tag:blogger.com,1999:blog-38600807.post1864396249045947897..comments2023-11-05T04:16:44.937-05:00Comments on Advanced Football Analytics (formerly Advanced NFL Stats): Team Efficiency Rankings - Week 13Unknownnoreply@blogger.comBlogger37125tag:blogger.com,1999:blog-38600807.post-87412050402635465662012-12-02T21:00:27.932-05:002012-12-02T21:00:27.932-05:00@ Mike M. lol@ Mike M. lolBoston Chrisnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-77847221522639329052012-12-01T23:59:36.002-05:002012-12-01T23:59:36.002-05:00Regression happens over a long period expecting it...Regression happens over a long period expecting it to happen in one or two games is foolish Mike...Season to season shows consistent inverse correlations with TO's providing excellent proof of To's randomness...<br />time to take a basic stats course?<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-1140918102277583532012-12-01T22:54:48.383-05:002012-12-01T22:54:48.383-05:00The big "regression to the mean" team th...The big "regression to the mean" team this week is the Bengals.<br /><br />The model has them favored, they will likely lose and maybe big.<br /><br />Another "regression to the mean" team the model has right is the Patriots who are due a regression to the mean but will likely pull-out a very close win but they will not cover the spread, however, a loss would not surprise me at all.<br /><br />A 'regression to the mean" team which should play much better is the Raiders.<br /><br />Something to think about, in regards to turnovers being random, Super Bowl winners almost always win the TO battle in the regular season and continue to win the TO battle in the post season.<br /><br />If TO's are random then why don't we find teams losing the TO battle in the regular season completely reversing coarse and winning the TO in the postseason and winning the SB on a regular basis ?<br /><br />Last season this was the Saints who were very highly rated on this site but won the TO battle in only 5 games in the regular season and then were taken out behind the woodshed for a TO beating by the 49ers in the playoffs when the 49ers had the most games winning the TO battle in the league but were only ranked like 12th or some such thing last season by this site ?<br />Mike Mnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-81571076134489476632012-11-29T18:28:16.037-05:002012-11-29T18:28:16.037-05:00First, there are no (including Brian's) active...First, there are no (including Brian's) active MAP-oriented models for ANY professional sport I am aware of. The whole argument about which model is "more forwards-looking" is silly.<br /><br />Second, it is relatively easy to look at prediction quality when win probabilities are involved (I mentioned this before and actually computed them for the ANS model for a week). This can be done using the normalized average negative log likelihood over a set of predictions. I have privately computed these numbers for the ANS models for a few more weeks, and compared them to some simple alternatives. I don't have the numbers handy, but ANS beat not only the 50/50 on all games predictor (i.e. it's better than random noise), but also a home-field based predictor with parameters optimized on the test set (in other words, as over-fitted as it could possibly be).<br /><br />Sadly, I don't know anywhere that that has historical numbers for straight-up win/loss bet payoffs from Vegas, and this is the "gold standard" everyone wants to compare against. It might be possible to convert lines into implied win probabilities, but I don't know how. If someone has these numbers, I'd be grateful for them, and happy to use them to compare models' accuracy.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-67912900755445206322012-11-29T11:54:47.959-05:002012-11-29T11:54:47.959-05:00The biggest flaw with Brian's model is that it...The biggest flaw with Brian's model is that it is actually not very predictive.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-58334510021974993832012-11-29T11:36:52.762-05:002012-11-29T11:36:52.762-05:00@Kulko
It doesn't matter. When you start out ...@Kulko<br /><br />It doesn't matter. When you start out with the absurd assumption, as FO does, that past turnover rates are solidly predictive of future turnover rates, then the rest of your model is worthless.<br /><br />The biggest flaw with Brian's model is that people don't understand that a team having WP of 60% over another team doesn't mean that they will win every time.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-23329281382672855212012-11-29T09:02:32.620-05:002012-11-29T09:02:32.620-05:00Is there any Proof, that this model is a MAP model...Is there any Proof, that this model is a MAP model?<br /><br />is there any Proof, that it is statistically better than FOs model when forward predicting?<br /><br />But from the way I see it, both models makes significant unproven assumptions what is relevant and what is not, and just play it from there. Both models include assumptions that i deem downright unlikely.<br /><br />Both models have year in year out teams, where there data not only looks wrong, but stays wrong during the whole season.<br /><br />Both models are significantly better than typical power ratings or won/loss.<br /><br />Any proof that one of them is statistically better than the other are notably absent.Kulkohttps://www.blogger.com/profile/17657346387956365135noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-13848780155518486162012-11-29T07:55:19.468-05:002012-11-29T07:55:19.468-05:00I'm wondering if you could tweak the model to ...I'm wondering if you could tweak the model to have different regression rates for different teams based on how consistent they areAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-44412379400165940812012-11-29T01:18:43.690-05:002012-11-29T01:18:43.690-05:00Nate There is a famous article/study written on ho...Nate There is a famous article/study written on how Books set lines to maximize profit. (sorry can't remember the name but I will find it) It explains that a "combination" of reducing outstanding liability & taking position on some games does this more than splitting action. To do this as I said they take advantage of betters psychology how they react to big wins big losses/ popular team etc.<br /><br /><br />As for proof on beating Vegas. There are some documented online Dr. Bob over huge sample a volume approach. Byeweekpicks using a selective smaller number of games approach I make a healthy second income over last 10 years my self.<br />I can give you other evidence. Underdogs on a whole cover more than favorites.This +ev increases when you look at teams with lower public betting on them. (ie fading the public) yes large underdogs are another +ev double digits are best. And, teams off large losses or off of powerful wins also has +ev in very large samples. Like I said its easy to find +ev opportunities over large samples the hard part is actually 'pulling the trigger on putting your hard earned cash on god awful teams (I did on Jacksonville last week And KC vs Pit a few weeks ago on Monday night. even though I knew analytically I had value and on right team. Vegas knows in a psychological sense it is very hard to back a loser.As they say everyone loves a winner and to be on a dominant team. Also betting certain teasers blind are proven to be +Ev. <br /><br />Also there are a large selection of people who lose well be one their expected return. This is because they dont flat bet they cut into edge by using star 1*2*3*5* systems (crazy) Or chase after losses by increasing bets.or play sucker bets. <br /><br />Mike<br /><br /><br /><br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-13148515268000959282012-11-29T00:52:41.238-05:002012-11-29T00:52:41.238-05:00Interesting--thanks, guys. Wasn't expecting a...Interesting--thanks, guys. Wasn't expecting a result so quickly. So it's true, but only barely so.<br /><br />Can you filter for heavy favourites playing on the road? I have a hunch that the average bettor does not adjust for that as they should.<br /><br />When I was doing this for real and making money, I also adjusted for a fuzzier and somewhat subjective sense of teams that had a big rep but were past their prime, aging and on a downward trajectory that hadn't yet trickled through to popular opinion. Helps too if they have an especially large and rabid fan base, like the Cowboys for instance. I'm not sure how you could quantify this...and of course with my tiny sample size I do understand it's entirely possible I just got lucky.SlackerInchttps://www.blogger.com/profile/08275358994906136088noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-65217619587113830212012-11-29T00:34:25.769-05:002012-11-29T00:34:25.769-05:00> Vegas is concerned with profits, not short-te...> Vegas is concerned with profits, not short-term risk.<br /><br />The Vegas line should still be influenced by the action. <br /><br />Consider an idealized model where the possible point differential in a game is continuous and has a cumulative probability distribution P: Reals->[0,1] and there's a betting pool of size 1, where the fraction that takes the under when the line is set at a particular value is U:Reals->[0,1].<br />Then the bookie's EV for setting the line at L is:<br />(2*P(L)-1) * (2*U(L)-1)<br />Natehttp://www.google.comnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-28007705482355783482012-11-29T00:29:09.520-05:002012-11-29T00:29:09.520-05:00Alan - Favorites of more than three points have go...Alan - Favorites of more than three points have gone 49.5% against the spread since 2007. Not sure if that is significant. Definitely not a market inefficiency you could make any money off of.<br /><br />Source: http://sportsdatabase.com/nfl/query?sdql=line%3C+-3+and+season%3E%3D2007&submit=S+D+Q+L+%21Michael Beuoyhttps://www.blogger.com/profile/03960600491528993233noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-44889717624921542112012-11-29T00:28:36.794-05:002012-11-29T00:28:36.794-05:00> For someone sharp with stats software, it sho...> For someone sharp with stats software, it should be easy enough to > put in all the cases where a team is favoured by more than a field > goal, and see whether (as I'd predict) they actually covered that <br />> spread less than half the time.<br /><br />Through 2000-2011 the team covered 1004 times, and failed to cover 1028 times, so I guess you're right. (Up to noise, the spread is correct.) This holds if I raise the threshold to 7, 10, or 14 points, though noise becomes more significant at those levels.<br /><br /><br /><br />Natehttp://www.google.comnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-31327657227272426602012-11-28T16:14:14.590-05:002012-11-28T16:14:14.590-05:00Anonymous, I too would be curious to see documente...Anonymous, I too would be curious to see documented support for the claim about beating Vegas consistently. Anecdotally, I did very well for a couple years before the site I used stopped taking American money. If I were in a state or country where I could bet fully above board, I'd be back in a heartbeat.<br /><br />As the poster said, it strikes me that the "masses" end up moving lines too far in the direction of favourites with big names and pedigrees, and are too slow to recognise potent, small market up and comers. And most people want to bet "for" a strong team, even when they must cover a spread, so whether the bookmakers shift a line to keep the money even on both sides, or whether they just want to take people for suckers, the incentive is the same: the point spread the strongly favoured team must cover is usually higher than it should be. <br /><br />For someone sharp with stats software, it should be easy enough to put in all the cases where a team is favoured by more than a field goal, and see whether (as I'd predict) they actually covered that spread less than half the time. I would be shocked if it weren't so!SlackerInchttps://www.blogger.com/profile/08275358994906136088noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-68233810428454890582012-11-28T13:36:10.512-05:002012-11-28T13:36:10.512-05:00hi brian,
in the article you just wrote for the wa...hi brian,<br />in the article you just wrote for the washington post (http://www.washingtonpost.com/blogs/football-insider/wp/2012/11/28/playoffs-bye-weeks-and-penalties-everything-you-need-to-know/ ), you said that if the skins win all their remaining games that they are guaranteed a playoff berth. i don't think that's the case. the giants could easily win enough to take the division. there are 4 teams ahead of the skins competing for the 2 wild card spots. the way i see it, at least two of them could win out too and stay ahead of the skins.<br /><br />it would be awesome if you could come over to the redskins "insider" blog and explain it.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-84303839231462987222012-11-28T12:59:07.132-05:002012-11-28T12:59:07.132-05:00"many can beat Vegas on a regular basis."..."many can beat Vegas on a regular basis."<br /><br />Mike, do you have any evidence to support this claim?<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-49403946937975683452012-11-28T12:55:43.704-05:002012-11-28T12:55:43.704-05:00Where would Jeff Sagarin's ratings fall on the...Where would Jeff Sagarin's ratings fall on the forward-looking scale?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-25745753035351563072012-11-28T12:44:09.597-05:002012-11-28T12:44:09.597-05:00Vegas...is most concerned with public perception.T...Vegas...is most concerned with public perception.There's is not the most 'forward looking system. THey use power rankings that are formed in large part to historical performance.i.e. what they did last year and before. This is done because the bettors remember successful teams & are slow to jump on perennial bad teams.(who is the leader in ATS this year surprise? if you are a successful gamble ryou will know its Cleveland. Could Brian improve his model to use portion of last years stats?Perhaps I have tested correlations and efficiency stats do have moderate correlation to year before.<br />The reason why Vegas make money is not because they have the best prediction ability but they can predict what the public will be on. They are experts on the psychology of gambling. With so many few games in one season. It is impossible to predict accurately what 'will' happen.<br />Vegas margin of error is really not very good. Brian has showed that ~75% is the upper end of a prediction system <br />Vegas falls way short of that.Also many can beat Vegas on a regular basis. Pick the necessary number of winners. But <br />get messed up with the psychology part.<br />mike<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-86820525417790054312012-11-28T12:15:09.087-05:002012-11-28T12:15:09.087-05:00[Sorry, forgot to add quote marks; would have edit...[Sorry, forgot to add quote marks; would have edited if that were an option.]<br /><br />"I'm convinced they [the Ravens] are probably significantly better than their efficiency numbers indicate."<br /><br />I really think they are. As I understand it, this model does not account for punting and punt coverage, something I consider one of the most underrated, overlooked elements (particularly in combination with strong defense and ball control offense) that can make a team very strong. And the Ravens seem for years to have focussed on this aspect of the game more than most teams.<br /><br />On turnovers, it is disappointing that nothing non-random can be teased out of the noise. Even if it can not be discerned in a statistical way, I don't believe they are truly random overall (even if a majority of them are). Some defensive backs have good instincts for where a QB will throw, combined with good hands to catch it when it gets there; others do not (though those others may "stay home" and make more solid tackles and give up fewer big plays). Some linebackers and safeties have a knack for stripping the ball when making a tackle (again with a tradeoff, as this at the possible expense of missing a tackle entirely), some do not.SlackerInchttps://www.blogger.com/profile/08275358994906136088noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-57844535004575799212012-11-28T12:14:31.678-05:002012-11-28T12:14:31.678-05:00Are there any examples of a MAP model for football...Are there any examples of a MAP model for football in action other than this siteAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-87884617649433777542012-11-28T10:39:37.590-05:002012-11-28T10:39:37.590-05:00It's easy to define backwards- vs. forwards-lo...It's easy to define backwards- vs. forwards-looking. Backwards looking systems use MLE (the model which makes the data most likely) and forwards-looking systems use MAP (the model which the data makes most likely). This is a well-understood, well-defined, and binary difference.<br /><br />Of course, this means that all major public sports prediction systems are backwards-looking, but (given that they were all created by sports-fans with minimal statistical backgrounds) didn't we already know that? Sometimes backwards-looking systems produce numbers very similar to forwards-looking ones...Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-53689649518700206072012-11-28T10:09:45.487-05:002012-11-28T10:09:45.487-05:00David, Re: interceptions. I would start here: http...David, Re: interceptions. I would start here: http://www.advancednflstats.com/2010/11/how-random-are-interceptions.html<br /><br />"This is something similar to a study Chase Stuart did last year at Footballguys. He created 25 notional QBs, each with an identical interception rate. He demonstrated that after 500 attempts each, there could be a wide range (from 9 to 22) in total interceptions due to sample error alone."Jameshttps://www.blogger.com/profile/01838293735141324662noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-41080356097305952092012-11-28T09:22:57.088-05:002012-11-28T09:22:57.088-05:00Nate, you're essentially saying that bookies w...Nate, you're essentially saying that bookies would rather have bonds than stocks. Vegas is concerned with profits, not short-term risk. The only game Vegas sets with the goal of splitting the action is the Super Bowl.Chase Stuarthttp://www.footballperspective.comnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-34704487278635677522012-11-28T09:12:46.206-05:002012-11-28T09:12:46.206-05:00I meat a graph of rating, not rank, but I think th...I meat a graph of rating, not rank, but I think the context covered it.<br /><br />>I had read in the past that Vegas preferred handicapping<br />>the game rather than the action, trusting that in the<br />>long run they'd have the same outcome.<br /><br />Let's assume - for the sake of discussion - that the bookie can calculate accurate odds for any handicap, and has good information about what the action would be like on them.<br /><br />If the action is (much) bigger than the bookie's reserves (not sure that's true in Vegas) then the optimum long term scenario for the bookie is to have maximum balanced action. That leaves the bookie with no risk and substantial return.<br /><br />Even if the bookie has infinite money, the most profitable line is going to split the difference between the action and the game. This is especially true because it may be impossible to handicap to exactly 50/50.Natehttp://www.google.comnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-36243279306005599682012-11-28T08:33:38.785-05:002012-11-28T08:33:38.785-05:00Is NE's defense as bad as it seems or is part ...Is NE's defense as bad as it seems or is part of it a product of the offense's efficiency? I.e. offensive efficiency results in more opponent possessions and so on...<br /><br />Also, is it Flacco or Harbaugh who really deserves credit for his ball security? Ken R.https://www.blogger.com/profile/13086158597256834776noreply@blogger.com