What is PredictaBot?

I've been a member of an online Steelers community for about fifteen years now, and a long time ago another member rated teams based not just on their records, but the number of wins of teams they had beaten. I thought it was interesting, and started working on a system to automate it. I had always thought that there was a little too much "emotion" in ranking teams. Multiple losses by a "good" team could be written off until all of a sudden they were recognized as falling apart. The system used a minimum of data to get all of its results. In fact, all it uses are the points scored and allowed for each game.

This eventually led to a program I had jokingly called "Big Blue", which I used for years to make predictions of NFL games. However, the predictions weren't all that amazing, and personally I liked looking at some of the data analysis more than the prediction aspect. It was mostly just something entertaining to do to look ahead to the next week's games.

Recently I overhauled the system, added 40 years of historical data, and basically re-wrote the entire thing. As a result of the analysis, I came up with some interesting results. Rather than just continue to send them in email, I decided to put up the blog so I can put in articles as well as weekly updates.

So, after all that, what really is the system and how does it work?

All in all, it's relatively simple. After each game, a team gains points that affect it's rating in terms of win toughness, offensive output, and allowed points. For offense and defense, the team merely gains or loses points based on how well they did against an opponent compared with what that opponent usually does. So, if a team scores 20 points against a team that normally allows 14, they gain six points for their offensive rating. Similarly, if they allow 24 points to a team that normally scores 20, they would lose four points from their defensive rating. The win toughness simply adds points for wins as well as the number of wins their opponent has. Beating a team with a lot of victories is a key measure of success, and is in fact the best correlation I've found to predict Superbowl winners. Almost 2/3 of the eventual Superbowl winners had the #1 or #2 "winrating" at the end of the regular season.

Those ratings are the basis for everything else. When two teams are scheduled to meet, their offensive and defensive scores are compared, home field advantage is applied, and a predicted score results. I'll provide more details later, but after looking at about 30 years worth of betting results, the program would have resulted in reasonably good performance. That is, betting $100 a game since the late 70s would have "only" lost you about $5000 dollars by now. Let's face it, if this system produced results that beat the odds, I wouldn't be posting about it.

Anyway, I don't take any of this too seriously. It's mostly just an excuse to think about football more. I should also point out that being a Steelers fan, there will be a focus on them and their divisional opponents.