Electoral estimations are measurements of the electorate's support to the set of eligible political agents of an electoral system at a given time. While they also take socio-political attitudes into account, they fundamentally differ from forecasts because they do not have the primary purpose of predicting the outcome of an electoral process in the near future. Many different approaches can be used for electoral estimation, but most of them are limited by their common focus on the electoral district. This focus is intuitive, because the outcome of electoral systems are decided at this level. However, this can only be implemented at the cost of accepting the difficult assumption of spatial homoskedasticity inside the unit of analysis. The wealth of studies on mechanisms such as gerrymandering provides solid empirical evidence about the importance of considering voter variability over the geographical dimension. The fact that heteroskedasticity violates the assumptions of linear regression illustrates the importance of explicitly addressing this issue by scholars relying on this class of models.

This study addresses the heteroskedastic challenge, focusing on implementing a computational technique capable to improve the methodological toolbox of electoral estimation. Our approach goes as follows. We start studying the characteristics of the electoral system of interest, with the purpose of identifying the geographical unit best suited to control for the heteroskedasticity that has  sufficiently available supporting data. We retrieve all electoral results at that level, and we make them available for programatic query using a database management system (DBMS) enabled with geographical information system (GIS) analytic capabilities. Finally, we use this database to train a statistical estimation algorithm implemented in R that computes elected seats by randomly casting votes over our geographical unit, using official electoral data to validate computational performance, removing the need for  homoskedastic assumptions. We complete the platform with scripts to automate tasks such as data retrieval, statistical analysis and visualization, that are common across elections.

In this communication we share the preliminary results of our research. We have applied our approach to the Spanish political system, identifying the ballot box ("mesa electoral") as the most valuable spatial dissaggregation. Our results show the utility of the model for researchers to produce two types of electoral estimations: forensic analysis, intended to explore voting behaviour over past elections; and projective estimations, where survey data from the Centro de Investigaciones Sociológicas (CIS), combined with expert panelling, has proved to produce the most accurate results. We illustrate the capacities of our model using the latest available election at the time of submission, the 2024 Election for the Galician Parliament. To provide a means of pre-registration of our projective estimations, and to enable transfer or research to society, we have enabled the website https://electo.gal/. In addition to this, we share our results using the scientific repository at Harvard University (United States) to allow reuse of our work.

We thank the Political Analysis Research Group (Equipo de Investigaciones Políticas) of the University of Santiago de Compostela , and researcher and co-founder of the Social Data Lab https://socialdatalab.org/ hosting the computational platform, Susana Sotelo Docío, for their continued support to this project since 2019.

Autor de correspondencia. Jesus M. Benitez-Baleato <jesusmanuel.benitez.baleato@usc.gal> es miembro del Equipo de Investigaciones Políticas y docente en la Universidade de Santiago de Compostela. Doctor Europeo en Ciencia Política, ha desarrollado su capacidad investigadora en las Universidades de Harvard (Estados Unidos), Konstanz (Alemania) o ETH Zürich (Suiza). Comparte regularmente os resultados de su investigación mediante repositorios abiertos y publicaciones evaluadas por pares, como la revista Science, contribuyendo al diseño de políticas públicas de digitalización en el contexto de la OCDE y el G7.