The NFL Combine and a Unified Theory of Athleticism
I’m only sort of mostly kidding about the overly ambitious title of this post. I do think this post will probably help inform your understanding of the athleticism of NFL players. I don’t actually have a unified theory of anything. Scratch that. I have a unified theory about Maggie Gyllenhaal’s alleged hotness, although I won’t sully these pages by getting into it.
Last night on Twitter I was discussing Rex Burkhead’s awful 40 time and saying that primarily I’m interested in a running back’s ability to generate power or momentum with their lower body, and Burkhead’s terrible 40 time didn’t really scare me given his outstanding performance on other Combine measures. If Burkhead has the lower body athleticism to post excellent tests in things like vertical leap, broad jump, 3 cone and shuttle, I’m fine with burning a flier pick on him even if I won’t ever get to see him score an 80 yard touchdown. Actually it’s premature to say I would burn a flier pick on him as I really shouldn’t speculate on that until I see what (if any) NFL team he ends up on.
The thing about Combine measures is that while they often get sliced and diced as if they indicate different athletic abilities, a number of them are correlated with each other. Below I’ve posted a correlation matrix which shows several of the Combine measures and their correlations to each other.
*Table created from data on 230 RBs and WRs who did all drills.
|HT||WT||40 Time||3 Cone||20 SS||Vertical||Broad Jump|
Let’s go through the table to see if we can makes heads or tails of it (as I typed that phrase I realized that it’s a very Grandpa thing to say).
- Vertical and Broad Jump are the two most correlated measures. That makes sense. They’re both measuring similar things. But vertical and broad jump also enjoy a decent correlation with 40 time. It’s negative, which is to say as 40 time decreases, vertical and broad jump should increase.
- The 3 Cone and Short Shuttle are correlated. They both test short area quickness and ability to change directions.
- Height and weight are correlated. People often ask why I include weight in many of my tables, but not height. I always first say that the two are generally correlated, but also I’m generally displaying the one that was more predictive in backtesting. But you can also see from the table that weight has higher absolute correlations than height on 3 of the 5 athletic measures. Height:Broad Jump is the only one where height significantly beats weight in terms of correlation with athleticism. That’s actually an interesting one and I think it bears more thought.
- Weight and 40 time are correlated. As weight goes up, so does 40 time. But weight and the 3 cone have almost the same correlation, which leads me to believe that perhaps we should be looking at adjusting 3 cone for weight in the same way that we adjust 40 times for weight.
- 40 Time enjoys correlation with a number of the athletic measures, which supports the notion that it can exist as some kind of short hand for athleticism. Or at least that would be true if we had enough observations to be confident we were getting the actual “true” number.
I have a lot more work planned on this, but initially I do think that this correlation matrix supports the kind of dimensions that a few of the writers have been talking about on the site, where they’re combining Combine tests to come up with things like Agility Scores and Explosion Scores. One of the problems with Combine measures is that they are all small samples. Most of the time we get one or two measurements and we all mentally discard the higher of the measurements. Combining measures like this gives you at least more observations to base an opinion or projection off of.
So the takeaways from this post are:
- We should probably look at adjusting 3 Cone for weight.
- Correlations support the kind of dimension scores that the RotoViz writers have been writing about.
- It might be worth looking at adding 40 times to both the Explosion Score and Agility Scores given the 40’s correlation with each measure. To do this you would probably need to standardize, or scale, each measure, which maybe would be a good thing to do anyway given the differing magnitudes of the measures.
There’s more wood to chop here, but I’m excited that the other writers have actually made a lot of progress on this stuff already.