Do elections constrain incumbent politicians’ policy choices? To answer this longstanding question, this paper proposes a novel identification strategy to separate electoral incentives from selection effects. Exploiting the unique institutional setup of lame-duck sessions in the U.S. Congress, where lame-duck incumbents who lost reelection vote on the same issues as their reelected colleagues, I use a close election regression discontinuity design to leverage quasi-random assignment of reelection-seeking representatives to lame-duck status, which is orthogonal to voter preferences and incumbents’ type. Comparing within-incumbent changes in roll call voting of barely unseated lame ducks to narrowly reelected co-partisans serving the same congressional term, I find that lame ducks revert to more extreme positions with lame-duck Democrats (Republicans) voting more liberally (conservatively). Consistent with lame ducks’ loss of re-election incentives driving the result, the effect of lame-duck status on roll call extremism is more pronounced among electorally vulnerable incumbents, representatives of districts with a large proportion of swing voters, and retirement-aged lame ducks who are less likely to rerun in the future. I also consider but ultimately dismiss several other mechanisms including emotional backlash, logrolling motives, party control, and selective abstention. Although quantitatively dominated by selection effects, electoral incentives to moderate positions are large enough to reduce polarization by 21--26% and to flip high-stakes legislative outcomes under plausible counterfactual scenarios motivated by ongoing debates over U.S. constitutional design.

Are politicians ideologically rigid, or do officeholders adjust policy strategically for electoral purposes? This paper sheds new light on this question by studying how U.S. House incumbents alter their roll call voting record prior to elections depending on their challenger’s platform. Estimating non-incumbent candidates' policy positions using pre-primary transaction-level campaign finance data, I classify as extremist the more liberal (conservative) of the top-two candidates in Democratic (Republican) challenger primaries. Leveraging a regression discontinuity design, I exploit the quasi-random assignment of incumbents to moderate or extremist challengers by close primary elections of the incumbent’s opponent party. I find that incumbents alter their roll-call voting record in the direction of their opponent’s position, committing to a more moderate policy when running against an extremist challenger and differentiating their position from more moderate opponents. Consistent with strategic responsiveness to electoral incentives, policy adjustment to challengers is confined to reelection-seeking incumbents and to incumbents defending a seat in a competitive district. I provide suggestive evidence that incumbents' reaction to challengers is conditioned by the presence of third candidates, and reflects a trade-off between persuading swing voters at the center and mobilizing core supporters. Importantly, incumbents' adjustment is not driven by a valence advantage of moderate over extremist challengers but by incumbents’ reaction to opponents’ policy positions, suggesting strategic complementarity of policy platforms.

We provide evidence of a causal effect of anticipated election closeness on voter turnout, exploiting the precise day-level timing of the release of Swiss national poll results for high-stakes federal referenda, and a novel dataset on daily mail-in voting for the canton of Geneva. Using an event study design, we find that the release of a closer poll causes voter turnout to sharply rise immediately after poll release, with no differential pre-release turnout levels or trends. We provide evidence that polls affect turnout by providing information shaping beliefs about closeness. The effects of close polls are the largest where newspapers report on them most; and, the introduction of polls had significantly larger effects in politically unrepresentative municipalities, where locally available signals of closeness are less correlated with national closeness. We then provide evidence that the effect of close polls is heterogeneous, with an asymmetric effect leading to a higher vote share for the underdog. The effect sizes we estimate are large enough to flip high-stakes election outcomes under plausible counterfactual scenarios.