The Non-Strategist’s Fallacy Autopsy
By Elliot Morris observes that any serious autopsy of the 2024 election has to start from the fact that Harris outperformed what could have been reasonably expected, given a climate extremely hostile to incumbents and a specifically highly unpopular incumbent president:
When we boot up the data, it’s obvious the main reason Harris lost — and the reason I am going to explore here, at this website, it being a data-driven website — is that 2024 simply had too much inflation-induced anti-incumbent sentiment for the incumbent party to overcome. This is curiously missing from its main diagnosis. The word “inflation” isn’t mentioned in the autopsy a single time (except in the context of inflation-adjusted ad spending).
The reality of the 2024 election is that it was going to be hard for a Democrat to win, regardless of who they were or how they campaigned. The broader economic and political conditions were so favorable to Republicans that you would have expected Trump to win about 90% of the time, regardless of campaign or candidate effects.
Political scientists have been pointing out for decades that you can predict presidential elections reasonably well using just two pieces of information: how voters feel about the incumbent president, and how voters feel about the economy. There are many variants of this model — such as the “Bread and Peace” model (Douglas Hibbs), the “Time for Change” forecast (Abramowitz), Ray Fair’s “Fair” model, and Wlezien/Erikson’s work with “Leading Economic Indicators” — but all use a similar set of economic and political “fundamentals” to predict the result of the election.
None of these models is perfect, but they’re a useful baseline for forecasting elections and understanding why certain results happen. One use of the predictions is to distinguish the part of an outcome that’s about candidates and campaigns from the part that’s structural.
The chart below shows a model I built for exactly these explanatory purposes. It uses Gallup’s net approval of the incumbent president measured in June and the trailing two-year average of the University of Michigan’s Consumer Sentiment Index to predict the incumbent party’s share of votes cast for the Democratic and Republican candidates for president — what we call “two-party” vote share. That’s it: two variables, both measured before Labor Day, both knowable months before any ad runs, and in some cases, before candidates are even picked.
The dashed line is a perfect prediction. The shaded line shows the 80% prediction interval for the 2024 election, and the solid blue dot shows the actual result of the race.
In 2024, Kamala Harris received about 49.3% of the two-party vote. The model — fit on data from 1956 through 2020, with 2024 held out — predicted she’d get about 48%, with an 80% prediction interval of 46.6% to 49.9%. Harris’s vote share lands on the upper end of this range, but still squarely inside of it.
In a sense, then, the surprise of the election is that Harris did as well as she did, considering the prevailing factors against her. Given Biden’s approval rating in June (deeply underwater, in the high 30s) and two straight years of the worst consumer sentiment readings outside of a recession, the Democratic nominee was on track to lose the popular vote by 4 points. She lost by 1.5.
IAnd, again, this trend was not just present in one liberal democracy, and the headwinds not only faced by one type of political faction:
If you only read the DNC report, you’d think 2024 was a uniquely American failure of Democratic messaging, organizing, and candidate definition. And maybe you don’t buy my “fundamentals” regression of election outcomes. Well, to that person I’d emphasize that 2024 wasn’t just bad for Democrats; it was bad for incumbent parties all across the globe.
As data journalist John Burn-Murdoch documented in the Financial Times, every single governing party in a developed democracy that faced voters in 2024 lost vote share. The Tories were obliterated in the UK. Macron’s coalition lost its majority in France. The LDP lost its majority in Japan. And on and on. Incumbents from Portugal to South Korea either lost outright or were badly weakened.
This isn’t to say that the Harris campaign didn’t make mistakes, or that it was impossible for the absolutely optimal campaign to win. But the idea that her campaign was some kind of disaster has nothing to be said for it outside of the general lazy pundit tendency to assume that any losing campaign must have been bad,
The biggest takeaway going forward, as Paul has also said recently,is that the idea that Democrats can’t afford to run anyone but a white man has no actual empirical basis. And it’s a particularly strange takeaway from the 2024 election, given that had the white male incumbent stayed in the race the Democrats probably would have lost Virginia and New Jersey in the electoral college and every swing Senate race.