Last year, I wrote a post on finding the "happy medium" between in-season Pythag and preseason PECOTA for MLB teams. I've always wanted to do something similar for college basketball, but that wasn't possible since Nate Silver doesn't do college hoops projections. In fact, I'm not aware of anyone that does preseason CBB projections in a manner that would work for these purposes.
So I figured I should go ahead and do it myself. Over the next three months (yes, I know most people are doing college football previews at this point) I'll be coming up with ratings for each major conference team, as well as some notable others.
When I tried to get a start on this last week, I realized it was going to be tough to just pull the numbers out of thin air. College hoops may be extremely difficult to project, but I figured that two metrics likely have a high correlation with future results: previous year's KenPom rating, and what percentage of minutes the team returns. This won't tell you everything (recruits, key players leaving, etc.), but it should be a start.
To determine just how important these two factors are, I looked at every major conference team from '06 to '09, and threw their previous year's KenPom and returning minutes into a regression to try to predict their KenPom rating in the following season. The results look like this:
Yes, the sole team in the final row is this epic failure.The ratings I'll be using for '09-'10 will use this formula as a baseline, but I'll be adjusting them manually (with some help). For example, just based on '09 KPom and returning minutes, UNC comes in at .862 for next season; that's pretty clearly unrealistic. (Yes, the ACC post will be the first one, hopefully in the next week or so.)
Just for fun, here are the ten teams in the data set (from '07 to '09, since you need Ret% from the previous season) with the highest projections:
Some hits there, particularly at the top, and only a couple big misses; it is kind of weird that '08 Arkansas and '09 Marquette declined despite returning such a large portion of the previous year's squad, but that's bound to happen occasionally. As noted above, these are the projections that should be close, since there's not much uncertainty with a team that returns 80-90% of its minutes.
Let me know in the comments if you have any questions about or concerns with any of this stuff, which I'm sort of just making up as I go along.
9 comments:
Good stuff as usual, VW, looking forward to the projections. I imagine Mississippi St will look solid with everyone coming back. Also have Renardo Sidney and John Riek coming in before they go on probation in 2010-11.
I've been doing some work attempting to project players's improvements from year to year, but I'm still in the data collection stage. If we knew that statistically a player improved the most from sophomore to junior year and that a team was returning a lot of rising juniors, that'd make a major impact on projections.
It's too bad you can't incorporate some kind of incoming class ranking, but I guess those numbers would be wildly inconsistent with actual KP-style worth in their first year at the school if they were useable numbers in the first place. I'm looking forward to the results.
Adam, I am working on incorporating recruiting right now. It is going surprisingly well.
JN - I don't know what the statistics say, but coaches generally will tell you that a player improves the most between his freshmen and sophomore year.
This might be completely off-base, but I know that KenPom ranks players on scales (ie go-to guys, major contributors, significant contributors, role players). I assume he uses a formula to determine this. Could you weight the returning minutes based on what role the player played last year to get a better idea of the quality of minutes returning, not just the quantity?
Nice analysis. I'll be waiting for the laughter that comes when you get to the Big east - both for unpredictability and for how my squad (St. John's), even with the minutes returning, might not be that good.
By the way - and this addresses John above - while I think quality minutes would be cool, I think there is enough variability between year-to-year performances, especially with small sample sizes (some "major contributors" received only 20-30% of the floor time, but jacked up shots/ took up possessions) that it might muddy the analysis... just my 2 cents.
- Pico
Have you thought about using returning minutes from seasons other than the previous? I think that might add a bit more accuracy for teams that could be bringing back players that have been on the team for a while and received minutes. Just a thought...
"Have you thought about using returning minutes from seasons other than the previous?"
I doubt I'll ever get around to it, at least not any time soon, but I like this idea.
Post a Comment