David,

If you google Andy Dolphin you'll find his article about clutch hitting. He's an astrophysics PhD who found a statistically significant effect for some hitters--but the likely effect over the course of a season might be a hit or two. That might have a Win Expectancy value on average of what--a quarter win or half a win. (Just a guess.)

For measuring "clutch", why not use Win Expectancy / Leveraged Index of Tango?

i remember someone doing a study like this like last year and he came up with the best clutch hitter in the past like 30 years being garth iorg

Ventura is one of my favorite players ever. I generally don't believe in clutch hitting but it is interesting to note the # of Grand Slams in his career. I believe he is in the top 3. The rest of the players around him in Career Grand Slams are all Hall of Fame #s. Perhaps Ventura was a clutch hitter.

Remember, too, that's only a small part of Ventura's career. This data goes from 1974 to 1992.

Nice study :-) The fact that RISP hits fall just outside the CI is probably further evidence to back your claim. And, I imagine the reason for so few hitters below expectation is that hitters who might have achieved that feat were sent back to the minors before they had a chance.

One problem that I have with clutch hitting studies is that they start with the assumption that a player's expected performance is a clutch situation is identical to the player's expected performance in all situations. That may not be the case. Bill James once noted that player performance "improved", as a whole, when comparing performance with no one on vs. performance with a runner on first, and there may very well be other such group norms when looking at other situations. If the group norm with RISP (for example) is for performance to decline 10%, then the fact that a player's performance did not change in RISP might very well be an indicator of true clutch performance.

Mike,

For the study, the batters listed hit .268 overall, .269 with runners in scoring position. My experience is that batters in general hit better with men on base, but it's not that big a difference.

Interestingly enough... if you look at the data and find a percentile value for each player - which I calculated as the point in their confidence interval in which their actual total falls - and call this PERCENTILE. And you divide RISPAB by total AB, which I call %RISPAB... you will find a significant correlation between PERCENTILE and %RISPAB (r=.145, p

Which says, in plain terms, that people who have more RISP AB's are more likely to perform above their average with RISP.

One the one hand, this could mean that these guys were deliberately put in as pinch hitters more than other people, due to their perceived tendency to be "clutch". However, the AB restrictions on this sample ensure that all players were pretty much regulars.

Perhaps this demonstrates a practice effect then? The more "crucial" AB's you get in the majors, the more "clutch" you perform.

Clutch simply does not exist.

I am not surprised to see Ventura at the top of the list. He has always been one hitter who I thought was really clutch. It seems like he has Grand Slams on tap and just orders one up whenever you need it.

your point may be correct, but your statistical method is not sound enough. A better method would be to get avg in risp and avg in non-risp and compare them. when we look at the table in this way, there are 13.6% of players whose average in risp is lower by .02 point (i.e. .260 (risp) and .280 (non-Risp), and 15.2% of players whose average in risp is higher by .02 point. I know the cutting point of .02 is somewhat arbitary, but it seems to me plausible.
Anyway, the top ten hitters are exactly identical. However, my method can show, for example, that the manager who enjoyed 8-game winning streak in April is one of top 20 clutch hitters in this list (.257 VS .294)

David,

With all the recent interest in the AL MVP race-down-to-the-wire this year, I was led to some postings about RISP-based stats. I found your posting most interesting and atypical in providing something concrete on which others could build.

Nice job. It really is interesting how similar these distributions are to each other. Your conclusions seem sound. It seems that additional conclusions are (a) contrary to popular belief, there is no such thing as a clutch hitter independent of simply being a good hitter, or (b) there may some other measure of clutch hitting which accurately reflects and validates popular notions about clutch hitting.

Just in case there is something really subtle, though, in RISP-based measures (e.g., RISP batting average minus overall batting average), it would be interesting to see how it correlates with subjective impressions of clutch hitting. I am a great believer in the ability of (knowledgable) individuals to come to somewhat accurate conclusions about complex situations. Of course, people can also be quite bad and even dangerously inaccurate. This makes comparison of subjective and "objective" assessments quite interesting and potentially illuminating.

If anyone is willing and able to do it, I would like to see some data collected on PERCEPTIONS of clutch hitters which could be compared to your data. This would require something like the following (the following is just an example for clarity):

(a) send your list of names, without the data, to a number of individuals in all baseball markets (I will leave it to the investigator to determine what is an appropriate number based on sampling theory);

(b) solicit responses about perceived clutch hitters from individuals who indicate that they were at least 18 years of age in 1974 and followed baseball somewhat closely from 1974 to 1992 (This is somewhat arbitrary. There certainly would be a problem with assessment of players whose careers did not overal well with this time period. Alternatives certainly would be okay, but should be as specific);

(c) solicit rankings from these individuals of a small number, say 5 or 10 players, who are perceived to the best and a similar number for the worst on your list;

(d) randomly select roughly equal numbers of respondents from all baseball markets (I will leave it to others to identify "baseball markets," adequate sample sizes, or alternative sampling methodologies.)

(e) one could compare the resulting data subset on perceived clutch hitters to your data on a putative RISP-based measure of clutch hitters (e.g. RISP BA - Overall BA)

(f) there are any number of statistical procedures which could be used for to assess the concordance between the data sets.

Additional statistics on data posted by David:

SUN 10/02/05 10:14:08 AM

SYSTAT VERSION 7.0.1
COPYRIGHT (C) 1997, SPSS INC.

------------------------------------------

BA RISPBA
N of cases 567 567
Minimum 0.210 0.187
Maximum 0.338 0.348
Median 0.262 0.264
Mean 0.264 0.265
95% CI Upper 0.266 0.267
95% CI Lower 0.262 0.263
Std. Error 0.001 0.001
Standard Dev 0.019 0.024
Skewness(G1) 0.253 0.027
SE Skewness 0.103 0.103
Kurtosis(G2) 0.338 0.634
SE Kurtosis 0.205 0.205

--------------------------------------------------

Kolmogorov-Smirnov One Sample Test using Normal(0.00,1.00) distribution

Variable N-of-Cases MaxDif Probability (2-tail)

BA 567.000 0.585 0.000
RISPBA 567.000 0.575 0.000

-------------------------------------------------------

Pearson correlation matrix

BA RISPBA
BA 1.000
RISPBA 0.779 1.000

Number of observations: 567

---------------------------------------------

Paired samples t test on BA vs RISPBA with 567 cases

Mean BA = 0.264
Mean RISPBA = 0.265
Mean Difference = -0.001 95.00% CI = -0.002 to 0.000
SD Difference = 0.015 t = -1.214
df = 566 Prob = 0.225

----------------------------------------------

Wilcoxon Signed Ranks Test Results

Counts of differences (row variable greater than column)
BA RISPBA
BA 0 274
RISPBA 293 0

Z = (Sum of signed ranks)/square root(sum of squared ranks)
BA RISPBA
BA 0.0
RISPBA 1.074 0.0

Two-sided probabilities using normal approximation
BA RISPBA
BA 1.000
RISPBA 0.283 1.000

--------------------------------------------

Sign test results

Counts of differences (row variable greater than column)
BA RISPBA
BA 0 274
RISPBA 293 0

--------------------------------------------