Question for Asteroid

As the one of the resident statistical wizards, I had a question about Sagain's model.

Currently WVU is still rather significantly ahead of us in Sagarin. Their Predictor is 94.29 to our 91.69, which translates to an "overall" of 93.69 to our 92.25 (which is bouyed by our "recent" number).

This seems fairly wide considering the following:

>SOS: WVU at 77.96 (ranked 41st) to KU SOS of 81.16 (ranked 4th).

>Record vs. Top 10: WVU is 5-2, KU is 6-2.

>Record vs. top 50: WVU is 7-4, KU is 11-3.

If we've split the series between us, KU is 37 spots higher in SOS, and KU has had more success vs. the top 10 and top 50, why oh why is WVU still ranked #2 overall in Sagarin.

Don't get me wrong, I'll take our 1st place in the conference and higher spot in the polls over WVU's 3rd place and lower spot in the polls. Still, I can't seem to figure out what factors are keeping them ranked where they are.

I have the DPPI which in my opinion is ever so slightly more robust than Sagarin's model. I also have a pure predictor model like sagarin. My pure predictor model is very similar in concept to Sagarin but my composite model operates from a slightly different (and as I said I think better) algorithm. The pur predictor model is intended to create a rating that will give you the best idea of what a future performance would be. It is based on a net scoring margin principle after adjusting for predictive factors (such as home/away). So it is not necessarily dependent in any way on SOS unless you deem that to be a predictive factor. In my DPPI I deem it to be a predictive factor. Thus I have an algorithm to net the scoring margin based on SOS. I am not sure what Sagarin does. In my DPPI West Virginia has been No. 1 for much of the season in pure predictor. They are currently No. 2 and KU is No. 10. In my composite ranking West Virginia is No. 3 and KU is No. 4. The problem with using a pure predictor rating is that is is not just a rating but it has two other numbers that are necessary to predict an actual scoring margin. Those numbers are the mean off the standard deviation and the standard deviation. So that sounds too complicated so let me simplify. KU has a raw predictor rating of say 92 (I do not have my data here so I am just using example numbers). So you calculate the expected scoring margin for all 26 of KU's games taking the difference in raw predictor ratings and adjusting for home away. You them compare this predicted scoring margin what the actual scoring margin. So say KU was predicted to beat Texas Tech by 5.5 but only beat them by 1. That would be a performance difference of -4.5 points. So you average this difference and that is your mean difference (usually less than 1). Then you look at all of these differences for all 26 games and calculate the standard deviation. KU's standard deviation is around 6.1 (which is very low). This means that the vast majority of KU's games they play within 6.1 points of the predicted margin. With these three numbers (rating, mean and standard deviation) you can calculate the probability of winning between any two teams. Now here is where it gets interesting. West Virginia has a higher raw predictor rating but also a much higher standard deviation (like 14). So on any given night West Virginia can play 14 points better or worse than the prediction. In other words they are wildly inconsistent. This is very important, here is why. FOr any given predicted scoring margin West Virginia will have a lower probability of winning than Kansas would for the same scoring margin. So say West Virginia was playing TCU and Kansas was playing Iowa St. And just for example let's say the difference in ratings of the each of these two matchups is the same meaning Kansas is favored by 5 over Iowa St. and West Virginia is favored by 5 over TCU. The probability of Kansas beating Iowa St. will be higher than the probability of West Virginia beating TCU. So if you now say let's let the computer do a round robin tournament where all 351 teams plays each other team. Even though West Virginia has a higher rating than Kansas, their probability of winning will not necessarily be higher due to West Virginia's inconsistency as measure by their higher standard deviation. So Kansas is rated 10th in the raw predictor meaning they would be the underdog to the 9 higher rated teams and the favorite to the other 341 teams whereas West Virginia is ranked 2nd in the raw predictor and would be favored against all but the No. 1 rated team. But when you add up all the probabilities of winning all 350 games in this round robin tourney, Kansas, because of their low standard deviation (high consistency) would have the 4th highest win percentage. That is the composite ranking. West Virginia would fall from 2nd in the raw predictor to 3rd in the composite. North Carolina is probably the starkest example. They are ranked 5th in the raw predictor but since they are one of the most inconsistent teams in the country, they fall to 12th in the composite. I realize this is a very cursory explanation for a fairly complicated probably. Basically, when you are looking at a rating, whoever creates it, it is not the whole story. Consistency is a huge factor in predicting outcomes.

Wow! Thanks. So to sum up: scoring margin and (in)consistency. Makes total sense.