A fun thing to do is to set that first week filter for past completed seasons. For example, if you do it for last season, you get Houston #1 (yay normal) and then UC Irvine #2 after they beat Oregon on the road by 13. Funnily enough, they ended their season losing on the road to Oregon by 26 in the NIT.
4 Likes
Did you learn that hot take on statistics at UVA?
I’m just messing with you!
3 Likes
Looking for everyone in the rotation Top 10 to get at least 10 minutes. Maybe only about 10 for Taine, but for most everyone else want to see the minutes spread equally 15-25 minutes per player. Really important pair of games this week to get everyone some meaningful run to build confidence ahead of a serious four game stretch starting next week (Wisconsin, SMU/WVU, Texas A&M, and Syracuse).
7 Likes
Yes! In fact I did — but that won’t stop many people (including myself) from forgetting it sometimes😉
1 Like
Before people get too crazy with Torvik’s tools, which are admittedly fun, every predictive system ever made (at least in basketball and football) is better with the preseason ratings included. And the improvement lasts for the entire season. They only get phased out because people would complain and I think kenpom has said it would be a little weird presenting a “2023-2024 rating” that includes data from previous years.
4 Likes
Genuinely curious as to why this is.
First, how are the preseason ratings calculated? Looking at our team specifically, there are an awful lot of question marks with guys who haven’t played at this level. How are they rated and factored into the algorithm? How do those preseason rankings predict minutes played? And if they are wildly off, for example, on BB vs Minor, doesn’t that end up skewing the numbers for the worse?
Second, doesn’t real game play against real teams provide a better assessment of how a team is actually capable of performing than all of the preseason assumptions?
Finally, what constitutes better? That is, against what metric are you judging them? And is there possibly a self-fulfilling prophecy element to the final rankings?
It totally does. But it’s not a question of either/or, it’s a question of “look at only one thing” or “look at both things.” The preseason data may have lower value than real game play, but it’s still incremental data.
For statisticians, better means more predictive. If you include this data in the modeling process, does it result in better predictions of future games?
Kenpom has written a few things about this (everyone yells at him about preseason data), here’s a recent post on it:
7 Likes
Most of this comes down to college basketball seasons just aren’t very long and even a whole season isn’t necessarily enough time for a team to show it’s true talent. It’s not all that uncommon for a good NBA team to have a mediocre/bad stretch of 5/10/20 games. You do that in college, that’s your season and you’re in the NIT. Conversely, a crap NBA team can run hot for like 20 games, but then comes back to earth. You do that in college, you get a surprise NCAA berth and your coach gets an extension.
As for how the preseason ratings are calculated, that varies by the system. It’s not just projected starters, but historical program performance, etc. I do think it will be interesting to see if the preseason ratings get worse with higher turnover.
As for how they’re judged, they just run the game predictions with and without preseason ratings included, and including the preseason ratings gets you closer to the scores (but obviously, none of these systems are crystal balls and there’s a lot of error no matter how they’re designed).
9 Likes
Not the same thing, but Ive read some articles about how fantasy football rankings take about 4-5 weeks before in season data is better than the pre-season rankings of players
1 Like
We’ll, there’s a reason I got a C in statistics.
Thanks guys. The Kenpom article was particularly helpful. Obviously, there are problems even with going back, say, five seasons for every team and assigning them the same weight. He used FAU, but you can look at a program like Louisville that has had huge turmoil, changes in coaching staffs, scandals, etc. I don’t know exactly how you factor those things in either.
But all-in-all, this is a great explanation of how these guys compile their preseason rankings. Thanks again.
2 Likes
The other thing to add about KenPom and his rankings is that it’s probably more accurate to say “historical data from the last four or five seasons (depending on how he’s adjusted his formulas to try to improve accuracy) that gets used to generate the preseason/inseason rankings” instead of just KenPom rankings, but the former is, um, very wordy.
Also, the historical data in the calculations never fully goes away, it just gets less important as it gets older and less relevant. At least, that’s my understanding of his substack post that was linked above that I read, like, two weeks ago.
Edit: or he might said something about this on a podcast interview. I honestly don’t remember, I just know I read/heard it from him.
1 Like
I think last year’s Monmouth team was the worst team I’ve seen us play.
By memory, but I think Ken has said that he’d like to keep the old data in, to make it more robust, but has decided to have it phased out.
I understand why KenPom and Torvik, et al, leave the preseason stuff in, and I also understand why they wouldn’t want to take it out, but I sort of find that it doesn’t really fit my use case – which is to try to get an in-season read on how good a team is, based only or mostly on in-season stuff. Now to be clear, 2 games in isn’t a great time to do this, but after about 5 games or so, that’s my preference.
My brain can account for the fact that Bill Self is a good coach and Wisconsin is a historically good program (for example). I don’t really need an algo for that. The algo is pretty good at telling me what to expect for random mid and low majors, I guess, but I don’t particularly care (I’m not using it for gambling).
Anyway, I understand why Ken has made his decisions, but the above is why I kind of prefer the functionality of Torvik as soon as possible, because I’m happy to rely on my cognitive biases to judge “programs” (and I tend to think that stuff matters less in the era of more movement, maybe …).
Edit - just realized you were talking about that Substack post. I didn’t see the exact point you alluded to, but I just skimmed.
Also, this is peak math guy BS from Ken, and as peak dumb guy myself, it’s why us dumb guys roll our eyes at math guys:
Even by the end of a 30-ish game season, there are teams who have been affected by random chance in a significant way that distorts the picture of who they are.
(that’s me rolling my eyes, Ken).
1 Like
Why, yes, now that you mention it, I did have a moment of PTSD upon seeing a Wyoming team led by Fennis Dembo on that list.
2 Likes
We’re going to lose this game.
1 Like
Goddamnit!
2 Likes
Yeah I wish Kenpom had a way to turn off preseason. I like that both he and Torvik use it in their “primary” algorithm – if it works, it works – but when I want to compare resumes, Torvik’s simple switch to turn off preseason is the best way to do it.
2 Likes
Look man. @WFS_HOO has gone soft, someone had to step up.
4 Likes
I prefer the metrics that have us the highest. That’s just me though 
3 Likes
Not the worst team but this was a pretty incredible beat down. Also saw the corner of my own head a few times in that group 1 side court student section. Good times
10 Likes