Predicting the 2024 U.S. Presidential Election : Extending a Logistic and Bayesian Logistic Approach
(2025) Understanding voter behavior with predictive modeling — ISBN: [9798337310374], p. 201-226, published
Files
No attached file found for this publication.
Details
- Authors
Gallina, Marta 
- Abstract
- This chapter addresses the persistent underestimation of Republican vote shares in U.S. presidential election forecasts, with a focus on the 2024 cycle. Despite methodological refinements, polling errors remain systematic. The chapter proposes a theory-driven, voter-level modelling strategy using logistic and Bayesian logistic regression applied to 2024 ANES data. Moving beyond poll aggregates and structural indicators, the models incorporate predictors such as ideological orientation, economic dissatisfaction, and sociodemographic traits. Results confirm that these factors substantially improve predictive accuracy, reducing the gap between estimated and actual vote shares. The weighted Bayesian model predicted Trump's national vote share within 1.6 percentage points—outperforming many major polling organisations. Overall, this approach demonstrates the empirical and diagnostic value of behavioural modeling in electoral forecasting and contributes to a broader methodological shift that centres voters' individual characteristics as key to understanding electoral outcomes.
- Affiliations
Citations
Camatarri, S., & Gallina, M. (2025). Predicting the 2024 U.S. Presidential Election : Extending a Logistic and Bayesian Logistic Approach. In Georgia Panagiotidou (ed.), Understanding voter behavior with predictive modeling (p. p. 201-226). https://doi.org/10.4018/979-8-3373-1037-4.ch007