Modeling choice and estimating utility in simple experimental games

Current experimental research seeks to estimate shape and parameterization of utility functions. The underlying experimental games tend to be "simple" in that best responses are salient and individual choices consistent, but the analyses arrive at diverging results. I analyze how precision of estimation and robustness of results depend on the choice models used in the analysis. Analyzing dictator and public goods games, I find that regression models overfit drastically when choices exhibit high precision (dictator games), that structural models with simple error structure (normal or extreme value) do not fit the curvature, and that random behavior and random taste models do not identify social motives robustly. The choice process is captured well through the random utility model for ordered alternatives ("ordered GEV", Small, 1987)