Pandering to the Future

Abstract

I study a model of delegated experimentation in which actions signal agents' underlying type---careerist or partisan. I show that careerist agents pander to the future by choosing the action most likely to be preferred ex post by the principal. I discuss when and why this behaviour differs from selecting the principal’s ex ante preferred action, which balances static and dynamic learning benefits. I conclude my analysis by considering implications for agents' responsiveness to changes in actions' popularity, for the benefits and costs of transparency, and for the role of partisanship structures.