Computational and algorithmic models of strategies in turn-based games

We study two different models of a turn-based game called the Marble Drop Game, which is an experimental paradigm designed to investigate higher-order social reasoning. Our first model is a computational-level description of the game, associating cognitive difficulty of a game trial with its structural properties. Our second model is an algorithmic-level model postulating a forward reasoning plus backtracking strategy for solving the game, rather than backward induction as prescribed by game theory. Our experiment shows that the algorithmic-level model is more predictive for the participants' reaction times. This research illustrates how various methods of logic and computer science may be used for building computational cognitive models.