ETD

• aggregate data
• ecological inference
• elections
• electoral data
• regression
• semiparametric

This thesis investigates the problem of ecological inference, which arises when the object of interest is the relationship between variables at the individual level, but data are only available at the aggregate level.
Section 1 presents the problem of ecological inference and its relevance in various contexts.
Section 2 introduces general notation and definitions for the problem of ecological inference, and possible assumptions that can be made to identify the individual-level relationships from aggregate data, with a particular focus on CAR assumptions and McCartan and Kuriwaki (2026).
Section 3 presents an adaption of the model for the specific context of electoral analysis.
Section 4 presents a synthetic example to illustrate different methods and their performance in a controlled setting, where the true individual-level relationships are known.
Section 5 presents an application of the method to 2020 US presidential elections in Texas.
Section 6 presents some concluding remarks and possible future developments.
Finally, Appendix A and B present respectively further synthetic examples and the construction of the dataset used in Section 5.