tag:blogger.com,1999:blog-38600807.post7714765321050073741..comments2023-11-05T04:16:44.937-05:00Comments on Advanced Football Analytics (formerly Advanced NFL Stats): Team Rankings: Week EightUnknownnoreply@blogger.comBlogger30125tag:blogger.com,1999:blog-38600807.post-63860198682114754442011-10-31T22:51:11.701-04:002011-10-31T22:51:11.701-04:00Why is Washington's defense ranked 6th above t...Why is Washington's defense ranked 6th above the 49ers (9th) when all of the 49ers' inputs are superior to Washington's? Is it a strength of schedule thing?Ian Bnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-20544915121937831412011-10-28T22:06:48.724-04:002011-10-28T22:06:48.724-04:00Hmm.. one more thing, the model coefficients you p...Hmm.. one more thing, the model coefficients you posted <a href="http://www.advancednflstats.com/2011/09/weekly-game-probabilities.html?showComment=1317404307706#c7815296964422940127" rel="nofollow">here</a> have adrunsr > 0, is that right, or did you drop a negative sign in your post?Sam's Hideouthttps://www.blogger.com/profile/17861031623526621055noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-15840011762519667512011-10-28T22:00:15.309-04:002011-10-28T22:00:15.309-04:00Brian, to check my understanding, the first ("...Brian, to check my understanding, the first ("input stats are regressed...") affect the input stats used to calculate the GWP for each team, and the second (all the team's final adjusted GWP...") affects the regression over past seasons used to compute the coefficients used to weight the (regressed) input stats in your model to compute GWP.<br /><br />Since I'm currently just using the computed GWPs from the table, I don't really have to worry about that (for now), or is my formula for computing game win probabilities incorrect? <br /><br />[Originally also included a question about what your current coefficient were but now I found them <a href="http://www.advancednflstats.com/2011/09/weekly-game-probabilities.html" rel="nofollow">here</a>.]<br /><br />OK, now I'm computing GWP from the efficiency stats and the values I'm coming up differ quite a bit from the GWPs from the above table. I'm guessing bugs in my code and/or I'm not adjusting (regressing) the efficiency stats (though.. I'm doing this in Octave and the code is so minimal it's hard to see where bugs could be hiding...)Sam's Hideouthttps://www.blogger.com/profile/17861031623526621055noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-47449200416356873022011-10-28T16:49:40.129-04:002011-10-28T16:49:40.129-04:00Hey Brian/Zach, could we please link these article...Hey Brian/Zach, could we please link these articles in the Stats > Team Efficiency section? I reference this series a lot and don't like scrolling through all the posts to find them.Ian Bnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-65276245530837191022011-10-28T16:07:58.500-04:002011-10-28T16:07:58.500-04:00SH-There are a couple additional things I do with ...SH-There are a couple additional things I do with GWP. First, and most importantly, the input stats are regressed according to how steady and reliable each stat tends to be league-wide. Second, all the teams's final adjusted GWP is adjusted to be exactly 0.50. This is because there are year-to-year fluctuations in the average team stats, but we know the average team must be average.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-38681203395008198202011-10-28T03:25:35.677-04:002011-10-28T03:25:35.677-04:00Well that's embarrassing, I discovered two bug...Well that's embarrassing, I discovered two bugs in my data tables that I feed my program, with the result that I now get pretty good agreement on the AFC South division winner probabilities.<br /><br />HOU: .8916<br />IND: .0001<br />JAC: .0213<br />TEN: .0765<br /><br />Unresolved ties: .0105<br />(currently only implement the head-to-head tie-breaker)Sam's Hideouthttps://www.blogger.com/profile/17861031623526621055noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-89417280073297596222011-10-28T03:04:10.606-04:002011-10-28T03:04:10.606-04:00Brian, if I'm understanding your model correct...Brian, if I'm understanding your model correctly, you compute game probability like so:<br /><br />P(Home Team A defeats Visitor Team B) = 1/(1+exp(-gl))<br /><br /> where gl = hf + logit(GWP_A) - logit(GWP_B)<br /> with<br /> logit(x) = log(x/(1-x)) (i.e. usual definition)<br /> hf = home field adjustment (near 0.5)<br /> GWP_A = Team A (adjusted) Generic Win Probability (from above table)<br /> GWP_B = Team B (adjusted) Generic Win Probability (also from above table)<br /><br />Following the above, I get roughly but not exactly the same game probabilities that you publish in the NYT (with hf=0.48).<br /><br />Does that look right? If so, there are some minor errors on the pages where you describe your model (I originally tried to implement exactly what was on your pages and failed, then went back to the definition of logistic regression...)<br /><br />I'd like to get this correct to go into my exact division championship calculator. By focusing on a single division at a time, I can run through all the remaining season permutations (roughly about 2^30 for each division at this point) instead of doing a monte carlo simulation. Currently for the AFC South this enumeration takes about 20 seconds.<br /><br />However, my computation for the AFC South division champion differs substantially from the playoff projection posts for week 7 and 8, so I'm trying to decide if this is due to a bug in my game probability calculation. (My code currently has a 0.67 probability that Houston wins the division which differs substantially from the 0.88 that the week 8 playoff post. Tennessee picks up about half the difference and almost all of the rest goes into tie-breakers my code doesn't yet resolve.)Sam's Hideouthttps://www.blogger.com/profile/17861031623526621055noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-51599835151502965912011-10-26T20:03:40.421-04:002011-10-26T20:03:40.421-04:00Interesting rankings, according to my power rating...Interesting rankings, according to my power rating method which includes ave gain per pass att of both off and def and carries the most weight, rushing yards of both off and def and turnovers, which I know you say are random, but I do count them.<br /><br />1. GB, 2. Houston, 3. 49ers, 4. Dallas, 5. Ravens<br /><br />3 of our top 5 are the same and the Ravens you have 7th, just out of the top 5.<br /><br />Very interesting since I count TO and you don't.<br /><br />The biggest difference is I have 49ers 3rd to your 16th mostly because the 49ers are 5-1 at winning the TO battle and you have NO 2cd where I have them closer to 16th mostly because they are 1-4-2 at winning the TO battle.<br /><br />It will be very interesting to see where these 2 teams go from here, I suspect the 49ers get further and deeper into the postseason than the Saints, but I could always be wrong.Mike Mnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-12335308039600067622011-10-26T19:13:21.705-04:002011-10-26T19:13:21.705-04:00James, thanks for the clarification. I still think...James, thanks for the clarification. I still think both teams will drop, based on my own modeling. But my prediction above wasn't all that strong, since Houston has nowhere to go but down. So I'll make it a little stronger: I predict Houston will drop out of the top 4 and Tenessee will drop even further below #18 before Week 17. Then again, the rest of the Texans schedule isn't exactly a murders' row, so maybe they'll continue to post good stats for this system.probablepicks.comhttp://probablepicks.comnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-54860433337257332452011-10-26T15:48:29.865-04:002011-10-26T15:48:29.865-04:00Brian,
yeah you are correct. They (PFR) made (sti...Brian,<br /><br />yeah you are correct. They (PFR) made (still make?)the same error when calculating the yearly league wide passer rating. I told them, but only got an arrogant answer. Otherwise a great site, they just should be open to critics and infos from readers.<br /><br />Karl, Germany.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-40439835592554303812011-10-26T14:59:35.426-04:002011-10-26T14:59:35.426-04:00Thanks Brian!Thanks Brian!weinsteiniumnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-57583316873585417112011-10-26T13:46:12.345-04:002011-10-26T13:46:12.345-04:00JinxJinxMichael Beuoyhttps://www.blogger.com/profile/03960600491528993233noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-38204383899787932912011-10-26T13:33:20.663-04:002011-10-26T13:33:20.663-04:00No. No bug. Everything looks correct. Let's lo...No. No bug. Everything looks correct. Let's look at NO as an example.<br /><br />I show NO with 2477 gross passing yds, 90 sack yds, 299 attempts, and 13 sacks. PFR, ESPN.com, NFL.com all agree. That comes to 2387 net passing yards on 312 dropbacks.<br /><br />2387/312 = 7.7<br /><br />PFR gets 7.9 because they are not including the number of sacks in the denominator:<br /><br />2387/299 = 7.9<br /><br />If you look at the PFR NO team page, the NY/A is listed correctly at 7.7. It's just their 2011 year page that's got the error.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-741996396373799962011-10-26T12:54:55.747-04:002011-10-26T12:54:55.747-04:00Hmmm. The table above is supposed to be each teams...Hmmm. The table above is supposed to be each teams true net YPA directly from game stats. It's before it gets regressed. There might be a bug somewhere in my code. I'll have to check. I might be leaving out sacks in the denominator or something like that.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-41471320555741776602011-10-26T12:45:05.429-04:002011-10-26T12:45:05.429-04:00Brian - I'm confused by how you are regressing...Brian - I'm confused by how you are regressing the offensive and defensive breakdown. I've focused on OPASS because I can get that from pro-football-reference's net yards per attempts. <br /><br />I can see that top teams have been regressed towards the mean:<br /><br />GB OPASS=8.8 NY/A=9.3<br />NE OPASS=8.5 NY/A=8.7<br />NO OPASS=7.7 NY/A=7.9<br /><br />But for teams with bad OPASS it seems that they have been regressed away from the mean:<br /><br />JAC OPASS=4.3 NY/A=5.0<br />CLE OPASS=4.9 NY/A=5.2<br />STL OPASS=4.9 NY/A=5.4<br /><br />Am I misunderstanding what you mean by "regresssing"?weinsteiniumnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-8494140642415573952011-10-26T11:39:57.796-04:002011-10-26T11:39:57.796-04:00Sorry, I should have written:
Six of the Rams fir...Sorry, I should have written:<br /><br />Six of the Rams first seven opponents are currently ranked in the top nine (and the seventh is ranked #14).<br /><br />(My mistake, as they don't play Pittsburgh until December.)<br /><br />Still a pretty ugly way to have to open a season.Eric Gnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-57784960387180992012011-10-26T11:27:02.990-04:002011-10-26T11:27:02.990-04:00Seven of the Rams first eight opponents this seaso...Seven of the Rams first eight opponents this season are currently ranked in the top nine (and the eighth is ranked #14).<br /><br />Brian, how does this kind of schedule brutality rate historically during the years you've been running your numbers?<br /><br />Regardless, the degenerate gamblers amongst us should be seeing a lot of opportunities to bet on St. Louis during the second half of the season.Eric Gnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-79228387598423671482011-10-26T09:38:44.951-04:002011-10-26T09:38:44.951-04:00Yes. Opp GWP is opponent average GWP. The plain GW...Yes. Opp GWP is opponent average GWP. The plain GWP is really "adjusted GWP"--it already incorporates to-date opponent strength.<br /><br />Also, yes. The input stats (the second table) are regressed slightly more than they have been in previous years at this point in the season. That's partly why the top teams aren't getting about 0.70 GWP.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-25982165571603489452011-10-26T08:53:10.876-04:002011-10-26T08:53:10.876-04:00Tarr, we basically saw that during the Colts-Saint...Tarr, we basically saw that during the Colts-Saints game on Sunday. #4 O vs #32 D, #19 O vs #11 D. Your next closest bet might be Cowboys vs Dolphins (#6 O vs #28 D, #24 O vs #8 D), or the Texans when they play the Bucs or Colts.<br /><br />probablepicks, Tennessee's ranking only marginally effects Houston's ranking. Even if Ten was the worst team in the league, Houston's Opp GWP would drop only 0.02 points. Houston's played really well, the only question is can they keep it up without Mario Williams.Jameshttps://www.blogger.com/profile/01838293735141324662noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-1252764046932147622011-10-26T01:08:59.359-04:002011-10-26T01:08:59.359-04:00I would like to see a team with New England's ...I would like to see a team with New England's offense and Jacksonville's defense play a team with Jacksonville's offense and New England's defense.Tarrhttps://www.blogger.com/profile/14368810359650066790noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-25650289264720919682011-10-25T21:56:51.885-04:002011-10-25T21:56:51.885-04:00Indy, Denver, Miami, OaklandIndy, Denver, Miami, OaklandAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-82676449614611403962011-10-25T20:29:03.770-04:002011-10-25T20:29:03.770-04:00I'm surprised the Packers are still #26. Mayb...I'm surprised the Packers are still #26. Maybe it's just me, but it seems like their defense plays way too conservatively with a big lead.<br /><br />Surprised the 49ers are #16, but I shouldn't be.<br /><br />If Indy gets Andrew Luck, just as Manning is headed out, then I'm going to be very angry :|Jonathannoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-89127229404533965782011-10-25T19:46:36.976-04:002011-10-25T19:46:36.976-04:00Houston is #1 because Tennessee is still ranked to...Houston is #1 because Tennessee is still ranked too high here, in my humble opinion. I predict that you will see both drop in the rankings in the coming weeks.probablepickshttp://probablepicks.com/noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-33975074890989243252011-10-25T19:03:56.348-04:002011-10-25T19:03:56.348-04:00I have to wonder about Atlanta's ranking here....I have to wonder about Atlanta's ranking here. I wonder if they are just an example of a team that doesn't fit the mold for your model and is therefore punished as they certainly seem to routinely over-perform according to this model.<br /><br />I understand that their OPASS rating is low and that is a major component of the model, but they do have a decent PassSR. Granted, I think they are generally overrated by the media but I don't see them as the 7th worst team in the league either.nottomhttps://www.blogger.com/profile/03991575500138449082noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-62551350898045612992011-10-25T18:53:23.714-04:002011-10-25T18:53:23.714-04:00Opp GWP is basically just strength of schedule fro...Opp GWP is basically just strength of schedule from what I understand.nottomnoreply@blogger.com