Predicting the 2024 Presidential Election

I built a model to predict the outcome of the 2024 presidential election. Here’s how I did it and what I found.

I began by compiling state-level voting margins for all past presidential elections beginning in 1976. However, simply using voting margins does not make for a good prediction model, so I made a series of adjustments, as outlined below.

Adjustment 1. First, I adjust for the popular vote. Candidates win electoral votes because of nation-wide political sentiments existing at the time of elections (great examples are Reagan in 1984 and Clinton in 1996). Because such political sentiments are unique to specific election years, I subtract the popular margin from each state’s raw margin. When I refer to a “state’s value” I mean this adjusted voting margin.

Adjustment 2. Adjusting for the popular vote is not good enough. There are also unique circumstances which must be taken into consideration. For example, in both 1992 and 1996 Bill Clinton fared better in Arkansas than another candidate would have (even after adjusting for the popular vote) because Clinton was from Arkansas and had previously won state-wide elections in Arkansas. Another example is 2016 spoiler candidate McMullin who received 21.5% of the the vote in his home state of Utah. For these unique cases, I manually adjust each state’s value to make it inline with what it was the years before and after the biased year. To do this, I graph a plot of the history of the particular state’s values, and set the adjusted value of the state in the biased year equal to the value which makes the plot the smoothest. It is important to adjust for these unique circumstances so that the political trending weights (Adjustment 3) are meaningful. You can see a list of unique circumstances and related adjustments in the fourth tab of my excel sheet, which is downloadable below.

Adjustment 3. Next, I take each state’s political trending into consideration. Specifically, I take a weighted-average of the change of each state’s values between presidential elections, weighing more recent election shifts more heavily. I attempted many different permutations of weights, and the best one (in my opinion) is assigning the shift from the 1996 election to the 2000 election a weight of 1.0, and increasing each subsequent shift by increments of 1.0. I decided not to include shifts pre-1996 in the model — or rather I assign pre-1996 shifts a value of 0.0 — because the political landscape was dramatically different after Reagan and Bush.

Adjustment 4. We are almost done. By this point I have estimated adjusted 2024 state values. I must unadjust these values so that I can estimate electoral votes per state to determine a predicted winner. Thus, this final step requires an estimate of how much I predict each state will be above or below the popular vote in 2024 which necessitates estimating the 2024 popular vote margin. In 2020 Democrats won the popular vote by a 4.4 point margin; I estimate the popular vote in 2024 will be Democrats winning by a 5.0 point margin because Biden is the incumbent and he has not (yet) had any major scandals.

Results of the 2024 Presidential Election | Original Predictions, August 2021

My model predicts Democrats will win the 2024 presidential election with a total of 309 electoral votes, while Republicans will have a total of 229 electoral votes:

This outcome is discussed in my post 2024 Presidential Election: Democratic Pathways to Victory (see Scenario 3).

Results of the 2024 Presidential Election | Revised Predictions, August 2022

It is now August of 2022, and much has changed over the past year. My revised prediction, as described below, is that Republicans will win the 2024 presidential election with 279 electoral votes, while Democrats will have a total of 259 electoral votes:

In my revised prediction I keep my first three adjustments the same, and only change the estimated 2024 popular vote margin from Democrats winning by a margin of 5 points to Democrats winning by a margin of 3 points. Why did I make this change? Biden has still not had any major scandals, so I still think the Democrats will win the popular vote. Additionally, the Democrats have won the popular vote seven out of the last eight elections, and I do not think this will change. However, the economy is suffering, and the Democrats have not yet made any meaningful changes. I think that the democratic nominee will perform better than Hillary did in 2016, when she won popular vote by a 2.1 margin, but worse than Biden did in 2020, when he won the popular vote by a 4.4 margin.

The 50-vote swing between my original prediction and my revised prediction is attributable to three states: Michigan, Pennsylvania and North Carolina, as these three states are trending Republican faster than other swing states, after taking into consideration the popular vote margin.

Lessons Learned from the Data

By studying how each state has voted over time relative to the popular vote, my model allows me to assess presidential voting trends on a state level. Based upon these trends, I believe that the swing states of the 2024 election will be, in increasing order of how Democratic I think they will vote, Georgia, Arizona, Nevada, Michigan, North Carolina, and Pennsylvania. My data suggests that Wisconsin and Florida are trending Republican too strongly for the Democrats to stand a change at winning these states in 2024, unless the nation as a whole has a dramatic shift left. My model predicts that Michigan will be the tipping point state; assuming each party wins states in the order my model predicts, the party which wins Michigan will win the election. More specifically, my model predicts that if the popular vote margin is less than 3.38 (for the Democrats), the Republicans win Michigan and they win the entire election. Conversely, if the poplar vote margin is 3.38 or higher, Democrats win Michigan and they win the entire election.

Model Benefits

Past presidential polls, both public polls and campaign funded polls, have large margins of errors. These errors are due in part to numerous issues, including sampling error and unpredictable turnout. These errors are not without costs — they can lead to massive misallocation of resources, both financial resources and human capital — which in turn can generate avoidable election results on both sides of the aisle. While the past cannot perfectly predict the future, with the proper analyses, I believe it possible to build models using past behavior to predict future outcomes.

Due to the flexible nature of the four adjustments in my model, my model is equally useful as a simulator to predict the 2024 presidential election if either Democrats or Republicans win a certain percentage of the popular vote. By capturing assumptions from a representative sample of voters, my model may be able to generate a more accurate prediction.

Model Limitations

My model uses past election data to predict the future, and places greater weight on more recent political trends. Thus, my model assumes 2024 will not be drastically different from 2020. However, if the future is nothing like the past, my model will not be as accurate. This can be adjusted by altering the weights used in Adjustment 3.

Another limitation of my model is that it does not account for a specific future candidate doing especially well (or poorly) with certain demographics. For example, the Rust Belt tends to vote in tandem, and my model does not currently adjust for the impact of voting regions.

Finally, while I manually adjust for past unique considerations (see Assumption 2), I cannot adjust for future unique circumstances with unknown future candidates.

What Do You Think?

You can download my data here.

You can make your own predictions in the Dashboard and see the results at the bottom of the page.

Please send me your thoughts/comments below. What to get in touch to learn more? Here is my email: gideon@heltzers.com.