Tim Miles trolls KenPom
Beneath the wholesome South Dakota boy exterior, he's a low-key loose cannon.
https://twitter.com/CoachMiles/status/968234209717387271
https://twitter.com/CoachMiles/status/968233938249486336
https://twitter.com/CoachMiles/status/968233395108147202
He's got to be careful with this one, though....
February 28th, 2018 at 3:31 PM ^
Ok, they're called "predictive" metrics? pic.twitter.com/nCNibhbynx
— Tim Miles (@CoachMiles) February 26, 2018
February 28th, 2018 at 3:32 PM ^
1/3 - I love @kenpomeroy and sometimes his computer... but here's my emotions of the season in three snapshots... 11-12-17 rankings - "Hey Kari - Let's wait on that kitchen remodel".... pic.twitter.com/SbDJuqDbyc
— Tim Miles (@CoachMiles) February 26, 2018
February 28th, 2018 at 3:33 PM ^
2/3... Damn... this does not look good. The computer has us 5-11 over the next two months... with a little luck 6-10. Realtor??? pic.twitter.com/PnLwMt3Vsg
— Tim Miles (@CoachMiles) February 26, 2018
February 28th, 2018 at 3:35 PM ^
don't think I understand what he is saying. I mean, I know he is going lose by like 35 on Friday, I get that. But I am not sure what he is saying with these tweets.
February 28th, 2018 at 3:39 PM ^
He calls out KenPom's predictive model for projecting Nebraska (back in November/December) to be a terrible team.
February 28th, 2018 at 3:43 PM ^
And they were terrible. Look at the 4 point win over Eastern Illinois.
It's also like predictive models are made to adjust for when team's change and more data becomes available.
February 28th, 2018 at 4:10 PM ^
Yeah but Miles is suggesting kenpom's model has been wrong about the whole time, not just the start of the season.
February 28th, 2018 at 4:42 PM ^
February 28th, 2018 at 4:46 PM ^
#51 still doesn't sound right to me. I think it's fair to say that certain algorithms will favor certain types of teams (Kenpom has always loved slow tempo efficient teams like UVA and Wisco) and have bias against others. Anyway, I guess we'll get another data point Friday.
February 28th, 2018 at 4:49 PM ^
February 28th, 2018 at 5:35 PM ^
What?
That's just... I don't even know what to say to that.
KenPom's numbers are tempo-free. That's pretty much the whole basis for his work. "Slow tempo efficent teams" are, well, efficient.
That's the whole point. They explicitly do not favor any particular tempo, and if you've read his blog, you'd know that he really prefers a fast-paced game. However, if you're trying to figure out who's more likely to win a game -- which is what he's trying to do -- the implicit bias against slow tempo teams in the standard metrics leads you to some wild conclusions.
KenPom's formula doesn't love Virginia because they're slow. The formula loves Virginia because they're very good on offense and absolutely lethal on defense. They're not 26-2 (15-1 in the ACC) by accident.
February 28th, 2018 at 6:00 PM ^
Are you trying to suggest KenPom just built the perfect algorithm with no systemic biases? Because it sounds like you are.
Kenpom overvalued Wisconsin for years if you go back and look at his rankings vs AP rankings or tournament success. That's because Wisconsin was an outlier with respect to pace. So yes I meant exactly what I said and it's correct. Thank you for your snark however.
February 28th, 2018 at 6:06 PM ^
February 28th, 2018 at 6:23 PM ^
Yes. Kenpom even had a post about the Wisconsin problem himself a few years back.
February 28th, 2018 at 8:06 PM ^
and yet at no point in that article does he say its because they play slow
February 28th, 2018 at 8:47 PM ^
Man, use a little intuition. Wisconsin was an outlier because of pace. Y'all would make some seriously shitty data scientists.
February 28th, 2018 at 9:18 PM ^
Also, that 2012 Wisconsin team finished #8 in his rankings and made it to the Sweet 16, where they lost by 1 to #1 seed Syracuse. And after that article he posted (which was about halfway through the season), the Badgers lost exactly 5 times the rest of the year, to the #2, #3, #101, #3, and #5 teams in KP, while beating #30, #9, #2, #9, and #16.
I don't know what your schtick is here, but "lazy arguments without factual merit" combined with "middle-school insults" is a strong choice.
you are using your intution to make a statement and then suggest those who don't make conclusions based on intuition would make poor data scientists?
You are probably correct that a slow team would be more likely to be "unlucky" and hence overrated by kenpom at the top end of the rankings since playing fewer possessions means that randomness plays a greater role in outcomes which would then have higher variance compared to expectations. It would follow that at the lower end of the rankings, slow teams would be systematically underrated if this hypothesis held (so it wouldn't exactly be correct to say slow teams are always overrated by kenpom).
But a data scientist would actually run the numbers, not make a claim based on intuition and assumptions.
Man, use a little intuition. ... Y'all would make some seriously shitty data scientists.
read it outloud but slowly
February 28th, 2018 at 7:01 PM ^
Well, I'm suggesting that you have a fundamental lack of understanding of tempo-free statistics and the way that they correlate to real-world performance, which you're reinforcing by trying to use the AP to prove a point. The AP poll in basketball, like football, is mostly about "who's gone the longest without a bad loss" and "what teams have I heard of?" It has very little to do with how good a team is.
And "tournament success" is an extremely noisy statistic. You could make a good argument that Michigan's 2014 team was better than the 2013 team, but the 2013 team made the NC game and the 2014 team "only" made the Elite 8.
The team that KenPom wrote that article about proceeded to finish 12-6 in the Big Ten, one game out of first place (three teams, including Michigan, tied at 13-5) and got a 4 seed. They finished #8 in his rankings, and they finished #14 in the final AP poll. That doesn't seem insane; in fact, it seems like pretty reasonable agreement.
(Virginia, BTW, was #33 in KenPom that season, despite having the same tempo and nearly the same defense as this season).
If you're not interested in having a better understanding of basketball, that's on you, and I don't care. But your argument basically boils down to "I don't like the results from KenPom's system, so I'm going to bad-mouth them."
As for whether or not KenPom's system has inherent bias -- it's a computer ranking. What kind of bias could it have? It's doing exactly when he asked it to do -- calculate tempo-free efficiency. It's up to you how much you value that. I value it quite a bit, because since I've been watching his site, he's done a pretty good job of predicting things based upon his system. I suspect that you don't value it because you don't enjoy efficient basketball -- from reading your posts, I get the sense that you would rather lose playing fundamentally inefficient physical basketball than win playing with finesse. You're welcome to your opinion, but I, for one, am damned glad that John Beilein doesn't share it.
February 28th, 2018 at 6:15 PM ^
Well, I would have said he's done a really good job this year, but since they're so good according to him, I now think he's done a terrible job since they're on the bubble
February 28th, 2018 at 5:57 PM ^
February 28th, 2018 at 5:59 PM ^
Kenpom has consistently underpredicted Nebraska the entire year. So yes, when it comes to Nebraska, there is a flaw because it's doing a poor job at processing the data available to think they are #51.
February 28th, 2018 at 6:19 PM ^
Well, T-rank has them 47th and BPI says 60th. So maybe it isn't Kenpom that's flawed. Maybe Nebraska just isn't as good as their record says it is.
February 28th, 2018 at 6:22 PM ^
February 28th, 2018 at 6:24 PM ^
Or maybe advanced stats have systematic biases? No no. Impossible. If you watch basketball and think Nebraska is the 60th best team in the country, eesh.
February 28th, 2018 at 6:27 PM ^
The computers must have programed themselves to hate Nebraska
February 28th, 2018 at 8:36 PM ^
You guys have like an elementary school level understanding of algorithmic data processing. Ask yourself, why are Torvik and Kenpom ratings different? They fundamentally use the same methodology, which I assume you and no one else reading this has bothered to investigate. The thing is they weight things differently.
Every single algorithmic based ranking system has biases. It's attempting to process innumerable data with a few simple rules. It has weaknesses. Nebraska is literally proof of that.
I'm not sure what you guys are circle jerking about, but I can guarantee you that your understanding of how these rankings is not fully formulated.
February 28th, 2018 at 8:52 PM ^
Or you could be wrong. Much like Tim Miles. But since you haven't provided any actual evidence to back your claims. Saying "look at Wisconsin and UVa, they are slow and KenPom loves them" ignores the fact that since 2014, UVa has won 30, 30, 29, 23, and 26 (and counting) games, while Wisconsin has won 30, 36, 22, 27, and 14 (and counting) games, and when they won fewer games KenPom ranked them lower.
And KenPom doesn't have any major issue with tempo teams; look at the teams that run a lot and you see a bunch of teams that are terribly inefficient offensively and, more times than not, are terrible at defense. Unless you believe former Michigan opponent adjusted tempo #10 North Florida is being unfairly ranked at #276.
If you want to be aggressive, at least put some effort into it. Of course, you also called people "uber cynical blog nerds" who now like to circle jerk about the mathematical approaches applied to ranking Nebrasketball, so this feels like it's not about sports anymore.
February 28th, 2018 at 9:58 PM ^
February 28th, 2018 at 10:47 PM ^
And as I responded elsewhere under said article, you didn't do any follow-up research and see that (a) his model accurately predicted Wisconsin would be a really good team that year despite the early-season losses (which actually works against your argument that it failed to accurately predict teams success), and (b) he never said it was due to pacing, which was and has been your ongoing argument regarding Wisconsin. And I guess third, the fact his model sorta pre-validated Wisconsin in 2012 has nothing to do with Nebraska in 2018, which plays a totally different style and doesn't have anything close to resembling that Wisconsin team's bonafides. I'm not ignoring your arguments for meaningless points; I'm pointing out where you are wrong and you are ignoring those comments with irrelevant responses or attempts to move goalposts.
February 28th, 2018 at 11:24 PM ^
Torvik has an explanation of the differences of the two systems right on his FAQ page: http://adamcwisports.blogspot.com/p/every-possession-counts.html
The current KenPom rating is "opponent-adjusted points per possession on offense minus opponent-adjusted points per possession on defense, times 100." That's it. There are two places Ken can (and does) put his thumb on the scale -- how much to weigh each game and how to adjust for the opponents. He's weighing recent games more heavily, which I imagine most people would like, and there are some adjustments for blowouts, home court advantage, etc. that are proprietary.
Torvik applies Bill James' pythagorean win expectancy forumula with a fudge-factor coefficient in order to come up with his rank. Pomeroy did that, but he stopped a couple of years ago. That's one additional fudge factor.
For a predictive system, the only meaningful question is "how well does it predict?" Pomeroy's predicts quite well -- he posts the results every day, along with the expected wins and losses. Is it possible that certain types of teams are frequently mis-predicted? Sure, and if that's the case, then average efficiency isn't a good predictor for them. I'm skeptical, however, not just because Wisconsin did turn out to be very good, but because of the small sample sizes involved. You can't really say "KenPom can't predict Nebraska," because he only gets about 30 chances a year to do so, and for half of those it's still using last year's data to help make the calculations. You'd need to be able to define a large class of statistically similar programs and show that, over time, KenPom did poorly in predicting how well they'd perform.
For example, if you truly believe that KenPom overrates slow teams (which has nothing to do with Nebraska, but whatever), you might take the bottom 35 teams in pace (~10%) for the last 10 years and compare the projected winning percentages to the actuals. If you can show a massive differnece, congratulations -- you've probably learned something critical about colelge basketball, and I urge you to take your newfound knowledge to Las Vegas. I'm guessing you won't.
PS: Please don't tell my boss about my inability to understand algorithms and statistics. :)
February 28th, 2018 at 6:32 PM ^
February 28th, 2018 at 6:38 PM ^
Besides Michigan (who Nebraska got at home when Michigan was on short rest), has Nebraska even beaten any other tourney teams? That's a major part of why their ratings aren't good. Their next best win is probably Penn State
February 28th, 2018 at 8:37 PM ^
What are you arguing? That it's impossible kenpom will systematically over value or under value certain team profiles? The first person who would tell you otherwise is Ken Pomeroy.
February 28th, 2018 at 9:08 PM ^
That's probably true. I've seen no evidence provided that his system under or over-values Nebraska, which is the original point of this post.
You want to know someone his system over-values? Auburn. They have one win over a team in the top-25 of KP, with losses to the #81, 80, and 56 teams in the country. He undervalues Clemson, a team with 3 top-25 wins and only one loss to a team outside the top-25 (#81 Temple). And yet, Auburn is #11 and Clemson is #18.
February 28th, 2018 at 7:06 PM ^
You're not just doubling down -- you're tripling or quadrupling down.
If you knew nearly as much about basketball as you think you do, you'd understand that Nebraska is a mediocre team; quite good at home, and not good at all on the road. Against Michigan's schedule, they wouldn't have sniffed the 4 seed in the tournament, and they are rightly on the wrong side of the bubble.
BTW: Nebraska isn't even receiving votes in your beloved AP poll. That makes them 40th at best, behind such stalwarts as Boise State, Seton Hall, and Loyola-Chicago. They are 33rd in the "coaches" poll, though -- 5 whole points ahead of Louisiana!
February 28th, 2018 at 8:40 PM ^
All I can do is continue to show you how you're wrong, with data. As I linked elsewhere, Ken Pomeroy himself acknowledged the Wisconsin issue. Every single algorithmic system will have biases and systematically undervalue or overvalue team profiles with respect to reality. That's just the nature of the business. Nebraska is a perfect example of an undervalued team. Where the interesting question lies, is why? What is it about their statistical profile that doesn't match reality?
February 28th, 2018 at 8:58 PM ^
Nothing. There is nothing about their statistical profile that doesn't match reality, because their statistical profile is, literally, reality.
By themselves, statistics aren't meaningful. They get their meaning when you put them into context to see if they explain things in the real world. If they're a poor model of the world, they should be replaced with something that's better. Nobody's arguing with that. But it's ridiculous to claim that statistics are somehow "not real" because you don't value what they measure.
There is an assumption inherrent in your argument above that Nebraska is undervalued.
The original Tweet pointed out an earely season prediction that was inaccurate. That is obviously a true thing. Nebraska won more games than they were predicted to win.If that defines undervalued, then you're correct. It is correct that based on November predictions Nebraska has more wins.
But if you are saying, as we sit here today, they should be considered a top 40 or top 30 team? I don't know that many people outside of Lincoln agree with that assessment. And you have certainly failed to meet the burden of proof for that.
The over-riding reponse the board seems to be making here is one of two things:
1) Tempo is not the sole proximate cause of this early season prediction error
2) It's not that Nebraska is a LOT better (although obviously they ARE better) than predicted, it's that a big section of their competition was a lot worse than predicted.
Also, you keep pointing out a 6 year old blog post from KenPom to support your claims. You repeatedly say that KenPom's model is flawed (you're *saying* biased, but you *mean* flawed). You know? He answered you in the article:
"Of course it’s flawed. The thing is, your knowledge is flawed, too."
and I agree that they are almost certainly underrated by these systems.
Two of their three best players transferred in so I'm not sure how their effect was considered by kenpom's early season rankings, and they were probably just assumed to be about as bad as they were last year (v bad).
As for the mid-season rankings, it's clear that they struggled to integrate those transfers early in the season and they weren't a very good team. They lost by 23 to St. Johns and 29 to MSU. They lost to UCF. They had a close games with LBSU and Eastern Illinois.
Then the transfers settled in and Isiah Roby emerged as an absolute beast and they turned into a different team. Kenpom's system is almost certainly putting too much weight on their early season performance, but again, it doesn't know the circumstances around the two transfers.
Torvik has them ranked 19th since the turn of the calendar. These systems use full season data because under normal circumstances, the first half of the season does matter even if you turn it around. But in their case, it probably matters a lot less than normal teams because of the transfers in.
February 28th, 2018 at 8:37 PM ^
At the beginning of the conference slate (well, the full-time start, not the 2 games everyone played), they hadn't beaten a team ranked higher than 44 and had lost to the #61, #79, #2 (by 29 points), #31, and #6 teams. Kansas was a good showing, but they also lost by 23 to St. John's. The model did not under-predict anything about Nebraska; they looked like a mediocre team. What happened was they won a number of close games that were deemed effectively toss-ups at the midway point as the teams like Minnesota, NW, Wisconsin, etc. went from "maybe they're still okay" to "nope, they are garbage."
On the season, Nebraska has three wins against teams that are ranked higher than them: Maryland, PSU, and Michigan. They have 9 wins against KP top-100 teams; broken-ass Wisconsin has 7, same with nowhere near the bubble IU. Marquette, the #50 team in KenPom has 10 (with a chance for 11 against Creighton) and is basically .500 in the Big East.
If anything, KenPom's numbers are a better indicator of true ability that conference record, which is what I think Miles is arguing about. the Big 10 is really down this year outside the top 4 teams, and so while being the talent midget is great as a talking point, the fact that nobody is buying Nebraska as anything more than a mediocre team beating up on the bottom feeders of a bad conference is just what the KenPom system actuall teased out.
February 28th, 2018 at 8:42 PM ^
No the B10 sucking is a myth. It just so happens that teams like Nebraska and PSU are decent this year while IU and Wisconsin suck, so the naturally people say the B10 is down. I would take either Nebraska or PSU over OU or Texas though without hesistation.
February 28th, 2018 at 9:02 PM ^
Match those four teams up against each other, and I'd probably take the home team in each case. On a neutral court, I think I'd rank them OK, PSU, Texas, Nebraska. But that's not because PSU or Nebraska are any good. They're all pretty mediocre.
Note that this doesn't mean that PSU or Nebraska should (or will) get into the tournament, because they both have terrible tournament résumés. PSU's soft scheduling hurt them quite a bit.
(If you're curious, KenPom has PSU, OK, Texas, Nebraska)
February 28th, 2018 at 9:29 PM ^
No, the Big 10 is just not that good this year. No amount of yelling and feelingsball will change that. PSU's best win outside of the conference is #77 Montana; they also lost to Rider at home, and closed the year losing 3 straight games. They are 9-9 in the Big 10, which is about right for, again, a mediocre team. Their sweep of OSU is the only thing even remotely making them relevant to bubble talk.
Nebraska is a .500 team most years. I've made this argument elsewhere, so I won't repeat it again, but they haven't beaten anyone of note except a supremely tired Michigan team and also lost to bad teams like Illinois. Both OU and Texas are ranked right about Nebraska, and they all have more impressive resumes than Nebraska or PSU (Texas has wins against top-25 teams Butler, TCU, and Texas Tech, plus a chance at another scalp against WVU before their conference tournament, while OU has Wichita St, TCU (2x), Texas Tech and Kansas). And both those teams have trended away from the bubble; everything I've seen recently says OU is barely in, and Texas is outside. And these teams are 7-10 in their conference right now, not 13-5 and 9-9. So again, same basic teams, but Nebraska and PSU benefit from being in a weaker conference.
February 28th, 2018 at 3:42 PM ^
I think he's trying to say he doesn't like his kenpom ranking of #51, but makes a fair point that kenpom has been wrong about Nebraska all season.
he's pointing out that at the midpoint of the season, they were predicted to win 7 big ten games and they nearly doubled that so he's saying the prediction was bad.
They played like the 97th or so best team in 2017 and they've played like the 19th best team in the country (per Torvik) in 2018. I'm not sure many people would have predicted that turnaround but given they were so reliant on transfers, it isn't as surprising that it took them a while to gel.
February 28th, 2018 at 3:42 PM ^
Tim Miles and I subscribe to the same website!
February 28th, 2018 at 3:42 PM ^
It is funny when any model predicts something worlds away from the actual result - at the 10-5 juncture, the analytics still were predicting a 16-15 finish to their season, which is obviously slightly different than the 22-9 outcome. The model - whichever model you want to deploy for predictive analytics - is never perfect, but it is kind of amusing when it is way off once the season is over (well, regular season in this case). I can especially see that if you're the head coach.
Of course, what he is sort of forgetting is that these models do adjust upward or downward based on in-season data, so if we had a later snapshot, it might be more in line. Still, I can see where you'd get a kick out of it emotionally.