Backward Induction in the Wild: Evidence from the U.S. Senate

Backward induction is a cornerstone of modern game theory. Yet, laboratory experiments consistently show that subjects fail to properly backward induct. Whether these findings generalize to other, real-world settings remains an open question. This paper develops a simple model of sequential voting in the U.S. Senate that allows for a straightforward test of the null hypothesis of myopic play. Exploiting quasi-random variation in the alphabetical composition of the Senate and, therefore, the order in which Senators get to cast their votes, the evidence suggests that agents do rely on backward reasoning. At the same time, Senators' backward induction prowess appears to be quite limited. In particular, there is no evidence of Senators reasoning backwards on the first several hundred roll call votes in which they participate.