XTRA! XTRA! READ ALL ABOUT IT: KIck Posterior QA by JNew and KENPOM...
-
Thought provoking take.
-
@JayHawkFanToo A few random thoughts:
-
In the KU RPI story above, KenPom himself says the best method to seed teams would be to take a composite of the top ratings systems. That would eliminate outliers.
-
I think “Top 50” wins is a bad way to look at quality teams. Why not Top 25? Or Top 101? Or Top 213? Where is the game played? Winning on the road at No. 51 is much more impressive than at home vs. No. 5o, This kind of analysis doesn’t add these factors. KenPom does.
-
I don’t see the world in black and white. To me, a team’s strength is independent of its schedule strength. In 2003, Syracuse won a national title without playing a road noncon game. We always hear that scheduling hard is best … but is it true? I have no idea, but I’m not painting myself in that corner. Can a good team play a good noncon schedule? Yes. Can a good team play a bad noncon schedule? I would say the answer to that is yes too.
-
KU is knocked because of margin of victory. Wins and losses of each game don’t give us enough data … the scores (along with pace) tell us so much more. We only get 35 or so data points each year. Throwing away some of the most valuable information — how much you outscored your opponent per possession — seems silly to me, especially when that info is available.
-
-
Good points., A few of my random thoughts as well…
-
I agree, an average of computer rankings should be a component of the overall ranking, like the BCS football rankings used to include…I am not sure if they still do with the new playoff system…and KU not being in the conversation.
-
You can use the top 25 or top 100 and will see the same trend; I picked the top 50 as an in-between compromise. KU has played only 6 of 13 games at home, I have not looked up the numbers for other teams but I will guess this is one of the lower numbers of home games of the top 25 or top 50 teams, so it should favor KU.
-
If you read my posts, I am a firm believer that the best team does not always win the NCAA. KU was not even close to the best team in '88 (lost 2 out of 3 to KSU and 2 out 2 to Oklahoma, teams it beat in the tournament) but on a 2-out-of-3 or 3-out-of-5 format, KU probably has 2 or 3 more titles when it was truly the best team, including the year Syracuse won. Rick Pitino said the same thing on the round table interview he did on ESPN with Coach K, Roy Williams and Jim Boeheim.
-
But then, shouldn’t coming back from 18 down in the second half (Florida) also be considered a big plus? A tougher SOS naturally results in smaller margins of victory than playing punch bags, wouldn’t you agree?
Interesting discussion with no wrong answers just different points of view.
-
-
Shoot, who disabled downvotes? I wanted to downvote the wise guru of KUhoopdom, the sabermatrician @Jesse-Newell.
Holy cow, UMiss has a chance to beat Kentucky! Late, breaking news!
-
@wissoxfan83
Here is a work around.
-1
-
@wissoxfan83
And you can even down vote yourself!
-
@jaybate-1.0 or
1 yourself
-
10^1000000000
-
@JayHawkFanToo Absolutely agree with you on No. 4 … a tougher schedule does mean you don’t have to win by as much to be considered a good team. I think what’s killing KU this year is its losses. Getting routed in two defeats (especially to Temple) drags down an otherwise strong resume.
-
Excuse me. I’m going back to the pic of the two guys with bras on their heads. Far easier for me to relate to than trying to crunch all the SOS, RPI and KenPom numbers. Might even Google and ooogle Ms LeBrock while I’m at it.
-
Look at the latest KenPom rankings…
KU and Michigan State did not even play this evening and yet KU dropped 3 spots and now sits behind MSU. KU has a better record in any way you look at, including overall record, SOS, record vs top 25, 50 and 100 teams, opponents O and D and more importantly KU beat MSU.How in the world can MSU be ranked ahead of KU? Same thing with Utah…and Baylor…and…I could go on…
Notre Dame just beat UNC, has a better record and yet it sits 4 spots below. Oklahoma at #7? Yes, they beat Texas but they have 3 losses 2 of them to mid-major Creighton and Washington, a team that is 0-2 in the PAC 12. I understand that the ranking will get better as the season goes on but at this point it seems grossly out of whack.
The RPI might not be the good but KenPom’s ranking seems to be even worse.
Again, just my take on KenPom’s ranking using the numbers he presents.
-
Fun Stats Game: Pick your point guard based on Statistics only:
Player 1 13 Games Played (14-15) 33.3 Minutes 12.4 Points 3.8 Rebounds 4.2 APG 1.7 SPG 0.3 BPG 2.2 T/O .491 FG .846 FT .514 3PT%
Player 2 39 Games Played (12-13) 35.3 Minutes 18.6 Points 3.2 Rebounds 6.7 APG 1.6 SPG 0.5 BPG 2.2 T/O .463 FG .801 FT .384 3PT%
Who would you pick and who do you think the players might be?
-
@Jesse-Newell I appreciate the explanation concerning UConn. All this time I thought he was cooking the books. I still think his rating on our defense is way, way off.
-
@JayHawkFanToo Kenpom’s numbers are illogical. 46th in defense…really?
-
I know the players are Frank Mason and Trey Burke of Michigan. Both records are for their sophomore years and Burke’s last year in college. Burke was a consensus 1st team All-American and much like Mason, the driving force in his team, so he would get the nod, although Mason’s stats are pretty gosh darn good and by the time he is a senior, he might well be a 1st team All-American as well.
-
@JayHawkFanToo Here’s the best way I can describe the way I look at it: Every team’s season is a book, and KenPom is able to read all the chapters.
A lot of the arguments that you made — and a lot of people make them — center on one team winning head to head or records against only certain teams. “Chapter 3 and 5 and 9 are good, so how can the book be bad?”
For me, KenPom’s computer is much better than me because it strips biases. Each chapter can add to the overall story. And teams can improve their stock even if they aren’t playing top 50 or 100 or 200 opponents.
In Michigan State’s case … it is the exact opposite of KU: It has lost every close game and blown out almost every other opponent. 30-point wins over top-100 KenPom teams should mean something. Losing an overtime game at Notre Dame (No. 14) also probably tells us MSU is better than its record.
We only get so many data points. Lots of people are content at stopping at head-to-head record or only looking at wins and losses. KenPom’s projection numbers, if you look, are almost a reflection of Vegas’ lines. And that’s because he takes each chapter of a team’s book into consideration.
-
@KUSTEVE Yes, 46th. Easily KU’s worst defense under Self. Jayhawks are 193rd in eFG% after never finishing worse than 64th in any other year under Self. This team doesn’t block shots like the past, and it never turns anyone over. Hard to make an effective defense when both those areas are bad.
Again, the numbers aren’t taking into account possible improvement (it usually happens under Self), but up to this point, I’d say that 46th number accurate reflects what this year’s team is: a good defense for normal team standards, a horrible defense for Bill Self standards.
-
You raise a intriguing point: the similarity between the Vegas Lines and KENPOM.
Having worked some with stats and computer modeling, I believe it is rather unusual for nonparametric statistical modeling to coincide closely in predictions with other non parametric statistical modeling, unless both are using essentially like modeling assumptions, and like non parametrically applied statistical models, and like data sets.
Is Vegas relying on KENPOM?
Or Vegas oddsmakers working with a stat model essentially the same as KENPOM?
How do they come out similarly?
If I recall correctly, and I am not absolute sure that I do, you and others have noted a lot of variance between KENPOM and other ranking and predictive systems regarding rankings and predictive spreads.
-
@Jesse-Newell I’m sorry, Jesse …I don’t see a bad defense like last year’s edition.
-
My 12th grade education is preventing me to fully appreciate jb’s last post, so I Google’d “non parametrically applied statistical models”. Even my computer crashed.
-
Parametric Models - I think I dated one in college.
-
-
@wrwlumpy fifi and lumpy! Ahhh
-
I agree that a computerized model treats all teams equally, but it also can introduce the developer’s own bias depending of how he sees different parameters. Many of the models have been developed by mathematicians with little knowledge of the game and who try to use numbers to account for everything when in real life you really cannot.
I did not based my comment on just the head-to-head result; you can look at all the parameters across the board and KU has better numbers in most if not all of them. Neither team played and yet MSU moved past KU in spite of all the numbers favoring KU. Like I said in my post, I understand that as the season moves on and more data is available the models get better, but at this point in the season, KenPoms seem to be unrealistic. You mention that MSU is losing close games and that is helping them, but you are completely overlooking the obvious…they are still losing! Do you think it is better to lose 20 games by one point each or to win the same 20 games by 1 point each? Under your premise, the teams that is 0-20 is rewarded for losing very close games while the teams that 20-0 is penalized for winning very close games…I understand that it is an extreme case and other factors are involved but the gist of it is still valid.
As far as the Vegas lines, you have to be careful on how you look at them. Initially they are set as close as possible to the most likely outcome with the sole purpose of having the bets balanced. If the betting is not balanced, and depending on the final outcome, Vegas can make a lot of money or lose a lot of money…Vegas does not like that uncertainty and the only guaranteed way for them to make money is to have the bets perfectly balanced and make money on the “vig” or commission it collects. Vegas will tweak the line not to better predict the outcome but to balance the bets, and on games where there are heavy emotions involved and one party is heavily betting one way, the final tweaked line is not close to what the outcome is expected to be but to the number that will bring the bets in balance, Vegas could not care less who wins a game, they only care about making money.
The gambling industry is a multi-billion business and I will guess they have access to resources that we don’t know and the models they use are considerably more sophisticated and confidential that the public ones, plus they have access to “inside information” such as confidential injury reports, that KenPom, Sagarin and Massey can only dream of.
Very interesting discussion and several perspectives that reflects our own bias. Good to have you here to keep us on our toes and get our thinking caps on.
-
Here’s this week’s update. KU makes a nice jump up the board.
-
I don’t usually like models, but Fifi has a serious chassis.
-
I would be interested in your personal comments about the KenPom numbers.
Obviously KenPom’s computer does not watch the actual games, and as I indicated before, the model is pretty flawed until close to the end of the season when the outcome is mostly obvious anyway.
Do you believe that OU that just lost to bottom feeder KSU at home has the best chance of winning the conference? It looks like once again, a team gets more credit for losing than winning a close game.
I watched ISU play OSU and WVU and it had to hold on for dear life to beat OSU at home and barely won at West Virginia; both wins were by 2 points . Does this mean that OSU is better and WVU worse than we know or that ISU is wildly inconsistent, and if the later is the case, should they really have the 3rd bets odds to win the championship? On the other hand, ISU is obviously in a better position than Oklahoma since it won two games it could have easily lost while OU lost a game it most certainly should have won, wouldn’t you agree?
Just a couple of observation by quickly glancing at the numbers, and again, I would be curious on what you think.
-
The response above I had to you “A few random thoughts” is how I feel about KenPom. Don’t have much to say other than that. We obviously aren’t going to agree on this … you say KenPom’s computer “doesn’t actually watch the games” while I would argue that KenPom’s computer is best at watching the games, being able to take every single possession of every game into account without human bias.
So yes, I believe OU is still a good team. I think there are many games (like the Texas road win) that prove that. Just as I wouldn’t say KU is a bad team because of the Temple game, I’m not going to use one game to end my judging for OU.
ISU barely beat OSU, IMO, because OSU is a really good team. I think you’ll see that tonight when the Cowboys play at Allen Fieldhouse.
Bottom line: From all the information we’ve gathered, OU is the best team in the conference, albeit just slightly. Knowing that, it remains the slight favorite at the top to win or share the title.
-
At this point of the season, does it take several games to significantly change a team’s stats and standings in KENPOM? I always wondered how many games into a season that things start taking several games to change. I guess I am asking about how many games before the stats and rankings get fairly stable.
And, hey, I am psyched up for the OSU-KU Live Blog!
Hope you and Blake and KHas will all be sitting in.
-
The issue with computer rankings is not that they look at teams differently, they most definitely have no bias in this regard as they treat all teams equally; however, the real issue is that they introduce the bias/vision of what the model creator thinks is important. If the computer models were perfect, then the results would be very close. They are not. For example look at the current ranking for KU by 3 of the top rankings:
- Ken Pomeroy - 16
- Jeff Sagarin - 12
- Kenneth Massey - 6
Many believe that Massey’s rankings are the most full-featured and scientific rankings. Even Pomeroy has spoken highly of Massey’s methods…and yet, all three have widely varying results that highlight the emphasis (read bias) each model developer has introduced into his own model.
Here is a link to a summary of all computerized rankings; you can see KU is ranked as high as #4 and as low as #19…and all of these models are using basically the same data, just looking at it differently and coming up with different results, so…which one is correct . See my point?
BTW, what is your prediction for tonight’s game? Are you taking OSU?
-
My suggestion is: the one that is closest to correct would be the one that averages the best forecasts of final game scores, predicted conference finishes, and final ranking in the NCAA tournament.
I suppose it wouldn’t be to hard for a quant with a budget to input that data for all the different services for the last 5 years and come to a conclusion.
-
@jaybate-1.0 Agree JB. Or even better, as Ken has suggested, take an aggregate of the top proven systems so that outliers are filtered out.
@JayHawkFanToo I didn’t think I was arguing Pom’s merits against Sagarin and Massey, which obviously are well respected as well. I can tell you that Pomeroy’s numbers almost directly reflect the Vegas line, which to me shows he’s right on the money in most cases. He also does test and retest and isn’t afraid to make changes if it improves his projections.
And I’ve got KU beating OSU by two.
-
As I mentioned before, the Vegas lines (and I am on the records saying that I like them a lot) are not predictor of outcome, they are set to make sure bets are evenly spread and they are tweaked, not based on any sports criteria, but uniquely to ensure even betting and profit for the house on the vig. Maybe, just maybe, the opening line could be…but it quickly changes to balance the bets. For example, against TTU, KU open favored by 15 and by game time it went down to 14-1/2 (might be off by 1’2 point) and KU ended up winning by how much? 32 points?
As far as tonight’s game, OSU has a short bench and in the last 6 games (as far back as I checked) they had only 5 players in each game score more than 2 points; the key is limiting Forte’s points. In the their 3 (not close) loses Forte had 5 points against South Carolina and 13 points against Maryland. BTW, Vegas opened at 6-1/2 KU and the consensus is now at 7-1/2; I will surprise you and take KU by 7.
-
@JayHawkFanToo The Vegas line is closer to reality than the actual game score. We’re trying to determine what is going to happen before the game. After it starts … well, stuff happens. You think Florida State is 30-plus points worse than Oregon? Or Oregon is 20-plus points worse than Ohio State? Of course not. Stuff happened. A lot of luck is involved. So if they played the games 100 times, who would win most? That’s what we’re trying to figure out, and the Vegas line (like you said, especially the opening line) is about as close as we can come to that.
-
The statistics explanation of how bet balancing works, at least assuming it is NOT corruptly engineered with managed hype and a lot of narco and intel money laundering distorting it massively, is that over large numbers of bets the error factors of each bet by each bettor cancel out to a probabilistic estimation of the most likely actual spread, especially if the betting process is biased with the initial line setting based on sophisticated initial prediction. Essentially betting becomes a huge series of tiny interactions around the initial line. So, statistically speaking, bet balancing is just another technique of predicting a probable outcome. The logic that bet balancing is not about the estimating the final outcome contradicts the reality of what all of the betters are in fact trying to do. They are each betting with their biases on what the actual outcome will be. The beauty of bet balancing is that it is continually interacting and canceling out the error factor of each individual bet and bettor. So the probability is that the Vegas line, with an enormous betting volume, is actually going to tend to be more accurate than any single estimation system, especially if it were to start with, say, KENPOM. Or at least that is how I have had it explained to me by a couple of folks that I would tend to think know quite a bit about this sort of thing.
-
“As I mentioned before, the Vegas lines (and I am on the records saying that I like them a lot) are not predictor of outcome, they are set to make sure bets are evenly spread and they are tweaked, not based on any sports criteria, but uniquely to ensure even betting and profit for the house on the vig.”
That’s it. That’s how the line works in Vegas. Because their money-making model is to reduce their risk to nothing while taking 10% fees from the losing side. The Vegas line is a measurement of how the betting world sees the game outcome at that very moment. That’s why lines change over time, even if most (or all) the real factors involved in the game remain the same, because bets come in and start changing the scales of dollar amounts on each side.
We can criticize computer prediction models all we want… because they only offer their results off the limited information they have.
If you want to put it to the ultimate test… try to have a computer predict specific player match ups. Well… if I wanted a machine to accurately predict results I’d want match up info in there and I’d want it to be accurate. How can that work?
Computer results tend to be really wild early in the year because there isn’t enough data yet.
Let’s all put ourselves in the shoes of these computers. Imagine a season where you didn’t see a single game. No visuals. And you don’t read blogs and news that give other information. The only thing you have to go off of is a stat book. Now you predict the future! We blast computer interpretations, but this is how a computer has to deal with it all, with some additional programmable formulas developed by someone… Whoever is the person (or people) writing the formulas and codes… there is nothing in this for them to develop inaccurate results. The big prize for them is to develop something that has a higher accuracy over time.
There will be improvements as we move forward, just like the fact that we will see improvements in all areas of computing. I saw a nerdy friend wearing this:
The only sure thing in life is:
Death
Taxes
Software Updates
-
Bottom line: From all the information we’ve gathered, OU is the best team in the conference, albeit just slightly. Knowing that, it remains the slight favorite at the top to win or share the title.
Do you still think OU is the best team in the conference? They just lost to WVY by 19…after losing to KSU at home.
-
@JayHawkFanToo ha ha!!
-
The Vegas line is closer to reality than the actual game score. We’re trying to determine what is going to happen before the game.
With all due respect, I don’t believe you understand how the Vegas line works. Although they seem to closely resemble margins of victory, the Vegas lines do not predicts game outcomes and they are not meant to do that, they simply predict the numbers that will force money to be bet evenly, which is the only guaranteed way it has of making money and this is why they are constantly changing, again, so money bets are balanced. Vegas does not care who wins or loses or what the margin of victory is, all they care is that the bets are balanced so they can make money.
BTW, nothing is closer to reality than the actual game score…the game score IS reality, right?
Again, no malice or disrespect intended.
-
Again with all due respect and no malice intended, you like Jesse are under the impression that the Vegas lines are trying to predict the outcome of games, the Vegas lines do no do such thing, they try to predict the number that will cause balanced betting, even when they resemble the predicted outcomes. I will leave it at that and you can research it for yourself.
-
I learned long ago not to quarrel about statistics, but I do want to take one more pass at what prediction and bet balancing are about, as far as I have been able to learn.
Accuracy of estimation in stochastic realm is all about canceling prediction biases and limiting variance to random error.
There is perhaps no better way to cancel biases than by bet balancing at high volumes of betting. It is a marvelously effective technique.
The larger the number the better the cancelling of biases.
And the “vig” doesn’t bias the outcome of the bet balancing at all. In fact, the vig is irrelevant to the final betting line. In effect the vig comes half out of the range above the betting line and half out of the range below the betting line.
I am not sure where this urban legend of bet balancing being nothing more than a meaningless indicator of what bettor’s think got started. It is not that anyone of the betters know diddly squat. It is that all together cancel out each other’s biases.
I know for a time that I learned this urban legend and assumed it was true for a time. But because I was trained some in stats I thought finally that there was something wrong with that logic and I worked through it and logically it is just an urban legend.
For the same reason that a Vegas line is a very effective way of predicting the outcome of games because of using high Ns of betting to balance, Ken Pomerory suggests using all the different estimation models, not just his, or Sagarin’s, or someone else’s, to make predictions. The more estimation models you use, if they are all based on the same data and varied, but sound algorithms, the more likely it is that the biases of their algorithms will cancel out and you will end up with a closer approximation of the actual outcome.
But when it comes to stats, again, I never argue. Just sharing some food for thought.
-
You obviously understand how the Vegas lines work; it is one of the most misunderstood concepts. I learned the concept myself the first time I visited Vegas 30+ years ago.
