|
Post by Chris Clement on Dec 14, 2019 17:20:29 GMT -6
You have to model it, but you can use a trivial model. Just using distance is going to get you a 90/10 solution (90% perfect, 10% effort). You can get much fancier. You can then adjust for selection/desperation bias at longer distances, but it’s tricky. The crux is that longer FGs in ordinary situations are only attempted by better kickers, inflating the true average. But late in the halves they’re attempted by teams with no other options, depressing the average. There’s an interplay here where selection bias dominates, then they sort of cancel out, then desperation bias takes over. If you want to get real cute you can incorporate lots of other factors - distance, wind, altitude... but then you need a more sophisticated model. Some are very “interpretable,” which is a fancy way of saying that I could explain how it works such that you’d understand and be able to recreate a small version on pen and paper, and some are less straightforward. The fanciest ones we call “black boxes,” because stuff goes in and stuff goes out and you have no idea what’s happening inside. But my research showed that you can use a simple kNN model and get excellent results. That stands for k-Nearest Neighbours, so you’re just averaging the results of the k most similar kicks in your dataset. Adding features also creates problems. Imagine an NFL model that considers altitude. It seems like a good idea, but only one stadium is at high altitude and second place is in the same division. So I’m trying to factor in altitude you end up massively over representing the Broncos kicker. If you want to consider weather you need to think about what constitutes “weather,” and find a way to get good weather information, because the NFL data only gives approximate weather at kickoff. Are you going to split wind into headwind and crosswind? You’ll need to know which way the field is oriented and what direction teams are kicking. Every decision has consequences and you need to know what the question is before you try to answer it. Since we’re looking at a 12 year sample of a team based at low altitude, I’d just use straight distance and assume the rest will average out. So, for someone who is less studied in statistics than yourself, we are tossing outliers and keeping the most common kicks, correct? Also, to back up a bit, why couldn't we just take the total FG attempts of each team over that 12 year period and compare them to Alabama's total FG attempts over that 12 year period? I mean, that was pretty much the crucial point of your original skepticism of Alabama's more kicks missed than anyone else stat, was it not? That we needed to have an idea of how many attempts Bama made compared to everyone else. Also, if a team attempts significantly more kicks than anyone else, in a sense they "get some slack" because generally speaking, the more attempts you make, the harder it is to keep a good success percentage, and the more impressive a high success percentage becomes. Consider that the New York Jets have a 100% success rate in the Super Bowl, while the Steelers have a 75% success rate in the Super Bowl, but the Steelers' record is more impressive because they have been to the Super Bowl 8 times and won 6 times, while the Jets have only been 1 time, even though they won. We’re considering ALL the kicks, there are no outliers, but we’re considering the distance. It’s not just that Alabama has more misses because they have more attempts, they may be systematically attempting longer field goals. But yes, sample size is an issue, but there’s an easy formula to figure out “confidence intervals,” which is basically saying “the true number is probably in this range,” so the Steelers’ confidence intervals is much smaller than than the Jets’. More properly, outliers exist, but you can’t just ignore them, they’re actually important. You can only dismiss them at the very last stage and then only if you have a specific reason for why they lie outside the norm. Weird things DO happen, just infrequently. It’s possible that once you did all the math you’d find that Alabama does have below-expectation field goal kicking, but that the confidence intervals are such that you’re pretty sure it’s just luck.
|
|
|
Post by tripsclosed on Dec 14, 2019 17:27:03 GMT -6
So, for someone who is less studied in statistics than yourself, we are tossing outliers and keeping the most common kicks, correct? Also, to back up a bit, why couldn't we just take the total FG attempts of each team over that 12 year period and compare them to Alabama's total FG attempts over that 12 year period? I mean, that was pretty much the crucial point of your original skepticism of Alabama's more kicks missed than anyone else stat, was it not? That we needed to have an idea of how many attempts Bama made compared to everyone else. Also, if a team attempts significantly more kicks than anyone else, in a sense they "get some slack" because generally speaking, the more attempts you make, the harder it is to keep a good success percentage, and the more impressive a high success percentage becomes. Consider that the New York Jets have a 100% success rate in the Super Bowl, while the Steelers have a 75% success rate in the Super Bowl, but the Steelers' record is more impressive because they have been to the Super Bowl 8 times and won 6 times, while the Jets have only been 1 time, even though they won. We’re considering ALL the kicks, there are no outliers, but we’re considering the distance. It’s not just that Alabama has more misses because they have more attempts, they may be systematically attempting longer field goals. But yes, sample size is an issue, but there’s an easy formula to figure out “confidence intervals,” which is basically saying “the true number is probably in this range,” so the Steelers’ confidence intervals is much smaller than than the Jets’. More properly, outliers exist, but you can’t just ignore them, they’re actually important. You can only dismiss them at the very last stage and then only if you have a specific reason for why they lie outside the norm. Weird things DO happen, just infrequently. It’s possible that once you did all the math you’d find that Alabama does have below-expectation field goal kicking, but that the confidence intervals are such that you’re pretty sure it’s just luck. Gotcha. Thanks for taking the time to write this stuff!
|
|
|
Post by fantom on Dec 14, 2019 17:36:47 GMT -6
There are many FBS coaches who recruit kickers-punters to scholarships rather than hoping that walk-ons pan out.
That doesn't mean they will work out any more than highly-recruited position players.
Or any less.
But when you're punting from the shadows of your own goal posts or trying a potential winning FG in a conference championship game, or one that will get you there - you better have both.
And a long-snapper.
If not, that's on you as coach. But if a position player doesn't pan out you can find something else for him to do: a position change, special teams, at least scout teams. If a kicker doesn't make it you're stuck. I remember when a new coach of a 1AA team was talking with the issues he'd have recruiting one problem that he mentioned was that they had three kickers on scholarship.
|
|
|
Post by pitt1980 on Dec 18, 2019 13:55:41 GMT -6
So, for someone who is less studied in statistics than yourself, we are tossing outliers and keeping the most common kicks, correct? Also, to back up a bit, why couldn't we just take the total FG attempts of each team over that 12 year period and compare them to Alabama's total FG attempts over that 12 year period? I mean, that was pretty much the crucial point of your original skepticism of Alabama's more kicks missed than anyone else stat, was it not? That we needed to have an idea of how many attempts Bama made compared to everyone else. Also, if a team attempts significantly more kicks than anyone else, in a sense they "get some slack" because generally speaking, the more attempts you make, the harder it is to keep a good success percentage, and the more impressive a high success percentage becomes. Consider that the New York Jets have a 100% success rate in the Super Bowl, while the Steelers have a 75% success rate in the Super Bowl, but the Steelers' record is more impressive because they have been to the Super Bowl 8 times and won 6 times, while the Jets have only been 1 time, even though they won. We’re considering ALL the kicks, there are no outliers, but we’re considering the distance. It’s not just that Alabama has more misses because they have more attempts, they may be systematically attempting longer field goals. But yes, sample size is an issue, but there’s an easy formula to figure out “confidence intervals,” which is basically saying “the true number is probably in this range,” so the Steelers’ confidence intervals is much smaller than than the Jets’. More properly, outliers exist, but you can’t just ignore them, they’re actually important. You can only dismiss them at the very last stage and then only if you have a specific reason for why they lie outside the norm. Weird things DO happen, just infrequently. It’s possible that once you did all the math you’d find that Alabama does have below-expectation field goal kicking, but that the confidence intervals are such that you’re pretty sure it’s just luck.
Good thoughts in this thread
How do you guys find kickers?
In theory kickers should be relatively easy to find, as its one of the more repeatable activities that go on in football
relative to most things that go on in a football game, there aren't that many variables
with finding college kickers, part of the problem you have is that HS kickers kick off tees, and lots of HS don't try many long FGs
wonder how much additional insight you can get tracking trajectories, guy who's 8/9 with all those down the center, is probably different than an guy who's 8/9 with a bunch of those just inside the uprights
does tracking slices vs hooks mean anything?
how much where the laces are on the hold matters (from holding back in high school, it seemed like a lot, seems like I could pretty reliably cause a hook or slice by holding the laces an inch or so off center, [how many misses are really on the holder?])
anyways, interesting thread
|
|
|
Post by fantom on Dec 18, 2019 14:46:56 GMT -6
We’re considering ALL the kicks, there are no outliers, but we’re considering the distance. It’s not just that Alabama has more misses because they have more attempts, they may be systematically attempting longer field goals. But yes, sample size is an issue, but there’s an easy formula to figure out “confidence intervals,” which is basically saying “the true number is probably in this range,” so the Steelers’ confidence intervals is much smaller than than the Jets’. More properly, outliers exist, but you can’t just ignore them, they’re actually important. You can only dismiss them at the very last stage and then only if you have a specific reason for why they lie outside the norm. Weird things DO happen, just infrequently. It’s possible that once you did all the math you’d find that Alabama does have below-expectation field goal kicking, but that the confidence intervals are such that you’re pretty sure it’s just luck.
Good thoughts in this thread
How do you guys find kickers?
In theory kickers should be relatively easy to find, as its one of the more repeatable activities that go on in football
relative to most things that go on in a football game, there aren't that many variables
with finding college kickers, part of the problem you have is that HS kickers kick off tees, and lots of HS don't try many long FGs
wonder how much additional insight you can get tracking trajectories, guy who's 8/9 with all those down the center, is probably different than an guy who's 8/9 with a bunch of those just inside the uprights
does tracking slices vs hooks mean anything?
how much where the laces are on the hold matters (from holding back in high school, it seemed like a lot, seems like I could pretty reliably cause a hook or slice by holding the laces an inch or so off center, [how many misses are really on the holder?])
anyways, interesting thread
It should be even easier to evaluate kickers coming out of college than HS but look at how many kickers who are drafted end up failing.
|
|
|
Post by Chris Clement on Dec 20, 2019 10:14:09 GMT -6
Yeah, but that's confounded by desperate coaches changing kickers to save their job. On the whole NFL kickers are basically indistinguishable (except Justin Tucker).
|
|
|
Post by fantom on Dec 20, 2019 10:28:25 GMT -6
Yeah, but that's confounded by desperate coaches changing kickers to save their job. On the whole NFL kickers are basically indistinguishable (except Justin Tucker). That's why I specifically mentioned kickers who were drafted rather than signed as FA's. It's easy for a GM to cut a FA. When you cut a guy who used a draft pick for you're admitting a mistake. When you cut a former high draft pick you're admitting a BIG mistake.
|
|
|
Post by Chris Clement on Dec 20, 2019 11:13:43 GMT -6
Sunk cost fallacy is huge in football.
|
|