The Frankish Reich Posted September 8, 2015 Posted September 8, 2015 The Bills finished 2nd in the division last year. They added: McCoy 2 Guards Harvin Clay Taylor Rex & Roman They lost: Freddy Alonso...but not really from 2014 as he wasn't here then either. Greggo Marrone & Hackett Orton I've probably missed some stuff here and there but at a basic level how does a 2nd place team with those additions/subtractions became a 4th place team? They lost DaNorris Searcy on D too. Not that that's a huge thing, but it matters, particularly if McKelvin is out -- Graham can't as easily move to CB if Darby struggles.
r00tabaga Posted September 8, 2015 Posted September 8, 2015 Reich you're reaching for straws brother...
Greg F Posted September 8, 2015 Posted September 8, 2015 I used to read a lot of Grantland so I'm familiar with Barnwell. He will always be the guy who tried to explain, using analytics ... From the article he links to a "statistical primer" also written by him. ...the Pythagorean theorem (or “Pythagorean expectation”) is a formula that translates a team’s points scored and allowed into an “expected” winning percentage. That formula isn’t exactly for the faint of heart: Points For2.37 / (Points For2.37 + Points Against2.37) The fact that he considers the "formula isn’t exactly for the faint of heart" tells me all I need to know about his math/statistical abilities. Can you say high school math? Just for fun I ran the formula from 2013 to predict the records for 2014 and then compared the results with the actual records. For 14 teams (44% of all the teams) the formula was off by 2 or more wins. Of those 14 teams, 8 (25% of all the teams) were off by 3 or more wins. Additionally, for 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. Seems a bit dubious to me but your mileage may vary.
Lurker Posted September 8, 2015 Posted September 8, 2015 From the article he links to a "statistical primer" also written by him. The fact that he considers the "formula isn’t exactly for the faint of heart" tells me all I need to know about his math/statistical abilities. Can you say high school math? Just for fun I ran the formula from 2013 to predict the records for 2014 and then compared the results with the actual records. For 14 teams (44% of all the teams) the formula was off by 2 or more wins. Of those 14 teams, 8 (25% of all the teams) were off by 3 or more wins. Additionally, for 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. Seems a bit dubious to me but your mileage may vary. Great work. Essentially a meaningless formula...
FireChan Posted September 8, 2015 Posted September 8, 2015 From the article he links to a "statistical primer" also written by him. The fact that he considers the "formula isn’t exactly for the faint of heart" tells me all I need to know about his math/statistical abilities. Can you say high school math? Just for fun I ran the formula from 2013 to predict the records for 2014 and then compared the results with the actual records. For 14 teams (44% of all the teams) the formula was off by 2 or more wins. Of those 14 teams, 8 (25% of all the teams) were off by 3 or more wins. Additionally, for 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. Seems a bit dubious to me but your mileage may vary. So for 56% of teams it was correct within 1 win? Better than a coinflip.
DC Tom Posted September 8, 2015 Posted September 8, 2015 From the article he links to a "statistical primer" also written by him. The fact that he considers the "formula isn’t exactly for the faint of heart" tells me all I need to know about his math/statistical abilities. Can you say high school math? Just for fun I ran the formula from 2013 to predict the records for 2014 and then compared the results with the actual records. For 14 teams (44% of all the teams) the formula was off by 2 or more wins. Of those 14 teams, 8 (25% of all the teams) were off by 3 or more wins. Additionally, for 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. Seems a bit dubious to me but your mileage may vary. Wow, that is truly !@#$ing stupid. His example, Kansas City, won four more games than his formula predicted...so Kansas City outperformed themselves? Or maybe the equation's nonsense. It may not be for the faint of heart, but it's certainly for the faint of head.
4merper4mer Posted September 8, 2015 Posted September 8, 2015 (edited) They lost DaNorris Searcy on D too. Not that that's a huge thing, but it matters, particularly if McKelvin is out -- Graham can't as easily move to CB if Darby struggles. Now I can see moving from 2nd to 4th. If we lost Robey we'd probably be the first team to ever finish 9th in a 4 team division. Edited September 8, 2015 by 4merper4mer
The Frankish Reich Posted September 8, 2015 Posted September 8, 2015 From the article he links to a "statistical primer" also written by him. The fact that he considers the "formula isn’t exactly for the faint of heart" tells me all I need to know about his math/statistical abilities. Can you say high school math? Just for fun I ran the formula from 2013 to predict the records for 2014 and then compared the results with the actual records. For 14 teams (44% of all the teams) the formula was off by 2 or more wins. Of those 14 teams, 8 (25% of all the teams) were off by 3 or more wins. Additionally, for 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. Seems a bit dubious to me but your mileage may vary. True, not rocket science. But you're all kind of missing the point, which is this: pythagorean W-L records in Year 1 have a stronger correlation with actual W-L records in Year 2 (as opposed to actual W-L in Year 1 to actual W-L in Year 2). This works better in baseball than in football -- think about it, one 48-7 blowout in football has a bigger impact than one 13-1 blowout in baseball -- but it's still worth looking at. And Barnwell did: http://grantland.com/the-triangle/the-nfl-statistical-crystal-ball-what-2014s-numbers-can-tell-us-about-2015/
4merper4mer Posted September 8, 2015 Posted September 8, 2015 True, not rocket science. But you're all kind of missing the point, which is this: pythagorean W-L records in Year 1 have a stronger correlation with actual W-L records in Year 2 (as opposed to actual W-L in Year 1 to actual W-L in Year 2). This works better in baseball than in football -- think about it, one 48-7 blowout in football has a bigger impact than one 13-1 blowout in baseball -- but it's still worth looking at. And Barnwell did: http://grantland.com/the-triangle/the-nfl-statistical-crystal-ball-what-2014s-numbers-can-tell-us-about-2015/ Are they really not taking into account that a team has players and coaches. These players and coaches change from year to year. So in theory if two franchises swapped entire organizations after going 16-0 and 0-16 respectively, the predictions would stick with the laundry, not the actual team. Asinine.
The Frankish Reich Posted September 8, 2015 Posted September 8, 2015 Are they really not taking into account that a team has players and coaches. These players and coaches change from year to year. So in theory if two franchises swapped entire organizations after going 16-0 and 0-16 respectively, the predictions would stick with the laundry, not the actual team. Asinine. Nobody ever said that. Of course we know that player and coaching personnel change. Prior season record is still relevant. It's not the whole story, but it's relevant. And pythagorean record is in many ways more relevant. Why have a discussion at all if you dismiss anything but seat of the pants feelings about the Bills?
YattaOkasan Posted September 8, 2015 Posted September 8, 2015 (edited) Nobody ever said that. Of course we know that player and coaching personnel change. Prior season record is still relevant. It's not the whole story, but it's relevant. And pythagorean record is in many ways more relevant. Why have a discussion at all if you dismiss anything but seat of the pants feelings about the Bills? What did his formula say we would do this year? I think I missed it unless its the 9.6 number at the top. Edit: I did minimal digging and found a calculator http://www.had2know.com/sports/pythagorean-expectation-win-percentage-baseball.html Using 16 games, 2014 points scored (343), and 2014 points allowed (289) from http://www.pro-football-reference.com/teams/buf/2014.htm I got 9 wins. The part I don't get is if my dataset is last year how do you not project 9 wins this year. I can understand some asymmetry for your predicted result but I think +1 and -5 requires a bit more explaining than he does. Models are great guides and I would use this one to say: The bills were good last year. I think they got better this year (I know I'm a fan but seriously we did upgrade almost everywhere). I will not have too much asymmetry in my results. Let's say +2 and -3 Best case scenario everything clicks and they are 11-5. Maybe they do some damage in the postseason Worst case scenario: Offense doesn't click and Defense regresses, 6-10. I wouldn't call that basement dwellers. I also think that's a more reasonable approach with a bit more grounding in the model. Let me know if I'm misunderstanding this. Edited September 8, 2015 by YattaOkasan
The Frankish Reich Posted September 8, 2015 Posted September 8, 2015 What did his formula say we would do this year? I think I missed it unless its the 9.6 number at the top. I don't think it's correct to say it's a prediction. What he did say is by points scored vs. points allowed the Bills "should have" had 9.6 wins. So they underperformed their pythagorean record by a bit. Which is a good sign for this year, since those things have a way of evening out. It wasn't a big gap like the Dolphins (2.5 wins under their pythagorean record) - that's one big reason why lots of "experts" have the Dolphins finishing ahead of the Bills.
PromoTheRobot Posted September 8, 2015 Posted September 8, 2015 (edited) I don't think it's correct to say it's a prediction. What he did say is by points scored vs. points allowed the Bills "should have" had 9.6 wins. So they underperformed their pythagorean record by a bit. Which is a good sign for this year, since those things have a way of evening out. It wasn't a big gap like the Dolphins (2.5 wins under their pythagorean record) - that's one big reason why lots of "experts" have the Dolphins finishing ahead of the Bills. Damn, which game did the .6 win slip away from us? Edited September 8, 2015 by PromoTheRobot
Greg F Posted September 8, 2015 Posted September 8, 2015 So for 56% of teams it was correct within 1 win? Better than a coinflip. Ummm ... not really. The probability of flipping a coin 32 times and getting 16 heads or tails has a probability distribution of 0.14. IOW, you only have a 14% chance that you will get 16 heads and 16 tails. The probability that you will get 15, 16, or 17 heads or tails is about 0.40 or 40% of the time.
Greg F Posted September 8, 2015 Posted September 8, 2015 True, not rocket science. But you're all kind of missing the point, which is this: pythagorean W-L records in Year 1 have a stronger correlation with actual W-L records in Year 2 (as opposed to actual W-L in Year 1 to actual W-L in Year 2). Perhaps you missed the point. I took the point differential from 2013 as dictated by the pythagorean equation and calculated the projected W/L record for 2014. The results were rather shabby. Doing the same using the W/L record from 2013 instead of the pythagorean equation the results were marginally better (and probably not statistically significant). For 13 teams (40.6% of all the teams) the formula was off by 2 or more wins which is one better than the pythagorean equation. Of those 13 teams, 9 (28% of all the teams) were off by 3 or more wins. For 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. In reality, using the records from 2013 to project the team records for 2014 produced a stronger correlation than using the dubious pythagorean equation.
The Frankish Reich Posted September 8, 2015 Posted September 8, 2015 Perhaps you missed the point. I took the point differential from 2013 as dictated by the pythagorean equation and calculated the projected W/L record for 2014. The results were rather shabby. Doing the same using the W/L record from 2013 instead of the pythagorean equation the results were marginally better (and probably not statistically significant). For 13 teams (40.6% of all the teams) the formula was off by 2 or more wins which is one better than the pythagorean equation. Of those 13 teams, 9 (28% of all the teams) were off by 3 or more wins. For 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. In reality, using the records from 2013 to project the team records for 2014 produced a stronger correlation than using the dubious pythagorean equation. Well, I stand corrected. But one season results are not sufficient to draw any conclusions. I haven't bothered to look at correlations between pythag and actual records with subsequent season performance -- as you suggest, you'd need to look at many seasons to draw a statistically significant result. Barnwell's article makes a more general point: when you see a big difference in actual record vs. pythag record, all other things being equal (and of course, they're not), expect the next year's record to fall back toward the current year's pythagorean record. It's a simple point, but one worth knowing, and interestingly, not one that was known before the analytical revolution that started with Bill James in baseball and moved on to other sports.
BillsFan-4-Ever Posted September 8, 2015 Posted September 8, 2015 Damn, which game did the .6 win slip away from us? Oakland
YattaOkasan Posted September 8, 2015 Posted September 8, 2015 Well, I stand corrected. But one season results are not sufficient to draw any conclusions. I haven't bothered to look at correlations between pythag and actual records with subsequent season performance -- as you suggest, you'd need to look at many seasons to draw a statistically significant result. Barnwell's article makes a more general point: when you see a big difference in actual record vs. pythag record, all other things being equal (and of course, they're not), expect the next year's record to fall back toward the current year's pythagorean record. It's a simple point, but one worth knowing, and interestingly, not one that was known before the analytical revolution that started with Bill James in baseball and moved on to other sports. However, we don't see a big difference in actual record vs pythag record (0.6) in the same year. That tells me we didn't get overly lucky or overly unlucky. So why does he think we should regress?
MDH Posted September 8, 2015 Posted September 8, 2015 From the article he links to a "statistical primer" also written by him. The fact that he considers the "formula isn’t exactly for the faint of heart" tells me all I need to know about his math/statistical abilities. Can you say high school math? Just for fun I ran the formula from 2013 to predict the records for 2014 and then compared the results with the actual records. For 14 teams (44% of all the teams) the formula was off by 2 or more wins. Of those 14 teams, 8 (25% of all the teams) were off by 3 or more wins. Additionally, for 4 teams (12.5% of all the teams) the prediction was off by more than 4 wins. Seems a bit dubious to me but your mileage may vary. I think I could get that close just guessing.
MDH Posted September 8, 2015 Posted September 8, 2015 Barnwell's article makes a more general point: when you see a big difference in actual record vs. pythag record, all other things being equal (and of course, they're not), expect the next year's record to fall back toward the current year's pythagorean record. It's a simple point, but one worth knowing, and interestingly, not one that was known before the analytical revolution that started with Bill James in baseball and moved on to other sports. The problem is turnovers. Turnovers can lead to a lot of extra scores for, or against, you and nobody can reliably predict them. Teams that lead the league in takeaways one year fall to the middle of the pack the next. A stable QB helps mitigate them on offense but overall they still fluctuate (unless you are the Patriots...) The other problem is sample size. In MLB the numbers will more closely reflect actual team's skill levels whereas in the NFL 16 games isn't much to go on. It's why you see sizable fluctuations year to year (though, again, solid QB play helps mitigate this.) So as much as someone might like to predict that numbers will regress towards the mean - and over a long period of time they would - it's tough to make those types of predictions with such a short season. Add the wild variable of turnovers and you have a sport that is very tough to predict with statistics.
Recommended Posts