PoSSUM USA 2024 Presidential Campaign Vote Share Predictions (October, 17-26 2024)

Digital fieldwork conducted from 17/10 to 26/10. We find Harris ahed nationally by ~2% points, but…

…importantly, this popular vote margin is not enough, according to our data and MrP model, to overcome Trump’s electoral college advantage (estimated at ~ 3%) ! Hence PoSSUM’s final estimates suggests Trump has a 0.65 probability of winning, with around 1 week left.

Harris has a 0.77 probability of winning the popular vote, and hence the most likely outcome for this election appears to be a popular vote / electoral college split favouring Trump (probability 0.42).

What is driving this outcome ? As I have written before, the AI is bullish on third-parties hurting Harris wherever they are running. This has strong consequences in swing states, where alternatives on the Left are assigned ~ [4,6]\% of the vote.

PoSSUM estimates on R-D margin are highly correlated with 538 forecast, agree on every call with the exception of MI and WI, which we assign to Trump. The estimates disagree with Uniform Swing model on NV and MI. Zooming it on swing-states, the AI is 1% R-biased v. 538.

We now await election day. How will we judge the AI’s performance? We seek establish AI Polling as a legitimate method of socio-scientific inquiry. To this end, we will look at whether PoSSUM’s performance can be statistically distinguished from high-quality polls at multiple levels of analysis (national, crosstabs, states) and place PoSSUM in a ranking according to Bias, RMSE, Coverage and Accuracy ratings for each of these. Future work will involve assessing the AI’s ability to generate accurate individual-level labels.


For more details on the polling numbers…

The PoSSUM Poll

Crosstabs Report


Previous forecast