In 2005, David Ortiz
finished second in the voting for AL MVP
, thanks to a strong
clutch performance. In Player Won-Lost
records, “clutch” performance can be equated to the inter-game win adjustment
which measures the extent to which a player’s context-dependent
percentage differs from his context-neutral
winning percentage. In 2005, David Ortiz’s inter-game win adjustment was +4.1%. This translates into an extra
1.3 net wins earned by David Ortiz due to his “clutchiness”. This accounted for approximately
one-fifth of Ortiz’s Context-Dependent
Wins over Replacement Level
(5.7), which was enough to push him up from
in the American League to
(behind Alex Rodriguez
) in this statistic. Based
on his 2005 season, together with back-to-back extra-inning game-winning hits
in games 4 and 5 of the 2004 ALCS, David Ortiz is regarded by many as the
greatest clutch hitter of his generation
There has been a great deal of research
within the sabermetric community on the
question of whether clutch hitting is, in fact, a skill. The most common way of testing whether things
are “skill” or “luck” in sabermetrics is to look at the extent to which
players’ clutch performances persist
. I constructed persistence equations for inter-game win adjustments
by each of the four basic types of Player won-lost decisions:
batting, baserunning, pitching, and fielding for all of the seasons over which I have estimated Player won-lost records.
The dependent variable in each case was the inter-game win adjustment
within a season on even-numbered plays. The explanatory variable, then, was the
inter-game win adjustment within the season on odd-numbered plays. In addition to inter-game win adjustment, I also included context-neutral win percentage (for even-numbered plays) as a second explanatory variable. I did this to consider the possibility that, perhaps, there might be some correlation between a player’s winning percentage and his inter-game win adjustment – that is, it could be the case that better
hitters are better (or worse) “clutch hitters” in general - perhaps simply due to measurement errors in my calculations. These persistence equations
were estimated using Weighted Least Squares
(WLS), where the observations were weighted by the harmonic mean
of context-neutral player decisions squared. Finally, winning percentages were subtracted from 0.500 so that all three variables (Inter-Game WinAdjEven
, and Inter-Game WinAdjOdd
) are centered around zero, and the constant term was set equal to zero by construction.*
*The weighting scheme and value of the constant term were both tested empirically. The results shown here seemed the most robust to me.
The results are summarized below.
The number n is the number of players over whom the equation was estimated, that is, who accumulated any Player wins and/or losses on both odd- and even-numbered plays within a particular season. The value R2
measures the percentage of variation in the dependent variable (Inter-Game WinAdjEven
) explained by the equation (i.e., explained by WinPctEven
and Inter-Game WinAdjOdd
). The numbers in parentheses below the equation are t-statistics
. T-statistics measure the significance of the regression coefficients, that is, the confidence we have that these coefficients are greater than zero. The greater the t-statistic, the more confident we are that the true value of a particular coefficient is greater than zero. Roughly, if the t-statistic is greater than 2, then we can be at least 95% certain that the true value of b is greater than zero (given that certain statistical assumptions underlying our model hold).
Persistence of Inter-Game Win Adjustments
Batters: n = 55,629, R2 = 0.0005
Inter-Game WinAdjEven = (-0.0041)•(WinPctEven - 0.500) + (0.0451)•Inter-Game WinAdjOdd + 0.0000
Baserunners: n = 55,054, R2 = 0.0328
Inter-Game WinAdjEven = (-0.1201)•(WinPctEven - 0.500) + (0.1023)•Inter-Game WinAdjOdd + 0.0000
Pitchers: n = 33,118, R2 = 0.0053
Inter-Game WinAdjEven = (-0.0217)•(WinPctEven - 0.500) + (0.0651)•Inter-Game WinAdjOdd + 0.0000
Fielders: n = 71,002, R2 = 0.0062
Inter-Game WinAdjEven = (-0.0557)•(WinPctEven - 0.500) + (0.0114)•Inter-Game WinAdjOdd + 0.0000
Is “Clutch” a Significant Skill?
So, is “clutch” performance a persistent skill? For Batting, Baserunning, and Pitching, the answer is a fairly unambiguous “Yes”, there is some small but significant persistence in clutch batting, baserunning, and pitching.
While the "persistence coefficient" on Inter-Game Win Adjustments is highly significant (> 10) for batting, baserunning, and pitching, in all three cases, this “skill” explains less than five percent of the overall variation in inter-game win adjustments in all three cases.
Interestingly, clutch batting, which has generated by far the most discussion in the debate over whether "clutch" is real, is actually the smallest (and least significant) of these three. To the extent that any "clutch skills" exist, clutch pitching and clutch baserunning are both more significant than clutch batting.
Exactly how significant is "clutch batting" as a persistent skill? The persistence coefficient in the batting equation above, 0.0451, suggests that if somebody has an inter-game win adjustment of, say, 0.050 on odd-numbered plays (which would improve them from a 0.500 hitter to a 0.550 hitter, say), we would expect the same player to have an inter-game win adjustment on even-numbered plays of 0.002 (0.05 * 0.0451), enough to improve from a 0.500 batter to a 0.502 batter.
In the case of Fielding, the persistence coefficient is significant, but far less so than in the cases of Batting, Baserunning, and Pitching, suggesting that "clutch fielding" is a less persistent skill.
What about David Ortiz, “the
greatest clutch hitter in the history of the Boston Red Sox”
As noted above, in 2005, David Ortiz
inter-game win adjustment
of +4.1% (1.3 net wins).
How clutch was David Ortiz
outside of 2005? It turns out that in David Ortiz's
regular-season career outside of the 2005 season, he accumulated an inter-game win adjustment
-0.2%, good for
-0.9 net wins.
All articles are written so that they pull data directly from the most recent version of the Player won-lost database. Hence, any numbers cited within these articles should automatically incorporate the most recent update to Player won-lost records. In some cases, however, the accompanying text may have been written based on previous versions of Player won-lost records. I apologize if this results in non-sensical text in any cases.
List of Articles