Abstract
<p>Pollsters have underestimated support for notable right-wing electoral outcomes, including Donald Trump’s vote share in 2016, 2020, and 2024, the UK Conservative Party in 2015 and 2024, and Leave in the 2016 EU referendum. Do these outcomes reflect isolated polling failures, or a persistent partisan asymmetry in polling error across democracies? We provide the first comparative analysis of partisan polling error. Drawing on over 38,500 election-poll-party observations across 32 countries and 372 elections between 1945 and 2025, we use a Bayesian hierarchical modeling framework to estimate latent election-party polling error. We find that, on average, support for parties on the left is overestimated, while support for parties on the right is underestimated. This pattern is evident across countries, party classifications, and vote intention polling dating to the 1940s. In spotlighting persistent partisan asymmetries, our findings contribute to understanding institutional and contextual correlates of pre-election polling accuracy.</p>