La prédiction des résultats électoraux au Canada : un modèle politico-économique sans sondage

Abstract

RésuméOn compte actuellement un grand nombre de modèles politico-économiques ayant pour objectif de prédire l'issue des élections au Congrès américain ou le sort des candidats à la présidence des États-Unis. Bien qu'un certain nombre de modèles aient vu le jour pour la France, l'Allemagne et le Royaume-Uni au cours des dernières années, le Canada, à l'instar de la majorité des démocraties, n'a reçu jusqu’à maintenant que bien peu d'attention. Cet article vise par conséquent à développer un modèle ancré dans une théorie du vote capable de prédire suffisamment à l'avance la part des voix récoltées par le parti sortant lors des scrutins fédéraux canadiens. Le Canada étant un système multipartite, différents modèles de régressions apparemment indépendantes sont également proposés afin de déterminer s'il est possible de prédire simultanément les scores de plusieurs formations politiques.

Parliamentary Debates in Canada (1901–2015)

“…gc.ca/n1/pub/11f0019m/11f0019m2017399-eng.htm. Unemployment: from Statistics Canada, based on Mongrain's (2019) data. Consumer price index: .…”

Playing the synthesizer with Canadian data: Adding polls to a structural forecasting model

“…Thus, their model is of the synthetic kind. Bélanger & Godbout's (2010) as well as Nadeau & Blais' (1993) models, Mongrain (2019) has recently proposed a new structural equation to predict the vote share of incumbent parties in Canada. This model, which explicitly underlines its structural nature, is composed of five variables, that is (1) the difference between the unemployment rates in Canada and the United States three months before the vote, (2) the natural logarithm of the number of consecutive months the incumbent party has been in office, (3) a dichotomous variable related to the substitution of the prime minister near an election, (4) the number of years of political experience gained by the prime minister in relation to his/her main opponent, and (5) a variable related to the province of origin of party leaders.…”

“…This model, which explicitly underlines its structural nature, is composed of five variables, that is (1) the difference between the unemployment rates in Canada and the United States three months before the vote, (2) the natural logarithm of the number of consecutive months the incumbent party has been in office, (3) a dichotomous variable related to the substitution of the prime minister near an election, (4) the number of years of political experience gained by the prime minister in relation to his/her main opponent, and (5) a variable related to the province of origin of party leaders. Mongrain's (2019) model integrates the economic voting theory as well as the notion of benchmarking (see Kayser & Peress 2012): Canadians are believed to react to the economic well-being of their country in relation to that of the United States, their neighbour to the south and most important trading partner. More precisely, if the American job market is doing better in comparison to that of Canada, Canadians voters should be more inclined to punish the incumbent party at the polls (see Carmichael 1990, p.719).…”

“…Although election polling and poll aggregation for federal (and provincial) races have greatly expanded in the last few years, this is not the case for fundamental-based models. In the last 30 years, only four such models were applied to Canadian federal elections (i.e., Bélanger & Godbout 2010;Mongrain 2019;Nadeau & Blais 1993;1995). 2 It goes without saying that synthetic models are also a rare commodity in Canada.…”

“…Consequently, this paper mobilizes Mongrain's (2019) structural model in an effort to assess the gains that can be derived from a synthetic approach in the context of Canadian federal elections. Mongrain's (2019) model is, to date, the only fully structural model predicting the incumbent's vote share in 1 Since Aggregators constantly add new polling data to their models, their final forecasts usually suffer from a very short lead time. As stated by Lewis-Beck (2005, p.151), " [w]hen the lead time is short, say a month or less, the forecasting exercise itself risks being trivial.…”