10/5 Ratings and Next Week's Lines

Last year I began to track my ratings' performance against the point spread for all games. (I use the ratings set that includes all teams instead of the Division I-A only ratings because more data is always better when evaluating true team strength.) This is where I should be launching into an infomercial about how awesome my picks are, but the truth is that I didn't save the data from last year. All I remember is that the highest confidence picks fared pretty well (65% range, although BRAGGonUT or Trips Left may remember better) down to the expected 52% range or so on lower confidence picks.

For this season I waited until last week to start tracking their performance in order to let them settle in a little better. In last week's games, the ratings went 6-1 in games with a confidence value of at least 14 points. The overall record was 33-17-2 on the week, but it was a weird week in that the best performances were the highest confidence picks and the lowest. The record was 18-8-1 in games with a less than 7 point confidence margin. So the record was only 9-8-1 for games between 7-14 points. Obviously that should settle down at some point and the records should trend down with confidence.

It's also important to note that the later in the season we get, the fewer games that will have very high confidence ratings as the books catch up with each team's true 2008 strength and rely less on name value and past season performance. The picks are made using a standard across-the-board 3.5-point homefield advantage, so you can adjust from there if you're so inclined. Yes, that is removed already for neutral site games like this week's Texas/OU tilt. The ratings are back-checked against year-to-date results to determine the scaling factor used for this week's games. The PMV column in my ratings, which is pure margin of victory, is similar to the Sagarin predictory and so that is used for this exercise. The current scaling factor is 2.395, so the difference between two teams' PMV ratings is multiplied by 2.395 this week to get a base margin, then the homefield advantage is applied. This factor fluctuates during the season and also from year to year based on scoring conditions.

With that said, here are this week's lines and picks. I'll use the first and last games in the list to explain the confidence column. The ratings say Tulsa should beat SMU by 42.68 points at home, on average. Therefore with a 25-point line, there is a 17.68 point cushion in this pick. Similarly, the ratings say that Florida Atlantic should, on average, lose to Troy by 3.92 points at home. So there is only a 0.08 point confidence number on that pick.

Finally, as always, these are only a starting point. I don't trust them myself in many cases, such as last week when I pointed out that my ratings picked Nebraska over Missouri against the spread and I would have bet it the other way.