Week 14 was interesting, for at least one game: Bears vs Vikings. These teams had played just two weeks previously and the result had been a relatively comfortable Bears victory. Chicago followed by dropping a tough OT loss to the Seahawks at home. Minnesota, meanwhile failed to match the Packers on the road. My algorithm predicted a safe Bears victory. Bear in mind that the model is still extremely primitive at the moment and I don’t expect it to catch any sort of subtlety. Still, even intuition suggested that the Bears stood a good chance against the Vikings, despite the Seattle loss.
I’ve started tracking weekly predictions from folks in my office and a few of them had picked the Vikings. Being curious about whether or not humans can reliably make accurate predictions, I asked them what was behind that pick. The response was that they weren’t sure about the Bears defense and that they hadn’t been playing well lately. So what is lately? Does it only includes the Seattle loss? Does it matter that that game had been tied up after 60 minutes? Did lately include the victory over Minnesota in some way that my superficial observation of score differences failed to capture?
Now here’s where it gets interesting. By many metrics, the Bears performed better than the Vikings. As measured by yards allowed, the Bears defense was actually better in their second matchup than they had been in the first allowing only 248 yards vs 258 two weeks prior. The Vikings air attack was half as strong as it had been in the earlier game, 77 yards vs 144. As for the Bears offense, it also performed much better. Chicago managed almost 140 additional passing yards. Cutler threw for 320 receiving yards, eclipsing the 230 they picked up against the Seahawks. Offensive yards on the ground were just a nose higher, 118 vs 113. The Vikings punted 7 times to the Bears 6. The Bears picked up 22 first downs while the Vikings managed only 17.
So what the hell happened? The answer is that I don’t know yet. However, there is one interesting difference and that’s turnovers. Minnesota turned the ball over 3 times on November 25, but only once this past Sunday. Chicago had just as many turnovers in both games. Is that what made the difference? Perhaps. Can I predict it? I doubt it. It is, nevertheless, an idea I’d like to explore. Is there a correlation between turnovers and upsets (if we may call the Vikings’ victory an upset)?
One additional factor may have played a role. The Vikings came out strong, scoring twice in the first quarter. The Bears only had two turnovers, but one of them came in their first offensive drive when they were in scoring position. Did this demoralize their offense? Did this keep them from completing their many scoring drives? Maybe. However, the more important question is whether or not this sort of event may be predicted in advance. Again, I’m sceptical.
In further conversation with a couple of the guys at the office, they touched on another idea that I’d like to explore. It’s similar to the idea of mean reversion. Their view of Chicago’s performance having been weak in the recent past was based on their a priori expectations of how the Bears would play. A positive score difference is indicative of a greater likelihood of winning. However, a score difference that’s less than expected may be indicative of a drop in performance. Mean reversion would suggest that they will return to form, however, in this case, we presume the opposite. A positive performance that was not positive enough, means that we have reduced expectations for a subsequent event. There’s probably some time series model that accounts for these kinds of short term trends that I’m not familiar with. Time to crack open the textbook.