Swing State Election Analysis

Recent elections at the national level of US government have resulted in poor outcomes for the Democratic Party, illustrated by their current absence in the White House and minority status in both houses of Congress. This erosion of power was precipitated by extremely narrow popular vote deficits in a handful of key "Battleground States". In an era defined by political polarization, Electoral College outcomes are no longer a matter of national appeal, but hinge on the voting patterns of geographic micro-factions. In a competitive and localized presidential election, small shifts in high-leverage counties can trigger disproportionate Electoral consequences for candidates.

The 2028 Presidential election offers a critical opportunity for a Democratic reversal; however, success requires the precise identification of the electorate factions that decide outcomes within these key states. Our goal is to address this challenge by developing a model that identifies swing-state counties with the largest and most impactful variance in voting patterns. Through the integration of historical election results (2016–2024) and county-level census demographics—specifically in Pennsylvania, Michigan, and North Carolina—we aim to isolate counties that offer the optimal combination of electoral elasticity and vote volume. Our analysis will utilize k-means clustering and association rule mining to uncover the specific socio-economic feature sets that characterize these high-volatility electorates.

US presidential elections are a common area of study for political scientists and data analysts alike. Alan Abramowitz (1988) famously weighed presidential approval, economic growth, and duration of incumbent party leadership in his "Time For a Change" linear regression model predicting popular vote distribution. More recently, Helmut Norpoth's Primary Model rivaled Abramowitz's, introducing variables surrounding national primary performance. After correctly forecasting a Trump Victory in 2016, the Primary Model incorrectly predicted the following two presidential outcomes. Both of the models described above are reliant on nationally scoped data (economic conditions, incumbents, and primaries) rather than local dynamics. Further, both are more interested in predicting outcomes than identifying the factions that dictate electoral results.

Recent work has shifted to more granular analyses. Chen and Patel (2021) demonstrated the efficacy of a localized approach by narrowing the geographic focus to 15 states through a "Swing State Index". By applying hierarchical clustering to 2020 election returns, they identified distinct state archetypes—distinguishing demographic-driven shifts vs policy-driven shifts. Qualitative analysis was employed to attribute these patterns to specific political and cultural events. Crane, Koch, and Lin (2024) focused on counties rather than state trends, finding that US counties with relatively high stock market participation were more likely to vote for the incumbent party when accompanied by strong stock market performance.

While relevant research has localized, a significant gap remains. Chen and Patel have described past volatility at the state level, dismissing shifts in behavior at more granular levels. Crane et al. describe a univariate voting dynamic at the county level; however, they fail to consider the impact of variables beyond stock market participation.