Analytics thread

[stable genius]

I meant to post this chart earlier, but here is some more perspective on how teams tend to do in valley play based on their kenpom.



Where we stand right now:

[Rambler63]

I meant to post this chart earlier, but here is some more perspective on how teams tend to do in valley play based on their kenpom.



Where we stand right now:

Loyola is at 95 in Torvik, which puts their average at 107.

[stable genius]

Loyola is at 95 in Torvik, which puts their average at 107.[/quote]


Thanks - looks like I was looking at their offensive efficiency there...let me check all these numbers again quick.

Kenpom had Bradley beating SIU by only 8 and they won by 19.. so they'll get a nice boost from that. I'll update everything once the Bradley game is in the system. Will probably just wait til the new NET comes out tomorrow.

[db1972]

I love that somebody is taking a deep dive into analytics on this site. We need more of it! However, the methodology being used in this particular case is flawed. In the initial post, you're listing a whole bunch of scenarios with MVC teams and their results in conference based on their Kenpom ranking. The problem is that the Kenpom numbers are biased based on how the team performed in conference. Obviously a team that keeps winning and ends up with 15+ wins in conference is going to be ranked higher (at least relative to where they started the conference season) than a team with 12 wins. This would (I think) be qualified as a variation of survivorship bias where teams that perform better in conference are retroactively assumed to be better entering conference play since there's no data from solely the non-conference slate.

Unfortunately, I have no answer on how to perform an unbiased analysis on how to expect a team like UNI to fare in conference play. Kenpom seems not to keep a record of the team rankings at a given date, so I can't go back and find his rankings of all 353 teams from this year on December 30 before conference play started, let alone from 2002 or 2012. This means any kind of retroactive analysis using Kenpom data to predict Valley play this season (or in future seasons) is going to have a huge asterisk.

[stable genius]

You're right that this is an issue and I thought about it. I think I mentioned this problem? Or I meant to but didn't?

You actually can to back and look at what the rating was at the beginning of conference play if you have a subscription. I close to go with the rating at the end of the year because it encapsulates the whole body of work as opposed to just half the season, so I think it's more accurate.

I would also say that the rating is not dependent on winning or losing. Just because the team wins doesn't mean the rating went up and just because the team lost doesn't mean the rating goes down. It's all about how a team performs vs how the system expected them to perform (efficiency/point spreads margin). I think what you argued is perfectly reasonable but I disagree.

Also a note: above I have posted average experience numbers. Kenpom had Austin Phyfe incorrectly listed as a junior. Got that fixed today and UNIs experience ranking went from 70ish to 119. So they're a little younger than what I had above.

[db1972]

You're right that this is an issue and I thought about it. I think I mentioned this problem? Or I meant to but didn't?

You actually can to back and look at what the rating was at the beginning of conference play if you have a subscription. I close to go with the rating at the end of the year because it encapsulates the whole body of work as opposed to just half the season, so I think it's more accurate.

I would also say that the rating is not dependent on winning or losing. Just because the team wins doesn't mean the rating went up and just because the team lost doesn't mean the rating goes down. It's all about how a team performs vs how the system expected them to perform (efficiency/point spreads margin). I think what you argued is perfectly reasonable but I disagree.

Also a note: above I have posted average experience numbers. Kenpom had Austin Phyfe incorrectly listed as a junior. Got that fixed today and UNIs experience ranking went from 70ish to 119. So they're a little younger than what I had above.

You have fair points. I think we'll have to agree to disagree on the best way to predict future conference performance though. I 100% agree that the rating at the end of the year is better because it encapsulates the team's entire body of work, but then you are comparing apples to oranges with current teams that have not finished their seasons yet.

I guess the main issue I have with the analysis is the comps between UNI and historical Valley teams. That's why I'm concerned with using end-season data to compare a mid-season team. If Northern Iowa were to finish 16-2 in the Valley, they will almost certainly finish the season rated higher than 69. If they finish 10-8, however, they would drop and finish rated lower than 69. Thus I believe it would be more beneficial to take all the teams rated comparably to UNI BEFORE Valley play started (within the 59-79 range, or whatever bracket you decide to use), and see their overall results. Then we're comparing apples to apples and have a better idea of how teams in this range before conference play performed in the Valley.

Again, just my 2 cents as I really appreciate what you're doing with this. Perhaps I should get myself a Kenpom subscription, would you say it's worth it?

Disclaimer: I know wins and losses are not necessarily equivalent to rating, but consistently winning games generally leads to higher ratings. To my point, UNI has 2 "bad" losses in conference play so far, but has still risen to 47 despite that.