Are election results more unpredictable? A forecasting test

Abstract

Changes in voters' behavior and in the campaign strategies that political parties pursue are likely to have increased the importance of campaigns on voters' electoral choices. As a result, scholars increasingly question the usefulness and predictive power of structural forecasting models, that use information from “fundamental” variables to make an election prediction several months before Election Day. In this paper, we empirically examine the expectation that structural forecasting models are increasingly error-prone. For doing so, we apply a structural forecasting model to predict elections in six established democracies. We then trace the predictive power of this model over time. Surprisingly, our results do not give the slightest indication of a decline in the predictive power of structural forecasting models. By showing that information on long-term factors still allows making accurate predictions of electoral outcomes, we question the assumption that campaigns matter more now than they did in the past.

The Changing German Voter

“…These claims may even be stronger in parliamentary campaigns than in presidential ones (Jennings and Wlezien 2016). Previous research on German elections has shown that predictions of vote shares have been reasonably accurate Gschwend 2013, 2017;Leininger 2016, 2017;Nadeau et al 2020) and that activation of factors like party identification does occur (Johnston et al 2014). In light of these studies, and in the spirit of "plus ça change, " one might expect that campaigns in Germany today resemble those in the US in the 1940s.…”

The predictive value of polls in a fragmented multi-party system: the Netherlands (1998–2021)

“…Moreover, the quality of opinion polls has come under pressure, as traditional sampling frames such as phone registers have become less complete, and response rates have declined (Keeter et al 2017). Nevertheless, there are no indications that election outcomes have become structurally less predictable over time Wlezien 2016, 2018;Nadeau et al 2020).…”

Using polls to forecast popular vote share for US presidential elections 2016 and 2020: An optimal forecast combination based on ensemble empirical model

“…Recent assessments of the Structuralists approach can be found in Graefe et al (2015) and Lauderdale and Linzer (2015). In a recent study Nadeau et al (2020), applying a structural forecasting model, find that information on long-term factors still allows making accurate predictions of electoral outcomes and, thereby they question the assumption that campaigns matter more now than they did in the past.…”