Who controls U.S. politics? An analysis of major political endorsements in U.S. midterm elections

In this study, we investigated the underlying patterns, rationales, and effects of major political endorsements in the United States midterm elections, specifically focusing on the endorsement patterns and impacts of Donald Trump and Bernie Sanders in the 2018 and 2020 midterms. In order to uncover such patterns, we examined five demographic variables (Cook's Partisan Voter Index (PVI), bachelor’s degree attainment, percentage white, percentage aged 65 and above, and median income) as well as two candidate-specific variables (incumbency and progressivity). Through employing density-based clustering methods and logistic regressions, we discovered a dichotomy of pragmatic and ideological endorsements in U.S. politics. We thus hypothesized that all political endorsements can be predicted by a combination of ideological and demographic factors, and fall somewhere along the dichotomy. Trump's endorsements were 97% predictable through logistic regressions on demographic variables alone, with PVI contributing the majority (55%) and education level contributing 28%; meanwhile, Sanders' endorsements were only 67% demographically predictable, but 80% predictable through the progressivity of individual candidates alone. We therefore concluded a trend of Trump basing his endorsements upon state demographics instead of individual candidates, and Sanders endorsing based on whether or not candidates align with a progressive ideology. Using a support vector machine model for binary classification, we found that Trump's and Sanders' endorsements contributed 11% and 6% to election victories of their respective endorsed candidates, significantly behind factors such as PVI (38%), bachelor’s attainment (25%) and incumbency (14%), justifying trends such as Congressional stagnation, widening educational disparities in electorates, and political polarization.