Chase Stuart
Footballguy
http://www.pro-football-reference.com/blog/wordpress/?p=304
http://www.pro-football-reference.com/blog/wordpress/?p=369Target stats — that is, the number of times a particular receiver was the intended target of a pass — are now widely available. But I’ve never been quite sure what to do with them. If two receivers have the same number of catches, but one of them was targeted much more often, which one is likely to have more catches in the future?
Occam would probably assume that the player with the lower number of targets — and thus the higher catch percentage — is probably the better player. After all, isn’t that what receivers are supposed to do? Catch the balls that are thrown to them.
On the other hand, the player with more targets is in some sense a bigger part of the offense. Either he’s open more often, or the quarterback is throwing in his direction even when he’s not open. At least if he’s staying in the same situation the following year, maybe some of those looks will turn into catches.
I have always suspected that neither of those explanations is in general correct, that while target numbers are probably relevant in certain cases, they aren’t worth anything unless you have more information about the particular situation. But I’d never really studied it before.
So I took all pairs of consecutive wide receiver seasons since 2002/2003 in which the player played at least eight games in each season and had at least 30 receptions in the first season (there were 264 such). Then I ran a regression of Year N+1 receptions per game against Year N receptions per game and Year N targets per game. Here is the resulting equation:
Year N+1 rec =~ .64 + .63*(Year N rec) + .07*(Year N targets)
The coefficient on Year N targets is positive, but it’s small, and not significantly different from zero in the “official” statistical sense. In other words, given the variation in the data, there is no real reason to assume the true coefficient on Year N targets isn’t zero.
And again, it doesn’t much matter whether it’s statistically significant or not. It’s too small to be very meaningful anyway. Last year, Chris Chambers had 3.7 catches per game on 9.7 targets per game. That’s a ton of targets for someone with so few catches. Our formula predicts him to have about 3.6 catches per game next season. If he had had the same number of receptions, but a more typical amount of targets, say 6 per game, last season, then the formula would project him with 3.4 catches per game next year. That’s a difference of only 3 catches over a 16-game season.
For what it’s worth, I also included various age controls in the regression and it doesn’t alter the conclusions.
Chambers’ low catch percentage last season was not an aberration, which leads me to the title of the post. Is Chambers’ consistently low catch percentage, like Eddie George’s consistently low yards-per-rush average, a sign that he’s not as good as he appears and that he’s only compiling raw numbers because he has been given a ton of opportunities? Or is the fact that he has consistently been given a ton of opportunities despite the seemingly poor production a sign that he must be pretty good, because no competent coach would make him such a major focus unless he had some real talent.
I don’t know how much this has to do with Chambers, but young running backs who get a ton of carries but have a low yards-per-rush average often turn into Hall of Famers. Here are the backs who had the most carries in their first three years despite a sub-4.0 average per carry:
Eddie George
Curtis Martin
Willis McGahee
Karim Abdul-Jabbar
Ricky Williams
Marshall Faulk
Jerome Bettis
Now there have been a lot of other young runners who failed to eclipse 4.0 yards per carry over their first three years. For example, Reggie Cobb, Antowain Smith, Johnny Johnson, and Leonard Russell. But those guys didn’t get as many carries as Curtis Martin and Marshall Faulk did. I think that might say something. A low yards-per-rush is bad. But a ton of opportunities over a reasonably long period of time despite a low yards-per-rush might just be a signal.
In the same way, Chris Chambers’ ability to remain a huge part of the offense through two entire coaching regimes (including several offensive coordinator switches) and numerous different quarterbacks, despite what appears on the surface to be sub par performance, might be a sign that he’s better than we think he is.
Targets — the number of times a player has been the intended receiver of a given pass — are becoming mainstream data these days. People generally cite a receiver’s impressive reception to target ratio as a reason to think he’s one of the best in the league. Conversely, critics will argue that a WR’s terrible catch ratio proves that he’s overrated. There hasn’t been a lot of target discussion on the PFR blog yet, outside of this post, but Doug and I have been discussing what to do with target data over the past few days.
Let me skip the discussion for a minute and just present some stats first. Note: targets are not an official NFL statistic, and there are always differences in target data depending on which site you use. Often, the sum of the targets for all the skill position players on a team is fewer than the total pass attempts for the team, due to human error. Regardless, all target data used here are from Footballguys.com, and I believe them to be as accurate as the target data available on any other site.
Here’s a list of the 25 most targeted WRs from last season:
Torry Holt STL 178Donald Driver GB 171Andre Johnson HOU 165Chad Johnson CIN 154Chris Chambers MIA 154Roy Williams DET 153Anquan Boldin ARI 152Laveranues Coles NYJ 151Terrell Owens DAL 151Marvin Harrison IND 148Mike Furrey DET 146Joey Galloway TB 142Steve Smith CAR 140Reggie Wayne IND 137Lee Evans BUF 137T.J. Houshmandzadeh CIN 133Keyshawn Johnson CAR 128Isaac Bruce STL 126Hines Ward PIT 126Jerricho Cotchery NYJ 125Javon Walker DEN 125Braylon Edwards CLE 123Plaxico Burress NYG 121Muhsin Muhammad CHI 117Marques Colston NO 115Now, here are the 25 WRs with the best catch-to-target ratio, minimum 90 targets:
Lots of intelligent people believe that receptions per target is a very good measure of the productiveness of a WR. But a metric that ranks Mike Furrey and Wes Welker as the 2nd and 3rd best receivers in the league is one that doesn’t come close to passing the smell test. Derrick Mason and Mark Clayton over Steve Smith and Terrell Owens? I’m not buying that one.One of the obvious problems is that guys like Welker and Furrey run short routes, while Owens and Smith are bigger play threats. It’s just like how David Carr had a great completion percentage last year but wasn’t very good. So perhaps yards per target, just like yards per attempt for QBs, will even the playing field? In theory this makes a bit of sense. If you are targeted five times on short passes, you might get four eight-yard receptions. If you’re targeted five times on longer passes, you could end up with two receptions for sixteen yards each. Your yards per target would be the same, which seems like a more appropriate way to explain how you did.Code:Player Team Targ Rec Rec/TargT.J. Houshmandzadeh CIN 133 90 67.7Mike Furrey DET 146 98 67.1Wes Welker MIA 100 67 67.0Jerricho Cotchery NYJ 125 82 65.6Marvin Harrison IND 148 95 64.2Larry Fitzgerald ARI 108 69 63.9Terry Glenn DAL 110 70 63.6Reggie Wayne IND 137 86 62.8Andre Johnson HOU 165 103 62.4Marty Booker MIA 90 55 61.1Marques Colston NO 115 70 60.9Derrick Mason BAL 112 68 60.7Reche Caldwell NE 101 61 60.4Laveranues Coles NYJ 151 91 60.3Lee Evans BUF 137 82 59.9Steve Smith CAR 140 83 59.3Mark Clayton BAL 114 67 58.8Isaac Bruce STL 126 74 58.7Hines Ward PIT 126 74 58.7Chad Johnson CIN 154 87 56.5Reggie Williams JAX 92 52 56.5Terrell Owens DAL 151 85 56.3Darrell Jackson SEA 112 63 56.2Rod Smith DEN 94 52 55.3Javon Walker DEN 125 69 55.2
But then again, that’s just in theory. Who were the league leaders in receiving yards per target last year?
Player Team Targ RecYd Y/TReggie Wayne IND 137 1310 9.6Terry Glenn DAL 110 1047 9.5Lee Evans BUF 137 1292 9.4Marvin Harrison IND 148 1366 9.2Marques Colston NO 115 1038 9.0Reggie Brown PHI 91 816 9.0Chad Johnson CIN 154 1369 8.9Larry Fitzgerald ARI 108 946 8.8Isaac Bruce STL 126 1098 8.7Javon Walker DEN 125 1084 8.7Roy Williams DET 153 1310 8.6Darrell Jackson SEA 112 956 8.5Eddie Kennison KC 101 860 8.5Steve Smith CAR 140 1166 8.3Marty Booker MIA 90 747 8.3Mark Clayton BAL 114 939 8.2Plaxico Burress NYG 121 988 8.2T.J. Houshmandzadeh CIN 133 1081 8.1Antonio Bryant SF 91 733 8.1Anquan Boldin ARI 152 1203 7.9Santana Moss WAS 101 790 7.8Terrell Owens DAL 151 1180 7.8Hines Ward PIT 126 975 7.7Jerricho Cotchery NYJ 125 961 7.7That looks a little better, I think. Welker and Furrey drop off the list, and the top guys are all very good. But seeing TO way behind Terry Glenn, Torry Holt (not listed) well behind Isaac Bruce, and Eddie Kennison above Steve Smith should raise some eyebrows. And I won’t forget 2004, when Brandon Stokley, Ashley Lelie and Eddie Kennison ranked 1st, 2nd and 4th in the league in receiving yards per target.So what do you think of targets? Are they worthwhile or not? Let me try and characterize Doug’s position:
For the most part, yards per target and receptions per target are meaningless statistics. Much more often than not, on an incomplete pass, the targeted receiver was the *best* receiver on the field that play. If you consider all “bad” passing plays — incompletions, interceptions, and also 4-yard-gains on 3rd-and-10 — the targeted receiver was the best (or at least, not the worst) receiver on on that play the majority of the time. There’s a positive correlation between how well the WR does his job and the probability that he gets the target.
I think Doug brings up a lot of good points. The fact that Bruce, Kevin Curtis and Steven Jackson all ranked ahead of Holt in yards per target last year is disturbing. Bryant Johnson has a better YPT average than Larry Fitzgerald or Anquan Boldin. Patrick Crayton, Terry Glenn and Jason Witten ranked ahead of Terrell Owens, too. A plausible explanation would be that Crayton only gets the target when he’s wide open (which is rare, because he’s not that good) whereas Owens gets the target when no one is open (since Owens is the best of the bunch). This is very much a Simpson’s Paradox problem. Consider this hypo:
The Cowboys run 200 pass plays with just Crayton and Owens as the targets. On 50 of them, Crayton is open and Owens is covered. On 100 of them Owens is open and Crayton is covered. On 50 of them, neither was open. When he’s open, Crayton will get 50 targets, and catch 36 passes. When Owens is open, he’ll catch 50 of the 50 passes thrown to him. So far, edge to T.O.. On the 100 passes when neither is open, Owens will get the target 90% of the time because he’s so much better than Crayton. Owens will convert 20 of the 90 passes into receptions, while Crayton will convert none of the ten targets into receptions. Clearly, Owens has outperformed Crayton. But Owens will have a catch-to-target ratio of 50%, while Crayton will have a catch-to-target ratio of 60%. So it’s certainly plausible that Simpson’s Paradox comes into play in target data, despite the relatively extreme hypothetical.
It’s also possible that strong WR2s and WR3s should do well, because of matchups. Assume a team has the 10th best WR1 in the league and the best WR2 in the league, but that team’s WR1 is still better than that team’s WR2. But since the WR1 is only a little better than the average CB1 he faces and the WR2 is a ton better than the average CB2 he faces, it stands to reason that he’ll have much better receptions and yards per target ratios.
However, Chad Johnson actually ranks ahead of T.J. Houshmandzadeh in yards per target, and both rank ahead of Chris Henry. Since many claim Chris Henry is the best WR3 in the league, you’d think he’d do well in this statistic. After all, Patrick Crayton did. And Lee Evans ranks behind Peerless Price and Josh Reed — two of the weaker receivers in the league — in receptions per target. I don’t think Price ranked well because he’s such a good WR2, but rather because Losman wouldn’t throw his way unless no one was anywhere near him.
So in the end, what’s the best use of target data to measure how good a receiver is? Yards per target? Receptions per target? Targets? I’ll throw out one last one, that I think has some validity to it: targets per team pass. Raw target data is skewed towards teams that pass a lot, but this gets at the key question: When your team is throwing the ball, who is the QB throwing it to? Obviously this isn’t perfect because Marvin Harrison would get a ton more targets if Reggie Wayne wasn’t around, and vice versa. The quality of your supporting cast factors in significantly here. But here’s the list:
Andre Johnson HOU 0.343 165 481Lee Evans BUF 0.318 137 431Laveranues Coles NYJ 0.310 151 487Torry Holt STL 0.301 178 592Terrell Owens DAL 0.298 151 506Chad Johnson CIN 0.293 154 526Anquan Boldin ARI 0.281 152 541Javon Walker DEN 0.275 125 454Donald Driver GB 0.272 171 629Marvin Harrison IND 0.266 148 556Joey Galloway TB 0.265 142 535Chris Chambers MIA 0.261 154 591Steve Smith CAR 0.259 140 541Jerricho Cotchery NYJ 0.257 125 487Roy Williams DET 0.256 153 598T.J. Houshmandzadeh CIN 0.253 133 526Reggie Wayne IND 0.246 137 556Mike Furrey DET 0.244 146 598Braylon Edwards CLE 0.241 123 510Hines Ward PIT 0.241 126 523Keyshawn Johnson CAR 0.237 128 541Plaxico Burress NYG 0.231 121 523Muhsin Muhammad CHI 0.229 117 512Eddie Kennison KC 0.226 101 447Drew Bennett TEN 0.220 98 445Donald Driver moves down, Javon Walker moves up. Lee Evans goes up, Chris Chambers goes down. And Cotchery, Houshmanzadeh and Reggie Wayne become the top WR2s on the list, ahead of Mike Furrey. That’s enough for me, at least.But what are your thoughts? What do you think of targets, and how should we use target data?
Chase
