≡ Menu

The predictive value of target data, part II

On Monday, I argued that target data has some predictive value. I wanted to update that post with a few observations.

Wide Receiver Targets

In the original post, I looked at year-to-year data for all players with at least 500 receiving yards in Year N and at least 8 games played for the same team in Years N and N+1. But it makes more sense to limit the sample to only wide receivers if we want to predict how wide receivers project in the next season.

There are 554 pairs of wide receiver seasons that meet the above criteria.1 The best fit formula to project future receptions based on prior receptions and prior targets is:

Year N+1 Receptions = 14.0 + 0.547 * Yr_N_Rec + 0.124 * Yr_N_Tar

The R^2 is 0.39, and while the receptions variable is statistically significant by any measure, the targets variable just barely qualifies (p = 0.044) as such. Still, this tells us that for every 8 additional targets a receiver sees in Year N, we can expect one more reception in Year N+1, holding his number of receptions equal.

If we want to project receiving yards instead of receptions, we get:

Year N+1 Receiving Yards = 180.3 + 0.434 * Yr_N_RecYd + 2.55 * Yr_N_Tar

The R^2 is 0.33, implying a slightly less strong relationships, which makes sense: yards are more variable to large outliers than receptions, so you would expect receiving yards to be slightly harder to predict. Another interesting note: the targets variable here is statistically significant at the p = 0.0003 level, and as expected, the receiving yards variable is statistically significant at all levels. Holding receiving yards equal, a receiver would need an additional 19 targets to increase his projected number of receiving yards by 50, so the practical effect may not be all that large.

Addressing the multicollinearity problem

In the comments to the original post, Danny Tuccitto wondered what the numbers would look like if we isolated the variables. Since targets and receptions are so closely correlated, perhaps using just receptions gets us nearly all the way there, which would imply that using targets doesn’t do all that much for us.

If we look just at receptions in Year N and receptions in Year N+1, the correlation coefficient between the two variables is 0.39. The best-fit formula is:

Year N+1 Receptions = 16.9 + 0.72 * Yr_N_Receptions

This means that if you want to predict a player’s receptions (subject to the thresholds mentioned earlier) in Year N+1, the best starting point isn’t his Year N receptions, but to give him 17 catches and then credit for 0.72 receptions for every Year N reception.

The R^2 is the same as it was when we used both targets and receptions to predict future receptions, supporting Danny’s theory that it was really the receptions variable driving the results.

But what if we use just targets as the input? We get the following formula:

Year N+1 Receptions = 12.7 + 0.45 * Yr_N_Targets

And the formula has an R^2 of 0.35. So yes, targets are predictive. They’re not as predictive as receptions (or using both targets and receptions), but they can help you predict future receptions.

What about future receiving yards? With an R^2 of 0.31, the best-fit formula is:

Year N+1 Receiving Yards = 251 + 0.673 * Yr_N_RecYds

If we use targets instead of receiving yards, the R^2 drops to 0.29, and the best-fit formula is:

Year N+1 Receiving Yards = 212 + 0.5.81* Yr_N_Targets

Welker is the king of catch rate.

Welker is the king of catch rate.

I did one more study. I looked at the top and bottom 20 seasons in terms of catch rate. The group with the high catch rates — featuring the Austin Collies, Wes Welkers, Kevin Walters, and Lance Moores of the world –averaged a 74% catch rate in Year N. The group with the low catch rates — — featuring a bunch of Raiders (Louis Murphy, Doug Gabriel, Denarius Moore) and deep threats like Nate Washington, a young Greg Jennings, and Justin McCareins — averaged a pitiful 43% catch rate. In Year N, the high-catch rate players saw 98 targets and caught 73 passes for 878 yards (9.0 yards per target), while the low-catch rate players saw 110 targets and caught 47 passes for 727 yards (6.6 Y/T).

In Year N+1, the high-catch group received 110 targets, catching 74 passes (67%) for 877 yards (8.0 Y/T), while the low-catch group saw 104 targets, caught 54 passes (52%) and gained 745 yards (7.2 Y/T). There are many conclusions one could draw from that small study, but one would be that the low-catch rate players had no problem holding onto their value, and even increased their yardage total the next season.

Since it’s Friday, and I’m in the process of tying up loose ends…

Post-script to the Calvin Johnson targets discussion

On Tuesday, I analyzed the leaders in targets in 2012 and noted that Calvin Johnson saw “only” 27.9% of the Lions’ targets last year, which placed him behind seven other wide receivers. Considering the lack of production from the other main targets — Brandon Pettigrew, Tony Scheffler, Titus Young, and Nate Burleson — I wondered why Johnson didn’t receive an even larger share.2 I concluded with the following:

If there’s a criticism of Johnson, it’s that perhaps he is not as complete a receiver as Brandon Marshall or Andre Johnson, who are able to be both possession receivers and big play threats. Arguably the Lions should have utilized Johnson more, which is pretty scary considering how many yards he picked up. But instead, Detroit wasted 286 targets on Pettigrew, Scheffler, Young, and Burleson.

Since that post, Mike Clay of Pro Football Focus has helped frame the discussion. The following table shows the depth of the targets Johnson saw (relative to the line of scrimmage) in 2012:

Depth% of TarTargRecYdsRec%Y/T
5 or less25%503733174%6.6
6 to 1023%463034665%7.5
11 to 1514%271728063%10.4
16 to 2019%382246558%12.2
21+19%381654242%14.3
Total100%199122196461%9.9

Without doing this for every starting receiver, I hesitate to draw any conclusions. But we can note that a full quarter of Johnson’s targets were within five yards of the line of scrimmage, so it seems like the Lions did a good job using him on shorter passes. Detroit did an even better job with Johnson on deep passes, as Megatron caught 16 passes that would be appropriately categorized as deep. But with only 14% of his targets coming in the intermediate range, perhaps this supports this idea that he wasn’t running the full route tree like Andre Johnson or Brandon Marshall.

  1. I have again pro-rated all seasons to sixteen games. []
  2. Again, I do note the inherent weirdness in asking why a player who set the single-season receiving record did not see more targets. But I must point out that he set the single-season record when his team set the single-season record for pass attempts, muting that argument. []
{ 4 comments }
  • sn0mm1s February 15, 2013, 2:34 am

    http://espn.go.com/blog/nfcnorth/post/_/id/53154/lions-the-calvin-johnson-effect

    Is an interesting article that could explain why CJ didn’t get many intermediate targets. I would suspect that if the Lions were routinely seeing defenses designed to stop the pass and that the extra defenders were generally used to double CJ that intermediate routes would be the most difficult for CJ to get open on. 11-15 yards is pretty much where a safety would be lining up to double CJ.

    Reply
    • Chase Stuart February 15, 2013, 11:48 am

      Thanks sn0mm1s. That leads me to another post I’ve got coming.

      Reply

Leave a Comment