U.S. Views on Epstein Disclosure and Responsibility

We estimate a machine learning ensemble that predicts opinion as a function of party, geography, age, sex, education, income, religion, race, and contextual variables. We generate predicted probabilities for detailed demographic cells and post-stratify them to the county, congressional district, state house, and state senate levels using turnout-adjusted population weights before aggregating to the national level. Reported differences (e.g., rural vs. urban) reflect differences in predicted support for those populations as they are composed—including their partisan and demographic makeup—rather than simple unweighted survey averages. As a result, rural–urban differences are not strictly ceteris paribus; they partly reflect differences in partisan composition. For ceteris paribus-style comparisons, we report within-group breakdowns and crosstabs (e.g., rural vs. urban within party).