Allan Lichtman, history professor at American University, Washington, D.C. made somewhat of a name for himself with his ‘13 keys’ (or Keys to the White House) model predicting the winners of US presidential elections based on thirteen true or false statements about the state of the country. He has been applying his keys since Ronald Reagan’s landslide victory in 1984; going into the 2024 election,he had a formidable record of nine out of ten correct predictions.

However, the problem with a prediction model for an event occurring every four years is that it would take more than a lifetime to collect statistically significant data to prove the model’s validity. Given that Professor Lichtman’s model only answers a binary choice, and gives no numerical values regarding the electoral or the popular vote, this problem becomes even more pronounced.

We have seen a similar prediction system completely implode recently: political scientist Helmut Norpoth’s primary model also only missed the extremely close 2000 election between George W Bush and Al Gore, decided by just 500 votes in Florida, between 1996 and 2016. In 2016, the model also projected Donald Trump to win the popular vote, which he did not, but given how unlikely his victory was thought to be by mainstream journalists and pollsters, Professor Norpoth’s model was still praised overall.

Unlike Professor Lichtman’s model, the primary model gives an actual numerical value for the percentage of chance of victory for a candidate—and this ended up being its demise. In 2020, the primary model gave President Trump a 91 per cent chance for re-election, but he ended up losing by a decent margin in both the Electoral College and the popular vote. He followed this up with another miss, giving President Biden, then Vice President Kamala Harris (who, by the way, did not win a single vote in the primary election…) a 75 per cent chance for victory. The eventual loss of Harris doomed Professor Norpoth’s model for complete irrelevancy.

It seems that the Keys to the White House model is heading towards a similar fate.

Lichtman himself has tried to argue that nine out of eleven correct predictions is a pretty good record. However, given that, as we wrote above, this model has only worked with a binary outcome, I believe it would need a damn near immaculate record to be considered significant or impressive. While an anomaly like the 2000 election, when the winner lost the popular vote and a single state decided the election by 500 votes, may be missed by a competent model and its interpreter, the 2024 election was not anywhere near as close.

All seven swing states went for the winning candidate, and only one of them was within a point. The winner also took the popular vote, and that margin was also not conspicuously close (such as in the 1960 election), 1.5 points.

'While an anomaly like the 2000 election may be missed by a competent model and its interpreter, the 2024 election was not anywhere near as close'

There are mountains of data to help predict the winner of a US presidential election: an immense amount of polling by numerous firms, both statewide and nationwide, primary turnout, early vote by registered party, economic data, or social media engagement. Any of these would have helped Professor Lichtman interpret his 13 keys, some of which are not entirely objective (such as the scandal key or the major foreign policy success key), to get the right outcome for the 2024 election.

And some people did do a lot better a job than him, a man who dedicated the majority of his life to this subject.

All major traditional bookmakers, as well as the decentralized prediction market Polymarket, had Donald Trump as the clear favourite to win on election day. Professor Lichtman, in all his wisdom, however, decided to highlight the one outlier prediction market that had it as an even contest and arbitrarily labelled it as ‘arguably the most reliable prediction market’.

In his final electoral map analysis going into the election, he had North Carolina going Democrat based on early vote data. Registered Republicans outright won the early vote in that state, and based on trends, Republicans were poised to win the election day vote as well. That is why even mainstream news outlets had reported that the Harris campaign was pulling resources out of North Carolina, believing it was out of reach, yet Professor Lichtman was for some reason unaware of these developments.

This all came after Professor Lichtman, in a perhaps even more embarrassing display, had made a series of statements across numerous media outlets that President Biden should stay in the race as he is poised for victory according to his model. In July 2024, President Biden’s approval rating by Gallup was just 36 per cent. Professor Lichtman likes to claim that, unlike his model, polls have no predictive capacities (even though he uses them at times to evaluate certain keys in his model…), but that is not the case for the Gallup presidential approval rating. It has been very much correlated with whether the incumbent is re-elected or not. At the time, President Biden was running nine points behind the President with the lowest approval rating in July of an election year that ended up winning re-election, Barack Obama.

Even worse for Lichtman’s reputation was the way he responded to failing his prediction for 2024.

He did publish a short video admitting he was wrong the day after the election. However, shortly after he most likely started to realize the implication of that statement—after all, he spent 40 of his 77 years on this Earth honing that model. So, soon after, he claimed in an interview on CNN that his model did not give an accurate output because ‘it relies on a rational, pragmatic electorate’ and blamed an abundance of misinformation for his failure. ‘Misinformation on the economy’ was an interesting point he was trying to make. It assumes that voters base their judgement on the economy based on statistics or talking points they are presented with, and not on the personal finances—which, evidently, is not the case.

So, Professor Lichtman, you can assume that the electorate is rational and pragmatic in your model in any election. You just need a rational, pragmatic model or a rational, pragmatic interpreter of that model now.

Related articles: