US election 2024: beware polling predictions as they can be wrong – but here's an approach which has often been on the money

Despite the avalanche of legal indictments, Donald Trump remains favourite to win the Republican nomination for the 2024 US presidential election. According to a poll reported by the website Real Clear Politics on August 19, he has a 41% lead over his main rival Ron DeSantis, the governor of Florida, in the Republican nomination race. His lead over the other Republican hopefuls such as former vice president Mike Pence is even larger.

If he does win the Republican nomination the question is: can he win the presidential election in November next year? Real Clear Politics also reported a poll showing that he is running neck-and-neck with Joe Biden, each having 44% in voting intentions. If the election took place tomorrow, he would have a real chance of regaining the presidency.

Forecasting presidential elections

There is a lively community of political scientists using a variety of different methods to forecast elections, with many focusing on the US. Most forecasting models use polling data, but since we are 15 months away from the presidential election, current polling should be treated with caution. This is because it reflects public opinion before the full-scale campaign has even begun.

It should be noted that US pollsters have had a mixed record in forecasting elections. This was particularly true in 2016, when pretty much everyone predicted a Hillary Clinton victory.

But there have been successes as well as failures. For example in the 2008 election, statistician and pollster Nate Silver, accurately predicted the winner in 49 of 50 states. Equally in the 2020 election, statistician Andrew Gelman working with the Economist magazine successfully called it for Joe Biden. But it is instructive that neither of these forecasts relied solely on polling data.

Electoral college

In the past, most US forecasting models have tried to predict vote shares in presidential elections – but this produces an additional source of error. The election is determined by who wins the electoral college, not the popular vote. In the 2016 election Clinton won a larger vote share than Donald Trump but lost the contest in the electoral college.

The electoral college was created by the US founding fathers, with delegates chosen to reflect voting support for the candidates in each state. There are 538 delegates altogether, made up of 435 from the constituencies in the House of Representatives plus two from each of the 50 states and three from Washington DC. A candidate must get the support of at least 270 to win.

The idea behind the electoral college was to create a firewall between the voters and the presidency and fill it with representatives elected by each state who then cast their votes according to the popular vote in their state. The concern was to prevent demagogues from becoming president resulting from a wave of voter enthusiasm for a particularly extreme candidate. Ironically, in the 2016 election it worked in reverse, delivering victory to Donald Trump who had lost the popular vote.

A forecasting model

There is an approach which does not rely on polls. Instead, it looks at voting in previous presidential elections to see what this tells us about the 2024 contest. The analysis uses a century of elections from 1920 to 2020, and a relatively simple model has a good track record in predicting elections over this period. It uses two variables to predict the Republican share of the delegates in the Electoral College, using a technique called multiple regression. The forecasts and the outcomes appear in the chart below.

The first and most important predictor in the model is the state of the economy, with an incumbent being rewarded for a good record on economic growth and punished for a poor one. The logic of this is simple: the voters will throw out an incumbent who fails to deliver prosperity and choose their rival instead.

Forecast and outcomes of the Republican share of the electoral college votes in US presidential elections 1920 to 2020

The second predictor is the ideological position of the Democrat candidate in the election. If they are very left wing, that will boost support for the Republicans, but if they are centre-left, this will reduce Republican support. Interestingly enough, the Republican candidate’s ideological position does not affect the party’s share of the Electoral College vote.

To give an example, the Democrat candidate in 1952, Adlai Stevenson, was very left-wing and he lost to Republican candidate, Dwight D. Eisenhower. Some 40 years later in 1992, centre-left Democrat Bill Clinton beat the incumbent Republican president George Bush senior.

The data for measuring ideology comes from the Manifesto Project, an international research programme. Its researchers use a technique called content analysis to translate the policy proposals made by party candidates in their election manifestos into scores on a left-right ideological scale.

The model takes into account unusual events occurring over the period that can distort the results if they are ignored. One is Franklin Roosevelt’s “New Deal”, which shattered support for the Republicans in the 1936 election when the latter party won only eight electoral college delegates. Another unusual event was the Watergate scandal, which led to the resignation of Republican president Richard Nixon prior to the 1976 election. Finally, Donald Trump was badly damaged by the COVID crisis in 2020.

The forecast for the 2024 presidential election appears in the chart. The prediction is that the Republicans will win 47% of the representatives in the Electoral College (253), and the Democrats 53% or 285. Needless to say this is uncertain since the model is not a perfect fit to the data and so subject to errors. In addition there are still 15 months to go until the election and, judging by the febrile state of US politics at present and the trials (if not the tribulations) of Donald Trump between now and next November, there’s a high level of unpredictability involved. But for now it looks like Joe Biden will be a two-term president.