. In the trust game, contrary to prior studies that report men are more trusting, we find no strong gender differences in trust, although we corroborate prior findings that women are more reciprocal. In the gift exchange game, we again find no significant gender differences in trust, but women appear to be less reciprocal than men.

Keywords:

• trust
• reciprocity
• gift exchange
• gender
• experiments

Notes

1. We should hasten to add the important caveat that while the trust game and the gift exchange game can both be used to study trust and reciprocity, they are not the same game and therefore differences in behaviour cannot be attributed solely to the presence of context. In order to focus on the role of context alone, one needs to run the same game using both neutral and loaded language. This is something that we are studying at the moment in related work using the Berg et al. (1995) trust game.

2. http://veconlab.econ.virginia.edu/admin.htm

3. Income is coded as follows: Income = 0 for weekly income less than $250, = 1 for $250–750, = 2 for $750–$1250, = 3 for $1250 +

4. This is available from the corresponding author upon request.

5. Each experimental dollar is worth NZ$0.10 so that the endowment of 10 experimental dollars in each rounds amounts to NZ$1.00. At the time the experiments were carried out the exchange rate was approximately NZ$1 = US$0.72.

6. Here and in what follows we report the results of specifications that best fit the data. Often we tried other models but refrain from discussing these in the interests of parsimony.

7. The introduction of the lagged dependent variable as a regressor in a dynamic panel data regression can produce biased estimates. Therefore, in these two models − (3) and (4) – we follow Wooldridge's (2002, pp. 542–543) tractable approach for dynamic panel tobit models. We include a set Z of variables corresponding to the initial value of the dependent variable and the values of continuous exogenous variables in all time periods. Note, however, that in this case, the Wooldridge correction produces qualitatively different results only when we consider the relevance of NZ Born and Income variables.

8. The round effect for men is estimated to be −0.672. When we consider the significant interaction term with female we find that the round effect for women is equal to −0.199 (−0.672 + 0.473).

9. A System GMM procedure is actually used, using the xtabond2 Stata command proposed by Roodman (2009).