Which election forecasting models can you trust?

WASHINGTON — As voters turn to pollsters and political analysts to figure out who might win Tuesday’s presidential election, the feud between the nation’s two leading election forecasters, Allan Lichtman and Nate Silver, will soon be put to the test.

Lichtman, a US university professor who has correctly predicted nine of the last 10 presidential elections, predicted Vice President Kamala Harris would win.

Silver, a statistician and pollster who founded FiveThirtyEight, recently wrote in the New York Times that the race is virtually a dead heat, but his gut tells him former President Donald Trump is likely to win.

Lichtman and Silver argue over methods

The couple sparred on social media over the validity of their methods.

In September, Silver questioned whether Lichtman had correctly assessed the “13 keys” he uses to predict election results, arguing that the professor’s system actually favored Trump. Lichtman responded that Silver, who has an economics background, “was not a historian or a political scientist” and had made mistakes in the past.

“At least 7 keys, and maybe 8, are clearly in Trump’s favor. Sorry brother, but that’s what the keys say. Don’t you admit that they are completely arbitrary?” Silver posted a message on social media.

So, whose forecast is more accurate? And how do they even come to such conclusions?

Forecasting approaches

Lichtman developed the metrics he uses for his election forecasts more than three decades ago with the help of an earthquake specialist and mathematician in Moscow named Vladimir Keilis-Borok.

Dubbed the 13 Keys to the White House, the system uses—you guessed it—thirteen true or false statements based on historical analysis of the state of the country, parties, and candidates to determine who wins.

It includes questions about whether there is a third-party challenger, whether the party is “avoiding a White House primary” and whether any of the candidates are charismatic.

This method does not take into account how campaign messages or major events such as debates influence voter sentiment. Lichtman often makes his assessment months before an election and doesn’t change it until major foreign policy events occur.

If six or more statements are true, the challenger is expected to win. If there are five or fewer false results, the incumbent party is expected to win. In 2024, Lichtman said at least eight key candidates favor Harris.

But Silver uses a completely different strategy and data set to study the state of elections.

It builds probability statistical models based on national and state surveys, economic data, likely voter turnout and other factors. The model also adjusts for discrepancies in the polls it aggregates and gives more weight to polls it deems more reliable.

Forecast records

Lichtman has correctly predicted the outcome of nine of the last ten presidential elections since 1984. Which one did he get wrong? 2000 presidential race in which George W. Bush defeated Al Gore.

Silver gained national recognition in 2008 when his statistical model correctly predicted the outcome of presidential elections in 49 of 50 states. Since then, his model has predicted the outcome of the presidential race in 2012 and 2020. During the 2016 election, Silver’s model projected a likely victory for Hillary Clinton but gave Trump a roughly 30% chance of winning, much higher than most other forecasters.

Which model is better?

It depends on who you ask.

Thomas Miller, director of Northwestern University’s data science program, argues that Silver and Lichtman’s strategies are “wrong in different ways.” Miller created his own election forecasting system by combining 60 years of historical analysis and Predict It betting market data.

He suggested that Lichtman’s model does not take into account how campaign messages and major events change public sentiment in the final months of an election.

Nothing the campaigns do matters much, Lichtman said. The messaging doesn’t matter, the positioning doesn’t matter… because it’s all kind of predetermined by history,” Miller said. He also questioned whether the economic indicators Lichtman uses, which look at the U.S. gross domestic product, accurately reflect beliefs about the economy.

This year, for example, inflation is a major concern for many voters. The US economy is doing relatively well, but voters aren’t necessarily feeling it.

But Lichtman refuted these claims and said his economic analysis was objective and rooted in history, dating back to 1860. Each key is clearly defined based on this analysis, Lichtman said. He claimed that the lack of campaign events in his keys is one of the reasons they were so successful.

“What some people say is the weakness of the keys… it’s the strength of the keys because they look at the fundamental factors rather than the ephemeral events of the campaign,” Lichtman said. He said the structural model reflects how American presidential elections actually work.

Miller also saw flaws in Silver’s approach, namely that he relied too heavily on polling data, which varies and can be flawed. If the polls are inaccurate, Silver’s forecasts will be inaccurate.

Weighting polls based on which groups of people are more likely to vote can also be difficult, Lichtman said. For example, polls may have underestimated the number of Democrats and Republicans turning out to vote.

David Wasserman, an elections analyst at the Cook Political Report, said that despite the variability, he found Silver’s approach “methodologically more rigorous.”

“Lichtman is ridiculously opinionated and doesn’t allow for any subjectivity in his method,” Silver said in late September, “but you do learn a lot about presidential elections from reading his work, and he at least puts himself out there. make testable predictions.”

Wasserman said he believes Silver’s approach is better suited to “communicating election results to the public,” in part because he “recognizes that there is inherent uncertainty in polls and future events.”

“I believe that campaigns matter… and that the choice of candidate influences voters’ thinking,” he said. “I have more confidence in Silver’s approach because he can take these factors into account.”

However, at their core, the models are completely different.

While Lichtman’s model relies on established patterns from past elections to predict future presidential votes, Silver’s model provides insight into how the views of the American electorate change over weeks and months.

Which election forecasting models can you trust?

Lichtman and Silver argue over methods

Forecasting approaches

Forecast records

Which model is better?

Archives

Categories