The NFL combine has come and gone and now we’re left to make sense of what we learned. Who are the risers and fallers? How will the results affect draft positions? And most importantly, does it even matter?
The combine regression trees were helpful to figure out which combine drills matter for wide receivers and running backs, and allowed us to identify the 2016 prospects who “won” the combine at the respective positions. That said, a major piece of what we currently know is missing from the combine regression trees: collegiate production.
I added collegiate production to the wide receiver combine model to see how production and measurables work in conjunction with each other.
This is the result:
I’ll save you some time looking through all the different decision nodes: none relate to the combine measurables. When I put age, combine measurables, and a boatload of different collegiate production stats into the regression tree, none of the combine measurables were seen as significant. This shouldn’t come as a total surprise, the Harvard Sports Analysis Collective found as much in its combine research, and I’ve conducted other research hinting that the combine didn’t matter all that much for wide receivers. But the regression tree gives us another angle to look at the question, and an easily interpreted tree to make sense of.
The first and most important split of the tree is at 29 percent of career market share of receiving yards, meaning that share is more important than raw numbers, and the arc of a prospect’s career is more important than only his final-year results.
However, the final year does matter. All the other splits relate to only final-year statistics, with the market share of receiving yards in the season of 42 percent or greater leading to the most successful node. There are 33 receivers who have accomplished this in the data set, including all-time-greats, like Calvin Johnson and Larry Fitzgerald; younger studs like Allen Robinson and Amari Cooper; plus RotoViz favorites who never panned out,1 like Marvin McNutt and Kenny Britt.2
In additional to high market shares of receiving yards, the model also prefers receivers who can stretch the field; at least that’s how I interpret its all-else-being-equal approval of receivers who average higher yards per reception, or have lower reception totals.
Age also makes it onto the tree, but not for the higher market share branches. If someone hasn’t been a dominant from a market share perspective, but meets the threshold of at least 933 receiving yards his final year, the model gives a higher success rate if he is 21 years old or younger. Age does matter, just perhaps not as much for those with gaudy production.
The power of the regression tree analysis isn’t just it’s ability to find what’s important, but also to identify what isn’t. As draft boards are reshuffled post-combine, a profitable approach should be to generally fade wide receiver prospects who rise, and buy those who are falling.