2024 Presidential Election Prediction
In this article, I will explain how I used an election simulator to predict the 2024 Presidential Election. The election simulator was initially developed as part of a political simulation game called The Political Process. In the game, players campaign for elected positions at the city, state, and national levels. In order to calculate realistic results for these elections, a sophisticated election simulator was developed. I thought that such a simulator would also be a fun tool for predicting real life elections, such as the 2024 presidential election. But before I attempt to make a prediction, it may be useful to learn a little bit about how the simulator works.
The Election Simulator
Essentially, the simulator is a complex algorithm driven by real life data. It is capable of calculating election results at all levels of the government. It does this by using demographic and electoral data gathered at the county level. Every county in the United States is simulated. Each county has unique variables for political party demographics, voter turnout, and voter registration. These variables were determined by using election results from different election types across multiple election cycles. This makes the simulator more flexible because it does not rely on a single source of data to make its predictions. It recognizes how election results can fluctuate under different circumstances and incorporates that into its algorithm.
The simulator also incorporates semi-predictable electoral patterns and voter behavior. For instance, it adjusts results based on party loyalty, party fatigue, and government approval ratings. Characteristics of each candidate are likewise factored into the prediction: policy positions, campaign platforms, character traits, approval ratings, voter enthusiasm, and so on. This complex grouping of factors works together to calculate results. All of these features combined put the simulator in a good position to predict real life elections. But, in order to maximize its accuracy, the simulator first needs to be calibrated.
Calibrating the Simulator
By default, the simulator has been balanced to create a fun and dynamic environment for players of The Political Process. The default settings work well for the game environment, maintaining balance and creating believable election results throughout the player’s political career. But to accurately predict real life elections, it may need some adjustments. Towards this end, I figured that the best way to calibrate the simulator would be to recreate the conditions of the 2016 and 2020 presidential elections. If the simulator could accurately recreate the outcomes of those elections, then it would be in a good position to accurately predict the 2024 election. That’s the idea. If you’re not interested in learning about the calibration process, you can certainly skip ahead to the prediction results, but I think the calibration process provides some interesting insights into the process of predicting elections. The short version: with a few adjustments, I was able to use the simulator to recreate the outcomes of the 2016 and 2020 elections. For the more thorough explanation, continue reading.
The 2016 Presidential Election
I started with the 2016 presidential election. All of the preliminary variables were adjusted to match 2016 conditions. The simulator needs to know things such as which party is in control of the presidency and how long they have held the presidency. It also needs to know the government approval rating at the time of the election. These background factors contribute to things like voter turnout and how likely voters are to support one political party over the other. I also provided the simulator details about each candidate, including: relevant policy positions, campaign platforms, character traits, and voter enthusiasm. With the inputs provided, the simulator was ready to make a prediction. The outcome of its prediction: an electoral college victory for Donald Trump (273 – 258). These results are promising; the simulator predicted the correct outcome, but it got the details wrong. If you recall, the real life results were 304 – 227. The simulation mistakenly gave Arizona and Pennsylvania to Clinton instead of Trump. Some slight adjustments were clearly needed.
I decided I would try increasing the “Party Fatigue” value for Independent voters. “Party Fatigue” is the term I use to represent a specific pattern of behavior where voters are less likely to support the current political party in power. Voters become fatigued with the current administration and want a change. The longer a political party holds the presidency, the more likely voters are to support the opposing party. By increasing the party fatigue value for independent voters, I made it so independent voters are more likely to support the opposing party candidate (in this case Donald Trump). After making these changes, I reran the simulation and the new results aligned with the real life results: an electoral vote of 304 to 227. Also, just like in real life, the simulator predicted a popular vote win for Hilary Clinton, despite her losing the electoral college.
The 2020 Presidential Election
With the calibrations for the 2016 election completed, would the calibrated simulator now be able to accurately calculate the 2020 results? The initial settings were adjusted to match conditions in the 2020 election. The presidential party was switched to Republican, Hilary Clinton was replaced with Joe Biden, and Donald Trump was left the same. Nothing else was changed. The Results: the simulator predicted a win for Biden: 290 – 248. This is very close to the actual results of 306 – 232. The only state this initial prediction got wrong was Georgia, which it gave to Trump instead of Biden. This is not too surprising considering how close the real life election was with a vote margin of approximately 0.2%. The simulator appears to be doing a decent job predicting elections.
After a few minor modifications, I recalculated the results and Biden was now predicted to win Georgia as well (by approximately 14,000 votes). The predicted results matched the real life results. Just to make sure these adjustments were not too biased towards the 2020 outcome, I recalculated the 2016 results using the same adjustments. The outcome was still accurate and I felt fairly confident that I could use these calibrations to make a reasonable prediction for the 2024 presidential election.
The final step before making a prediction was to input the current 2024 conditions into the simulator, replace Joe Biden with Kamala Harris, and make some minor adjustments to Donald Trump’s character. With all of that complete, the simulator was ready to make a prediction. Continue scrolling to reveal the results.
Prediction Results
Prediction Analysis
So what do you think about this prediction? Will it be right? Will it be wrong? Is it too close to predict? I will attempt to analyze the details of this prediction below. But first, let’s take a general overview of the results. Kamala Harris is predicted to win with 287 electoral votes to Donald Trump’s 251. As a reminder, a candidate needs 270 votes to win. So, we can first tell that the simulator is not predicting an overwhelming win for Harris. If one or two states were flipped in Trump’s favor, it would alter the outcome. Let’s look at that possibility. The predicted margins are very close in Arizona and Wisconsin. The simulator predicts that Kamala Harris will win Arizona by only 3,375 votes. It predicts that she will win Wisconsin by only 5,403 votes. These margins are even closer than the results in 2016 and 2020. And with margins that close, it is easy to see how this prediction could be wrong. Any small adjustment could change the outcome. For other swing states such as Michigan and Pennsylvania, the predicted margins are also very close: 0.6% and 1.6% respectively. So what were the major factors that contributed to these predictions and if they were adjusted, how would it alter the results?
Mood of the Country
The first factor I will analyze is the “Mood of the Country”. In the simulator, this metric is referred to as Government Approval. But, it essentially represents whether voters are satisfied with the direction of the country. Generally, if satisfaction is low, the simulator will reduce voter turnout for whichever political party is perceived to have power at the time of the election. In the 2024 election, Democrats have control over the presidency and the senate. This means that turnout should decrease among Democratic voters. The question that needs to be answered is: how dissatisfied are voters with the current direction of the country and how much will that influence turnout? To try to answer this question, I looked at some historical poll data from Gallup [Source]. In October 2016, 28% of voters were satisfied with the direction of the country. In October 2020, only 19% of voters were satisfied. And in October 2024, the percentage of voters satisfied with the direction of the country is 22% (slightly better than 2020).
Based on this information, the mood of the country seems fairly similar to how it was in 2020. If that is the case, the government approval (mood of the country) metric should stay the same as it was in the 2020 calculation. Right? If I had chosen to do that, the margin of victory for Kamala Harris would have been much better: a victory of 319 to 219. She would have gained two additional states: North Carolina and Georgia. But I did not choose to keep it the same. Instead, I decided to lower it, significantly.
The main reason for this decision was the inflation rate. In 2020, the inflation rate was approximately 1.2% right before the election. In 2024, it has doubled to 2.4% [Source]. In terms of election outcomes, this may be an insignificant difference, but I felt like it should have an impact. How much people pay for everyday things like groceries seems like a generally good indicator of how people feel about the direction of the country. If that is true, the government approval rating should be lowered. How much is unclear. I did not want to risk underestimating the mood of the country in this prediction.
Voter Enthusiasm
Another deciding factor in this prediction is voter enthusiasm, which has a significant impact on voter turnout for each political party. Candidates are assigned a voter enthusiasm value relative to an average candidate from their political party. If a candidate’s voter enthusiasm is below average, it decreases turnout among party voters. If it is above average, it increases turnout. When setting these values, it is a matter of determining whether Kamala Harris and Donald Trump have more or less voter enthusiasm relative to average Democratic and Republican candidates.
In the case of Kamala Harris, I decided to give her a slightly above average enthusiasm rating relative to a generic democratic candidate. But the adjustment was small, only 2% more than the average politician. If she was given the standard enthusiasm, she would still win the election, but Wisconsin would flip to Trump. What do you think? How enthusiastic are democrats to vote for Harris relative to a generic democratic candidate? This is certainly a point where the prediction could get things wrong.
In the case of Donald Trump, I decided to continue using the voter enthusiasm levels used during the 2016 and 2020 predictions. The thought process here is that republican enthusiasm for Trump is unwavering. For comparison purposes, Trump’s enthusiasm rating is 22% above the enthusiasm for an average Republican candidate. (That is a lot more than the 2% used for Kamala Harris.)
Some readers may think this level of enthusiasm is too high. Given everything that has happened since the 2020 election, it’s reasonable to assume that Republican enthusiasm for Trump could decrease in the 2024 election. But I don’t want to risk making that assumption, especially given how much loyalty the Republican party has continued to show to Trump.
Regardless, I decided to do a test. If Trump’s enthusiasm was reduced to that of an average Republican candidate, the simulator would predict him to lose Georgia, which he is currently predicted to win by 86,000 votes.
Candidate Likability
A third major factor contributing to the prediction results is candidate likability. There are many components to this, including a candidate’s campaign platform, age, and personality traits. The campaign platform represents’ a candidates policy positions and priorities on the campaign trail. Voters are more likely to support a candidate whose policy positions align with those of the voters.
Age has a small impact when calculating results: if a politician is perceived to be too young or too old for a position, it could sway some voters. In the case of Trump, his age causes him to lose votes. The impact is minor, but it is enough to change the results. As an example, we can look at the results in Wisconsin. If age was not a factor, Trump would gain an additional 6,800 votes. That is enough to flip Wisconsin in his favor, where he is predicted to lose by 5,400 votes.
Candidate traits are another meaningful factor in the prediction. In the simulation, traits can be positive or negative. They are used to influence how likely voters are to vote for a candidate. This is only a small influence compared to all of the other factors, but it can add up. For the 2024 election, I added some additional traits to Trump’s character. I decided to add two generic “Scandalous” traits to account for everything that has happened since the 2020 election. These traits are used to represent the negative voter perception associated with things like the January 6th Capital Attack, the multiple legal cases against Trump, the felony conviction, and so forth. These two traits alone are enough to determine the outcome of the election. If both are removed, the simulator would predict a Trump win of 276 – 262. For reference purposes, removing both traits increased Trump votes in Wisconsin by only 25,000 votes. As you can see, this has only a small effect, but in an election this close, it could make a difference.
Third Party Candidates
The impact of third party candidates in this simulation is very interesting. But, before I get into the details, we should think about the role that third party candidates play in elections. It is tempting to think of third party candidates as “stealing” votes from the two major candidates in an election. But this may not be an accurate way to think about it. Realistically, these voters know that they are “wasting” their votes by voting for a third party candidate, and yet they are choosing to do it anyway. It seems reasonable to assume that, without the presence of third party options, these individuals would be non-voters. They do not like Harris or Trump enough to vote for either of them and so would choose to vote for neither.
With that consideration in mind, let’s look at the impact of third party candidates in this prediction. As part of the simulation, I included the Libertarian party candidate, the Green party candidate, and Robert F. Kennedy Jr. Technically, Kennedy has withdrawn from the election and endorsed Donald Trump, but his name still appears on the ballot in multiple states, so I have included him in the simulation. His character has been adjusted to reduce his impact on the election.
As a thought experiment, I wondered what would happen if third party candidates were not included in the simulation. I did a test. Without the inclusion of the third party candidates, the simulation would predict Trump as the winner with 287 electoral votes to 251. The simulation assumes that these voters would switch their vote to a different candidate, but it is unclear whether that would actually happen in real life. They may simply choose to not vote for the presidency. Regardless, the third party candidates are not going to be excluded from the real life election, so they should not be excluded from the simulation. That being the case, it is important to evaluate whether they are being over or under represented in the results.
Robert F. Kennedy Jr. is predicted to win approximately 0.3% of the vote, in states where he is on the ballot. These votes come mostly from Republican and Independent voters. In a state like Wisconsin, he is predicted to win over 9,500 votes. If you recall, Harris was predicted to win Wisconsin by only 5,403 votes. It is a significant consideration. How many voters will actually continue to vote for Kennedy now that he has withdrawn from the election, and will any of them switch their votes to Trump?
The Libertarian candidate is predicted to win 0.8% of the vote. Most of those votes will come from Republican and Independent voters. For comparison, the Libertarian candidate won approximately 3.28% of votes in the 2016 election and 1.18% of votes in the 2020 election. So the simulator is underestimating the Libertarian candidate in this prediction. If it was more aligned with the real life results of previous elections, it would take even more votes from Trump.
The Green Party candidate, Jill Stein, is predicted to win 1.3% of the vote. Most of those votes will come from Democrats and Independents. In the real life 2020 election, the Green party candidate only received 0.26% of the vote. But in 2016, when Jill Stein was the nominee, she received 1.07% of the vote. Given this information, the simulator’s prediction of 1.3% for Stein might be too high. If that is the case, you could expect slightly more votes for Kamala Harris.
A fourth independent candidate, Cornel West, was not included in the simulation. But if he were to take votes from any candidate, it would likely be Harris. It may also be the case that he could take votes from Jill Stein, as he was originally a candidate in the Green Party’s primary elections.
Conclusion
As we’ve seen throughout this article, there are many factors that could shift the outcome of the election. Small changes could make a difference. I’ve attempted to provide the simulator with reasonable inputs for predicting the election, but it’s certainly possible that I have gotten these inputs wrong. Additionally, the simulator may not be specialized enough to account for all of the factors that end up deciding this election. For one thing, it has not factored in the campaign process: what voters are the campaigns targeting, where are campaigns using the majority of their resources, how much effort is going into voter registration. Ultimately, this article was just a fun way for me to think about the factors that influence elections. I hope you have enjoyed it as well. I’m looking forward to seeing the real life results and comparing them to this prediction.
For Your Consideration
If you liked this article, and you are interested in predicting elections for yourself, the election simulator used in this prediction is available as part of The Political Process game. Additionally, if you are the type of person who enjoys thinking about politics and government policy, you might like playing the game’s campaign mode. The campaign mode allows you to create a custom character, run for political office, write legislation, balance budgets, ascend the political hierarchy, and watch as your policies shape the nation. You can learn more about the game here.