Estimation of social preferences in generalized dictator games

- Statistical analysis of social preferences requires econometric modeling of choice.
- Different choice models estimate different utility functions, preventing consensus.
- I analyze precision (in-sample fit) and robustness (out-of-sample fit) of standard models.
- Random utility model for ordered alternatives fits well, in-sample and out-of-sample.

To statistically infer the motives underlying pro-social behavior, econometric models of choice are required. Such inference is comparable across studies if the choice model yields estimates that are precise in-sample and robust out-of-sample. Analyzing two extensive dictator game data sets, I find that structural models of choice prevent significant overfitting (contrary to regression models), structural models with generalized error structure fit the choice pattern, and random utility models yield robust identification of subject types (contrary to random behavior and random taste models). The random utility model for ordered alternatives provides robust estimates across games and is therefore suited for analyses.