tag:blogger.com,1999:blog-38600807.post4841324014664196633..comments2023-11-05T04:16:44.937-05:00Comments on Advanced Football Analytics (formerly Advanced NFL Stats): Response to Luck and Belichick Article CommentsUnknownnoreply@blogger.comBlogger10125tag:blogger.com,1999:blog-38600807.post-11830793429227535572010-01-29T15:52:04.343-05:002010-01-29T15:52:04.343-05:00On the contrary, I think it's far more naive t...On the contrary, I think it's far more naive to believe whatever it was they were doing gave them no advantage.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-79585567534553822032010-01-29T15:29:24.173-05:002010-01-29T15:29:24.173-05:00Stealing signals isn't cheating. Filming them ...Stealing signals isn't cheating. Filming them with a camera from the sideline is cheating. You seriously think the Patriots success can be explained by this fact? You are naive. Stick to number crunching.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-83650220502774000982009-01-06T16:54:00.000-05:002009-01-06T16:54:00.000-05:00David:I will, he's still cheating.David:<BR/><BR/>I will, he's still cheating.Unknownhttps://www.blogger.com/profile/12380982973878242269noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-48108312750211131032008-12-18T13:05:00.000-05:002008-12-18T13:05:00.000-05:00Hm. Belichick's Patriots are second in the league ...Hm. Belichick's Patriots are second in the league this year, after going 2.1 over their expectation last year. Care to revisit this article's allegations in light of two seasons post-scandal whose results show a consistency with the team pre-scandal? I would think this would be worth your time.David Kociembahttps://www.blogger.com/profile/11392285155892307582noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-89281127429746516212008-01-17T00:25:00.000-05:002008-01-17T00:25:00.000-05:00While the difference from expected wins is signifi...While the difference from expected wins is significant, the cheating theory is probably the most fantastical of explanations, and also, a somewhat derivative one.<BR/><BR/>That is, if you think "knowing what the other team will do" is significant, that suggests that the strategy PERIOD is significant to the result -- even if you know from some other legal means (tendency analysis of the other coach, etc.).<BR/><BR/>The point isn't that the "knowing" is significant, but that changes in strategy that don't appear to be extreme can have such a dramatic effect on the results.<BR/><BR/>For example, one of the other qualities that is well known about the Patriots is that they change their game strategy more than any other team from game to game and half to half, or even situation to situation. They do this in "subtle" ways as well, such as going all spread on offense all game long and rarely running it, or playing dime coverage all game, then switching to something totally different for the next opponent. They are not resorting to the kinds of extremes in strategy that many stats try to account for - such as never punting or kneel downs to eat the clock in end game situations. <BR/><BR/>The real question this exposes is how much these more "subtle" variations in game strategy can impact the outcome of the game.<BR/><BR/>In short, using your analogy on luck, it would appear that in Football, you CAN control how you bunch your performances (at least from game to game, bunching all your passes against a weak passing opponent, etc.)<BR/><BR/>It's just this subtle adjustment of strategy that makes it seem like a good poker player is cheating - "bunching" his best hands with his highest bets.Vernhttps://www.blogger.com/profile/17749548749451100874noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-43985931658215841892008-01-12T03:40:00.000-05:002008-01-12T03:40:00.000-05:00Re 1: So doesn't this indicate that their cheatin...Re 1: So doesn't this indicate that their cheating had negligible effect? Therefore the 2.33 is almost entirely (if not completely) due to Belichick (according to your interpretation of the statistic). You could calculate how much is likely due to cheating by treating years 2003-2006 as control group data, 2007 as experiment group data. <BR/><BR/>Re 2: What I find 'lame' is the probability that Gibb's 2.11 was due to rule changes as you had suggested earlier. I wasn't speaking about the changes in their ratings between their stints, as you seemed to think.<BR/><BR/>Re 3: See Re 1 above.<BR/><BR/>Re 4: True, but you do need to quantitatively account for the sample size differences to say how much of a difference there really is between Dungy and Belichick. <BR/><BR/>I really do think that your interpretation of this statistic is at least somewhat flawed, though. Anything that has Dan Reeves, Mike Martz, and Dave Wannstedt (!) above Bill Walsh has to be wrong.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-33099795855617246512008-01-11T20:19:00.000-05:002008-01-11T20:19:00.000-05:001. Yes, NE has exceeded their expected wins again....1. Yes, NE has exceeded their expected wins again.<BR/><BR/>2. Gibbs coached in two different eras. Changes to the passing rules have altered the balance between run and pass. Gibbs had a long discontinuity between his two stints. He was out of coaching completely for years. Belichick had a very short period between his two stints as a head coach, and his took place in roughly the same era. I'm not sure why you find that 'lame.'<BR/><BR/>3. I agree Gibbs' number from the 80s is quite remarkable. Perhaps he was good and lucky. Belichick might be good and lucky and play games with QB helmet frequencies.<BR/><BR/>4. Dungy has had some overperforming years mixed with underperforming years. It's the fact that Belichick has <I>never had an underperforming year</I>, i.e. it's the several consecutive overperformances by significant amounts that puts him in the 3SD range. Dungy's pattern isn't like that at all.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-68343963119609775412008-01-11T19:14:00.000-05:002008-01-11T19:14:00.000-05:00Doesn't the Pats 2007 data showing them yet again ...Doesn't the Pats 2007 data showing them yet again outperforming your models' expectations contradict the cheating hypothesis? The 2.1 is very much in line with data from 2003 on.<BR/><BR/>Also, I find it hard to believe that Gibb's 2.11 with the 80s/90s Skins was due to (a) talent advantage and (b) new rule exploitation. If it's talent, why don't you see similar results for Seifert or Noll, for example? And you could obviously argue that Gibbs really didn't have much of a talent advantage (if any) compared to other NFC teams of the era on a year-in, year-out basis. Also, shouldn't a talent advantage be captured in their Expectancy? Do you mean to suggest a 'clutch talent' advantage? If so, good luck hunting that unicorn.<BR/><BR/>Rule exploitation is a really lame hypothesis - the rules (and their enforcement) are always changing, so you're basically saying that of all the rule changes that have occurred, that's the only one where one coach (Gibbs) effectively exploited it while no one else did. I don't buy it.<BR/><BR/>Hell, if you go that route you could say Belichick exploited the lack of enforcement of illegal contact (until 2004) and thereafter had a significant clutch talent advantage, particularly over divisional rivals. <BR/><BR/>Side note: I'd like to see p-values calculated to account for sample size differences between coaches. It could be that Dungy's 1.73 over a larger sample size would be less statistically likely than Belichick's 2.33 with the Pats.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-38746776013701494052008-01-09T23:49:00.000-05:002008-01-09T23:49:00.000-05:00Mike-Unfortunately, the available data before 2002...Mike-Unfortunately, the available data before 2002 is not as robust as post 2002, thus the discontinuity. For example, the data identifies interceptions and fumbles, instead of simply 'net turnovers.'<BR/><BR/>Regarding your points.<BR/>2. The 3 SD observation regards data from 2002-2006, which is more detailed. The Gibbs 2.11 number, and the Belichick 2.33 numbers are based on the pre-2002 standards. Even though I had more data for 2002-2006, I stuck with the same model to compare apples to apples.<BR/><BR/>3. So no, I would confirm the opposite. The 3 SD claim regards 2002-2006 data, and not the first 2 years of his NE stint. I wish I had some more turnover data.<BR/><BR/>Gibbs was obviously cheating. He has no character. Just kidding--he's actually a neighbor of mine. I think the difference between his two stints here in DC is due to two things. First, he was blessed with a talent advantage in the 80s/early 90s. Second (and this is what pertains to my research), in the 80s he was playing against teams who had not yet adapted to the new passing rules (o-line blocking/ interference/ etc.). So his old-school style of "run out the clock with Riggo as soon as you get a lead in the first half" didn't work against the modern Eagles and Cowboys.<BR/><BR/>But that's a good point. The Gibbs 2.11 number in his first stint is, as you said, a once-in-500 yrs number or so. But there's an important difference between Gibbs and Belichick.<BR/><BR/>Belichick was caught cheating. It's only a question of the extent. Gibbs has never been suspected of anything like that.<BR/><BR/>The Gibbs 2.11 number might be a classic case of a Type II error. We see significance when there really is none. Unless we have a prior reason to suspect foul play, we can't make the same inference about Gibbs as we can about Belichick.<BR/><BR/>Good luck on Saturday.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-26780258562853598552008-01-09T22:59:00.000-05:002008-01-09T22:59:00.000-05:00Few comments. First, this is a terrific site, and...Few comments. First, this is a terrific site, and this issue in particular is really fascinating; thanks for the effort on this.<BR/><BR/>1. I agree that the observation about the Pats' consistently outperforming their win expectancy is interesting and supports the cheating hypothesis. (It also supports various other hypotheses, e.g. Belichick is an exceptionally good coach, etc. For what it's worth, I'm a Pats fan.)<BR/><BR/>2. In the "Rating Gameday Coaches" post, when you break down Joe Gibbs' record into his two stints with the Redskins, he comes out with 2.11 wins added above expectation in his first stint. In view of this, Belichick's 2.33 in the current stint with the Pats looks less out of whack. If, as you say, the 2.33 is a once-in-a-millenium number, it looks like there have been two of those in the last twenty years. (Though they have been in different millenia...) Unless 2.11 only counts as once in 500 years or something. I do wonder if your remark, that Belichick is 3 standard deviations above the mean, is based on the broken out version of Gibbs, or the unified Gibbs. Also, do you have a conjecture about the difference between the two Gibbs stints? Did he cheat in the '80's, then lose his cheating mojo for his current stint? (No offense, Skins fans; if there's evidence in the one case, there's evidence in both cases.)<BR/><BR/>3. In the graph on the "Rating Gameday Coaches" post, 2000 and 2001, Belichick's first two years in NE, are omitted, but I would suppose those years are included in claims like the 3-standard-deviations one; could you confirm that please? (It does look like the graph would support a number less than 2.33, though it's a little hard to tell where exactly a couple of the plotted points lie.)Anonymousnoreply@blogger.com