Testing game theory without the social preference confound

We propose an experimental method whose purpose is to induce selfish behavior in games for a broad class of social preferences. It provides a theoretical framework for testing game theoretical predictions by confronting subjects with a commonly known payoff matrix actually representing their preferences. The paper describes the empirical tests of this method based on the comparison of results from several popular experimental games played with and without our methodology. Apart from it being a test of validity of the method, our experiment helps answer the question of how useful social preferences could be in explaining commonly observed deviations from selfish rationality. Results suggest that our method does induce more selfish behaviors: a substantial part of the difference between predictions based on selfishness and observed behaviors seems indeed driven by such preferences. But they also indicate that a considerable share is left untouched, perhaps giving weight to alternative explanations.