Exit poll surveys during elections are conducted to predict the outcome of actual elections. However, such polls historically have been controversial, particularly for multi-phase elections, because they can influence the behavior of voters in the later rounds of voting. If subsequent voters are more likely to vote for the predicted frontrunner, the effect is known as the bandwagon voting phenomenon, whereas if they vote for the predicted trailing candidate, the phenomenon is known as underdog voting. To avoid such issues, in 2009 the election administration in the world's largest democracy (India) introduced a blanket ban on publishing exit polls in the media until all rounds of an election are completed. Exploiting the potentially exogenous timing of this reform, and using administrative data to compare states that went to elections before and after the ban, we find that in response to the policy, vote shares increased for the frontrunner and declined for others. The result implies that in the counterfactual, without the ban, fewer people would have voted for the frontrunner. The evidence is suggestive of underdog voting.