A Mobile Experiment on Perception in Dynamic Games

Abstract

While a game-theoretic tree is an unrivaled way of representing dynamic interactions, no real-life interaction comes with its tree representation. Before backward inducting, players must understand the interaction they participate in – we say that they go through a process of tree construction. We test whether people correctly construct trees; i.e., whether their perception of interactions is consistent with how theory models these interactions. We design a test that reveals subject's unobservable perception through her observable choices. Replicated in 27 different interactions with 5,001 subjects from 133 countries, the test shows that between 13% and 63% of subjects fail at tree construction. Data comes from Blues and Reds, a mobile app that functions as an experiment. We argue that our result is detrimental to the internal validity of testing backward induction. We also argue that tree construction is more challenging than backward induction.

Keywords: dynamic games, tree construction, perception, backward induction, experimental game theory, mobile experiment

JEL Classification: C72, C99, D91