Two important elements that often lack sufficient treatment in inductive equilibrium selection works are initial conditions and final outcomes. Without these two important ingredients, models of adaptive dynamics do not yield inductive equilibrium selection. We present experimental data from repeated games, where an explicit and rigorous treatment of both initial conditions and final outcomes is shown necessary for meaningful predictions. To gauge the predictive ability of a model, we propose a graphical-statistical tool based on kernel density estimation of simulated final outcomes. Applying this technique, we find that mainstream approaches predict poorly in some coordination games.

Co-author Dale Stahl