Par Richard Nadeau
Résumé : The outcome of the 2016 election made it abundantly clear that victory in U.S. presidential contests depends on the Electoral College much more than on direct universal suffrage. This fact points to the importance of using state-level models to arrive at adequate predictions of winners and losers in U.S. presidential elections. In fact, the use of a model disaggregated to the state level and focusing on three types of measures—namely, changes in the unemployment rate, presidential popularity and indicators of long-term patterns in the regional strength of the Democratic and Republican parties—has in the past enabled us to produce fairly accurate forecasts of the number of Electoral College votes for the presidential candidates of the two major American parties. In this article, we bring various modifications to this model in order to improve its overall accuracy. With Joe Biden out of the race, this revised model predicts that Donald Trump will succeed in winning back the presidency with 341 electoral votes against 197 for Kamala Harris.