If a Statistical Model Predicts That Common Events Should Occur Only Once in 10,000 Elections, Maybe it’s the Wrong Model (Danny Ebanks, Cal Tech)

Political scientists forecast elections, not primarily to satisfy public interest, but to validate statistical models used for estimating quantities of scholarly interest. Although scholars have learned a great deal from these models, they can be embarrassingly overconfident: Events that should occur once in 10,000 elections occur almost every year, and even those that should occur once in a five-quadrillion elections are sometimes observed. We develop the first generative statistical model of all US federal legislative elections for the U.S. House (1946–2024) and Senate (1954–2024) and validate it with extensive out-of-sample tests. We find that even as the incumbency advantage rises and falls, voting patterns nationalize, and heresthetic maneuvering remake candidacies and campaigns, the fundamental quantities of interest related to American legislative elections remain constant over eighty years, including but not limited to the probability of incumbent defeat, the magnitude and dispersion of candidate effects, the midterm penalty, and the partisan bias of electoral maps. Most striking, we show that the same set of structural parameters describes both U.S. House and Senate elections, implying a universal data-generation process for American legislative elections in the postwar era.

Join Zoom Meeting

https://pomonacollege.zoom.us/j/82624384355

Meeting ID: 826 2438 4355