from Amazon
Friday, May 07, 2010
Max on Clutch
Interesting what he did here. He ranked all the PA by LI and all the PA by run value, and then ran a correlation between the ordinal rankings of the two.
I suspect that it would be almost impossible to get a high correlation. For example, say that someone actually did have a true talent level that perfectly correlated to LI. What kind of sample performances are we going to get? For example, say that someone’s true wOBA goes from .450 to .250, from LI of 10.0 to 0.01. What sample wOBA are we going to get at LI=10.0? Well, the mean is going to be .450, but the observed can be anything, even a strikeout.
• Sabermetrics
• Clutch
At the risk of repeating myself ...
You can’t judge the strength of the relationship by just how big the correlation coefficient looks to your eye. You have to use the equation. It could be that .050 is a huge clutch effect.
It could even be that .050 is a huge clutch effect but not statistically significant because your sample size is too small.
So, show the equation! If Thurman Munson was a .050, what does that mean in terms of runs? Is that the equivalent of 50 points of wOBA in a leverage 3.0 situation? Is it only a few points and not signficant? What does the .050 MEAN?
It’s like ... if I give you a gift card for 100,000 Elbonian Pesos. Is that a lot of money, or not? There’s no way to tell from just the number. You have to translate!