We collect data of one-shot play for a representative selection of two by two games with unique and completely mixed strategy predictions, to compare the predictive power of theories of one-shot play ``out of treatment:'' competing theories are calibrated with pre-existing data using different games and subjects. We find that all theories, except Nash equilibrium, have predictive power; no theory is uniformly best; and taking into account risk aversion significantly improves predictive power. Finally, Nash equilibrium with risk aversion is among the best predictors of play, except for one player position in games of a matching pennies variety. .
Here you can find the data, instructions and screenshots of the experiment and more information about theories used in the paper.
Please contact me to get access to the the code, the data in a machine readable format or anything else.