Are We Overvaluing Power Hitters?
Aaron Judge and Jose Altuve were seemingly neck and neck in MVP voting this year (even if they are neck and belly button when standing next to each other). Judge had the edge in FanGraphs WAR, while Altuve held an edge according to Baseball Reference. Altuve had gaudy batting average and stolen-base totals, while Judge reached the coveted 50-home-run plateau to go along with his jaw-dropping Statcast numbers. Heading into awards season, the American League MVP was hyped as a two-man race that could go either way.
But, then the voting happened, and Jose Altuve got 27 first-place votes to Aaron Judge’s two. There were a lot of reasons for this from storyline, to traditional numbers, to team record. One of the most prominent among them for sabermetric voters was Aaron Judge’s clutch performance. According to the Clutch metric found on this site, he was the least clutch player in baseball this year. Actually, he was the least clutch player in baseball this entire millennium. Wait no, actually, he had the single least clutch season in the history of the metric (since 1972).
Up until now, Clutch has not been shown to have predictive value, even if it is important in deciding things like MVP races which are based on things that have already happened. But, as you may have guessed from the article title, I think there may be evidence to suggest otherwise. Here a list of the least clutch players in history for their entire career according to the clutch metric.
| Rank | Name | Games | HR | BB% | K% | Clutch | WAR |
| 1 | Sammy Sosa | 2354 | 609 | 9.4 % | 23.3 % | -14.67 | 60.1 |
| 2 | Mike Schmidt | 2404 | 548 | 15.0 % | 18.7 % | -13.45 | 106.5 |
| 3 | Lance Parrish | 1988 | 324 | 7.8 % | 19.6 % | -12.90 | 43.4 |
| 4 | Jim Thome | 2543 | 612 | 16.9 % | 24.7 % | -11.66 | 69.0 |
| 5 | Chet Lemon | 1988 | 215 | 9.5 % | 13.0 % | -11.03 | 52.0 |
| 6 | Jermaine Dye | 1763 | 325 | 8.3 % | 18.1 % | -9.67 | 14.5 |
| 7 | Alex Rodriguez | 2784 | 696 | 11.0 % | 18.7 % | -9.53 | 112.9 |
| 8 | Andre Dawson | 2627 | 438 | 5.5 % | 14.0 % | -9.49 | 59.5 |
| 9 | Gary Carter | 2295 | 324 | 9.4 % | 11.1 % | -9.25 | 69.4 |
| 10 | Barry Bonds | 2986 | 762 | 20.3 % | 12.2 % | -9.13 | 164.4 |
This list is populated by a certain type of player: good ones. The difference in WAR between Jermaine Dye and Lance Parrish in 9th place would be a fantastic career for almost anyone. But, more importantly, it is populated by high-strikeout, power-hitting sluggers. Every single player on this list has a double-digit strikeout rate and everyone but Chet Lemon has at least 300 career home runs. The list of the most clutch players in history, on the other hand, is not made up of power hitters.
| Rank | Name | Games | HR | BB% | K% | Clutch | WAR |
| 1 | Tony Gwynn | 2440 | 135 | 7.7 % | 4.2 % | 9.49 | 65.0 |
| 2 | Pete Rose | 2179 | 57 | 10.6 % | 5.8 % | 9.07 | 43.5 |
| 3 | Scott Fletcher | 1612 | 34 | 8.6 % | 9.1 % | 8.61 | 24.9 |
| 4 | Mark McLemore | 1832 | 53 | 12.1 % | 13.6 % | 8.51 | 17.4 |
| 5 | Ichiro Suzuki | 2636 | 117 | 6.0 % | 10.0 % | 8.25 | 58.2 |
| 6 | Dave Parker | 2466 | 339 | 6.7 % | 15.1 % | 7.64 | 41.1 |
| 7 | Omar Vizquel | 2968 | 80 | 8.6 % | 9.0 % | 7.54 | 42.6 |
| 8 | Ozzie Guillen | 1993 | 28 | 3.4 % | 7.2 % | 7.48 | 13.1 |
| 9 | Lance Johnson | 1447 | 34 | 6.1 % | 6.6 % | 6.89 | 26.4 |
| 10 | Jose Lind | 1044 | 9 | 5.4 % | 9.2 % | 6.71 | 3.3 |
| 11 | Mark Grace | 2245 | 173 | 11.6 % | 6.9 % | 6,.58 | 45.5 |
This list is made up of a very different kind of hitter. Tony Gwynn, Pete Rose and Ichiro are perhaps the three most well-known contact hitters of all time. Only three players on this list have double-digit strikeout rates, and only one has 300 career home runs. Chet Lemon, dead last in home runs on the other list, would rank second on this one.
Aaron Judge fits right into the pattern of these lists, as one of four qualified players with a 30% strikeout rate or higher. If you sum the clutch score of the top 10 players in strikeout rate this year, you get -9.05, or nearly one win per player lost due to clutch performance. If you remove Aaron Judge, the sum is a still gaudy total of -5.41.
I charted Strikeout rate against clutch score for all players qualified in 2017, and there is a small but definite trend. Below the chart, you can see the regression equation along with the P value for the coefficient and the R^2.
CLUTCH = 0.593 – 3.736*KPCT
R^2=.045
P=.0101 pic.twitter.com/msrDUblKtJ— Arron Kruse (@arron_kruse) December 22, 2017
Ultimately I don’t have the tools or the time to fully explore this idea, but it would appear that there is an actual relationship here. The effect may be minuscule as the R^2 indicates, but the general trend seems to indicate that clutch players are more contact-oriented. This makes sense, because the most clutch situations in a game happen with men in scoring position, where the difference between a strikeout and a fly out or ground out can be an entire run. Further work needs to be done, but I would not be surprised to find that batted-ball type or walk rate also has an impact. For example, hitters with higher fly-ball rates may be more clutch because, with runners on base, a fly ball avoids a double play with a man on first, and may drive in a run with a man on third. With nobody on base and nobody out, the way a batter gets out does not make a difference. But in clutch situations, all outs are not created equal.
Good article but I think it is more simple than that. Voters got smarter but they still don’t like sub 300 hitters to win the mvp and also not historically high strikeout hitters.
Stanton only was at 280 too but the vast majority of mvps hit above 300.
What stats are those? Pete Rose played like twice that many games
Those stats are post-1972, when the clutch metric became available.
I actually looked at this a while back and found that the relationship between strikeout rate and clutch disappears entirely when one includes batting average in the regression. In other words, it’s batting average, not strikeout rate (unless of course there is some other omitted variable that is primarily responsible).
This is noteworthy because, for one, the correlation between strikeout rate and batting average is not perfect. Some players manage a decent batting average despite a high strikeout rate, etc. It also changes the analysis of whatever mechanism might be going on here, if in fact the relationship is one of cause and effect. Why might batting average, specifically, increase “clutchness?”
I also found the relationship between clutch and ISO to be negative, and the relationships with BB% and wRC+ to be statistically significant but comparatively tiny in terms of the actual effect. To be clear, unlike BB% and wRC+, the strikeout rate coefficient was not only small but also statistically insignificant.
It is also essential to question the clutch metric itself. Does it really measure perfectly what it intends to measure? And if it does, then what might be the resultant implication on how we value players? These questions are not necessarily all that easy to answer. But if “clutch” is in fact flawed in some manner then the above analysis could be meaningless. There definitely could also be something here, though at this point I’m not sure.
I think the question of the validity of the clutch metric is an interesting one which I am not equipped to really explore. I don’t know how it could be skewed unless WPA or LI are systematically biased, but I admittedly don’t completely understand the logic of how it is calculated.
I haven’t explored batting average or WRC+, which may be good indicators, but I was intrigued by the idea of a players batted ball profile being important for clutch so I ran a regression that included walk%, strikeout%, groundball% and pull% to focus on contact type rather than results. Strikeout% and Pull% were significantly negative and Groundball% was significantly positive while walk% was not significant. To be clear, based on this regression clutch players are generally low strikeout, high groundball spray hitters.
I don’t really know what to make of this, but I think this may have to do with the ability to hit tough relievers. Players with high pull percentages and low groundball rates (who should be less clutch based on this regression) are probably more likely to have big platoon splits and have a tougher time catching up to high velocity relievers.
When I get a chance I want to look at platoon splits and clutch, which I have a feeling explain much of these findings. I may be wrong, but I bet high strikeout rates (and also low batting averages) are highly correlated with platoon splits.
I think balls in play are probably the driver here that binds K% and batting average. Good players with few Ks almost by definition have high batting average, if you have few Ks AND power then you’re an elite hitter with a high average, but that profile isn’t particularly common (young Albert Pujols maybe? Hank Aaron? Miguel Cabrera? Edgar Martinez?). What hitters who don’t strike out do in clutch situations with a high leverage index is they put the ball in play and advance their team’s win probability in a positive direction even when they fail. Failure for Ichiro is generally a groundout to second base. With a runner on third and one out down a run, that groundout to second base just tied the game. For Judge by contrast, failure is a strikeout an overwhelming percentage of the time. So when Judge fails (and he will 65% of the time) the runner stays at 3rd and there are two outs, Judge’s failure did nothing to help his team win, while Ichiro’s failure did. I think the precise statistical mix in a regression model for this is probably complicated, but I suspect that’s the basic formula, ability to put the ball in play advances win probability even in failure.
A methodological note: I would try z-scoring the strikeout rates (and other stats) to their year to make Ichiro’s K% more comparable to Pete Rose’.
Nice work, really interesting observation that should open up some further research.
I think that an interesting question to explore is players who run 20-25% or more strikeout rates, but also high BABIPs allowing them to still have reasonably high averages. This year the poster boy would be Judge, who was clearly not clutch indicating to me that strikeouts are more important than batting average. This makes sense with the logic Corey2 is following.
What an excellent comment! (Have you considered applying for FG’s new writer position?)
The broad point I take from your comment is that the more proxies we use in a study — “high-SO%” as a proxy for “high-power,” “low-SO%” for “high-BA,” “Clutch” rating for whatever it is we really mean by “clutch” — the stronger the found correlation must be to have significance, and the greater the chance that it’s all just noise.
Something that the “clutch” stat ignores is how many batters the pitcher has faced leading up to the observed at bat. It doesn’t factor into WPA or Leverage Index. I would be interested in seeing how the league fairs in clutch situations when they are the first batter compared to if they are not the first batter facing the pitcher. The first batter is at a disadvantage for at least these two reasons:
1) The pitcher probably has favorable matchups against the first batter they face. I presume this relationship doesn’t hold up terribly well after the first batter.
2) The batter hasn’t been able to see observe the pitcher’s game plan, what’s working/not working that night, etc.
This would also explain why all the least “clutch” hitters are power hitters: because they probably experience pitching changes preceding their at-bats at a higher frequency, especially in clutch situations! If this is the case, then the “clutch” stat is indeed flawed as @wilmerr suspected. Perhaps controlling for this “first batter” effect could lead to a more useful “clutch” statistic….
A point about those top-10 and bottom-10 “clutch” ratings — When you choose a sample that small, and still find that two or three players from each list would fit the perceived trend of the other list (SO% in this case), then those lists don’t really show anything, and are potentially misleading.
That’s especially true when the lists don’t account for context, i.e., the league SO% during those careers. For example, A-Rod’s 18.7% and Dye’s 18.1% are both slightly better than Parker’s 15.1%, compared to the non-pitcher SO% during their career span. Bonds and Carter, despite “double-digit” SO%, were both contact hitters for their day, respectively 3.4% and 2.2% below the league rate.
The larger study is more interesting.
Btw I’m a believer of overall value. If ISO and walks are the same lower K hitter is better but those components are just as important,albeit walks usually are the smallest factor unless they are extreme.
But if you have two power hitters with decent walk rates lower K is better.
I don’t like mid 00s the Ks don’t matter analysis either because always was juan pierre vs adam Dunn. Of course dunn wins that because he was historically great in iso and walks while pierre vs a zero in both.
But on the other hand dunns iso and bb was the same as trout but the hitting clearly wasn’t. So Ks clearly matter they can just be compensated by other stuff.
I don’t care about Ks if production is the same though.
To what extent are IBB’s responsible for this?
My gut tells me they play a role. Most IBB’s occur in high leverage situations but result is a negligible change in WPA. Meaning for every IBB a batters WPA and WPA/LI remain about the same, but their pLI goes up.
Since Clutch equals [WPA]/[pLI] – [WPA/LI] this implies that every IBB reduces a player’s Clutch Stat. Yet they have literally no control over this. That’s why IBB’s are removed from wRC+ and mostly removed from WAR. Hell, you can’t even Miggy your way out of an IBB these days…
My gut tells me these high power players are more likely to get IBB’d and this biases their clutch scores down. I’m not sure this would explain the entire discrepancy, but I bet it is a factor.
Someone test this!