This post will cover a few small changes that have been made since the posting of this article, concerning how I will be using this model throughout the rest of the 2025 season.

To start, I mentioned I would start posting predictions in Week 4. However, due to a much busier September than I anticipated, I am only now releasing predictions just ahead of Week 6. I will also release the model’s Week 5 predictions and its results shortly after this post — I had time last week to code the predictions script, but not enough time to publish the articles themselves.

On that note, due to time constraints, predictions articles will be very formulaic — there will be very little, if any, journalistic aspects to them. I will be treating them mostly as a means for documenting the model’s performance, and not much more than that.

Next, there is one small issue with the models — both the large “base” model, and the slimmed down model. Early in the season, some teams simply won’t have accrued any stats in a particular category that the model uses, which causes the model to output “NA” for its prediction. Simply put, a small handful of games won’t be predicted early in the season.

Lastly, I am aiming to finish coding the neutral site logistic regression model during the week ahead, in advance of the Red River Shootout. I am also planning to build a rankings system off of this neutral site model, but that is likely going to have to wait (probably until Texas’ next bye week, or later).

Any further updates will be added to the end of this post.

Updates

October 5, 2025: Beginning with Week 6, a link to a Google Sheet containing the full model breakdown will be attached to each predictions and results post. This is due to the lack of plugin functionality associated with the Personal WordPress plan (makes it extremely difficult to create data tables).

October 9, 2025: I’ve decided that the “Predictions” post will simply include the running total, a visualization of the predictions, and a link to a Google Sheet containing all predictions for that week (with some slight formatting). My rationale for this is there really isn’t much to highlight until after the games are played. However, I still want to publish the model’s predictions ahead of the games that weekend.

The “Results” post will be separate, but will largely pull from the “Predictions” post with additional content. The formatting for these posts is likely to change a few times throughout the season as I feel out what works best.

October 15, 2025: All predictions and results posts will now include a visualization. These visualizations will use the base version of both the home/away model and the neutral site model. In effect, this means that two models are being rolled into one visualization. However, there are generally only 0 to 3 neutral site games per week. It also doesn’t really matter outside of breaking a graphing convention, which is necessary in order to include neutral site games in these visualizations, given how the models are set up.

Additionally, each results post will now have a “biggest nailbiters” section: games in which a team with >= 80% win probability wins by <= 10.