≡ Menu

The predictive value of target data

Demaryius Thomas made the most of his targets in 2012.

Demaryius Thomas made the most of his targets in 2012.

In 2007, Doug Drinen and I wrote a pair of articles discussing our views on targets. I’m working on a wide receiver project this off-season, and a complete discussion of receiving statistics includes a discussion of targets.

Let me start with the prevailing few: targets are important, and if two receivers have the same production on a different number of targets, the one who produced on fewer targets is better/more valuable. Similarly, if all else is equal, the receiver with a higher catch rate — calculated as catches/target — is the better/more valuable one.

There are some problems with the prevailing view. By placing targets in the denominator of a formula, we’re implying that targets are a bad thing, or at a minimum, an opportunity wasted. But targets aren’t like pass attempts. Pro Football Focus has a stat called yards per pass route run, and that actually is the receiver version of yards per pass attempt.1

But targets don’t help identify the player who deserves blame: on a random incomplete pass, assume three receivers are running routes, and one of them is targeted. Absent a drop, I have a hard time saying that of the three wide receivers, the targeted one did the worst of the three. If we grade a receiver by his yards per route run, each receiver is equally penalized with one route run on the play; if we grade a receiver by yards per target, the two wide receivers that did not get open are not penalized, while the one that was targeted is penalized. That seems fundamentally wrong to me.

Here’s another problem: In a broad sense, the player with more targets (or percentage of his team’s targets) is in a very real sense a bigger part of his team’s offense. Either he’s open more often, or the quarterback is throwing in his direction even when he’s not open (whether because the coaches call more plays for him or because he’s earned the quarterback’s trust). In any event, the target itself is an indicator of quality, and penalizing a player — which is what you do when you place targets in the denominator — for an event that is highly correlated with quality is not something I’m comfortable doing.

I ran a regression to predict the number of catches a receiver would record in a season based on his number of receptions and targets the prior season. I looked at all receivers who:

  • From 2000 to 2011, had at least 500 yards and played in 8 games; and
  • Played for the same team in the next season and played in at least 8 games

There were 726 receivers who met those criteria. I then pro-rated all player seasons to 16 games. Running a regression to predict Year N+1 receptions, the best-fit formula (with an R^2 of 0.37) was:

13.3 + 0.484*Rec_YrN + 0.164*Targ_YrN

That indicates that targets are a decent predictor of future receptions, and would imply that penalizing receivers for failed targets may not be wise. Let’s use Larry Fitzgerald as an example, because he represents an extreme example of high number of targets/low number of receptions. Fitzgerald had 156 targets but only 71 receptions. This formula would predict him to have 73 receptions next year. If Fitzgerald had instead caught his 71 passes on a more typical number of targets — say 120 — he would be projected to have 67 receptions in 2013. On the other hand, Andre Johnson, who had 112 catches on 162 targets, is projected for 94 catches. If Johnson had caught his 112 passes on 200 targets, he’d be projected for an even 100 catches next year.

What if we look at yards instead of receptions? Using the same sample of receivers, the best-fit formula (also with an R^2 of 0.37) was:

124.0 + 0.577*RecYd_YrN + 1.690*Targ_YrN

As above, the p-value on the target variable was statistically significant at the p = 0.01 level. Let’s use Indianapolis’ Donnie Avery as an example. Avery had 781 receiving yards on 123 targets, for a poor 6.3 yards/target average. This formula projects him to have 783 yards next year. Had Avery instead averaged 9.0 yards per target in 2011, he would be projected for only 721 yards in 2013.

What about Denver’s Demaryius Thomas? He had 1,438 receiving yards on only 141 targets; yes, it sure helps to have Peyton Manning throwing you the ball (Thomas averaged 7.9 yards per target in 2011 with Tim Tebow). Because he averaged over 10 yards per target, this formula doesn’t love his 2013 prospects: he’s projected to have only 1,192 yards next year. Much of that is due to general regression, but had he been targeted 170 times instead of 141 times, and gained the same number of yards, he’d be projected for 1,241 yards next year. Not a significant difference, but a small effect.

So what do we make of this? Let’s look at Tampa Bay’s Mike Williams. In each of his first three seasons (despite wildly different years from Josh Freeman), Williams has finished the year with a roughly 50% catch rate. I don’t think this means that Williams is a bad receiver. A low catch rate could be a sign that a player is being targeted frequently because he’s doing his job better than the other receivers, and he’s being targeted even when he’s not open. Other times a low catch rate is simply a reflection of how he’s used in the offense. Or the side effect of being saddled with an inaccurate passer.

Vincent Jackson had a 49% catch rate last year; among the other Buccaneers with over 200 yards, Doug Martin had a 70% catch rate, Dallas Clark was at 63%, and Tiquan Underwood was at 51%. The obvious takeaway is that the Bucs tried to use their receivers to make big plays, but used the running back and tight ends on safer plays — both Clark and Martin averaged fewer than ten yards per completion. Mike Williams averaged fewer yards per target (7.9) than Vincent Jackson (9.4), and saw fewer targets as well. In that case, it’s safe to say he’s not as good or as valuable as Jackson. But trying to extrapolate Williams’ catch beyond lands you in risky territory.

  1. Unfortunately, yards per pass route run is not going to help us if we want to grade receivers on a historical basis. []
  • I’m sympathetic to the view that targets, yards/target, catch rate, etc. are overrated statistics. Lots of learned fantasy people swear by these kinds of stats; I’m not one of them. Nevertheless…

    a) I think the main problem is not with targets, per se, but with the fact that applying it in the way people do assumes the cause of variation in catch rate is random across players. As you discuss in depth here and in the PFR blog post, it clearly isn’t. Basically, except as an indicator of a WR’s level of involvement in his team’s passing game, targets are useless without some second level of analysis, be it game charting or a readily available like “air yards,” both of which get at WHY a target was not a reception.

    b) Having said all that, I do have an issue with your regressions. Just a hunch, but almost certainly suffer from ridiculous amounts of multicollinearity. By definition, receptions and receiving yards are dependent on targets, so including the latter with either of the former in a regression model is going to gum up the works. I’d just run bivariate correlations for the three stats (i.e., receptions, receiving yards, targets), compare the target-receptions correlation with the receptions-receptions correlation, and compare the target-receiving yards correlation with the receiving yards-receiving yards correlation.

    • Chase Stuart

      Thanks Danny. Good comments as always. I’ll look into this — probably won’t get to it until tomorrow, though.

    • Chase Stuart

      Some other thoughts…

      Agreed that some fantasy people swear by target data, although as usual, the logic is a bit flimsy. Some think targets are hugely important, which would imply that yards/target is not important. Others look at yards/target to see the leaders to get a sense of who the next breakout could be. This could be good, or it could land you with Ashley Lelie.

      a) The problem is when you take targets as face value. Or, like me, when you’re trying to grade receivers across many years, you simply can’t get to the next level of analysis. I agree things like air yards/depth of target would help. Drops is a stat that I haven’t really worked with, but I’m sure there’s some value in that as well.

      b) Putting aside the question of how to measure the relationship, what do you think of targets?

      • Richie

        I think for fantasy purposes, targets is telling us something different than how good a receiver is – they are just telling us how often a QB is throwing to a receiver. When I need to decide between starting Steve Smith and Miles Austin, I don’t necessarily care who is really the better receiver. I know their fantasy stats might be similar, but if one guy has more targets than the other he might be a better play just because he’s getting more opportunity. (Again, I don’t care if that opportunity comes from being better at getting open or being the only good receiver on his team, or being the QB’s favorite target.)

  • GMC


    Two thoughts:

    1. Targets (and snaps/pass routes run) are an opportunity indicator. As such they in the first instance represent a positive indication: That X receiver receives more targets is an indication of their performance in practice, the trust their QB has, and a large number of other factors, all of which represent good indicators of a quality receiver. Opportunity data, because opportunities come only to the best-performing athletes, are a good indication of future performance (for an extreme example, think of the “college starts” element of the Lewin Career Forecast). Targets aren’t a positive indicator because of opportunity; they’re an indicator that the player was good enough to be given an opportunity.

    2. However, between and among receivers who are in similar circumstances, efficiency is a good metric of performance; a high Catch Rate or Y/T is a good indication of the difference between similarly situated receivers. Some of the best DVOA seasons in Football Outsiders’ database (which values Y/T and catch rate type information highly) are third receivers in top-flight passing offenses. But no one thinks Az-Zahir Hakim was a better receiver than Isaac Bruce.

    3. A variable for % of team targets (as you’ve previously discussed) is a start on looking at this problem.

    • Chase Stuart

      Thanks GMC. Agree that the devil is in the “similarly situated receivers” caveat.

  • Chase Stuart

    Also, this comment from Doug in my old post still has some validity to it:

    I’m not much of a hoop-a-metrics guy, but I know that Horace Grant and Dennis Rodman had higher field goal percentages that Michael Jordan most seasons. That’s not because they’re better shooters. In fact, it’s precisely because they are worse shooters. It’s because, on the significant percentage of possessions where there is simply no good shot to be had, it was Jordan who took the tough shot, because he had the best chance of making it.

    • GMC

      Nice. It’s helpful (to me, only) that the basketball analogy happens to be to the one NBA team (the 2nd set of Jordan championships) I actually know anything about. And that is exactly the point I was making above. Calvin Johnson doesn’t have a lot of incomplete passes because he rubs butter on his fingertips.

      Of course, in today’s NBA, there is an ongoing attempt (google “STATS LLC”) to use cameras to create data on players directly. I’d be curious to know if this was something football had looked at; certainly such an effort would seem to make sense as a means of quantifying players who are fast “in pads” and so on. Not to mention figuring out what types of throws different QB’s can and can’t make…

  • Shattenjager

    “He had 1,438 receiving yards on only 141 receptions . . .” should be 141 targets, right?

    This project is off to quite an interesting start, Chase–I look forward to its continuation.

    • Chase Stuart


      Thanks Shattenjager. Glad you’re enjoying.

  • Danish

    My mind goes straight to Kobe Bryant (or Carmelo): They hawk the ball and tak a rediculos number of shots. Partially because they like shooting. But also because if Kobe doesn’t shoot that means Metta or Derek Fisher taking those shots – not clearly a better option. So when theres nobody open for a good shot, a bad shot must be taken, and Bryant is probably the best Laker to do that.

    Similarly when Jay Cutler doesnt find anyone open, he chucks it to Brandon Marshall since he’s clearly the best Bear at catching hard passes.

  • Hello, I log on to your blogs on a regular basis.
    Your humoristic style is awesome, keep up the good work!