Is there such a thing as a clutch player? Sabermetrics says no but I've already found a couple problems with commonly accepted Sabermetric theory, so I thought I'd let the data speak on this.
First, it's important to define what I mean by clutch. I'm going to define "clutch" player as someone who performs better in high leverage situations. That is, a clutch player is more likely to get a single with a runner on third than a non-clutch player.
I'm going to use RE24 to get a better look at this. The idea here is that a clutch player will have a statistically higher RE24 than a non-clutch player because he will get more hits than average with runners on base or in two-out situations -- both of which are likely to drive RE24 up.
I started by collecting data on every hitter in the game for the last 20 years. (Pithcers were dropped.) I then regressed RE24 on PA, singles, doubles, triples, home runs, sacrifice flies, sacrifice bunts, double plays, walks, and strikeouts. The idea here is that if a given player is as likely to, say, hit a single with the bases empty as he is with a runner on third, these statistics should define much of the variance in RE24. The regression results are quite good. My regression explains 85% of the variance in RE24 and every individual statistic has the expected sign and is significant. (That is, strikeouts detract from RE24, home runs increase it, etc.)
This regression allows me to predict what their RE24 would be if there was no luck involved in when they happened to get their hits. I define "clutch" as the difference between their actual RE24 and this predicted value. It's important to note that, over the course of a season, some players will be lucky and have RE24s well above the predicted value and others well below the predicted value. However, in the normal course of events, a player will have both good luck and bad luck over the course of his career. Some years will be well above average, others significantly below average. So, to identify a clutch player, we need someone who put up an RE24 significantly above the predicted value over a career.
To test this, I average the "clutch" statistics for every player who played four or more seasons over this stretch. This comes out to 1293 different players. Only five of them had statistics significantly greater than zero. The list is a surprising one: Brandon Larson, Yamaico Navarro, Andrew Romaine, Peter Bourjos, and Kevin Kouzmanoff. Of the five, Kouzmanoff has by far the best record, outperforming his predicted RE24 by almost 7 runs per season. On the flip side, four players have RE24s statistically below their predicted value: Jason Bates, Matt Walbeck, Mike Gallego, and Adam Piatt. Bates is the worst of these, averaging an RE24 almost 4.5 runs a year under the predicted value.
When dealing with statistics, it's well known that sometimes statistical noise causes "false positives." Looking at the data, I'm pretty convinced that's the case for all 9 of the players mentioned above. Thus, it turns out that Sabermetrics is right: there are no clutch players.
Filed under: Uncategorized