How Many Games Do Clutch Hitters Really Win?
This was originally posted in 2006 at
This might seem like an easy question to answer. Just check how many times a player drove in the winning run in his team's last AB (like David Ortiz, who seems to do it regularly). But very quickly problems arise. What about the guy who got on base and who scored the winning run? Should he get part of the credit? And what about times when a player fails to get a hit that would have won a game? Should that count as some kind of loss or be subtracted from the games he did win? None of these concerns even deal with whether or not players really can hit better than they normally do when the game is on the line. But I assume, for the sake of argument, that whatever difference in performance in the clutch that we find for players is real. Then I estimate how many wins (or losses) this brings to his team.
First, I start with a somewhat imprecise method and also a somewhat crude measure of clutch performance. The clutch stat is hitting in "close and late" or CL situations. These are situations when the game is in the 7th inning or later and the batting team is leading by one run, tied, or has the potential tying run on base, at bat or on deck. If a player hits better (or worse) in these situations, it will raise (or lower) his team's chance of winning. I use that increase (or decrease) to estimate the increase (or decrease) in wins over a full season due to a given hitter's CL performance.
Here is an example. Suppose a player has an OPS of .750 in non-CL situations (OPS is on-base percentage + slugging percentage). Let's say he raises it by .072 when it's CL (that is a pretty big increase). Spread over nine players in the lineup, it raises his team's CL OPS by .008. But how many wins does that add over the course of a season? For that, I will use a regression generated equation from some previous research I have done called "Does Team Clutch Matter?" A team's winning percentage is estimate by
(1) PCT = 0.501 + 0.918*NONCLOPS + 0.345*CLOPS - 0.845*OPPNONCLOPS - 0.421*OPPCLOPS
where CLOPS means how the team hit in CL situation and OPP refers to how their opponents did. NONCLOPS is how the team hit in non-CL situations. If a team's CL OPS goes up by .008, its winning percentage rises by .345*.008 or .00276. Over 162 games, the team wins .447 more games (.00276*162). I then found all the players from 1987-2001 who had 6,000 or more plate appearances and found out how much their OPS went up or down in CL situations as compared to non-CL situations. Then I estimated how much their team OPS went up and then calculated its impact on team wins over a season. There was one additional step. I raised each player's CL OPS by 4.5% because that is about the normal drop off in those situations (this might be due to not having the platoon advantage or just facing better than average pitchers when the game is on the line). If a player simply maintained his normal OPS in CL situations, he would essentially be a clutch hitter since OPS normally falls.
The table below shows how the players ranked.
Let's take Tino Martinez. His non-CL OPS was .812 while his CL OPS was .889. But increasing that CL OPS by 4.5%, we get .929. This is .117 higher than normal and spread over 9 hitters in a lineup, it raises the team OPS when it is close and late by .013. How many wins does this add? We first multiply .013 by .345. This is .00449. That is the increase in team winning percentage. Then multiplying that by 162 we get about .727. So the best CL performer added .727 wins a year for his team. Also, there were only 5 players who added or subtracted even .5 or more wins (and well over half made a difference, whether positive or negative, of less than .25 wins per season).
So it is rare to find a player whose clutch performance will have much impact. That is something GMs should probably keep in mind when making personnel decisions. It is unlikely that clutch performance would be so different between two players that you would clearly pick one over the other. Even the biggest difference is only 1.286 (between Martinez and Carter). If two players were identical in every way, sure, you pick the better clutch player. But in most cases, there is not much difference. In fact, in only about 20% of the cases, the difference between any two given players is .5 wins or more. If a GM was thinking about using clutch performance to help decide which player to sign, trade or trade for, it is not likely it would matter much. And this assumes that these differences are real, not due to chance.
Close and late is only one situation and it only makes up about 15% of all plate appearances (PAs). So it may not tell the whole clutch story. Some situations might be nearly as clutch or crucial as CL. The 6th inning with the score tied could be clutch. Or maybe your team is leading by only 2 runs in the 7th inning or later. Those cases are obviously less clutch, but it might be wrong to say they have no clutch significance. One way around this is to use every PA in calculating a clutch stat and weight each PA by how much impact its outcome might have on your team's chance of winning. This kind of stat has been done many times, probably first by the Mills brothers around 1970. For my purposes here, I use data from Ed Oswalt's site The Baseball Player Value Analysis System. He refers to this stat as "Player's Win Value" or PWV. So a guy gets more value or more added to his score if he hits a HR with the score tied in the 8th inning than if he hits one in the 9th inning with his team up by 10 runs. So the better you do, the closer and later the game, the better your PWV. Hits with runners on count for more than with none on. All events mattered: hits, strikeouts, sacrifice flies, etc.
To see who were the most clutch hitters, I found the PWV for each player from 1972-2002 with 5,000 or more PAs (284 players). I then converted into PWV per 700 PAs or about a full season. Then I ran a regression in which PWV per PA was the dependent variable and each player's OBP and SLG relative to the league average were the dependent variables (I used relative to the league average because Ed Oswalt adjusted PWV based on the average number of runs scored in each season). The regression finds an equation that tells how much impact OBP and SLG have on PWV (for more details on this, go to my study The Problem With "Total Clutch" Hitting Statistics. I found that those two stats do very good job of explaining PWV. That is, the higher the OBP and SLG, the higher the PWV per PA.
Here is the regression equation
(2) PWV/PA = -.0246 + .000149*OBP + .000097*SLG
Remember, that both OBP and SLG were relative to the league average for each player. Take Barry Bonds, for example. His OBP relative to the league average was 128 (meaning his was 28% above the league average) and for SLG it was 145. Plugging those values into equation (2) gives 0.0084825. That multiplied by 700 leaves about 5.938. His actual PWV per 700 PA was 5.578. So he actually did not quite do as well as expected. His clutch performance in that sense was negative. He did about .36 wins fewer than expected. So he gets a -.36. This was done for all 284 players. What I do here is similar to what Dick Cramer did in his 1977 article "Do Clutch Hitters Exist?" Also, see Is David Ortiz really Mr. Clutch? By Nate Silver
In the table below, I show how the players ranked in terms of how much better or worse they did in actual PWV as compared to what equation (2) predicts. About 82% of the players were within .5 PWV or wins of what equation (2) predicts. Only 5 were more than 1 win better or worse. So at most, the best clutch hitter can add about 1.38 wins a season above what you would expect them to.
There is another table after this one that shows where all the players ranked in actual PWV and predicted PWV. Some players ranked higher in one than the other. If we did cross player comparisons, about 11% of the comparisons would be reversed if we used actual PWV as opposed to predicted PWV. That is, if we compared player A to every other player, in about 11% of the cases actual PWV would reverse who predicted PWV said was the better hitter. That does not seem like much change. In fact, the correlation between where a guy ranked in actual PWV and predicted PWV was about .96 (a perfect correlation is 1-that would happen if there were no change in ranks at all between actual PWV and predicted PWV). Also, about 80% of the players saw their rank change less than 28 places. That does not seem like much change either since 28 places is about 10% of the total number of players.
So even if there are clutch hitters, very few of them increase their team's win total much more than what we could expect by conventional, non-situational stats like OBP and SLG.
Sources: Other sources include the cnn/si site. A few years ago they had career data on CL situations (I am not sure they still have that). I also used some of the STATS, INC "Player Profiles" books.