Suddenly, The Presidential Race is a Tossup Again
Kamala is surging, but I wouldn't say she is "favored" yet
When I last posted, the Democratic party was in turmoil after Biden's unexpected withdrawal from the presidential race. Most people seemed to agree that stepping down was the right move for him, given the circumstances. I’ll cite a passage from my last post here that I think summarizes well why Biden was in such a bad situation and give myself a pat on the back:
Directionally, I think it’s good for Democrats that Biden dropped out. He already was in deep deep trouble electorally, with my latest forecast giving him only a 23% chance of winning (a projection mostly aligned with Nate Silver's). This decline wasn't due to a mere gaffe or temporary polling setback. Rather, Biden's performance in the first debate failed to meet basic expectations, making a substantial comeback seem improbable given his current state.
The past three weeks have been pretty great for Democrats. Trump's odds of winning the presidency on Polymarket have plummeted from a high of 72% to just 46%. The party swiftly united behind Harris, demonstrating somewhat unexpected solidarity in their focus on defeating Trump. Campaign donations (and general enthusiasm) are surging for Democrats, while JD Vance already seems like a weak VP selection for Republicans. Moreover, Trump seems to be panicking and reverting to his less disciplined persona, exemplified by his undisciplined appearance in front of the National Association of Black Journalists. Kamala is now polling ahead of Trump nationally, and just got some super good numbers in a NYT poll as I was writing this up.
Initial Harris v. Trump Forecast Launch
To cut to the chase, my initial forecast gives Kamala Harris a 47% chance of defeating Donald Trump this November. This is a substantial improvement from the 23% chance I gave Biden on the day he dropped out. While conventional wisdom has swung dramatically in Harris's favor, with many now considering her the frontrunner, it’s important to note that current data doesn't quite support this optimism. If the election were held today, it would be essentially a tossup.
The forecast outlines a battleground of 7 key states, 6 of which are tossups (MI, WI, PA, NV, GA, and AZ) and one of which leans towards Trump (NC). Harris's clearest path to victory involves securing all reliably Democratic states plus Michigan, Wisconsin, and Pennsylvania. This combination would yield exactly 270 electoral votes – the minimum required to win the presidency.
Republicans have a 48% chance to retain control of the House and an 83% likelihood of flipping the Senate. Their odds of winning a federal trifecta – control of the White House, Senate, and House – stand at 40%. In contrast, Democrats face a mere 17% chance of securing their own trifecta. The most probable outcome, with a 43% chance, is split government control. This scenario likely involves either a Trump presidency with a Republican Senate or a Trump White House coupled with a Democratic House.
Methodology Stuff
I waited a little longer than some to launch my initial Harris v. Trump forecast, just because I wanted to get a critical mass of polling to base a forecast on before putting anything out there. By disregarding hypothetical matchup polls from before Biden's withdrawal, I aimed for a more accurate prediction based on more predictive voter sentiment. Now that the forecast is live, I'll update it bi-weekly for the next few months, transitioning to daily updates in the weeks immediately preceding the election.
The forecasting model combines fundamental data (incumbent performance, economic indicators, past election results, national political climate) with polling data at national and state levels to generate point estimates for each US Presidential, House, and Senate race. The model then runs thousands of correlated simulations daily, accounting for individual race variability and inter-race result correlations. The center of the process is a 524x524 covariance matrix encompassing all federal races. The covariance matrix, containing over 250,000 cells, is populated through a combination of historical election data, backtesting results, and expert knowledge. This approach helps avoid overfitting, as seen in some past election models.
Electoral College Bias: Expected Republican Edge Narrows
Based on a lot of conversations with people I’ve had recently, conventional wisdom is probably overestimating Kamala Harris' chances of winning the presidency compared to data-driven forecasts or betting markets. There are two main reasons for this:
Positive momentum: Harris has recently gained more visibility and enjoyed positive coverage, which naturally leads people to be overly optimistic about her prospects.
Overlooking electoral college dynamics: Many people still don't internally fully account for the current bias in the electoral college system that tends to favor Republican candidates. This pro-GOP bias means a Democrat like Harris would need to win the popular vote by a significant margin to secure an electoral college victory.
Gun to my head, I guess I would take Harris over Trump at this point, but I think it’s exceptionally close. I am not afraid to break with betting markets (I think the Balance of Power one is absolutely ridiculous and would *hammer* NO on Democrats sweep) but I think Harris/Trump at 51-46 is almost exactly right.
While most reasonably politically engaged individuals understand the concept of electoral college bias, especially after it cost Hillary the election in 2016, it's easily overlooked amid a barrage of headlines from major outlets like the New York Times and CNN proclaiming "HARRIS NOW LEADS TRUMP." Sure, Harris going from losing by 2% to Trump in national polls to winning by 2% in the past month is really good for Democrats, but she still would be more likely than not to lose if she only wins nationally by 2%. Quantitative election forecasts and betting markets provide valuable context and nuance, helping us see beyond these simplistic national poll narratives.
The pro-Republican Electoral College bias is largely due to random chance—unlike the Senate, where the Republican advantage stems from their consistently stronger performance in lower population states (creating a structural bias that's likely to persist indefinitely). Currently, Democrats face a disadvantage in the EC because seven pretty large states (combining for 93 electoral votes) are a couple points more Republican than the nation as a whole. This was especially notable in 2016 when Hillary Clinton won the popular vote but lost the Electoral College, but also in 2020 when Biden needed to win a 3.8% popular vote margin to reach 270 electoral votes), Simulations suggested the Electoral College bias would decrease from 3.8% to between 2.5-3% in 2024, primarily due to natural variations in key battleground states.
The initial Harris-Trump forecast predicts only a 2.4% pro-GOP structural bias in the Electoral College - the first time this cycle it's dipped out of the 2.5-3% range. This shift is largely due to promising Democratic polling in key swing states, particularly Wisconsin, Michigan, and Pennsylvania. If these closely correlated states trend left relative to the nation as a whole, it could significantly reduce the Electoral College bias. However, I remain cautious about Harris's ability to maintain such an efficient Electoral College coalition. I expect her to slip (in a relative sense) with older, whiter, and more socially moderate or conservative voters in the upper midwest after her honeymoon period ends, but it's getting hard to ignore the mounting pile of evidence that Democrats will run well in these three states. Current polling averages from FiveThirtyEight and DecisionDeskHQ/The Hill suggest just a 0.9% and 0.4% Republican bias respectively, but the forecast is more conservative after considering fundamental factors.