Horse race coverage in American elections has shifted focus from late-breaking poll numbers to sophisticated meta analytic forecasts that often emphasize candidates’ probability of victory. We place this “probabilistic horeserace” in the context of Riker and Ordeshook (1968), and hypothesize that it will lower uncertainty about an election’s outcome (perceived potential pivotality), which lowers turnout under the model. After demonstrating the prominence of probabilistic forecasts in election coverage, we use experiments to show that the public has difficulty reasoning about the probability of a candidate’s victory. Critically, when one candidate is ahead, win-probabilities convey substantially more confidence that she will win compared to vote share estimates. Even more importantly, we show that these impressions of probabilistic forecasts cause people not to vote in a behavioral game that simulates elections. In the context of the existing literature, the magnitude of these findings suggests that probabilistic horse race coverage can confuse and demobilize the public.