Forecasting two-horse races in new democracies: Accuracy, precision and error

Abstract

The purpose of this article is to explore electoral forecasting in two-horse races in new democracies. Specifically, it applies a Bayesian dynamic linear model (coined the Two-Stage Model, TSM) to look at the 2020 Chilean two-question national plebiscite. The ultimate objective is to test the TSM in terms of accuracy (how close the forecast is to the election results), precision (how close the forecast is to other methods of prediction) and error (how the forecast deviates from perfect accuracy/precision). The article finds that while the TSM does appear to be a stable estimator, its accuracy and precision is affected under certain conditions. Using the difference in the results for each of the two questions, the article discusses how sharp and unexpected shifts in electoral preferences can affect forecasts.

Keywords

• bayesian inference; election campaigns; new democracies; public opinion; plebiscites