In their 2007 work “The Book” (as in “Managing by …”), Tango, Lichtman and Dolphin used hard statistical analysis to debunk any number of notional ideas about baseball players and teams, among them that certain players are “clutch” performers. Their analysis indicated that whatever clutch tendencies players might exhibit in a given season would “correct” over time such that performance levels over a career would be much the same in “clutch” situations as in any other.
But, that doesn’t stop us from looking at those one-season tendencies, which I’ll explore next in looking at the players (like the Rockies’ Nolan Arenado to the left) who were best in the clutch in the 2016 season.
Baseball-Reference.com offers affordable subscriptions (highly recommended) to their “Play Index” search engine for accessing the wealth of statistical data hosted on the site. Among the tools available with Play Index is something called Split Finder which, as the name indicates, provides statistical breakdowns based on an array of different “splits” across a broad range of categories. So, it is to Split Finder that I turned to find these traditional clutch splits:
- Men On Base
- Runners in Scoring Position
- Runners in Scoring Position with 2 Outs
- Late and Close
The first three are self-explanatory, while “Late and Close” includes all PAs from the 7th inning on with a score difference of one run or less, or with the tying run on base, at bat or on deck (thanks David P for this explanation). Looking at the 146 players with a qualified (502 PA) batting season last year, the best and worst OPS results in the above splits look like this.
So, many names at the top of these lists that are not a surprise and a few (Stephen Piscotty, Travis Shaw, Logan Forsythe) that are. At the bottom of the list, Zack Cozart takes the prize of a bottom 10 result in all four splits, with Curtis Granderson close behind with three bottom 10’s (Granderson’s struggles to produce RBI were much remarked on by HHS readers last year; now, we know why).
Looking at how much better or worse those raw OPS scores were compared to these players’ overall OPS results yields this leaderboard.
The numbers shown are the difference between the players’ OPS results in these splits and their overall season OPS. We see many of the same names, but also some new ones. It’s notable that the magnitude of the OPS differences increases markedly from Men On to RISP and, to a lesser degree, from RISP to RISP – 2 Outs. Are these players really that much better at producing when runs are available? Or, is some other factor at play?
My sense is that, indeed, there is another factor at play that I’ll term “selection bias”. What I mean by that is that, for the first three of these splits, the plate appearances represented will occur more frequently in games when these players and their teams are “having their way” with their opponents’ pitching offerings. So, while the numbers may indicate, to some degree, a tendency for some batters to come through (or not) in the clutch, the positive results are likely also indicating that these players were simply taking advantage of inferior pitching.
To help to counter the selection bias inherent in the first set of metrics, I also found some additional splits that are more neutral or even skewed to games when the pitchers were having the upper hand.
- Leading Off an Inning
- Two Outs
- Behind in the Game
The first split is neutral in the sense that there is the same number of these events in every game, with a slight bias to games where the pitching is dominating in that the leadoff PAs in those games represent a larger proportion of all of the PAs in the game. The second split is also mostly neutral, with a slight bias in the other direction in that there will be more of these events in games with more PAs (i.e. when the batters are dominating). The third one is reserved, of course, only for batters whose teams are trailing in a game, which most often means batters who are facing tougher pitching; in 2016, OPS for all of MLB was .718 when behind, .746 when tied, and .755 when ahead.
It may not be intuitively obvious why these splits would evidence clutch characteristics, so let me explain. Leading off an inning and batting with two outs were chosen because of the significant difference in run expectancy that results depending on whether a player makes an out or not. Going back to “The Book”, the difference in run expectancy between making an out and reaching first base as the first batter of an inning amounted to almost two-thirds of a run (.657 to be exact) for the years 1999-2002, or about .600 runs for 2016. That’s not chump change! Similarly, when batting with two outs, while the run expectancy change between an out or reaching base will depend on which bases are occupied (it ranged from about one-tenth to three-quarters of a run in 2016), what is known with certainty is that an out reduces that expectancy to zero every time. So, in the sense that the difference between an out or reaching base is so clear cut in these splits, doing the latter can be characterized as a clutch performance. Batting when behind most often does not change the ultimate game result (in 2016, teams trailing after one inning posted a .306 winning percentage that, of course, reduced the later in the game a team was trailing), but doing well in this split is nevertheless important if a team harbors any hope of making a comeback.
While this second set of splits don’t seem like clutch stats in the same way the first set of splits did, I think it’s worthwhile looking at them, if only to provide a bit of counter-weight to the selection bias inherent in that first group of metrics. Here is that leaderboard for OPS.
Notably, these results are less extreme, at both ends of the spectrum, than what was seen for the fist set of metrics, a result that would be expected for splits with less inherent selection bias. That pattern holds for OPS difference as well, as shown below.
After those dismal results for the first set of metrics, good to see Zack Cozart doing well in the leadoff split (where we finally see a chink in Nolan Arenado’s armor). Russell Martin‘s OPS differences are almost mirror images of each other, for the Leadoff and 2 Outs splits.
You might be wondering just what proportion of PA’s are represented by these splits. Here are those results, together with the composite totals for the two sets of splits, which I’ve termed the “hard” (first set) and “soft” (second set) metrics.
While it wasn’t planned that way, the hard and soft metrics each approximate about half of an “average” player’s PAs. So, just for fun, lets throw the whole bunch of splits into a blender and see what we get. Here’s that leaderboard.
To compile this overall metric, the seven different splits were weighted individually for each player based on the percentage of PAs each split represented for that player. So, the result is really showing how each player did in his own personal context. It’s just a fun list, so not to be taken too seriously, but I probably would have been guessing for a really long time before coming up with the bottom three names for the OPS Diff list. The last column is the average of all %PAs for the seven different splits – Toronto may want to rethink its batting order next year, as Josh Donaldson is probably not the guy you want bringing up the rear in this category.