Abstract
•Game-theoretic trees vary in complexity and explains why subjects might deviate from the behavior prescribed by the theory (backward induction).•The paper proposes an empirical measure of complexity based on response times.•The proposed measure of complexity provides a control variable for empirical research.•We also find that adding rounds to a tree (length) makes it more complex than adding actions per round (width).•We use a novel methodology to collect data: mobile experiment.•Our sample consists of 6,637 subjects from 143 countries who played 44,113 puzzles.•Our results are robust to using multiple subsamples of the data.
Game-theoretic trees vary in complexity. This paper introduces the concept of graph-based complexity and relies on the subjects’ behavior to empirically derive a measure of tree complexity. Data comes from the mobile app Blues and Reds, designed specifically to conduct experiments. The sample consists of 6637 subjects from 143 countries who play 27 different dynamic games. Based on subjects’ response times, we find that two measures – the average response time spent at the first round and the average total time spent solving the tree – are the best candidates for the empirical measure of tree complexity. We focus on two-person, finite, zero-sum dynamic games with perfect and complete information.
