Apologies. The computer and I have been a bit under the weather and took a little unscheduled vacation from the world. But we’re back (mostly), and ready to tackle the rest of the season. This seemed like an excellent time to look back and see how the computer has fared through this season and see how the changes I made a couple weeks ago have worked out.
After the Illinois game, I noticed that the computer chronically under-predicted score lines, sometimes to the tune of 20+ points per team. In fact, against Nebraska it missed Iowa’s score by 20 under and Nebraska’s by 43 under. These were some large gulfs, and as a result, I tweaked the inputs for the model, believing that the changes I made would make the score predictions more accurate. Now, it has only been 2 games under this new model, so the evidence is weak at this point, but it looks like a huge improvement overall. Below are some diagnostics that I have run.
|Measure||Old Model||New Model|
|Avg Abs IA Diff||10.3||8.5|
|Avg Abs Opp Diff||17.4||9.5|
|Avg Abs Total Diff||27.7||18|
|Avg O/U Diff||-15.6||1|
After just two games in the new regime, H.A.W.K.E.Y.E.S. is predicting Iowa’s scores nearly 2 points better than before and opponents scores by a full 8 points for a net score per game improvement of almost 10 points. Perhaps the most important point here is that the improvement to the over/under (the total difference, rather than the absolute difference), where the last two games the model has only been wrong by an average of 1 point, compared to almost -16 previously. Yes, it is too soon to say a lot here, but the preliminary diagnostics to seem to tentatively say that the model changes have been justified (H/T to BoilerHawk for suggesting some sort of post-estimation diagnostics. I apologize that they aren’t in exactly the form we’d like, but it is still better than nothing). On to the next game.
Poor Sparty. In basketball and football this year Michigan State has not looked at all like its normal self. Not that I’m complaining. Tom Izzo practices the dark arts, and so does Mark Dantonio and it’s fun to see them fail. But that’s not to say that Michigan State is bad at basketball this season. They just aren’t dominating like the usually do. Their strengths are clearly on the defensive side, while they struggle to consistently get their offense going (Iowa’s doppelganger?).
So what’s the computer telling us? Expect a fast and dirty game, with both teams playing in transition a lot but having to settle for low-percentage shots. I think Iowa’s key to victory here is going to be free throws. If they can shoot better than 70% as a team at the line, then I think that’s going to be the difference in this game. The model predicts an Iowa win in this contest, 80-75 and I am inclined to agree with this completely.
Iowa’s Current Record: 14-11 (6-6)
Predicted Score: 80-75 (W)