Running the Numbers: Analytical View of Joseph Randle
At DallasCowboys.com, I broke down running back Joseph Randle’s pros and cons from a statistical perspective, including some analysis on player comps.
Each year prior to the draft, you can find dozens of scouting reports on each prospect. Many of them contain a “Player Comparison” section in which the writer compares the prospect to a current or former NFL player. I think these can be misleading, for two reasons. The first is that such a comparison can sometimes imply that a particular prospect has a rather narrow career outlook.
The truth is that most prospects have a very wide range of potential career paths, so we should really compare each to multiple similar players. The most comparable players can be weighted the strongest, but it would be wrong to insinuate that a particular player will without a doubt have a comparable career to someone else.
The second problem with most player comparisons is that they typically emphasize the wrong traits. As I’ve mentioned in the past, we should search to see which traits have been predictive of NFL success in the past, then weight the most important characteristics more heavily than those that haven’t been great predictors. If college receptions didn’t matter at all for running backs when predicting their futures, for example, there would be no reason to factor them into a search for comparable players. The most similar players are the ones who have near-matching numbers in the metrics that matter, i.e. those that can accurately predict a career.
However, how many times do you see player comparisons with two prospects who went to the same school? We saw that last week in Bryan Broaddus’s player comparison post; Broaddus asked some scouts around the league to compare the Cowboys’ draft picks with current NFL players. Two of the players, Travis Frederick and Terrance Williams, were provided with comps who played at their colleges. In these situations, it’s likely that the scouts were suffering from the availability heuristic – a mental shortcut through which people make judgments based on how easily they can think of examples. It might be easy to compare Williams to fellow Baylor wide receiver Josh Gordon, for example, but they’re pretty different players in regards to traits that appear to matter in the NFL. The fact that they both attended Baylor isn’t one of those important traits.
Let me be clear that NFL scouts are really good at what they do; for the most part, their player grades are pretty accurate, and many of them do it without the aid of analytics. But it’s really difficult, perhaps impossible, to generate meaningful comps just from memory. There are all kinds of biases involved in that sort of process. That’s really why we use data and advanced stats in the first place; no matter how great a scout’s memory or how well he knows a prospect, there’s no way he could recall a list of player comps faster or more accurately than a computer. In effect, algorithms can help us eliminate what we think we know to tell us what’s really there.
Head over to the team site to see why I’m not too high on Randle.
No related posts.
Related posts brought to you by Yet Another Related Posts Plugin.