The Polls Show Harris and Trump Are Tied — My Model Shows a Clear Favorite
- *This model will not be updated after June. The last update was on June 29th**
- *Article update since Biden has dropped out: this model is designed for the incumbent party, not the incumbent. Meaning this still holds for Harris despite the model being constructed while Biden was still running**
The 2024 presidential election is heating up in mid-October, and Kamala Harris and Donald Trump find themselves in a close race. Harris is slightly ahead in national polls, but the battleground states, including Arizona, Georgia, and Pennsylvania, remain toss-ups. Voter sentiment has fluctuated, with both campaigns focusing heavily on key issues like inflation and immigration. Polling data shows Harris with a narrow lead, yet with weeks to go, these numbers could easily shift depending on late-breaking developments.
While established models like those from FiveThirtyEight incorporate vast amounts of polling and economic data, your simpler logistic regression model cuts through the noise. Focusing on the unemployment rate, the incumbent party’s approval rating in June, and the number of years the incumbent party has held power, this model hones in on the fundamental indicators of electoral success. This approach could provide a clearer prediction by reducing complexity and focusing on the metrics most likely to influence voter decisions in November.
Model
The beauty of this model lies in its simplicity and the strategic selection of variables. Despite its simplicity, it provides a powerful and accurate prediction of election outcomes. The model’s variables are:
- Unemployment Rate: A key economic indicator that reflects the country’s economic health.
- Incumbent Party’s Approval Rating in June: This variable captures the public’s satisfaction with the current administration and serves as a proxy for overall voter sentiment.
- Years in Power of the Incumbent Party: This accounts for the potential fatigue or continued support for the incumbent party.
These three variables encapsulate significant aspects of the political and economic landscape, providing a distilled yet effective prediction framework. The data spans every presidential election from 1952 to 2020. The pseudo-R² value for this model sits at around 70%, and robustness tests (two linear regressions with the same variables testing the margin of victory in the popular vote of the incumbent party (predicted value: +3.055%) and electoral college vote share of the incumbent party (predicted value: 277.08)) came back with similar results in terms of R² value and its predicted winner.
Model Analysis
The Logistic Regression Model
The logistic regression model used is as follows:
logit(p)=3.8277−0.4637×Unemployment rate+0.1372×Approval rating in June−1.3183×Years in power of the incumbent party
Here, p represents the probability of the incumbent party winning the election.
Let’s consider the following values for the 2024 election:
Using these values, the linear predictor (LP) is calculated as:
Calculate the LP:
LP = 3.8277–0.4637 × 4.0 + 0.1372 × 39.9–1.3183 × 4
LP = 3.8277–1.8548 + 5.46678–5.2732
LP = 2.16648
Calculate the Odds of Incumbent Party Victory:
Odds = e^(LP)
Odds = e^(2.16648)
Odds = 8.724
Calculate the Predicted Probability:
Probability = Odds / (1 + Odds)
Probability = 8.724 / (1 + 8.724)
Probability ≈ 0.898
This means that, given the current conditions, the predicted probability of an incumbent party victory in the 2024 election is approximately 89.8%.
Now along with the two robustness tests performed for this model (popular vote win: 3.055%; electoral college vote total: 277.08), we can apply to the map to try to predict how the 2024 map should look like using current polling data and some subjective calls.
Insights from the Model
While traditional models incorporate a vast array of polling data, this model demonstrates that focusing on a few, well-chosen variables can yield highly accurate predictions. The approval rating, in particular, is a powerful variable as it captures the essence of voter preferences and sentiments. Economic conditions, as reflected by the unemployment rate, and the historical context of the incumbent party’s tenure also play crucial roles.
Conclusion
In summary, this logistic regression model provides a straightforward yet powerful tool for predicting election outcomes. By focusing on three key variables, it offers a clear and concise analysis that is easy to interpret. As we approach the 2024 election, this model suggests a high probability of an incumbent party victory given the current economic and political conditions.
So, from this model (which is now predictive due to Biden’s June approval rating), Harris looks likely to win.
