NCAA Selections from the Big XII

I do not really understand why the seeding committee gave ISU a birth but not OU. They were tied in the league and OU has been play much better the last month than ISU. ISU had some wins early in the year but the lack of offense at ISU is really apparent. Hopefully they will show out and make the sweet 16 run but I know OU played really well in the Big XII tournament. I check a great source for power rankings, the DPPI, and it has OU ranked 43 and ISU ranked 59.

Iowa State has 9 Quad 1 wins and Oklahoma has 4. As for Quad 2 wins, Iowa State has 1 and Oklahoma 6.

Corpus, two questions:

1. Do you have the Q1 and Q1+Q2 data?
2. Can you explain why your model has such a large ranking for Michigan compared to the others? That is, what factors influenced that result?

Thanks!

Of course I could easily calculate the Q1, Q2 data. I do not as a matter of course. What I do have in my DPPI standard output is the record against Top 10, Top 25 and Top 50 teams. Baylor and Kansas are tied with 12 wins each against top 50 teams. Texas Tech is 3rd with 11 wins.

As for the Michigan discrepancy, I don't have a specific diagnosis but I will give you hopefully some general insight.

1. The spectrum of algorithms is from pure predictor to some form of pure retrodictor. All models are predictive to some degree but some models (such as what the NCAA does for seeding) is based on past performance and not so much how that performance plays into future predictions. So considering things such as Quad 1 wins is not so much a factor that aides in predicting future outcomes but rewards teams for some factor of performance. My model is intended to be a pure predictor so I don't factor in anything that I haven't judged to be efficacious at predicting future performance. Some models use more reward based algorithms.

2. There are two factors that I have not yet found reason to include in my DPPI that some other models include. My reasoning (right or wrong) is I have not found them to be purely predictive. The first is rewarding for wins. My model is based purely on scoring margin. So obviously winning is rewarded since it has a positive scoring margin. But some models give extra benefit for a victory. I do not. I have researched this and every year I consider adding it but have not yet convinced myself that simply winning has some benefit beyond the scoring margin.

3. The 2nd factor I do not use that many other models use is SOS adjustment. Now by this I mean that after all the ratings have been calculated based on the actual games played (which obviously includes the actual opponents and thus their relative strength) some models adjust ex post facto a pure SOS adjustment. I do not do this. I think it is double dipping since the calculated rating is based on the opponents relative strength already. There are 2 common modifiers to this ex post facto SOS adjustment.

3a. The first modifier is adjusting for the overall strength of the conference a team is in. There is some merit to this adjustment since one of the weakness of any rating system is that you are dealing with relatively few data points. Most teams only play 13 non-conference games or so and that is really minimal to get all the teams statistically "connected". Since most teams play more than half of their games against a limited set of opponents (in conference games), some models do a conference adjustment to get a better statistical connection of all teams. I will probably end up doing something like this at some point in the future. I have noticed in my DPPI compared to other models, the top non-power 6 teams tend to be rated slightly higher which means lower rated power 6 teams can be rated slightly lower.

3b. The second modifier to the SOS adjustment is to weight the games against top teams more heavily. I do not do this at least in terms of weighting and ultimately assigning a rating. I consider the relative performance against top tier vs bottom tier teams in a separate function that goes into calculating the predicted scoring margin. In other words, I consider this but not in the rating. That is why two teams rated the same may have different predicted scores against a common opponent. Remember, my rating system is purely predictive and thus, it is impossible to distill all the valid predictive factors into a single rating number. I have several adjustments that are applied at the predictive level and not at the rating level.

I hope this gives some general insight. I have Michigan currently rated 53rd. But they have the 355th rated consistency. This "consistency" does not impact the actual rating but it is significant in predicting outcome of future games. Since they have a tendency to play much better or worse in any given game, their probabilities of winning and losing are impacted. I won't go into that here as that gets to be a little overly nerded, but just as a summary since it is an important point, no predictive rating system can distill everything into a single rating. There are simply too many factors in the manner of predicting. But the world writ large wants a single rating so you have to do your best to come up with a rating that is meaningful.

HE, I forgot to directly address OU's snub. No matter how you slice it, the NCAA Selection committee dropped the ball. Or at least they used selection criteria that are somewhat whack. Consider this. OU is 33rd in the DPPI, better than 19 other teams that were selected at-large. Even using the Massey Composite index OU is ranked 38th and that is the average of all of the computers (including the DPPI). Below is the DPPI rating of the teams ranked 33rd (OU) to 89th (Rutgers who was the lowest rated al-large selection). Included is the record against top 50. The shaded teams were not selected. ^ teams jump out that should never have been selected. But one thing that also is apparent is OU lost 12 games against top 50 teams. But that seems to be a straight up punishment for playing in the Big 12. Yes, OU did not win a lot of games but they played well and thus achieved a high rating. KU's two games against OU are testimony to how good they are. Incidentally, the three teams with the most top 50 losses are Oklahoma, Kansas State and West Virginia. Anyone see a relationship of these three teams. Rutgers was ranked 89th but they were 7-5 against top 50 teams. Creighton was ranked 68th but they were 5-5 against top50 teams. Likewise Miami (FL) ranked 64th but 7-5 against top 50. And they are ranked 68th in the Massey Composite. But Marquette? WTF? Ranked 75th in DPPI but to be fair they are 46th in the Massey composite. And Wyoming is ranked 62 in DPPI and 56th in the Massey and they were 2-6 against top 50.

Basically I can make some sort of argument to support most of the NCAA's selections, but these 4 are real head-scratchers

Rutgers
Wyoming
Miami (FL)
Notre Dame

Meanwhile BU is getting rolled by UNC. KU needs to be on its game today, not a good look for B12 so far today.

BU is making a game of it. Back to within 3.