The paper can be found here.

Abstract
We collected data on one-shot play for a representative selection of two-player two- and three-strategy games with unique and completely mixed strategy predictions and propose a new method to compare theories of one-shot play “out of treatment” that allows a direct likelihood comparison: Competing theories are calibrated with pre-existing data using different games and subjects. Most theories in most treatments have predictive power. No theory is uniformly best. Considering risk aversion significantly improves predictive power. Nash equilibrium with risk aversion is among the best predictors of play in four of six treatment groups: the exceptions are the behavior of individuals in the asymmetric own-payoff player positions in games of a matching pennies and rock-scissors-paper variety.