*3.5. Initial Attractions*

We estimate initial attractions to strategies for all learning models. This adds two additional parameters to estimate for all models, for the row player the initial attraction to strategy Up and for the column player the initial attraction to strategy Left. This seems sensible, because, empirically, it does not appear that subjects choose strategies randomly in the first period of play, and, *a priori*, there is a systematic difference between payoffs when considering the expected play by the opposing player. We can compare the actions in the first round of the experimental data to the estimated initial attractions as a sensible test of the learning model. We fit all learning models using the stochastic choice rule and appropriate learning theory and then predict the choice for each period (round) and subject in the dataset.
