tag:blogger.com,1999:blog-38600807.post391658492756044975..comments2018-06-02T14:19:34.554-04:00Comments on Advanced Football Analytics (formerly Advanced NFL Stats): Week 13 Game ProbabilitiesUnknownnoreply@blogger.comBlogger10125tag:blogger.com,1999:blog-38600807.post-39629383609014927822008-12-04T10:55:00.000-05:002008-12-04T10:55:00.000-05:00Hi Brian,It's freely adapted from a method I've us...Hi Brian,<BR/><BR/>It's freely adapted from a method I've used to model goals scored in football (soccer) matches. <BR/><BR/>First a few stats ,which I hope I've got correct.A average NFL game has 3.52 turnovers,therefore if two perfectly average teams met at a neutral venue they'd each give away and force 1.76 turnovers.<BR/><BR/>Now looking at NYG and Washington before the week 13 game.NYG were averaging 1.73 takeaways per game and 0.91 giveaways per game.The respective numbers for Washington were 1.18 and 0.91.<BR/><BR/>(Ive used nfl.com for the stats,they had NYG in for 19 takeaways (fumbles lost plus interceptions)and 10 giveaways for the season to week 12 and 13 and 10 respectively for the 'skins).<BR/><BR/><BR/>Now for NYG divide their givaways (0.91) by the league average(1.76);that tells you that NYG commit giveaways at 0.91/1.76 or 0.517 times the league average.Washington's takeaways are 1.18/1.76 or 0.67 times the league average.<BR/><BR/>Next there's a few ways to combine NYG's giveaways with Washington's takeaway,but the simplest is two just multiply the two rates.So NYG's giveaways (and therefore Washington's takeaways) at a neutral venue would be 0.517*0.67 or 0.347 times the league average.<BR/><BR/>Next step is to make the rate venue specific.Home teams average 1.71 turnovers per game compared to 1.81 for the vistors,so visitors allow takeaways at 1.81/1.76 or 1.028 times the league average.You need to multiply this rate by NYG's neutral venue rate to take the game to Washington.<BR/><BR/>So finally we have that NYG playing Washington on the road would allow turnovers at 1.028*0.347 or 0.356 times the league average.We know the league average is 1.76 turnovers,therefore NYG would average 0.628 turnovers if they repeatedly played the 'skins in Washington.<BR/><BR/>Do the same calculation for Washington and you'd expect Washington to average 0.87 turnovers if they repeatedly faced NYG at home.<BR/><BR/>This gives you an average turnover expectancy for each team derived using both teams takeaway and giveaway numbers from the season as a whole.If you pop these two averages in a poisson you can predict the probabilities of Washington having exactly no turnovers,1 turnover,2 turnovers etc.Same for NYG.You can also work out the probability that NYG will win the TO battle by exactly one by summing all the most likely outcomes.(Washington commit 2 turnovers,NYG just one.....Washington committing 8 and NYG 7 has a vanishingly small likelyhood). <BR/><BR/>If you'd done this for the NYG/Wash game there was a 25% chance of Washington committing exactly one more TO than NYG,10% for two more,3% for 3 more,0.6% for 4 more and 0.1% for 5 more.<BR/><BR/>Figures for NYG were 18% for NYG committing exactly one more TO than Washington,5% for 2 more,1% for 3 more,0.2% for 4 more and 0.02% for 5 more.<BR/><BR/>If you use the fairly well accepted figure of a TO costing you 3 points Washington's 10% chance of exactly 2 more turnovers would cost them (0.1*3*2) or 0.6 points on the scoreboard.<BR/><BR/>If you do this calculation for every turnover outcome,then NYG's expected turnovers would cost them 0.98 points and Washington's would cost them 1.7 points.So NYG's expected turnover superiority is worth 0.72 points to them before the game's played.<BR/><BR/>Mmarknoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-78375354367443310822008-12-03T08:13:00.000-05:002008-12-03T08:13:00.000-05:00Not sure where to put this, but from Audibles at t...Not sure where to put this, but from Audibles at the Line on FO I posted:<BR/><BR/>Elias Holman: On their last meaningful possession, the Packers got stopped on the 1-yard line on two successive plays and decided to kick a go-ahead field goal, which seems like the safe strategy, but it seemed to me that they should have gone for the touchdown on fourth down. The probability of converting is very high, and they had been horrible on kick return coverage, so burying the Panthers at the 1 without a kickoff to return was a good idea. Given that the Panthers went long kickoff return, Steve Smith for 50-odd yards, DeAngelo Williams touchdown on the subsequent series (for the second time, I might add) it seems like my intuition was correct, but I don't know if the value of being ahead by any amount late in the game outweighs the field goal-versus-touchdown differential. I would assume this needs a football commentary-style analysis.<BR/><BR/>http://www.advancednflstats.com/ might be able to helpAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-31041503495691596152008-12-02T22:27:00.000-05:002008-12-02T22:27:00.000-05:00Mark-That's a really interesting idea. I've though...Mark-That's a really interesting idea. I've thought of trying to model scoring with Poisson distributions, but never turnovers. I know that sports like hockey and soccer lend themselves to Poisson-distributions.<BR/><BR/>Just curious, how did it workout that the Giants would be favored to win the t/o battle? It seems the Redskins have the leagues best t/o rates.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-44434979983736815882008-12-01T13:42:00.000-05:002008-12-01T13:42:00.000-05:00Brian,regarding turnovers being responsible for yo...Brian,regarding turnovers being responsible for your model favouring Washington over NYG on Sunday,have you tried modelling turnovers using a Poisson.<BR/><BR/>You can follow a similar path used to predict goals scored in English soccer.Basically you use season long turnover rates for both offense and defense for both teams to calculate an average number of turnovers expected in your match up.Then you stick those averages into a Poisson to get probabilities for each team committing zero,exactly one,2,3 etc turnovers.That allows you to try to predict the probability of say each team committing the same number of TO's,one team committing one more TO than the other etc,etc.<BR/><BR/>If you know the odds of one team committing say two more turnovers than the other and if you know,on average how many points a turnover cost a team (3 would be a good enough guess imo),you can then see how many points each team's predicted array of turnovers will possibly cost them in an up coming game.<BR/><BR/>I crunched the numbers for the Washington/NYG game and his is what I got.<BR/><BR/>There was a 36% chance that each team would commit the same number of TO's.A 42% chance that NYG would commit less TO's than the 'skins and a 23% chance that the Redskins would win the TO contest. <BR/><BR/>Those numbers broke down as a 26% chance that NYG would win the turnover battle by exactly one TO,an 11% chance they'd win it by exactly two,3% for three .0.7% for four and 0.1% for five.The respective percentages for Washington were,17% for one,4.4% for two,0.9% for three,0.1% for four and 0.01% for five.<BR/><BR/>If you total up those percentages and assume a TO on average cost a team three points (either in points they don't score or points the opposition does),then Washington's TO's would in a matchup at home to NYG cost them 1.9 points on the scoreboard.NYG's by contrast would only cost them 0.9 of a point.So on average each team's TO record would likely leave NYG one point in credit on average over a long series of matchups.<BR/><BR/>Washington do better if you just look at interceptions as opposed to all TO's,then they shade the Giants by on average half a point a game.<BR/><BR/>(I've just thrown together the spreadsheet,but the numbers look fairly ok).marknoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-76961974379106489532008-11-26T15:56:00.000-05:002008-11-26T15:56:00.000-05:00results hereresults <A HREF="http://www.advancednflstats.com/2008/11/week-11-efficiency-rankings.html#comments" REL="nofollow">here</A>Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-11510514482804268302008-11-26T15:07:00.000-05:002008-11-26T15:07:00.000-05:00Trying my hardest not to track it each week. I ten...Trying my hardest not to track it each week. I tend to obsess. At the end of the year I'll evaluate. I know the model is not doing as well as the past 2 years. Someone else out there is probably tracking it.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-36151049312812902322008-11-26T14:54:00.000-05:002008-11-26T14:54:00.000-05:00last week 10-6. week before...10-5-1...thats not e...last week 10-6. week before...10-5-1...thats not exactly 10-4....what is your overall record Brian?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-32268016473473716882008-11-26T09:37:00.000-05:002008-11-26T09:37:00.000-05:00Anon-I feel the same way. The Skins have incredibl...Anon-I feel the same way. The Skins have incredibly great turnover stats, which keep them ranked very high. But they're too careful with the ball now. Their passing game has been too cautious in recent weeks, and they've suffered for it.<BR/><BR/>Anytime I see a whacky prediction like this one, I go to the most recent <A HREF="http://www.advancednflstats.com/2008/11/week-12-efficiency-rankings.html" REL="nofollow">efficiency ranking post</A> and look up each team's stats (the second table) and strength of schedule.Brian Burkehttps://www.blogger.com/profile/12371470711365236987noreply@blogger.comtag:blogger.com,1999:blog-38600807.post-30201960319766640542008-11-26T09:25:00.000-05:002008-11-26T09:25:00.000-05:00Anonymous - feel free to browse back to the model ...Anonymous - feel free to browse back to the model pick for WAS/DAL gameGeoffnoreply@blogger.comtag:blogger.com,1999:blog-38600807.post-44256148198359235942008-11-26T09:22:00.000-05:002008-11-26T09:22:00.000-05:00im a big redskins fan, and i hate to say that the ...im a big redskins fan, and i hate to say that the 63%/37% makes the model look silly, 50-50 at best for the skinsAnonymousnoreply@blogger.com