/cdn.vox-cdn.com/uploads/chorus_image/image/46304840/usa-today-8555210.0.jpg)
Way back when I was trying to figure out if there was a place for me to write about baseball, I wrote a post in the FanGraphs Community Research section that looked into clutch hitting. It's a subject I've always wanted to return to, because it's one that won't go away. Despite tons of research by really smart people suggesting that clutch hitters only exist in our fertile imaginations, it's one of those notions that remains a staple of baseball discussion.
This Tableau data viz includes every player with at least 2500 plate appearances in complete careers since 1938 (which is as far back as Baseball-Reference has fairly complete play-by-play data), or 2000 PA for current players. The data is generally complete going back to 1973, but prior to that has gaps due to incomplete play-by-play data. I included data for batting with a runner on first, because these three states (empty, RISP, first) represent every plate appearance, from which the plate appearances with no data can be derived. This is a pool of 1,405 players, or around twelve percent of position players in this time span. I've set the sliders to players whose careers began in 1980 or later with at least 5000 PA, but they can be moved to view the data any way desired.
<script type='text/javascript' src='https://public.tableau.com/javascripts/api/viz_v1.js'></script><div class='tableauPlaceholder' style='width: 700px; height: 763px;'><noscript><a href='#'><img alt=' ' src='https://public.tableau.com/static/images/Cl/ClutchHitting/Scatter/1_rss.png' style='border: none' /></a></noscript><object class='tableauViz' width='700' height='763' style='display:none;'><param name='host_url' value='https%3A%2F%2Fpublic.tableau.com%2F' /> <param name='site_root' value='' /><param name='name' value='ClutchHitting/Scatter' /><param name='tabs' value='yes' /><param name='toolbar' value='yes' /><param name='static_image' value='https://public.tableau.com/static/images/Cl/ClutchHitting/Scatter/1.png' /> <param name='animate_transition' value='yes' /><param name='display_static_image' value='yes' /><param name='display_spinner' value='yes' /><param name='display_overlay' value='yes' /><param name='display_count' value='yes' /><param name='showVizHome' value='no' /><param name='showTabs' value='y' /><param name='bootstrapWhenNotified' value='true' /></object></div>
The chart has two axes, with batting average when the bases are empty plotted on the horizontal axis and batting average with RISP plotted on the vertical. My definition of clutch is any occasion with runners in scoring position. In addition, I also contend that score of the game is irrelevant--whether is 0-0 in the top of the first or 14-0 in the top of the ninth, hitters want to drive in runners on base.
In my FanGraphs post, I suggested the ridiculous standard of a difference of twenty points in batting average to test for clutch. What I didn't know at the time was how huge a difference this is.
*** WARNING--boring (and quite possibly incorrect) math ahead (and apologies to the coiner of the actual phrase, Russell Carleton, but I love it so much) ***
In the sample, career batting averages range from .195 (Jeff Mathis, who must be the best pitch framer in the history of pitch framing) to .344 (Ted Williams), but averages tend to clump around the .250-.300 range:
These t-test results tested if the difference between RISP batting average and batting average with the bases empty was statistically significant:
t-test: Two-sample assuming unequal variances | <>5000 PA | 5000+ PA | |||
---|---|---|---|---|---|
eBA | rBA | eBA | rBA | ||
Mean | .255 | .261 | .269 | .277 | |
Variance | .0004 | .0005 | .0003 | .0004 | |
Observations | 775 | 775 | 630 | 630 | |
Hypothesized Mean Difference | 0 | 0 | |||
df | 1510 | 1240 | |||
t Stat | -5.706 | -7.050 | |||
P(T<>=t) one-tail | 0.000 | 0.000 | |||
t Critical one-tail | 1.646 | 1.646 | |||
P(T<>=t) two-tail | 0.000 | 0.000 | |||
t Critical two-tail | 1.962 | 1.962 |
eBA=batting average with bases empty rBA=batting average with RISP
Just for fun, I stratified the data, using 5000 PA as the cutoff, and as expected, there is an increase in career batting average for players with 5000 or more PA.
For those still reading, these differences in batting average are statistically significant, which means the tests detected a difference in average between hitting with RISP and the bases empty due to more than just chance. As such, if someone were to state, "You know, clutch hitting does exist--hitters really do hit for higher average with runners in scoring position than when the bases are empty!", they're correct. When the range of batting average is around fifty points, a difference in six to eight points will have an impact.
Clutch hitting does exist, at least in the manner in which I set this up--do clutch hitters? In other words, are there some players who excel in clutch situations, or is this a statistical anomaly that doesn't translate into meaningful impact? Eight points of batting average is around five hits a year for players who get 600 at-bats, or about one a month. That's not very many.
That's batting average--what happens when slugging percent is measured?
t-test: Two-sample assuming unequal variances | <>5000 PA | 5000+ PA | |||
---|---|---|---|---|---|
eSLG | rSLG | eSLG | rSLG | ||
Mean | .393 | .393 | .425 | .427 | |
Variance | .003 | .003 | .003 | .003 | |
Observations | 775 | 775 | 630 | 630 | |
Hypothesized Mean Difference | 0 | 0 | |||
df | 1546 | 1257 | |||
t Stat | -0.045 | -0.396 | |||
P(T<>=t) one-tail | 0.482 | 0.346 | |||
t Critical one-tail | 1.646 | 1.646 | |||
P(T<>=t) two-tail | 0.964 | 0.692 | |||
t Critical two-tail | 1.961 | 1.962 |
eSLG=slugging percent with bases empty rSLG-slugging percent with runners in scoring position
So long, statistical significance! These hitters, the best in baseball over the past seventy years or so, show a marginal increase in batting average and no difference in slugging percent in RISP situations.
What might cause the increase in average in RISP situations? Defense adjusts with runners on base--as runners are held on base, defenses may be stretched in such a manner that balls that might be reached with the bases empty slip through for a hit. Pitchers might attempt to be more precise in their pitching, thus consequently serving up a fat ball to be hit. Hitters may adjust their approach, determined to swing only at pitches that are exactly what they want.
So, does clutch hitting exist? Of course it does--it occurs every time a player delivers a hit with runners on base. What don't exist are clutch hitters--there are good hitters in baseball, and there are not-so-good (the bad are generally not represented in this sample). Adrian Gonzalez is pictured at the top of this post and labeled in the data viz because of these numbers:
BA | SLG | |
---|---|---|
Bases Empty | .281 | .489 |
RISP | .329 | .558 |
These are phenomenal numbers, a seventeen percent increase in batting average and a fourteen percent increase in slugging percent in RISP situations. This doesn't necessarily make Gonzalez a clutch hitter--what he is is a very good hitter, and good hitters are good hitters in all situations. He's also quite the outlier. Click on some of the other points around Gonzalez and see if those names come to mind in the greatest clutch hitters in baseball discussion.
Click on some of the data points on the opposite side of the line from Gonzalez--these are the players whose averages are lower in RISP situations. Robinson Cano is marked because his batting average is around ten percent lower in RISP situations than when the bases are empty. Does this make Cano a bad hitter? Of course not--his career slash line of .309/.357/.497 is among the best for current players. He "only" bats .284 with runners in scoring position, which recognizes that very good hitters rarely have anywhere to go but down when looking at the data this way.
Scroll over the data points, and it becomes readily apparent that players bat near their career averages regardless of the situation. There are good hitters and not-so-good hitters, and clutch situations and non-clutch situations. Good hitters perform better than not-so-good hitters no matter what the situation is. Clutch hitting is real and happens every day--clutch hitters are more a phantom of the data than a real phenomenon.
All data from Baseball-Reference. Any errors in gathering and processing the data are the author's.
Scott Lindholm lives in Davenport, IA and writes for the Baseball Prospectus Cubs site, BP Wrigleyville. Follow him on Twitter @ScottLindholm.