Writers of Pro Football Prospectus 2008

11 Nov 2006

FO in the NY Times: NFL Kickers Are Judged on the Wrong Criteria

The idea behind this column is simple: NFL kickers are more consistent kicking off than they are kicking field goals. But until I did the research for this article, I had no idea just how dramatic the difference really is.

Measuring every kicker from 1999-2005 who had at least 10 field goal attempts in two consecutive years, the year-to-year correlation of field goal percentage is .03. POINT ZERO THREE!!! That sure explains why the KUBIAK kicker variables are almost all based on team offense. The measures we use at FO, which compare each field goal to the average from that distance, aren't much better: the correlation there is .08.

On the other hand, the year-to-year correlation of average kickoff distance -- same time period, same minimum of 10 kickoffs -- is .60. That makes average kickoff distance one of the most predictable individual stats in the entire NFL, at any position. And sure, some of these players are switching teams and home stadiums from one year to the next, but that affects field goals and kickoffs equally.

I'm sure that some kickers truly are better at field goals than others, but you would need to look at a lot of years of data to tell who those guys were. So isn't it better to spend your money -- or a fourth-round pick -- on a guy who you know can boom kickoffs? And if field-goal percentage one year does nothing to predict field-goal percentage the next year, what about first half vs. second half? Should field goals just be removed from DVOA, like fumble luck? Clearly I've got some more work to do on this.

Click the comments for some more fun data.

Posted by: Aaron Schatz on 11 Nov 2006

59 comments, Last at 08 Jan 2010, 1:35pm by Jameszeee

Comments

1
by admin :: Sat, 11/11/2006 - 11:28pm

I think we're using something else as the graphic in tomorrow's Times, but I put together a list of the top 12 players in field goal percentage and kickoff distance in Weeks 1-9 of 2005, and then Weeks 1-9 of 2006.

Here's the field goal percentage Top 12 for Weeks 1-9 2005:
Neil Rackers
Todd Peterson*
Jay Feely
Phil Dawson
Nate Kaeding
Joe Nedney
Mike Vanderjagt
Rian Lindell
Lawrence Tynes
Olindo Mare
Josh Brown
Jeff Wilkins

Here's the same list, Weeks 1-9 2006:
Robbie Gould
Matt Stover
John Carney
Rian Lindell
Sebastian Janikowski
Nate Kaeding
Adam Vinatieri
Rob Bironas
Shayne Graham
Phil Dawson
Jeff Wilkins
John Kasay

Only four players are on both lists: Dawson, Kaeding, Lindell, and Wilkins. Remember how badly Sea-Bass sucked last year? He's 9-for-10 this year.

Now, here's kickoff distance Top 12 for Weeks 1-9 2005:
Neil Rackers
Olindo Mare
Michael Koenen
Rob Bironas
Jeff Wilkins
Josh Scobee
Kris Brown
Josh Brown
Jose Cortez*
Todd Sauerbrun*
Adam Vinatieri
John Kasay

Same list, Weeks 1-9 2006:
Stephen Gostkowski*
Paul Ernster*
Rob Bironas
Josh Brown
Neil Rackers
Jason Hanson
Nate Kaeding
Michael Koenen
Kris Brown
Dave Rayner
Josh Scobee
Adam Vinatieri

This time SEVEN players are the same on both lists. But note that four players CAN'T be on both lists because they only played in one season or the other.

2
by Kaveman (not verified) :: Sun, 11/12/2006 - 12:00am

This is interesting, for sure. But, I'm not convinced by your conclusion that choosing a kicker solely for kickoff distance is justified.

Consider that Jason Elam has attempted 35 field goals over 50 yards as compared to Mike Vanderjagt's 22 and Adam Vinatieri's 14 (since 1998). That's got to have an effect on each kicker's success percentage?

Seems like the useful stat here would be variance over different distance ranges, but I'm no statistician...

3
by Becephalus (not verified) :: Sun, 11/12/2006 - 12:02am

Interesting stuff. So would DVOA just reward teams for the attempt a set amount then. And just ignore the result? That seems correct.

4
by turbohapy (not verified) :: Sun, 11/12/2006 - 12:12am

Re: 2

I bet that attempted field goal distance has a lot to do with where they play. Elam plays a lot in Denver. Vanderjagt played a lot in a dome. Vinateri didn't play much in either until recently.

5
by admin :: Sun, 11/12/2006 - 12:16am

Re: 2. The FG PTS+ stat we use for special teams here at FO adjusts for the issue of field-goal distances, and as noted above, the correlation is barely better than standard FG%.

6
by PMD (not verified) :: Sun, 11/12/2006 - 12:17am

I suppose DVOA could do some sort of reversion to the mean estimate instead of totally throwing out field goal makes or misses. Either way, it seems that some kind of alteration is needed. This gives us a good idea into the intuition behind the pessimistic projection for the Bears in the second half.

7
by Josh (not verified) :: Sun, 11/12/2006 - 12:47am

Good interesting article Aaron. Just a minor point, but I don't think what you say above about kickers switching teams, that it should affect kickoffs and FGs the same, is entirely true, because when a kicker changes teams he's working with a new holder on FGs.

8
by jimmo (not verified) :: Sun, 11/12/2006 - 1:07am

small nit, which doesn't obscure the point, but wasn't Rayner Indy's KO specialist for most or all of '05?

9
by Smeghead (not verified) :: Sun, 11/12/2006 - 1:19am

Ever consider trying to refine the concept of FG accuracy by looking at how close misses are? A 51-yarder that slithers just wide still reflects a pretty high degree of precision on the part of the kicker. A Vanderjagt playoff honk doesn't.

I'm just pulling this out of the air, but it might be the case that there's a higher level of accuracy correlation year to year if you consider "close misses" ... almost like fumbles, the question of whether a kick from a distant hashmark hooks a foot inside the post or kisses off the outside of it is at some level nearly random.

Or, that might be total nonsense.

10
by jonnyblazin (not verified) :: Sun, 11/12/2006 - 2:10am

On the one hand I buy the idea that FG kickers are fungible, but on the other hand as a Raven fan its a nice feeling to have knowing Stover is always money. Is the proper analogy FG kickers are akin to relievers in baseball (theoretically you should gather scraps and see what works out, yet its nice to have a Rivera)?
Re: 7
I looked over Stover's career and found some interesting info. In his first 3 years in CLE he kicked FGs 72.3% overall, 51.7% from 40+. The next 2 years in CLE he kicked 90.4% overall, 78.9% from 40+.
He then went to BAL, and his first years he kicked FGs 75.8% overall, and 53.7% from 40+. The next eight years he kicked 88.7% overall, and 75.5% from 40+.
It should be noted that Stover has pretty much been automatic under 40 yards (maybe 1 miss a year).
So it seems that in each instance Stover kicked about 50% long field goals the first 3 years in a city, then from then on kicked them at 75%. So in this instance its not being familiar with a new team, but familiarity with the home stadiums (wind patterns) that might help out kickers.

11
by admin :: Sun, 11/12/2006 - 2:14am

Correct on Rayner. I was thinking Cortez, but I forgot Cortez was in Dallas early in 2005, Indianapolis late in 2005. Now edited above.

12
by Zac (not verified) :: Sun, 11/12/2006 - 2:17am

"Ever consider trying to refine the concept of FG accuracy by looking at how close misses are?"

That information is not part of the play-by-play. And even if they used information from the game charters, it seems like it would be pretty subjective.

13
by throughthelookingglass (not verified) :: Sun, 11/12/2006 - 2:42am

"I’m sure that some kickers truly are better at field goals than others, but you would need to look at a lot of years of data to tell who those guys were."

We have years of data on most NFL kickers. 21 of the 32 starting kickers have been in the league since 2001, and half have been active since the '90s, so it would seem that there would be enough data to find their true level of performance.
With the kickers with only a few years of experience, could a DAVE-like formula could be used to combine track record with regression to the mean?

Weather Adjusted Kicker Opportunities (WAKO)

14
by BillWallace (not verified) :: Sun, 11/12/2006 - 3:17am

I fully believe that kickers at the highest level could be very very fungible. But obviously that still doesn't mean that every person who can average 60+ on kickoffs is going to fluctuate between 70-100% on FGs. Some kickers will just suck and won't reach the level where they're fungible with other kickers... so just saying the the NE kicker will be fine because he booms the kickoffs is premature I think.

15
by BillWallace (not verified) :: Sun, 11/12/2006 - 3:20am

Oh a question.
How does FG accuracy compare in importance in proportion to kickoff distance? If the NE kicker were say 8% worse on average on FGs, but 5 yards better than average on every kickoff, is he a good kicker or bad?

16
by DavidH (not verified) :: Sun, 11/12/2006 - 3:40am

Is there some kind of cutoff level where poor performance begins to actually be predictive? i.e. kickers can bounce around from 65% to 100%, but if they have a 50% season, it really tells you they're no good?

Also, the Stover data makes me wonder - if you only include players that have been with a certain team for, say, at least 3 years, does the correlation get better?

Lastly, I'm guessing the sample sizes are too small to be usefull, but are the correlations any better for certain distances? Are some people money from less than 30, while others have a few shanks every year?

17
by Mnotr (not verified) :: Sun, 11/12/2006 - 3:50am

One thing I always want to know about kickoffs and punts:

how does hang time affect the return? Distance is great, especially on kickoffs, but if I were an NFL team I'd measure hang time and use that to help me judge kickers.

18
by DavidH (not verified) :: Sun, 11/12/2006 - 4:12am

Because FG's are either made or not, there's a lot of variance in the numbers just due to the quantitization. I wanted to get an idea of how that might look, so I made 10 imaginary kickers in Excel. Each kicker played 10 years, and each year he kicked between 23 and 33 FG's. Every FG had an 80% chance of being good; you could call this the kicker's "true ability." Here are the year by year FG% for the 10 imaginary kickers:

75 , 71 , 86 , 73 , 74 , 83 , 73 , 90 , 70 , 72 ….. Career: 76.7
69 , 82 , 71 , 78 , 91 , 74 , 77 , 72 , 93 , 81 ….. Career: 78.8
72 , 77 , 89 , 74 , 71 , 72 , 79 , 59 , 90 , 83 ….. Career: 76.6
78 , 81 , 83 , 86 , 84 , 77 , 85 , 83 , 89 , 86 ….. Career: 83.2
88 , 78 , 74 , 76 , 69 , 79 , 83 , 80 , 81 , 93 ….. Career: 80.1
89 , 75 , 87 , 86 , 79 , 77 , 77 , 73 , 87 , 94 ….. Career: 82.4
71 , 80 , 87 , 83 , 86 , 77 , 83 , 74 , 88 , 83 ….. Career: 81.2
72 , 66 , 88 , 85 , 81 , 81 , 82 , 81 , 81 , 81 ….. Career: 79.8
74 , 70 , 81 , 90 , 82 , 81 , 56 , 87 , 82 , 85 ….. Career: 78.8
65 , 94 , 87 , 77 , 91 , 72 , 87 , 76 , 85 , 77 ….. Career: 81.1

As you can see, there are a couple times where the FG% goes from below 80 to above 90, back to below 80.

And it just dawned on me as I type this to run a year-to-year correlation, like Aaron did with the real data... *Excel-ing*

Well, what do you know, not much better than real life:
0.095

Keep in mind this would be like a kicker that got to kick from the exact same spot (which has a degree of difficulty of 80%?) every single time. Once you factor in the differing numbers of long kicks vs. short kicks, the weather, and day-to-day human variability, it's no wonder the FG% stats are all over the place.

Did this make any sense at all, or am I just rambling. It IS 2 AM.

19
by DavidH (not verified) :: Sun, 11/12/2006 - 4:14am

Oh, you know what, those career numbers are wrong. I just averaged the yearly FG%'s. As if ANYONE cares.

20
by Kibbles (not verified) :: Sun, 11/12/2006 - 6:39am

I just realized how well Ernster's really been doing in Denver. Even adjusting for altitude, Denver's been 6th best in the league on kickoffs, and above-average (10th in the league) on punts. Really remarkable, when you consider that the guy he replaced was a pro bowler and a performance enhancer who actually had negative value in both categories last year. Maybe NFL coaches actually know what they're doing after all...

Re #19: Oh, you know what, those career numbers are wrong. I just averaged the yearly FG%’s. As if ANYONE cares.

Oh, whew, that explains it. I thought Kicker 10's career average looked a touch low. I remember that amazing sophomore season of his very clearly. It's a shame he never played for an offense capable of moving the ball beyond the hypothetical point where he had an exact 80% chance to make field goals.

21
by Becephalus (not verified) :: Sun, 11/12/2006 - 9:32am

LMAO...this is one of those good little threads. So interesting and humorous.

Since it is so random maybe kickers should unionize? They could arrrange it so they all get paid the same and only have to play for the teams they want :)

22
by Sam B (not verified) :: Sun, 11/12/2006 - 11:44am

18. Great post.

23
by fyo (not verified) :: Sun, 11/12/2006 - 12:05pm

Yeah, Atlanta had a lot of luck with Koenen. That really worked out well.

24
by Zac (not verified) :: Sun, 11/12/2006 - 7:53pm

Koenen was 8-8 in the preseason, including 6 kicks over 40 yards. Maybe his leg was getting tired from doing punts, FG kicks, XP kicks, and kick-offs. His drop-off was rather inexplicable.

25
by Peter Libero (not verified) :: Sun, 11/12/2006 - 8:30pm

I'm reluctant to believe that field goal kicking is random. The only thing I can think of to explain all this data is that a lot of kickers have short careers and are never heard from again after a few too many misses. Maybe the "really good" kickers like Akers or Stover (and one day Kaeding?) or whoever are more consistent than the overall picture, which includes the bad ones that come out of nowhere and leave quickly. It's a little depressing to think that it's a complete crapshoot no matter who's kicking.

26
by Peter Libero (not verified) :: Sun, 11/12/2006 - 8:31pm

One other question, what's the correlation like from first to second halves of seasons?

27
by Yaguar (not verified) :: Sun, 11/12/2006 - 9:50pm

We're not saying kicking is random, and all the kickers are equal.

What we're saying is that even if one kicker is much better than another, the 30 or so attempts they get each season is such a small sample size that we can't tell very much from the stats, and that in any given season Phil Dawson can outkick David Akers.

28
by Peter Libero (not verified) :: Sun, 11/12/2006 - 11:34pm

Um... I think that's definitely the the implication. What Aaron showed was that no matter how well or poorly a kicker does, there is no connection between that result and the next year's. Why would you want a "good" or "bad" kicker, if really they're just getting lucky/unlucky in stringing together a few good/bad years? Take a look at the randomization in post 18... Kicker 7 never went below 71% or above 88%, a 17 point range, while Kicker 10 had a 29 point range.

Aaron suggested removing field goals from DVOA, and I think that might not be that crazy an idea. Shouldn't it be easy to check if doing so improves DVOA correlation with victory?

It's certainly counterintuitive to think that field goal percentage is basically random at the highest level, but I guess that might be how it is if there's no explanation or subdivision of the stats that does provide a correlation yearly.

29
by MRH (not verified) :: Mon, 11/13/2006 - 12:23am

This raises lots of questions. Is it the PK who is inconsistent or the situation in and from which he kicks that is?

For example, I'm sure the correlation on EPs is pretty high. Easy kick at consistent distance.

But FGs can be at all sorts of distances and are (probably) more affected by weather. In season 1 he might have a high percentage of attempts over 40 yds in windy conditions and in season 2 he might have a high percentage of attempts under 30 yds in warm and dry or dome conditions.

It would be interesting to see the correlations on FGs under 30 yds, 30-40 yds, and over 40 yds. The smaller sample size might be a killer.

Anyhow, very good article and interesting analysis.

30
by Jim A (not verified) :: Mon, 11/13/2006 - 1:32am

There was an article published in Chance Magazine about 10 years ago that also argued that there were very little skill differences among NFL field goal kickers. As I recall, they found that at shorter distances the skill differences were almost insignificant. What really separated one FG kicker from the next was his ability to convert at long distances, but that kickers have so few (yet widely varying) opportunities at long distances that it takes many years of data on kickers to distinguish reliably between their levels of FG-kicking skill.

31
by Yaguar (not verified) :: Mon, 11/13/2006 - 2:02am

Again, I don't think it's entirely random. Some people must be innately better at kicking a football through uprights than others, even if those differences don't manifest themselves clearly over, say, 100 attempts.

It's fair to say that David Akers is flat-out better than, say, Billy Cundiff. But I'd agree that if you're looking at Nedney and Vinatieri, there won't be any noticeable difference in field goal kicking ability.

32
by Peter Libero (not verified) :: Mon, 11/13/2006 - 2:55am

If Billy Cundiff has as good a chance of having a perfect season as Akers, he's no worse. If the difference between kickers can only be seen reliably over a few hundred attempts, then they don't matter, and Aaron is right that only kickoff distance matters.

33
by Yaguar (not verified) :: Mon, 11/13/2006 - 3:13am

My answer is that he doesn't. He sucks.

There are the very good NFL kickers, who are more or less interchangeable on field goal attempts. And then there are the guys on the margins. And they are there for a reason.

There is some correlation between seasons, especially with the FO stats, and I bet if you looked at longer stretches, the correlation would be stronger. If there were no correlation at all, you could look at a kicker who goes 0/12 on field goal attempts one year, and say he's as likely as Matt Stover to be perfect next year. Obviously, that's silly.

34
by DavidH (not verified) :: Mon, 11/13/2006 - 8:47am

someone should lump together 3-year stretches and calculate the stretch-to-stretch correlation.

Somebody who is not me.

35
by Jim A (not verified) :: Mon, 11/13/2006 - 10:49am

If there were no correlation at all, you could look at a kicker who goes 0/12 on field goal attempts one year, and say he’s as likely as Matt Stover to be perfect next year.

But that 0/12 kicker is not an NFL kicker. He'd never last that long. We're only talking about kickers good enough to have at least 10 attempts in two consecutive seasons.

36
by azibuck (not verified) :: Mon, 11/13/2006 - 11:36am

Does anyone else think that if you took this thread, and substituted "relief pitcher" for "kicker" that everything would still be as true (and debatable)? Even, or perhaps especially if you narrowed it down to "closers". There are a few elite closers/relievers/kicker, and even if their numbers fluctuate you'd still consider them elite and not fungible. But the majority are closer to the margins and are fungible. I haven't thought about it for more than 5 seconds, but my first question about this is, so kickers are more consistent at kicking off, so what? What effect on the team does it have? Can you measure the value of a 65 yard-average kickoff guy vs. Vanderjagt's 58 yard average? And I agree with the post about hang time. I can't believe hang time is considered so important for punts yet is almost universally ignored for KO.

37
by Scott de B. (not verified) :: Mon, 11/13/2006 - 11:47am

Again, I don’t think it’s entirely random. Some people must be innately better at kicking a football through uprights than others, even if those differences don’t manifest themselves clearly over, say, 100 attempts.

One thing to take into consideration is that teams see these guys practice. I don't know whether practice FGs are a substitute for game FGs, but if a kicker takes 20 practice kicks each week of training camp and preseason, and 10 practice kicks each week of the season (low estimates, I would think), then you've got a database of several hundred kicks each season.

Now, whether any team pays attention to that sort of thing (except on the broadest level) is another question.

37
by stan (not verified) :: Mon, 11/13/2006 - 11:47am

36,

It's not ignored by the coaches.

39
by Malene, cph (not verified) :: Mon, 11/13/2006 - 1:03pm

re 36:
go read a transcript of a Belichick conference call. He'll address things like that for 15 minutes while reporters try to get him to talk about missing Vinatieri. There's a reason they keep Gostkowski, and it's not stubborn pride about "wasting a draft pick".

40
by Malene, cph (not verified) :: Mon, 11/13/2006 - 1:11pm

re 37:
yes, those are VERY low estimates. Here's Morten Andersen's pre-game routine since returning to the Falcons (same as ever, btw) starting at 1:15 before kick-off:

9 FGs in one end, 4 from each hashmark and one in the middle, starting from XP distance, ending at 50 yards. Then switch ends, 9 more from xp to 50. Then 8-10 kickoffs to warm up the return units.
Then finally a 28-yarder down the middle. He'll usually miss one of the 50 yarders, if not, you'll see him kicking from 50 rather than Koenen in that game.

That's a lot of kicking: on game-day!
The ST coaches have a HUGE number of kicks to evaluate. And Mike Vick might not watch film, but the FG units do - evaluating all kicks on height, distance, accuracy etc.

41
by Zac (not verified) :: Mon, 11/13/2006 - 1:39pm

"Can you measure the value of a 65 yard-average kickoff guy vs. Vanderjagt’s 58 yard average? And I agree with the post about hang time. I can’t believe hang time is considered so important for punts yet is almost universally ignored for KO."

Yeah, the difference in kickoff distance correlates with starting field position. Starting 7 yards farther back on every position following a kick-off is a big deal, especially in a high-scoring game.

And hang time is ignored because no one is keeping track of it.

Someone in the know, does the game charting project keep track of hang-time?

42
by Zac (not verified) :: Mon, 11/13/2006 - 1:46pm

And you guys, take a look at the graphic in the article
http://graphics8.nytimes.com/images/2006/11/12/sports/12score_GFX.jpg
It gives the FG% and Yards Per Kickoff for Vinatieri, Vanderjagt, Longwell, Rackers, and Akers (the top kickers in the league). The FG% fluctuates for all of them, but the yards per kickoff are consistent (except for when Longwell, Vinatieri, and Rackers switched teams).

And yes, Akers and Vinatieri have been pretty consistent. But 2 guys out of 32 seems like a fluke.

43
by zlionsfan (not verified) :: Mon, 11/13/2006 - 3:35pm

No, we aren't being asked to track hang time, which is just as well. Trying to get the actual time of the kick based on what we'd see on the screen would be a nightmare - I doubt you'd want to take the network's word for their own clock, and half the time the camera would miss either the punt or the catch (kick or catch for a kickoff).

Thinking about it, this whole concept makes sense to me. Kickoffs are pretty much the same thing every time: while the conditions might change for outdoor kickers, everything else stays the same. Kick from the 30, kick it deep, give your coverage team time to get down there. But FG attempts are pretty much different every time: left or right hash or in the middle, different yard lines. I can see that it would be a lot easier to identify weak or strong kickoffers, um, kickers who are weak or strong at kickoffs, but much more difficult to identify kickers who are good or bad at FGs.

The other part about that is what Kaveman mentioned first, that you would really want to focus on accuracy within each range, rather than overall accuracy, but that's the point: if you have only 10 attempts from a certain range in a particular year, it's much more difficult to tell if that kicker is missing kicks because he's bad or because he kicked into a 30-mph wind that day.

Just for fun, I looked at all the kickers from the last four years who attempted 10 or more FGs from 40-49. (By the way, I love how ESPN's sortable stats give you a 404 error when the page doesn't come back. A page not on your site? Hello? I just clicked on a link you gave me. Hey, I'm a web programmer, I know how this stuff works.)

There are exactly two kickers who appear in the list all four years: Stover and Janikowski.

Stover's data, latest first:
11/14, 9/10, 11/14, 7/10.

Janikowski:
7/12, 8/10, 9/10, 7/12.

Not much variation for Stover. Janikowski? All over the board.

And this is mostly offense-driven. You only get 40-49 FG attempts if you get inside the 35, but not inside the 20 (roughly), so in a sense, you need an offense that moves the ball decently, but not great. It's hard to sustain that from season to season.

There's probably a larger sample from 30-39, but then you get into the lack of variance between kickers, I think. After all, most NFL kickers will hit most of the time from that distance, or they won't be NFL kickers for long.

44
by Bill Barnwell :: Mon, 11/13/2006 - 3:38pm

If you refresh the sortable stat link that gives you a 404, it usually comes up.

Yes, I know, that's not the point.

45
by DGL (not verified) :: Mon, 11/13/2006 - 3:44pm

#41: Seven yards of field position between your own 15 and your own 40 (where 65% of drives start, and I'd guess where the overwhelming majority of drives after kickoffs start - though I don't have the "how posession was obtained" in my charts) is worth, on average, 0.2 points per drive.

In 2005, there were 2,439 kickoffs (excluding onside kicks); that's 4.76 kickoffs per team per game. So seven yards of field position per kickoff equates to about 0.95 points per game, or 15.2 points over the course of the season.

Also in 2005 (sorry, I'm being less rigorous than Aaron and only looking at one year), there were an average of 30 FG attempts per team. Across the league, 81% of kicks were made, so your league-average kicker would make 24 of 30 kicks. The worst FG kicker (Janikowski) made 67% of his kicks, so would have (and in fact did) make 20 of 30, costing his team 12 points compared to the average kicker. The best FG kicker (Rackers) made 95% of his kicks, so would have made 29 of 30, gaining 15 points compared to the average kicker.

So if we consider the worst FG kicker in the NFL in 2005 to have been "replacement-level" (a very rough-and-ready determination of replacement level, I grant, but I'll be consistent later), the best FG kicker in 2005 was worth 27 points over replacement level.

But if we're comparing best to worst to average, we can't look at a seven-yard delta; the best KO average (Rackers) was 68.8 yards, the worst (Novak in WAS) was 56.8 yards, and the league average was 62.2 yards. That 12-yard swing in field position between best and worst is worth about .36 points per kickoff, or about 1.7 points per game -- or, how about that, 27 points per season.

So based on this single-season sample, it would appear that FG accuracy and kickoff distance are roughly equal in importance. Which would tend to support Aaron's thesis - a kicker who can consistently boom deep kickoffs will get you that 22-27 points every year, while on FGs a kicker is likely to bounce around from just above replacement-level to Rackersian accuracy from year to year.

15: One yard in the Own 15-Own 40 range is worth about 0.03 points. Over 76 kickoffs per year, that's about 2.28 points per year per yard of kickoff distance. A 2.5% increase in FG accuracy is worth 2.25 points per year if you ignore quantization noise, which isn't a good idea because quantization noise underlies Aaron's whole thesis on FG accuracy not correlating year-to-year. But if you wanted to trade off the two, there's your equivalence - one yard on kickoffs = 2.5% FG accuracy.

46
by Led (not verified) :: Mon, 11/13/2006 - 5:34pm

"...it takes many years of data on kickers to distinguish reliably between their levels of FG-kicking skill"

This is an interesting point. In any given year, a kicker's number of attempts is a small enough sample size that it is probably almost meaningless as a predictor. I mean, how predictive is a strecth of 30 at bats in baseball on the next stretch of 30 at bats. Take a kicker that has kicked for enough years to have a sufficiently large sample of attempts, and it would not surprise me to find that (once you normalize for distance, weather conditions, etc.) all the ups and downs from year to year are within normal random distribution given the kickers career average.

47
by DGL (not verified) :: Mon, 11/13/2006 - 6:03pm

For 2005, the mean accuracy on all field goal attempts was 81.2%. The standard deviation of field goal kicker accuracy was 13.6%, so Janikowsky (at 66.7%) and Rackers (at 95.2%) were the only "regular" kickers (more than 10 attempts) out of 32 who were more than one SD away from the mean.

Unfortunately, I don't remember enough statistics to calculate how much of the variation in FG accuracy is attributable to random chance - if anyone else does, I'd be glad to provide the raw data.

48
by BBWC (not verified) :: Mon, 11/13/2006 - 6:19pm

Has anyone tried lengthening kickers' sample periods by adding in their college stats?

49
by zlionsfan (not verified) :: Mon, 11/13/2006 - 6:57pm

Well, one problem with using college kicks is that the environment is considerably different - wider hashmarks for one. (No kicking tees any more.)

lol at Bill. Yeah, that's what makes it funny. It's almost like a Python sketch:

"Hello, I'd like to see your kickers' book."
"Here it is, right here."
"Oh, thank you. Is there a page that is sorted by attempts?"
"No, I'm sorry, that would be in the kickers' book."
"But I thought this was the kickers' book!"
"No, it's not."
"You told me it was!"
"I did? Well, it must be, then."
"Well, then where is the page?"
"What page?"
"The page sorted by attempts!"
"Oh, it's right here, see."

50
by admin :: Mon, 11/13/2006 - 7:02pm

Glad everyone has enjoyed this so much. Just wanted to point out two things.

First, this has come up in a lot of the comments and in a lot of the blog posts that reference this article, so let me repeat: The FG PTS+ stat we use for special teams here at FO adjusts for the issue of field-goal distances, and as noted above, the correlation is barely better than standard FG%.

Second, there is a reason why my post above ends with the words "Clearly I've got some more work to do on this." I don't want to go jumping to any conclusions about just how unpredictable field-goal kicking might be. All I said is that it doesn't correlate from year-to-year, so there's a pretty good chance that you won't get what you paid for. There's no doubt that some guys just plain aren't good enough to kick field goals, no matter how well they hit kickoffs, though I doubt Steve Gostkowski is among them. I still need to look at correlation between the first half and second half of the season, and the correlation between career field goal percentage and field goal percentage in the player's next season, and a few other things.

As for measuring the value of kickoffs, we do that all the time here. Click "Our New Stats Explained" and go down to the special teams part.

51
by Larry (not verified) :: Mon, 11/13/2006 - 10:15pm

Re: 45

You're probably overstating the difference, longer kicks probably also have longer returns on average, so you can't count the whole 7 yards per kick against the kicker, it's some fraction of that. But good to see that, order of magnitude, it is clearly pretty close between Kickoffs and FGs.

Quantization errors are pretty clearly a big deal here, and correlation coefficients may not be the best assessment tool in the presence of such small number statistics. The analysis in #18 shows that pretty nicely, the sample size is too small to find even a noiseless real effect. Not being able to see evidence of a real effect is a lot different than finding evidence of the absence of the effect.

52
by Dan (not verified) :: Mon, 11/13/2006 - 10:39pm

Nice article, Aaron. I ran some numbers, and I'm getting that, given the abilities of the current crop of NFL kickers, finding a kicker who is good at kickoffs is over three times as important as finding a kicker who is good at field goals.

How much is it worth to have above average FG performance? Over the last four seasons, having a FG performance 1 standard deviation above average (which is usually about 5th best in the NFL) is worth about 7.2 points (over average). But most of that, as Aaron's article shows, is due to chance, not the kicker's ability.

What is the relationship between a kicker's ability and his performance in a single season? We can't directly measure a kicker's true, underlying ability, but we can get at it indirectly. What we can do is to assume that any year-to-year consistency in a kicker's performance is due to his underlying ability. The year-to-year correlation of .08 exists only because performance in each year is correlated with the kicker's true ability. That means that the correlation between true ability and performance in a single season is the square root of .08, which is .28.

A kicker whose performance is one standard deviation above average would be earn his team about 7.2 points over average, but a kicker whose true ability is one standard deviation above average (which means, roughly, the fifth best FG kicker in the league) would be expected to earn his team only .28 x 7.2 = 2.0 points over average.

If you run the numbers the same way for kickoffs, you find that a kicker whose true, underlying kickoff ability is one standard deviation above average (which would be about the fifth best kickoff kicker in the NFL) would be epected to earn his team .77 x 8.1 = 6.3 points per over average per season.

So having a guy who's good at kickoffs is about 3.1 times as important as having a guy who's good at FGs. And having a guy who is worse than other kickers at kickoffs is 3.1 times worse than having a guy who is worse than other kickers (to the same degree) at FGs.

Obviously, it's important to keep someone who's way worse (i.e. several standard deviations worse) at kicking FGs than most other NFL kickers from having that job on your team, but other than that a good kickoff guy does seem to be a lot more important than a good FG guy.

53
by Scott (not verified) :: Tue, 11/14/2006 - 12:17am

One other thing that makes kickoff ability more important than FGs:

A team can adjust its offensive strategy to reduce its reliance on a bad FG kicker (go for 1st down conversions more often in opposing territory, or pooch punt, instead of attempting dicey long field goals). The marginal offensive value of attemping a field goal is not (3 points x prob of making field goal), it is (3 points x prob of making - expected points from attempting 1st down conversion) or (3 points x prob of making - reduced opponent points from resulting punt position).

On the other hand, kickoffs are pretty unavoidable (in fact they become more important the more successul your offense is). You could probably reduce them some by instituting a ball-control offense, but that would really be the tail wagging the dog!

54
by Scott (not verified) :: Tue, 11/14/2006 - 12:22am

Forgot that my oversimplified formulas in #53 should at least have included another pretty big offsetting term: (- increased opponent points from better starting field position after field goal miss x prob of missing field goal)

It is a wonder to me sometimes that so many teams attempt long field goals on 4th and short.

55
by Steve Greenwell (not verified) :: Tue, 11/14/2006 - 5:20am

So, whatever happened to the kickoff specialist the Chargers had, Kurt Smith? Last I can find of him on Google, he worked out for the Saints in early September.

In light of this article, it seems like he could be useful for a team, since he did boom more than 50 percent of his kicks for touchbacks in college. Is a kickoff specialist more important than a deep bench player?

56
by David B (not verified) :: Tue, 11/14/2006 - 5:33am

Just thinking out loud here, but we could probably use panel
analysis techniques to isolate cross-sectional facts over a time series of
multiple kickers and get us a much better sense of what variables really
matter. In other words, we can create a time series of all kicks by all kickers
over the past ten years and then analyze the impacts of certain variables
cross-sectionally across that data set.

The problem I see though is that there are some significant
variables which we have no record of. Some key ones to think about:

1) Ball spot, and I don't mean distance. Is the ball on
dirt, grass, wet grass, snowy grass, or turf? Is the ball centered, or on the
extreme left or right?

2) Snap and hold. Minor glitches in snaps and holds don't
show up in play-by-plays, but can significantly affect a kick.

3) Pressure on the kicker from field goal blockers and\or
from the game situation.

In other words…it’s harder to kick a 50 yard field goal in
the snow, after a bad snap, and with 3 defenders busting through the line to
win a playoff game than to kick a 50 yard field goal in practice. Since we only
currently have good data on the distance, mediocre data on weather conditions,
and minimal information on the rest, I’m not sure how accurate any current
model could be.

Other than those missing data points, there are two other
issues I see:

1) I'm not sure that a standard regression analysis will be
totally accurate in analyzing kicking because kicks may not be randomly
distributed; i.e., the result of one kick may be affected by the previous kick.
For example, a kicker who misses one important kick in a game may be down on himself, and thus kick more poorly for the rest of the kicks
that game. If this pattern exists, it would be very important to take into
account the non-randomness in any panel type model.

2) Because of the low sample size (especially at any given
spot, with any given weather condition, with any given game condition, etc...)
we may not be able to get statistically significant results.

All that aside, it still seems like a worthwhile project. We
could at the very least analyze the value of stadiums, new holders, new teams,
some weather (cold\indoors), and some game situation information on kicks.
Anyone want to get the raw data to do this? I've got the panel
regression software...I just need the raw data to input into it.

57
by DGL (not verified) :: Tue, 11/14/2006 - 11:31am

#50:

"There’s no doubt that some guys just plain aren’t good enough to kick field goals, no matter how well they hit kickoffs..."

Yes, and those guys generally won't make NFL regular-season rosters; if they do, they won't last long. It's like David Lewin's QB projection system - you don't need to look at all the data, because you can look at the outcomes. NFL coaches do a good job of weeding out the guys who plain aren't good enough to kick field goals; whoever's left will be "replacement-level" at worst.

So coaches need to keep doing the same "filtering" they do now on basic competence in kicking field goals, but given the choice between two "good enough" field goal kickers, they should take the one with the longer kickoffs.

58
by Winston Spencer (not verified) :: Wed, 11/29/2006 - 4:42pm

Sample size is a big issue, particularly considering the number of variables that can affect a kick. My guess is that there are variables that are simply not tracked in the stats (wind, cold, mud, Ye Olde Buffalo Bills Bruce Smith, etc...) that affect the kick more significantly.
I would like to make the assumption that kicking field goals is like tossing coins (a binomial event), except that for each kicker (coin) the number of makes to shanks is more like 4 to 1, instead of 1 to 1 (a weighted coin).
The math I used: Probability of making a kick = made/attempted, standard deviation=sqrt(P*(1-P)/attempts), assuming you use a Gaussian approximation of the binomial distribution: 90% confidence is 1.6449*s.
The math tells us that lowest meaningful sample size is probably about 10 kicks or so. But even then, assume we are measuring an 80% kicker (But we don't know. That's why we are measuring). A sample of 10 kicks will be 90% likely to produce a number between 58% and 100%. Stated another way, we can be 90% confident (in our measurement) that our measurement is within 22% of actual (long-term) performance.

I did little look at the Gary Anderson's stats. Interesting upshot is that the trend lines show that his accuracy actually went up from about .75 to .85 over his career. Surprisingly, his per season number of attempts actually went down over his career.

59
by Jameszeee (not verified) :: Fri, 01/08/2010 - 1:35pm

My son is working on a science project for school and we need to find out how to Calculate a kicker's Field goal percentage.

Can someone please let us know how?

I dont know if we add up the total attempts made with the total field goals made, or if we need to take into consideration the distance of the field goal.

Thank you