Working Paper: NBER ID: w15610
Authors: Steven D. Levitt; John A. List; Sally E. Sadoff
Abstract: Although backward induction is a cornerstone of game theory, most laboratory experiments have found that agents are not able to successfully backward induct. Much of this evidence, however, is generated using the Centipede game, which is ill-suited for testing the theory. In this study, we analyze the play of world class chess players both in the centipede game and in another class of games - Race to 100 games - that are pure tests of backward induction. We find that world class chess players behave like student subjects in the centipede game, virtually never playing the backward induction equilibrium In the race to 100 games, in contrast, we find that many chess players properly backward induct. Consistent with our claim that the Centipede game is not a useful test of backward induction, we find no systematic within-subject relationship between choices in the centipede game and performance in pure backward induction games.
Keywords: backward induction; game theory; centipede game; chess players
JEL Codes: C9; C91; C92; C93; D01
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| world-class chess players (C72) | backward induction ability in race to 100 games (C73) |
| world-class chess players (C72) | stopping at first node in centipede game (C73) |
| stopping behavior in centipede game (C73) | performance in race to 100 games (L83) |
| variations in race to 100 game (C73) | player behavior (C72) |
| cooperative behavior in centipede game (C72) | backward induction failures (D89) |