Is it possible to predict electoral abstention on the individual level? A preregistered test on forecasting the effects of abolishing compulsory voting in Belgium

There is a vast literature on determinants of electoral turnout that allows us to forecast which groups of the population will turn out to vote and which will not. Here we report on a rather unique forecasting experiment at the individual level. In June 2024, elections were held in Belgium with compulsory voting. In October 2024, another election was held, but this time without compulsory voting. Simultaneously, a panel survey was conducted, spanning from April to November 2024. The information in the first two waves of the panel were used to forecast the likelihood of individual respondents turning out again in October, which we preregistered. The forecasting models were indeed successful in predicting who would turn out to vote, but they tended to give relatively elevated turnout likelihood scores to non-voters. The prediction models tended to underestimate the effect of political interest in explaining actual electoral turnout.