A discrete choice model of voters’ decision in the general election 2019 using LASSO regression

Abstract

Statistical modelling faces problems when too many variables are involved in the model. Using all variables in the model will result in overfitting which in turn produces unstable predictions. LASSO (Least Absolute Shrinkage and Selection Operator) is a method that can be used to overcome this problem. LASSO selects more important variables with the goal of improving the prediction accuracy and interpretability. This paper identifies factors those influenced voters in 2019 presidential election using LASSO penalized logistic regression. This logistic model is also known as a discrete choice model. The data used in this study is the result of an exit poll conducted during the election day with a total sample of 2,289 respondents. The response variable in this study is a vote choice. The independent variables include sociological, psychological, political economy, and campaign variables. The results showed that there were 17 of 27 variables having nonzero coefficients. These variables include candidates’ personal qualities which show the largest effect, followed by the incumbent performance variable. On the other hand, variables such as religion, education, and ethnicity have much less effects. This implies that the psychological and political economy factors played more important role than the sociological factor in determining voters’ decision. Based on this finding the prospective candidate in the future election should focus their campaign strategy on enhancing positive images (empathy, integrity and capability) as well as offering better programs to convince voters.