Analysing Spatio-Temporal Voting Patterns in Brazilian Elections Through a Simple Data Science Pipeline

Abstract

Since 1989, the first year of the democratic presidential election after a long period of a dictatorship regime, Brazil conducted eight presidential elections. Short and long-term shifts of power and two impeachment processes marked such a period. This instability is a research case in electoral studies, mainly regarding the understanding of citizens' voting behavior. Comprehending patterns in the population behavior can give us insight into phenomena and processes that affect democratic political decisions. In light of this, our paper analyses Brazilian electoral data at the municipal level from 1998 to 2018 using a simple data science pipeline, which consists of five steps: (i) data selection; (ii) data preprocessing; (iii) identification of spatial patterns, in which we seek to understand the role of space in the election results employing spatial auto-correlation techniques; (iv) identification of temporal patterns, where we investigate similar trends of votes over the years applying a hierarchical clustering method; and (v) evaluation of results. We study the presidential elections focusing on the right and left-wing parties most relevant for the period: the Brazilian Social Democracy Party~(PSDB) and the Workers' Party~(PT). We also analyse the congressman election data regarding parties ideologically to the right and left in the political spectrum. Through the obtained results, we found the existence of spatial dependence in every electoral year investigated. Moreover, despite the changes in the political-economic context over the years, neighboring cities seem to present similar voting behavior trends.

Learning beyond the spatial autocorrelation structure: A machine learning- based approach to discovering new patterns and relationships in the context of spatially contextualized modeling of voting behavior

“…As depicted in this figure, we used the dataset prepared in agreement with subsection 4.3.1 (Step 1) to assess the approaches parameterized according to subsection 4.3.2 (Step 2). We applied an SCV technique, which da BATISTA, 2021) explicitly proposed for the application in question, to estimate the performance of the models (Step 3). We reported these results via the evaluation metrics described in subsection 4.…”

“…Title: Analyzing Spatio-Temporal Voting Patterns in Brazilian Elections Through a Simple Data Science Pipeline.This chapter is based on a paper with shared co-first authorship (JACINTHO et al, 2021). This chapter is based on a paper published on the Journal of Information Data Management.…”

“…While it is the most suitable procedure for assessing machine learning models built from spatial data, it generates unseen correlated distributions in the test set. This scenario happens because spatial boundaries determine the folds, and a removing buffer region is defined as a strategy to diminish the spatial dependence between the test and training sets (SILVA; PARMEZAN; BATISTA, 2021;PLOTON;MORTIER et al, 2020). Consequently, the test set distribution is not observed in the training set, and there are only correlated distributions.…”

“…Rising allegations of political polarization and electoral fraud have boosted concerns about the integrity of democratic elections (NORRIS, 2013). In this context, multidisciplinary studies point to a growing body of research interested in understanding the factors influencing electoral results (JACINTHO et al, 2021). Scientific efforts are concentrated, but not limited, to investigating external factors that impact the popular vote (REID; LIU, 2019), analyzing ideological trends (FAUSTINO et al, 2019), and studying elections on social networks (RECUERO; SOARES; GRUZD, 2020).…”

“…Considering how people are geographically contextualized and the data's spatial characteristics can enrich our understanding of electoral processes. With this in mind, few researchers have been dedicating themselves to developing non-linear models capable of predicting electoral features from thousands of explanatory features, taking geographic and spatial aspects intrinsic to the data into account (JACINTHO et al, 2021). Such models, when deployed, need to be fairly evaluated to understand their prediction error in future test cases.…”