How accurate are citizens' political predictions, and how well do they remember what they predicted? We report results from two pre-registered survey experiments examining prediction accuracy and recall bias in the context of Donald Trump's second presidential term. In Study 1 (N=1,000, January 2025), conducted immediately before Trump's inauguration, respondents predicted the likelihood of various political events occurring within the first 100 days. Participants were randomly assigned to incentivized or non-incentivized conditions. Contrary to expectations, financial incentives did not improve prediction accuracy relative to Polymarket forecasts. Moreover, we found no partisan differences in predictions between Democrats and Republicans, suggesting minimal motivated reasoning in prospective forecasts.

Study 2 (N=668, May 2025) re-contacted the same respondents after the 100-day period to assess recall accuracy. Participants were asked to remember their original predictions from January. The control group recalled their predictions immediately, while the treatment group was first ask to report whether each event actually occurred before recalling their initial forecast. Results reveal substantial hindsight bias: both groups' recalled predictions were systematically biased toward actual outcomes. Critically, this bias was significantly amplified in the treatment condition—actively considering what happened before recalling predictions increased memory distortion. These findings have important implications for democratic theory, suggesting that retrospective evaluations of individuals' political judgment may be fundamentally unreliable due to reconstructive memory processes. When citizens evaluate their own past predictions, they shade them towards what actually occurred.