Models of aggregate election results should be treated very cautiously. But this leaves out:
1) The most prominent U.S. models: "Time for Change" and "Bread and Peace" were first published in 1988 and 1987, respectively. There has been new data since then.
1) The most prominent U.S. models: "Time for Change" and "Bread and Peace" were first published in 1988 and 1987, respectively. There has been new data since then.
2) Even when the prominent models are old, when other researchers have overfit a model, you can use model averaging to try to account for that, which has been done. https://www.jstor.org/stable/1350304
3) Too many pundits ignore that, which there are a small number of presidential elections in the U.S., ***economic voting models have been tested all over the world***. When you consider elections in democracies worldwide, there is a good deal more evidence.
There is a huge literature on "comparative economic voting," and there are scholars more up-to-date on it than I am, but these are good places to start:
https://www.amazon.com/Economic-Vote-Political-Institutions-Condition/dp/0521707404
and https://www.annualreviews.org/doi/abs/10.1146/annurev.polisci.3.1.183
https://www.amazon.com/Economic-Vote-Political-Institutions-Condition/dp/0521707404
and https://www.annualreviews.org/doi/abs/10.1146/annurev.polisci.3.1.183
And as @david_darmofal points out, Silver also misstates the problem he's concerned about. If a variable which had little variation in the previous dataset all of a sudden varies a lot (plague deaths), the problem with the previous lit wasn't overfitting. https://twitter.com/david_darmofal/status/1246845943670026242
If there is one take-away from this, it is, election pundits please remember: political science doesn't only study the United States!
Worldwide, there is more evidence for economic voting than you might think, including studies of when the effects are stronger or weaker.
