Acceso usuarios
There are two approaches to predicting election outcomes: (1) a historical approach, which uses past election results alongside macroeconomic and political variables to forecast election results up to a year in advance, and (2) a campaign-oriented approach, which uses current campaign trends to forecast vote shares at the end of the campaign. They are in some way at odds—one approach says the campaign doesn't matter, the other focuses entirely on the campaign. This article considers whether the two approaches might be usefully combined; it considers whether the prediction errors in historical models may be related to trends during the campaign. That possibility is tested here using 17 elections in the US, UK and Canada, combining historical predictions and automated content analyses of campaign-period media content. Results suggest that campaigns do not account for errors in the historical predictions; but there may be other ways in which campaigns matter in conjunction with historical models.
