The Internet is saturated with explanations of the 2024 election. Some of these “quick takes” are dispassionate, while others take the form: If only Harris had done what I know is right, she would have won.
The challenge is epistemic: it’s virtually impossible to explain a single past event that involves many decision-makers (in this case, about 150 million of them).
Explaining the decisions of a few powerful people is hard enough, but at least then we can use evidence about their individual values, goals, and personalities. For instance, we can investigate why Napoleon ordered the main assault at the Battle of Borodino in 1812. However, says Tolstoy,
It was not Napoleon who directed the course of the battle, for none of his orders were executed and during the battle he did not know what was going on before him. So the way in which these people killed one another was not decided by Napoleon’s will but occurred independently of him, in accord with the will of hundreds of thousands of people who took part in the common action. It only seemed to Napoleon that it all took place by his will (War and Peace, 10:28)
It’s easier to explain the pattern displayed in a large set of cases (inductive reasoning). John Burn-Murdoch observes that every incumbent government in the world that has faced an election in 2024 has suffered major setbacks. John Sides argues that inflation lowered Biden’s approval rating, and the incumbent’s approval predicts reelection.
But these generalizations cannot explain the single event of the 2024 US presidential election. Generalizations inevitably involve variance, and 2024 is obviously anomalous. Should we even categorize Harris as the incumbent, when she was a vice president stepping in for a president and running against the previous president?
We can also look at patterns within the population to try to explain why individuals voted. For instance, Michael Tesler assembles evidence that few American women vote from gender solidarity and race consistently trumps gender as an explanation of voting.
This is a valid approach that will yield more precise insights once we have voter files and better survey analysis for 2024. But this method also has limitations for the purpose of explanation. As the (true) cliché reminds us, correlation is not causation. Besides, individuals vary in ways that are not captured in generic surveys. And we must distinguish carefully between two tasks: explaining why large numbers of people voted for each candidate, versus explaining the marginal change since 2020. Big blocs of the electorate vote predictably, yet much of the conversation is about changes at the margin. Our whole discussion would be different if Harris had won by 4 points instead of losing by less than one point, but either way, most people would have voted the same.
To emphasize the last point: I strongly suspect that a male Democrat would have fared no better than Kamala Harris, or even possibly worse. One of many pieces of supportive evidence is the fact that people whose survey answers indicated sexism already tended strongly to oppose Joe Biden in 2020 (Spencer 2021). I doubt that sexism explains the marginal change between 2020 and 2024, yet that hardly makes sexism irrelevant, since it helps to explain the 2020 baseline. Whether you feel that sexism is at stake may reasonably reflect your own depth of concern about misogyny in our society; this is not simply a statistical question. Put another way, whether you explain the result in terms of sexism depends on whether you are trying to a) combat misogyny or b) win an election. The explanation is relative to its purpose.
We might conclude that it is fruitless to make a model to explain any particular case. But that is exactly what we must do before we act. Even if there is no way to know now what would have happened had Harris acted differently, Harris and her team had to do something. In September, they needed a prospective model of the single case that confronted them: the election.
In 1903, Charles Sanders Peirce coined the term “abduction” (or “abductive judgment”) for the logic that explains a single case. Abduction is a pragmatic necessity because we always act in specific circumstances. In my view, valid abduction never depends on a single claim. There is no way to test whether one premise caused a given outcome in a given case. Rather, a good abduction consists of many linked components: a whole model. And it is appropriate for the model to contain facts, values, and strategies.
Thus, if you were Kamala Harris in September, you needed a coherent account of the current US electorate (facts), what you sought to achieve as a president (values), and how various messages and methods would affect the outcome (strategy). You had to guard against biases (believing facts because they confirmed your values), but you were entitled to bring your self into the analysis. For one thing, this was a model for how you should act, so it had to motivate you and your team and sound authentic coming from you.
We cannot tell which parts of Harris’ implicit model were right or wrong–and it remains possible that her model was as good as it could have been. But what we need now is a model to guide our own next steps.
Since I am not running for president, my model should not be designed for that purpose–although I might start armchair strategizing in 2026 or so. For now, I need a model that guides my actions as a concerned citizen during the Trump Administration. To a limited extent, my model might be guided by my retrospective assessment of the 2024 presidential campaign–but not by much. The main question, as always, is what should we do?
Sources: Spencer, Bettina. “Impact of racism and sexism in the 2008–2020 US presidential elections.” Analyses of Social Issues and Public Policy 21.1 (2021): 175-188; Peirce, C.S. 1903. Lectures on Pragmatism, Lecture 1: Pragmatism: The Normative Sciences. See also: using a model to explain a single case; overestimating the impact of leaders; What Should We Do? A Theory of Civic Life