Data and Code for: Complexity and Procedural Choice

We test the core ideas of the ``automata'' approach to bounded rationality, using simple experimental bandit tasks. Optimality requires subjects to use a moderately complex decision procedure, but most subjects in our baseline condition instead use simpler (often suboptimal) procedures that economize on ``states'' in the algorithmic structure of the rule. When we artificially remove the mental costs of tracking states by having the computer track and organize past events, subjects abandon these simpler rules and use maximally complex optimal rules instead. The results thus suggest that the main type of complexity described in the automata literature fundamentally influences behavior.<br>