Clutch hitting is back in fashion in the baseball research community. For years, many of us had looked in vain for the existence of some persistent clutch hitting ability and, failing to find it, had come to the conclusion that such an ability must not exist. The pioneer of this approach was Dick Cramer, who wrote an article on this subject in the 1977 Baseball Research Journal, but many of us have done similar studies over the years. First, you determine who performs better than normally one year in "clutch situations" (and the definitions of these situations change from study to study) and then you see if these players have a tendency to repeat their performance the next season. They don't, which has led a generation of baseball researchers to roll their eyes whenever announcers start rhapsodizing about Joe Blow's ability to come through when it counts.
In "Underestimating the Fog", an article in the 2005 Baseball Research Journal, Bill James argues that we were wrong to think that such an approach "proved" anything. There is so much random noise inherent in this method, so much "Fog", that we shouldn't expect to see anything when looking for clutch ability in this manner. Well, I might get around to testing this hypothesis at some point, but for now I thought I'd take a different tack. I thought it might be interesting to compare a player's ability in both clutch and non-clutch situations over the course of his career. So I'm not really looking for persistence in results from one year to the next, but rather I'll be looking for results that are not what we'd expect to see if there were only random forces at work. Hopefully, dealing with much larger groups of at-bats will help to thin out the fog somewhat.
The first problem facing anyone undertaking a study like this is that we don't really know what clutch means. Or rather it seems to mean something different whenever it's used, depending upon the point we are trying to make. Ted Williams was once accused of not being a clutch player based upon his performance in a handful of games, selected both because they had a significant impact on his teams chances to win a World Championship and because he performed relatively poorly in them. Games in the middle of a tight pennant race weren't clutch, only a couple at the very end of a few seasons. Others have defined terms like "Late Inning Pressure Situations" to identify players who perform well or poorly in a handful of at-bats near the end of close games.
One lazy way out of this problem (hint: it's the one I'll be taking) is to define a clutch situation as an at-bat with runners in scoring position. In a sense, this is nonsense: a lead-off hitter in the late innings of a tie game is usually a much more clutch situation than a batter at the plate with a runner on second and a 15-2 lead. Still, it's often what we mean by "clutch". I don't know about you, but when someone talks about how well this or that player has hit in the clutch, I'll usually test the statement by checking to see how the man has hit with men in scoring position. These at-bats may not all be the most pressure-packed of the season, but they probably come close enough for our purposes.
There is a problem with taking this approach, however: batters do not hit equally well in all situations. Here is a breakdown of batter stats for each of the 24 game situations from 1960 to 2004. Note: this data is not complete for these years and for the purposes of this article, we will be ignoring any games for which we are missing play-by-play information. For a list of these missing games by league and team, please see the article on the Value Added method.
FST Out AB H 2B 3B HR BB IBB HBP SF OBP SLG OPS
--- 0 1532622 397872 70352 10052 41281 124600 6 9597 0 .319 .399 .719
--- 1 1089667 272485 47627 6764 26633 93420 24 6859 0 .313 .380 .693
--- 2 855785 211439 37083 4936 21779 82275 91 5645 0 .317 .378 .696
x-- 0 321722 91402 14907 1921 8458 21371 3 2354 0 .333 .421 .755
x-- 1 396177 111618 18431 2313 10860 27732 36 2658 0 .333 .422 .755
x-- 2 400916 103338 17374 2518 10836 32030 131 2531 0 .317 .395 .711
-x- 0 98402 24968 4104 654 1967 10171 621 862 6 .329 .369 .698
-x- 1 184504 45058 8137 1278 4269 30453 8618 1494 35 .356 .372 .727
-x- 2 218447 52285 9214 1509 5045 45860 17097 1757 0 .375 .365 .740
xx- 0 75884 20950 3420 411 2045 5327 16 660 7 .329 .413 .742
xx- 1 154245 40316 7178 961 4073 11667 52 1175 26 .318 .400 .718
xx- 2 199902 47270 8384 1422 4753 17909 168 1540 0 .304 .364 .668
--x 0 17713 5348 953 195 388 2276 181 178 2835 .339 .443 .783
--x 1 53374 17321 3045 514 1356 9344 2363 663 9639 .374 .477 .851
--x 2 90284 21777 3907 628 2075 16646 3975 799 0 .364 .367 .731
x-x 0 29489 9865 1653 192 865 2459 340 309 5351 .336 .492 .828
x-x 1 60054 19653 3278 460 1728 5483 786 598 11295 .332 .483 .816
x-x 2 91638 23143 4109 667 2179 8579 472 773 0 .322 .383 .705
-xx 0 16965 5153 920 165 437 3173 1304 203 2882 .367 .455 .822
-xx 1 36700 10873 1957 319 809 16311 11963 452 6302 .462 .433 .896
-xx 2 49866 11563 2089 347 1062 14106 7438 476 0 .406 .352 .757
xxx 0 19001 6248 1064 128 599 1367 0 216 3466 .326 .493 .818
xxx 1 48260 14911 2646 384 1406 3396 0 503 8762 .309 .467 .776
xxx 2 68338 16292 2938 553 1710 5491 1 529 0 .300 .373 .673
Where: AB - at bats
H - hits
2B - doubles
3B - triples
HB - home runs
BB - walks
IBB - intentional walks
HBP - hit by pitch
SF - sacrifice flies
OBP - on-base percentage (H + BB + hit by pitch) / (AB + BB + hit by pitch + SF)
SLG - slugging percentage (H + 2B + (2 * 3B) + (3 * HR)) / AB
OPS - on-base plus slugging percentage
I thought it might be easier to see some trends if I compressed the data in a few ways. First, the aggregate performance by outs:
Out AB H 2B 3B HR BB IBB HBP SF OBP SLG OPS 0 2111798 561806 97373 13718 56040 170744 2471 14379 14547 .323 .405 .728 1 2022981 532235 92299 12993 51134 197806 23843 14402 36059 .328 .397 .725 2 1975176 487107 85098 12580 49439 222896 29373 14050 0 .327 .378 .705
And by men on:
FST AB H 2B 3B HR BB IBB HBP SF OBP SLG OPS --- 3478074 881796 155062 21752 89693 300295 121 22101 0 .317 .388 .705 x-- 1118815 306358 50712 6752 30154 81133 170 7543 0 .327 .412 .739 -x- 501353 122311 21455 3441 11281 86484 26336 4113 41 .360 .368 .728 xx- 430031 108536 18982 2794 10871 34903 236 3375 33 .313 .385 .699 --x 161371 44446 7905 1337 3819 28266 6519 1640 12474 .365 .412 .777 x-x 181181 52661 9040 1319 4772 16521 1598 1680 16646 .328 .434 .762 -xx 103531 27589 4966 831 2308 33590 20705 1131 9184 .423 .397 .820 xxx 135599 37451 6648 1065 3715 10254 1 1248 12228 .307 .423 .730
And with runners in and out of scoring position:
AB H 2B 3B HR BB IBB HBP SF OBP SLG OPS
Not RISP 4596889 1188154 205774 28504 119847 381428 291 29644 0 .319 .394 .713
RISP 1513066 392994 68996 10787 36766 210018 55396 13187 50606 .345 .392 .737
With men in scoring position, batters have just about the same slugging percentage and a higher on-base percentage than they do in their other at-bats. So they hit somewhat better in these situations. Except, of course, that they don't.
There are two deceptive things about comparing situational statistics in this manner. First of all, sacrifice flies only occur with a lead runner on third or (very rarely) on second. These are about the same as run-scoring ground outs and the decision not to count these as at-bats is a mistake first adopted in 1889 (when, who knows, perhaps it made sense), a mistake which has gone in and out of fashion over the years. I'm not sure how many of the sacrifice flies hit from 1960 to 2004 were actually struck with a sacrificial intent, but I'd be surprised if all but a handful of these were merely failed attempts at getting a hit. So the first thing we're going to do in our study is to treat sacrifice flies as at-bats.
The second thing that's misleading is walks. Walk rates vary quite a bit from situation to situation. With men on second and third and one out, a batter is nearly five times more likely to get a walk than he is with the bases loaded and one out. Most of this difference is due to intentional walks, which are easy to remove (and we will), but large differences in walk rates still remain even without them. Not only are these differences significant, but they vary quite a bit from batter to batter. The reason for this, of course, is that in some situations (most frequently with first base open) some batters are "pitched around" and sent to first via a non-intentional intentional walk. The decision to pitch to a batter in this manner is largely made based upon the his reputation, the relative handedness of the pitcher and batter, and often simply because of a manager's hunch.
As a result, in addition to treating sacrifice flies as outs in this study, I'm also going to ignore walks. This is not to say that walks aren't important, or are not in many instances the outcome of "clutch" at-bats, only that it is difficult to level the playing field with respect to walks and I don't want a batter's reputation inflating (or deflating) his apparent performance in clutch situations.
And finally, I'm also going to remove hit by pitch. Not that it can't be clutch (and painful) to take one for the team, but I'd like to concentrate on the hitting aspect of batting, rather than the getting hit.
With these changes made, here are the new situational breakdowns:
FST Out AB H 2B 3B HR AVG SLG BPS
--- 0 1532622 397872 70352 10052 41281 .260 .399 .659
--- 1 1089667 272485 47627 6764 26633 .250 .380 .630
--- 2 855785 211439 37083 4936 21779 .247 .378 .625
x-- 0 321722 91402 14907 1921 8458 .284 .421 .705
x-- 1 396177 111618 18431 2313 10860 .282 .422 .704
x-- 2 400916 103338 17374 2518 10836 .258 .395 .652
-x- 0 98408 24968 4104 654 1967 .254 .369 .622
-x- 1 184539 45058 8137 1278 4269 .244 .372 .616
-x- 2 218447 52285 9214 1509 5045 .239 .365 .604
xx- 0 75891 20950 3420 411 2045 .276 .413 .689
xx- 1 154271 40316 7178 961 4073 .261 .400 .661
xx- 2 199902 47270 8384 1422 4753 .236 .364 .600
--x 0 20548 5348 953 195 388 .260 .382 .643
--x 1 63013 17321 3045 514 1356 .275 .404 .679
--x 2 90284 21777 3907 628 2075 .241 .367 .609
x-x 0 34840 9865 1653 192 865 .283 .416 .699
x-x 1 71349 19653 3278 460 1728 .275 .407 .682
x-x 2 91638 23143 4109 667 2179 .253 .383 .636
-xx 0 19847 5153 920 165 437 .260 .389 .648
-xx 1 43002 10873 1957 319 809 .253 .370 .622
-xx 2 49866 11563 2089 347 1062 .232 .352 .583
xxx 0 22467 6248 1064 128 599 .278 .417 .695
xxx 1 57022 14911 2646 384 1406 .261 .395 .657
xxx 2 68338 16292 2938 553 1710 .238 .373 .611
Where: AB - at bats plus sacrifice flies
AVG - batting average (H / (AB + SF))
SLG - slugging percentage (H + 2B + (2 * 3B) + (3 * HR)) / (AB + SF)
BPS - batting average plus slugging percentage
The performance by outs:
Out AB H 2B 3B HR AVG SLG BPS 0 2126345 561806 97373 13718 56040 .264 .402 .666 1 2059040 532235 92299 12993 51134 .258 .390 .649 2 1975176 487107 85098 12580 49439 .247 .378 .624
And by men on:
FST Out AB H 2B 3B HR AVG SLG BPS --- 3478074 881796 155062 21752 89693 .254 .388 .642 x-- 1118815 306358 50712 6752 30154 .274 .412 .686 -x- 501394 122311 21455 3441 11281 .244 .368 .612 xx- 430064 108536 18982 2794 10871 .252 .385 .638 --x 173845 44446 7905 1337 3819 .256 .382 .638 x-x 197827 52661 9040 1319 4772 .266 .398 .664 -xx 112715 27589 4966 831 2308 .245 .365 .610 xxx 147827 37451 6648 1065 3715 .253 .388 .641
And with runners in and out of scoring position:
AB H 2B 3B HR AVG SLG BPS
Not RISP 4596889 1188154 205774 28504 119847 .258 .394 .652
RISP 1563672 392994 68996 10787 36766 .251 .380 .631
So with these adjustments, it's clear that batters actually hit worse with runners in scoring than they do otherwise.
Since batters hit best with a man on first, I thought it might be interesting to see how right-handed and left-handed hitters do in these situations.
Righties:
FST Out AB H 2B 3B HR BAVG SLG BPS --- 2090674 523625 92577 11626 54934 .250 .385 .635 x-- 677801 181155 30450 3739 18246 .267 .404 .671 -x- 300296 72633 12563 1876 6802 .242 .364 .606 xx- 263504 65779 11630 1539 6715 .250 .382 .632 --x 105321 26474 4667 733 2358 .251 .377 .628 x-x 120675 31523 5479 714 2932 .261 .391 .653 -xx 70013 17130 3107 482 1490 .245 .367 .611 xxx 92149 22948 4151 579 2244 .249 .380 .629
Lefties:
FST Out AB H 2B 3B HR BAVG SLG BPS --- 1387400 358171 62485 10126 34759 .258 .393 .651 x-- 441014 125203 20262 3013 11908 .284 .425 .708 -x- 201098 49678 8892 1565 4479 .247 .374 .621 xx- 166560 42757 7352 1255 4156 .257 .391 .647 --x 68524 17972 3238 604 1461 .262 .391 .653 x-x 77152 21138 3561 605 1840 .274 .407 .681 -xx 42702 10459 1859 349 818 .245 .362 .607 xxx 55678 14503 2497 486 1471 .260 .402 .663
As expected, left-handed hitters are able to take more advantage of the man on first situation, since holding the runner on opens up a hole on the right side.
Since I want to do away with as much of the fog as possible in this study, I'm going to only consider those players with at least 3000 at-bats (including sacrifice flies). This group of players should be significantly better hitters than the ones with less than 3000 at-bats for two reasons. First of all, requiring a significant number of at-bats will eliminate all pitchers from the mix. And secondly, I'm assuming that batters with longer careers are better than those with shorter careers.
Before going much further, then, I wanted to see if my target group showed a similar decline with runners in scoring position. Here are the statistics for the two groups of batters:
1-2999 AB H 2B 3B HR BAVG SLG BPS Not RISP 1748765 414965 70688 9459 35513 .237 .349 .587 RISP 588724 135614 23479 3614 10466 .230 .336 .566 >=3000 AB H 2B 3B HR BAVG SLG BPS Not RISP 2848124 773189 135086 19045 84334 .271 .421 .693 RISP 974948 257380 45517 7173 26300 .264 .406 .670
The percentage declines were about the same for the two groups. This isn't what I would have expected if clutch hitting is a talent that some players have and others don't. I would have assumed that the more talented group of hitters would have done better. Of course, there's no reason why talent and clutch ability have to go hand in hand.
Still, I was surprised that batters, both good and bad, hit worse with men in scoring position. Much of this is due to the big spike in performance that occurs when there's a man on first. Another reason is the presence of force-outs and fielder choices that aren't available with no one on.
Still, the single worst hitting situation is second and third with two outs. One reason for this could be a selection bias: good hitters are often walked in these situations. As a result, the quality of hitters batting at these times is lower than at others. I thought this might be something we could look at. Here are the average BPSs of the hitters up in each of the 24 game situations:
MenOn Number of Outs FST 0 1 2 --- .646 .640 .647 x-- .655 .659 .647 -x- .656 .651 .648 xx- .664 .652 .634 --x .648 .662 .651 x-x .664 .658 .642 -xx .656 .643 .636 xxx .651 .641 .630
There's something to this theory, as the quality of hitters at the plate with men on second and third and two out is the among the worst.
Later on, we will explore some other possible explanations for the dropoff in performance with runners in scoring position.
Enough talk. So who were the greatest clutch hitters from 1960 to 2004, the players who were able to raise the level of their game when it mattered most (or at least when runners were on second or third)? Here they are:
- NO-RISP - --- RISP --
Name B AB AB BPS AB BPS DIFF
Bill Spiers L 3430 2548 .607 882 .722 .115
Mike Sweeney R 3760 2673 .764 1087 .867 .103
Pat Tabler R 3948 2815 .626 1133 .725 .099
Jose Valentin B 4882 3678 .666 1204 .765 .099
Wayne Garrett L 3308 2557 .557 751 .643 .087
Sandy Alomar B 4748 3831 .519 917 .592 .073
Tony Fernandez B 7972 6100 .665 1872 .736 .071
Rennie Stennett R 4554 3520 .612 1034 .682 .070
Joe Girardi R 4150 3117 .596 1033 .666 .070
Rick Miller L 3910 2991 .599 919 .668 .069
Larry Parrish R 6848 5075 .679 1773 .747 .068
Carlos Beltran B 3508 2587 .748 921 .815 .068
Tony Taylor R 6587 5304 .597 1283 .663 .067
Scott Fletcher R 5294 4014 .583 1280 .649 .066
Johnny Edwards L 4585 3471 .575 1114 .638 .063
Brent Mayne L 3652 2701 .590 951 .649 .059
Troy O'Leary L 4043 2917 .700 1126 .758 .058
Miguel Tejada R 4277 3115 .726 1162 .782 .057
Orlando Merced B 4028 2883 .682 1145 .738 .056
Henry Rodriguez L 3054 2243 .719 811 .776 .056
Edgardo Alfonzo R 4981 3700 .702 1281 .758 .056
Where: DIFF - BPS with runners in scoring position minus BPS without.
Just who I expected to see: Bill Spiers, Wayne Garrett, Rennie Stennett, Rick Miller....
And the other side of the coin:
- NO-RISP - --- RISP --
Name B AB AB BPS AB BPS DIFF
Richard Hidalgo R 3193 2252 .821 941 .614 -.207
Jermaine Dye R 3863 2750 .772 1113 .611 -.161
Al Martin L 4269 3233 .753 1036 .598 -.155
Larry Brown R 3472 2729 .574 743 .423 -.151
Earl Williams R 3058 2186 .701 872 .554 -.147
Hal Morris L 4037 2952 .769 1085 .624 -.145
Jim Edmonds L 5139 3739 .868 1400 .727 -.141
Jim Morrison R 3414 2494 .708 920 .570 -.139
Dean Palmer R 4953 3586 .753 1367 .617 -.137
Pete Ward L 3088 2253 .690 835 .553 -.136
Lee Maye L 3849 2978 .708 871 .577 -.130
Mark Kotsay L 3756 2898 .734 858 .611 -.123
Don Slaught R 4101 3020 .721 1081 .599 -.122
Pat Borders R 3183 2364 .662 819 .541 -.121
Todd Walker L 3704 2772 .749 932 .630 -.119
Reggie Smith B 7119 5306 .797 1813 .680 -.117
Shawn Green L 5566 4102 .814 1464 .700 -.114
Tony Bernazard B 3735 2808 .671 927 .558 -.114
Kevin Young R 3944 2814 .721 1130 .608 -.113
Phil Bradley R 3716 2836 .730 880 .618 -.112
Warren Cromartie L 3958 3022 .705 936 .593 -.112
Lee Lacy R 4582 3502 .718 1080 .606 -.112
For lack of a better term (and so I don't have to keep writing "the difference between a batter's BPS with and without runners in scoring position"), I'm going to call this difference ("DIFF" in the charts above) Clutch Percentage. I know it's not really "Clutch" and not really a "Percentage", but it's the best I could come up with.
The poor Clutch Percentages are more extreme than the positive ones, partly because the median of the group is not zero, but rather -.027.1
I'm not sure what I expected to see here. I doubt that if I had presented these two lists of players to you and told you that one was a list of the best clutch hitters and the other the worst, you could have figured out which was which.
One of the things that bothers me about the last list is that 12 of the 20 players on it have less than 1000 at-bats with runners in scoring position. Of the 727 players in the study, only a little more than 30% (222) fell into that category. If the differences we're looking at were caused more by chance than talent, you'd expect to see players with small sample sizes at the two extremes.
The raw data for the players in the study is here.
Could these results have been random? The way I usually approach this kind of question is with brute force. Rather than attempting to finesse the issue with mathematics, I run over it with simulation. My approach this time is perhaps best shown by example.
In the games we have, Vada Pinson had a runner on second or third in 2114 of his 8954 at-bats, or 23.6096%. So to simulate his random career, I generated 8954 random numbers (one for each at-bat) between 0 and 1. If the number was less than .236096, I counted it as an at-bat with runners in scoring position. When I was done, I had randomly selected around 2114 at-bats that I'm considering to be clutch. Using these two pools of at-bats (the ones selected by this process and the ones not selected), I computed his simulated Clutch Percentage.
One problem with this approach is that we already know that the data is not random. Players on average hit worse (in terms of BPS, 27 points worse) with men in scoring position. Our random tests will not reflect this. Since we're doing these simulations to see how much random variation there will be in the data, this problem might not be fatal, but it does complicate things. For example, we will want to compare the amount of spread in both the real and simulated data. This spread will be centered around -.027 in the real run and .000 in the simulated runs.
I did 1000 of these simulations. What did I find out? Well, there was nothing terribly unusual in the spread of the real data. In the random run the average distance from each player's Clutch Percentage and the expected value varied from a low of .2872 to a high of .3512. The actual values differed by .3314, which was a little high but nothing out of the ordinary (117th place out of 1001). In addition to looking at the spread, I also broke the range of values into 20 groups (each .015 wide except for the first and last) and saw if the distribution of the players were similar in both the real and the simulated worlds. Note that the mid-point in the two worlds is different, since the expected Clutch Percentage is -.027 for the actual values and .000 for the simulated ones. In the chart below, then, group A contains the count of players from .000 to .015 over the expected value, B contains the count of players from .015 to .030, and so on. Not too surprisingly, -A contains the count of players from .000 to .015 below the expected value, -B contains the count of players from .015 to .030 below, and so on.
-J -I -H -G -F -E -D -C -B -A A B C D E F G H I J
Real 1 3 6 5 15 21 46 76 77 115 105 88 69 41 31 13 10 1 3 1
Fake 1 2 3 7 14 26 45 70 94 109 109 92 68 45 26 14 7 3 2 1
Our real distribution is very similar to the average of the fake ones. But it is important to note that this doesn't prove anything. While a very different spread and distribution could be used to demonstrate that Clutch Percentage is not random, the fact that these results are similar is not evidence that only random forces are at work here.
This section will explore some factors that might complicate things, causing batters to hit worse (or better) with runners in scoring position.
The first thing that occurred to me is that batters might be facing a platoon disadvantage more often with runners in scoring position than they might otherwise. To test this, I looked at batters who hit right, left and from both sides of the plate and determined how well they did against right and left pitchers. I next computed what types of pitchers they faced both with and without runners in scoring position and used that information to generate an expected BPS (batting average plus slugging percentage) given the mix of pitchers they saw in both situations. Here's the data:
- NO RISP - --- RISP --
Count AB BPS BPSvR BPSvL PLAT% ExBPS PLAT% ExBPS
Right 402 5247 .679 .662 .714 36.8 .679 34.1 .677
Left 219 5298 .693 .718 .612 75.9 .694 71.8 .689
Both 106 5220 .652 .651 .647 50.9 .652 50.0 .651
Total 727 5259 .679 .677 .6732 50.6 .679 47.8 .677
Where: Count - the number of batter included in sample
AB - the average number of at-bats in group
BPS - the average overall BPS
BPSvR - the average BPS against right-handed pitchers
BPSvL - the average BPS against left-handed pitchers
PLAT% - the percentage of times having the platoon advantage
ExBPS - the expected BPS given the mix of pitchers faced
This table presents a lot of unfamiliar information so it might be a good idea to go over a sample line. There are 402 right-handed hitters in our study. The average righty in the study had 5247 at-bats and an overall BPS of .679. As expected, he hit lefties better than the righties (.714 to .662), but had a platoon advantage only 36.8% of the time with no runners in scoring position. Now, I didn't assume that all right-handed hitters had a platoon advantage against left-handed pitchers. Instead, I determined which type of pitcher each batter performed better against over the course of his career. Most of the time, hitters did better against pitchers who threw from the other side, but not always. Given the percentage of pitchers of each type our hitters faced with no one in scoring position, and how they hit against these pitchers, righty hitters had an expected BPS of .679 in these situations. When runners were on second or third, the platoon advantage and BPS drop slightly to 34.1% and .677 respectively.
You should not assume from the chart above that switch-hitters had no platoon advantage or disadvantage. The reason why they hit almost the same against both righties (.651) and lefties (.647) is that the platoon differentials of switch-hitters tended to cancel each other out. To illustrate this, here are the players with 3000 or more at-bats with the greatest platoon differentials:
Name B P BPSvR BPSvL Diff
Rob Deer R vs L .599 .801 .202
Adrian Beltre R vs R .750 .667 .083
Tom Goodwin L vs L .599 .621 .021
Randy Bush L vs R .666 .369 .298
Dave Hollins B vs L .608 .806 .198
Wally Backman B vs R .652 .365 .287
Where: B - the handedness of the batter (R - right, L- left, B- both)
P - the handedness of the pitcher (R - right, L- left)
BPSvR - the batting average plus slugging percentage against right-handed pitchers
BPSvL - the batting average plus slugging percentage against left-handed pitchers
Diff - the difference
The average platoon differential is greatest for the lefties in our study (.107), and just about the same for right-handed hitters (.058) and switch-hitters (.057). People often assume that just because a batter hits from both sides of the plate that he hits equally well from each side. This is not the case, although it isn't always obvious which side is their weakest (unless it's someone like Wally Backman).
Platoon advantages by themselves are not sufficient to explain the fact that hitters tend to perform worse with runners in scoring position. The average dropoff is about 23 points of BPS (.693 to .670), and the expected dropoff due to platoon disadvantages is only 2 points for right-handed hitters, 5 points for lefties, and 1 point for switch-hitters. Of course, this effect is different for each player. Frank Howard, for example, punished lefties so much that he seldom faced them with men in scoring position, causing him to have a platoon disadvantage of 17 points. Tony Batista, on the other hand, is a right-handed hitter who has hit righties better than lefties over the course of his career. As a results, he has a platoon advantage of 3 points with runners in scoring position.
Another factor we might want to take into account is that the quality of pitchers is often worse in these situations. This makes sense. After all, when you're up with men in scoring position, you are usually facing the pitcher who permitted those runners to reach base, something that happens a lot more frequently with a Jaime Navarro on the mound than a Roger Clemens. To determine how much worse they are, for each at-bat by one of the batters in our study, I calculated the pitcher's opponents BPS, taking into account the handedness of the batter. I found that the the average pitcher when runners are in scoring position is about 3 to 4 points worse (in BPS) than those on the mound when there aren't. Not a big deal and a result that seems to balance out the platoon disadvantage, except, as with the platoon disadvantage, there are differences from player to player. The most extreme cases among the players in our study are Hal Morris, who has faced pitchers 17 points worse with runners in scoring position, and Larry Walker, who has faced pitchers 11 points better. All in all, I think it's a good thing to check before anointing someone either a great or a poor clutch hitter.
The last thing we want to look at is any possible park effects. After all, there are more runners in scoring position in good hitting parks. So I calculated the average park factor for the two situations and, to make a long story even longer, here's what I found:
AB PFact
NO RISP 2848124 1.0048
RISP 974948 1.0084
Where: PFact - the average park factor.
Since there are more at-bats in a typical game in a hitter's park then there are in a pitcher's park, it's not too surprising that the average park factor in both groups would be greater than 1. Note that the advantage with runners in scoring position is slight. Still, this is not insignificant for all players. The two extremes:
NO-RISP RISP
Name PFact PFact PRat
Lee Maye .980 .970 .990
Todd Helton 1.214 1.259 1.037
Where: PRat - the RISP park factor divided by the NO-RISP park factor.
It is perhaps not too surprising that a member of the Colorado Rockies got the biggest park factor boost with runners in scoring position.
I wanted to take one last look at the players at the top and bottom of our lists, this time with their platoon, strength of opposition and park factors included.
- NO-RISP - -- RISP --
Name B AB AB BPS AB BPS DIFF PlatF OppF PRat
Bill Spiers L 3430 2548 .607 882 .722 .115 -.006 .003 .999
Mike Sweeney R 3760 2673 .764 1087 .867 .103 .003 .013 .998
Pat Tabler R 3948 2815 .626 1133 .725 .099 -.001 .008 1.005
Jose Valentin B 4882 3678 .666 1204 .765 .099 -.001 .013 1.007
Wayne Garrett L 3308 2557 .557 751 .643 .087 -.004 .008 .994
Sandy Alomar B 4748 3831 .519 917 .592 .073 -.000 .005 1.008
Tony Fernandez B 7972 6100 .665 1872 .736 .071 .000 .008 1.003
Rennie Stennett R 4554 3520 .612 1034 .682 .070 -.004 .011 1.004
Joe Girardi R 4150 3117 .596 1033 .666 .070 .001 .005 1.011
Rick Miller L 3910 2991 .599 919 .668 .069 -.010 .012 1.002
Larry Parrish R 6848 5075 .679 1773 .747 .068 -.001 .003 1.001
Carlos Beltran B 3508 2587 .748 921 .815 .068 -.001 .006 1.008
Tony Taylor R 6587 5304 .597 1283 .663 .067 -.001 .007 1.007
Scott Fletcher R 5294 4014 .583 1280 .649 .066 -.003 -.005 1.001
Johnny Edwards L 4585 3471 .575 1114 .638 .063 -.003 .010 1.004
Brent Mayne L 3652 2701 .590 951 .649 .059 -.002 .012 1.024
Troy O'Leary L 4043 2917 .700 1126 .758 .058 -.005 -.001 1.003
Miguel Tejada R 4277 3115 .726 1162 .782 .057 -.000 .008 1.002
Orlando Merced B 4028 2883 .682 1145 .738 .056 -.001 .009 1.006
Henry Rodriguez L 3054 2243 .719 811 .776 .056 -.004 .003 1.007
Edgardo Alfonzo R 4981 3700 .702 1281 .758 .056 -.000 .008 1.010
Where: PlatF - is the platoon advantage or disadvantage with runners in scoring position
OppF - is the expected BPS increase or decrease based upon the quality of pitchers
Some of these hitters (Mike Sweeney, Jose Valentin, Rennie Stennett and Johnny Edwards) got a bigger than average boost by facing weaker pitchers with runners in scoring position, and Brent Mayne had the advantage of both facing weaker than normal pitching and hitting in these situations in friendlier parks. My feeling is that Mayne would not have been on the list without this help.
The bottom list revisited:
- NO-RISP - -- RISP --
Name B AB AB BPS AB BPS DIFF PlatF OppF PRat
Richard Hidalgo R 3193 2252 .821 941 .614 -.207 .000 .006 .996
Jermaine Dye R 3863 2750 .772 1113 .611 -.161 -.001 .005 1.003
Al Martin L 4269 3233 .753 1036 .598 -.155 -.017 .002 1.011
Larry Brown R 3472 2729 .574 743 .423 -.151 -.004 .005 1.001
Earl Williams R 3058 2186 .701 872 .554 -.147 .000 .002 1.000
Hal Morris L 4037 2952 .769 1085 .624 -.145 -.002 .017 1.006
Jim Edmonds L 5139 3739 .868 1400 .727 -.141 -.008 .005 1.001
Jim Morrison R 3414 2494 .708 920 .570 -.139 -.001 .004 1.000
Dean Palmer R 4953 3586 .753 1367 .617 -.137 -.006 -.001 1.003
Pete Ward L 3088 2253 .690 835 .553 -.136 -.002 .013 1.005
Lee Maye L 3849 2978 .708 871 .577 -.130 -.006 -.005 .990
Mark Kotsay L 3756 2898 .734 858 .611 -.123 -.001 .006 1.006
Don Slaught R 4101 3020 .721 1081 .599 -.122 -.003 -.004 .995
Pat Borders R 3183 2364 .662 819 .541 -.121 -.001 .005 .999
Todd Walker L 3704 2772 .749 932 .630 -.119 -.007 .007 1.011
Reggie Smith B 7119 5306 .797 1813 .680 -.117 .001 .001 1.006
Shawn Green L 5566 4102 .814 1464 .700 -.114 -.006 -.001 1.006
Tony Bernazard B 3735 2808 .671 927 .558 -.114 .000 .005 1.006
Kevin Young R 3944 2814 .721 1130 .608 -.113 -.002 .008 1.000
Phil Bradley R 3716 2836 .730 880 .618 -.112 -.003 .000 1.005
Warren Cromartie L 3958 3022 .705 936 .593 -.112 -.004 .001 1.001
Lee Lacy R 4582 3502 .718 1080 .606 -.112 -.003 -.001 1.001
It looks like only Al Martin (with a bad platoon factor) and Lee May (who seemed to have everything go against him in these situations) could claim to owe their spots on this list to forces beyond their control. Hal Morris, on the other hand, had reasonably good factors and still hit poorly with runners in scoring position.
The data for all the players in the study is here.
So did I find evidence of clutch hitting? Not really. I did come up with lists of players who performed well and poorly in this area. Along the way, I presented quite a bit of data on situational hitting, platoon advantages, opposition pitching strength and park effects, and I attempted to both understand and explain what I found. At the end of all this, however, I guess I'm still not convinced that the players owe their inclusion on these lists of mine to talent rather than luck. Even when dealing with sample sizes of several thousand at-bats, the amount of random variation that I found in my simulations was very close to what I found in the real data. As I mentioned before, this doesn't necessarily mean that there isn't some real differences buried in all that noise, only that I'm not sure I found them. One could argue that the forces at work here, if they exist, must be awfully weak to so closely mimic random noise, and if they are really that inconsequential perhaps we could assume they don't exist without much loss of accuracy.
Earlier I had mentioned the players as a whole hit an average of 23 points worse with men in scoring position. But the average Clutch Percentage is -27 points. This might seem confusing but hopefully won't after an example.
Let's say their are three players: Moe, Larry and Curly. Here are the performance both with and without men in scoring position:
- NO RISP - -- RISP --
Name AB BPS AB BPS CLP
Moe 50 .600 20 .200 -.400
Larry 150 .600 30 .500 -.100
Curly 200 .700 50 .800 .100
Total 400 .650 100 .590 -.060
So as a group, they hit 60 points worse with men in scoring position, but this counts Curly's contribution much more heavily than Moe's. If we average their respective Clutch Percentages (and so count each player equally), we get .633 in the "NO RISP" group ((.600+.600+.700) / 3), .500 in the "RISP" group ((.200+.500+.800) / 3), and an average difference of 133 points.
Some of these averages look weird. For example, the overall BPS for switch-hitters (.652) is greater than the players' averages against BOTH righties (.651) and lefties (.647). This would seem, on the face of it, to be a mathematical impossibility. It is caused by the manner in which I determined the averages, and is best shown by example. Let's say we have two players, Moe and Larry, who have the following right-left splits:
-- VS R -- -- VS L -- -- TOT --
Name AB BPS AB BPS AB BPS
Moe 10 .200 90 .600 100 .560
Larry 90 .600 10 .200 100 .560
Total .400 .400 .560
So when I average these, I do not weight them by at-bats, which would cause the players with more at-bats to influence the results more than someone just over the 3000 at-bat minimum, but just take an average of the averages. And since players tend to have fewer plate appearances when they do not have the platoon advantage (as I showed in an extreme example above), the results can look a little strange at times.