tag:blogger.com,1999:blog-38600807.post1647009181919009721..comments2023-11-05T04:16:44.937-05:00Comments on Advanced Football Analytics (formerly Advanced NFL Stats): Belichick 4th Down Follow-UpUnknownnoreply@blogger.comBlogger118125tag:blogger.com,1999:blog-38600807.post-11854284490004653672009-11-25T14:20:35.052-05:002009-11-25T14:20:35.052-05:00Hey another long pass on short yardage during mon ...Hey another long pass on short yardage during mon nite football. <br /><br />I think it was 1.09 left in 3rd quarter and they threw a legitimate long pass on 3rd/one.<br /><br />Dont tell me that no one covers long passes on short yardage plays. It happens every week in the NFL.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-28949391828350252132009-11-24T21:44:09.977-05:002009-11-24T21:44:09.977-05:00Brian,
thanks. Per my explanation above, ideally ...Brian,<br /><br />thanks. Per my explanation above, ideally we should strip out "failed" attempts due to offensive penalty and apply that percent of total attempts to the probability of Colts scoring after Pats punt from 5 or 10 yards back (weighted average or just 7.5 yards back if probably close enough), rather than lumping in that category of results with actual failed attempts and applying it all to probability of Colts TD from Pats 28. <br /><br />Depending on the data and on one's assumptions, this more appropriate approach could either make a significant difference or not.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-22061127288682316782009-11-24T20:18:38.431-05:002009-11-24T20:18:38.431-05:00Yes, offensive penalties are included too as long ...Yes, offensive penalties are included too as long as they occur during the play. Pre-snap penalties are not included, and any other penalty between plays, because there is still the option to punt following the penalty.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-59208444040436111572009-11-24T19:27:01.133-05:002009-11-24T19:27:01.133-05:00Brian,
It sounds like you are saying that conversi...Brian,<br />It sounds like you are saying that conversion via Defensive penalties are included in your data. Are Offensive penalties included, too (i.e., offensive penalty during the play such as holding)?Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-25825485864971132192009-11-23T20:59:59.179-05:002009-11-23T20:59:59.179-05:00To elaborate on my comment above and lay out the a...To elaborate on my comment above and lay out the algebra and an illustration:<br /><br />If "failing" with (i.e., due to) offensive penalty were not a possibility, for Pats to LOSE after going for it, two things would have to happen, (1) Pats fail to convert, and (2) Colts TD starting from Pats 28. So, if P (fail) = 0.42 and P(Colts TD from Pats 28) = 0.53, than P(Pats losing if go for it) = (0.42)(0.53) = 0.223.<br /><br />But since "failing" with offensive penalty IS a possibility, and would result in (at worst for Pats), a punt from back 5 or 10 yards (let's just say from back 7.5 yards assuming even number of 5 yard and 10 yard offensive penalties), we have to separate out that possibility. <br /><br />Suppose 1 in 7 failures to convert are with offensive penalty. Then P(failing to convert without offensive penalty) = 0.36 and P("failing" to convert with penalty) is 0.06.<br /><br />And let's suppose that P(Colts scoring from own 34.5) = 0.32<br /><br />Then P(Pats losing if go for it) = (0.36)(0.53)+(0.06)(0.32) = 0.21.<br /><br />So properly separating out actual failure to convert from "failure" due to offensive penalty gives us probability of Pats losing if go for it of only 21%, rather than the erroneously higher 22.3% we get if we inappropriately treat "failure" due to offensive penalty as if it has the same consequences as actual failure (consequences would be quite different: Colts ball on own 34.5 vs. Colts ball on Pats 28).<br /><br />So a proper treatment of actual fail vs. "fail" due to penalty results in going for it being more attractive for Pats (lower chance of losing) than if we inappropriately lump the two scenarios together.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-79689718575991107682009-11-23T18:11:09.479-05:002009-11-23T18:11:09.479-05:00Additional notes re: penalties:
- Converting with...Additional notes re: penalties:<br /><br />- Converting with Defensive penalty should be considered equal to converting without Defensive penalty, because result is the same.<br /><br />- Optimal equation for WP in Go For It scenario would consider probability of converting, probability of failing without penalty applied to probability of Colts scoring from Pats 28, and probability of "failing" with Offensive penalty applied to probability of Colts scoring from own 32 or 37 (corresponding to penalty of 5 or 10 yards). And yes, after a "fail" with Offensive penalty the Pats could go for it again from 5 or 10 yards back instead of punting, but assuming rationality they would only do so if doing so provided a higher probability of winning vs. punting, which is why I said in previous comment that "at worst" (from Pats perspective) a "fail" with Offensive penalty means Colts ball on own 32 or 37.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-74677740693207206402009-11-23T17:14:20.477-05:002009-11-23T17:14:20.477-05:00If I set up this Google doc correctly and pasted t...If I set up this Google doc correctly and pasted the link correctly, at this link http://spreadsheets.google.com/ccc?key=0Am3TwX_gP-RCdHk1dDhFNmlNQ3FjQ1EyaWxxY29jMHc&hl=en is a spreadsheet tool I've created that anyone can use to plug in your own assumptions (and tinker with them) to assess Punt vs. Go For It.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-27160498432597360952009-11-23T16:00:35.281-05:002009-11-23T16:00:35.281-05:00Typo: I meant to say punting after Offensive penal...Typo: I meant to say punting after Offensive penalty would mean Colts ball on own 32 or 37.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-12990410044736569262009-11-23T15:55:48.390-05:002009-11-23T15:55:48.390-05:00Brian,
Re: penalties, we have 4 categories:
1. Con...Brian,<br />Re: penalties, we have 4 categories:<br />1. Convert without pentalty on Defense.<br />2. Convert with penalty on Defense.<br />3. Fail without penalty on Offense.<br />4. "Fail" with penalty on Offense.<br /><br />Including penalties on Offense in the final figure, thus including it as a factor in the weighted average of conversion percentage (probability) artificially decreases the attractiveness of going for it (not necessarily lower than punting; I just mean directionally). "Failing" due to Offensive penalty means (at worst) punting from 5 or 10 yards back, meaning Colts ball on 32 or 27, which would mean much lower probability of Colts TD than Colts starting on Pats 28 after an actual failed attempt. In other words, each incident (data point) of "failing" due to offensive penalty shouldn't be factored in with nearly as much weight as each incident of actually failing (without penalty) and turning ball over on the spot. In effect, counting the "failing" due to Offensive penalty equally erroneously implies that there is no difference in winning probability between Colts starting on Pats 28 and Colts starting on either their own 32 or 37.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-3100619593044910642009-11-23T15:10:38.034-05:002009-11-23T15:10:38.034-05:00The data I have comes with a code for any play tha...The data I have comes with a code for any play that makes a first down. It says P if it was a passing play, R if it was a run, and X if was a penalty. So yes, 1st downs by penalty are considered in the data, but only if the penalty happened on the play. Penalties between plays do not factor in.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-38730021117801752422009-11-23T15:03:06.235-05:002009-11-23T15:03:06.235-05:00Brian - You mention stats that most people see on ...Brian - You mention stats that most people see on nfl.com and espn.com may not include 4th downs converted via penalty. On the other hand, do the stats in which you rely on, include offensive penalties on 4th down conversion attempts as failed attempts? Should they be included and if so, is it possible to get an adjusted sample?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-34722899744761639162009-11-22T23:49:32.451-05:002009-11-22T23:49:32.451-05:00Edward -- thanks.Edward -- thanks.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-50073537514141331732009-11-22T22:44:55.834-05:002009-11-22T22:44:55.834-05:00"Also, what about factoring in cost of losing..."Also, what about factoring in cost of losing vs the benefit of winning? Given the playoff implications, I think you assign a very high cost to losing here. In other words, a win gives the Patriots marginally improved playoff prospects from an already good position. A loss causes a significant hit to their playoff prospects."<br /><br />If you're concerned that the cost of losing is greater than the benefit of winning, the appropriate decision is to not play the game in the first place. Once you show up and the game starts, you're either going to win or lose. Every decision that increases your chance of winning by 1% also decreases your chance of losing by 1%, and vice versa.<br /><br />Obviously, there's plenty of room to argue about how to give yourself the best chance of winning, but the importance of the game is entirely irrelevant.<br /><br />(The importance of the game does have a bearing on what the media backlash will be like after a loss, so an argument can be made that the "conventional wisdom" is the better choice in terms of protecting the coach's reputation. That's what's so infuriating about all this.)James Sinclairhttps://www.blogger.com/profile/10213045233649924060noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-82056067364998446652009-11-22T20:16:16.748-05:002009-11-22T20:16:16.748-05:00Gordon:
http://www.nfl.com/teams/statistics?team=...Gordon:<br /><br />http://www.nfl.com/teams/statistics?team=ARI<br /><br />Repeat for all 32 teams and add up the results (which by now have been updated to include today's games).Edward Leehttps://www.blogger.com/profile/09858930644002883626noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-63715320626636308042009-11-22T18:26:29.198-05:002009-11-22T18:26:29.198-05:00Brian -- Can you please provide a link to your sou...Brian -- Can you please provide a link to your source for that 4th and 3 or less data for the NFL in 2009?<br /><br />Edward -- Can you do the same for your source for ALL 4th attempts in NFL in 2009?<br /><br />thanksGordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-32937377989732575342009-11-22T17:16:40.539-05:002009-11-22T17:16:40.539-05:00Edward and Brian,
Thanks for the correction re: 2...Edward and Brian,<br /><br />Thanks for the correction re: 2009 4th and 3 or less (assuming you're correct). As I said, the figure I presented was from a friend. I didn't verify it, so if it's incorrect, my apologies.Gordonnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-13721548899347356782009-11-22T15:27:28.464-05:002009-11-22T15:27:28.464-05:00Dr. Duru,
First, everyone knows that the statisti...Dr. Duru,<br /><br />First, everyone knows that the statistics discussed here are estimates. We'll never know the "true" probability of winning via going for it on fourth down, but neither will we ever know the "true" probability of winning via punt. There is no reason that punting should be considered the "safe" option. That being said, the 9% edge that going for it is pretty big, as you can check if you try to manually adjust the probabilities that go into the computation.<br /><br />Second, the effect of a win or loss doesn't change whether Belichick goes for it or punts (unless you put a lot of stock in all the pop-psychological arguments about undermining his defense, etc).Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-8682708389934500972009-11-22T12:58:42.177-05:002009-11-22T12:58:42.177-05:00Sorry! I just realized that the cost of losing vs ...Sorry! I just realized that the cost of losing vs winning is probably immaterial in your analysis! You just want to increase the chance of winning, period.Dr. Duruhttp://www.drduru.com/onetwentytwonoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-62904316032033361322009-11-22T12:56:53.975-05:002009-11-22T12:56:53.975-05:00Your two posts did not address whether the nine pe...Your two posts did not address whether the nine percentage points is significant enough to warrant going for it. Also, what about factoring in cost of losing vs the benefit of winning? Given the playoff implications, I think you assign a very high cost to losing here. In other words, a win gives the Patriots marginally improved playoff prospects from an already good position. A loss causes a significant hit to their playoff prospects. Of course, this is subject to debate (more statisitcal analysis).Dr. Duruhttp://www.drduru.com/onetwentytwonoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-79328369025650411272009-11-22T09:34:37.797-05:002009-11-22T09:34:37.797-05:00One other thing. The data most people see, like on...One other thing. The data most people see, like on nfl.com or espn.com, may not include 4th downs converted via penalty.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-59173752530243589982009-11-22T09:32:17.619-05:002009-11-22T09:32:17.619-05:00Yeah, that 2009 4th down data someone posted is wa...Yeah, that 2009 4th down data someone posted is way off. If we just take 2009 (which I still think is a bad idea) for all 4th and 3 or less, the conversion rates are 107 for 183 (58.5%).<br /><br />That includes attempts inside the red zone and on the goal line, which would be more difficult than the 4th down we're discussing. There have only been 27 4th and 2s outside the red zone this year.<br /><br />Sorry to sound all 'logical.'Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-73144790366496852582009-11-22T01:54:26.594-05:002009-11-22T01:54:26.594-05:00Gordon:
I just checked NFL.com's stats for ea...Gordon:<br /><br />I just checked NFL.com's stats for each team, and assuming I'm not an idiot, NFL teams are 146-for-304 (48%) on *all* fourth downs in 2009 (up to and including Thursday's MIA-CAR game).Edward Leehttps://www.blogger.com/profile/09858930644002883626noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-34614296602666512042009-11-22T01:19:48.414-05:002009-11-22T01:19:48.414-05:00Gordon:
Your estimates would be reasonable.
I...Gordon:<br /><br />Your estimates would be reasonable.<br /><br />I'm still puzzled that the 4th-and-3 or less rate from the first half of 2009 would be so low, though. Brian has 9 seasons of data vs. 1/2 a season (18 times as many games), so one figures that he'd have around 18/3 = 6 times as many plays in his 4th-and-2 figure. Maybe only 4x as many if we think that 4th-and-2s make up less than 1/3 of the 4th-and-3-or-less set.<br /><br />It could be just a statistical anomaly, it could be that one or the other data set is biased, or it could be that the defensive rule changes really make a difference.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-43997704630457313382009-11-22T00:44:49.706-05:002009-11-22T00:44:49.706-05:00Good question. The recency of the data is far less...Good question. The recency of the data is far less important than the size of the data set. QUOTE: "If we constrain ourselves to looking at just 1 (or less) season of data, we're going to have a very large sample error. There just aren't that many 4th and 2s where teams go for it...."<br /><br />I love it how you purport to be all logical with your analysis of this and then someone challenges you with 300 data pts and you claim it's a large sample error. <br /><br />Have you heard of standard deviation? This is probability class 101. 300 data pts is certainly plenty of data pts to form the type of statistical estimates that you based your entire analysis on..<br /><br />QUOTE: "To have a reliable estimate, we need to look back further and gather more instances of that situation...."<br /><br />Sez who? You? Why because the recent data doesnt support your conclusion you need to go back further. <br /><br />That's real logical.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-61353344477771139012009-11-22T00:41:44.260-05:002009-11-22T00:41:44.260-05:00testtestAnonymousnoreply@blogger.com