from Amazon
Friday, August 11, 2006
OPS and Clutchiness
First off, see the second post here:
http://www.baseball-fever.com/showthread.php?t=48531
Nice to see that 1.8 * OBP + SLG has the highest correlation to team runs scored.
Next up is Cy Morong’s look at clutchiness:
http://www.beyondtheboxscore.com/story/2006/8/10/111723/993
I just want to focus on one thing, which is:
(2) PWV/PA = -.0246 + .000149*OBP + .000097*SLG
Now, in actual fact, that line should read as:
(2) PWV/PA = -.0246 + .000149*(OBP/lgOBP*100) + .000097*(SLG/lgSLG*100)
PWV/PA is the win value per PA.
I don’t know what the lgOBP and lgSLG was for this time period, but I’ll just use .33 for lgOBP and .40 for lgSLG. Feel free to rework my numbers with whatever is more accurate.
So, expanding the equation, we get:
(2) PWV/PA = -.0246 + .0149/.33*OBP + .0097/.40*SLG
Which becomes:
(2) PWV/PA = -.0246 + .0452*OBP + .0243*SLG
Which becomes:
(2) PWV/PA = .0246(-1 + 1.84*OBP + .99*SLG )
Or the similar:
(2) PWV/PA = .0246 * (1.84*OBP + .99*SLG - 1)
A couple of things to see:
1 - The relationship between OBP and SLG is around 1.85 weight for OBP and a 1 for SLG. Hence, the reason to use 1.8 * OBP + SLG, and not OPS.
2 - What’s also nice is that the league average of 1.8 * OBP + SLG is pretty much 1.00. What can be nicer than that?
3 - Now, these days, OBP and SLG is a bit higher than the historical. And, since it doesn’t make much of a difference, I like to keep the weight of SLG as “1”, and just float the OBP coefficient. It’s not right of course, but, it’s kinda nice to keep the baseline at 1.00. And, as we saw with the first link, the correlation between plain old OPS and 1.8 * OBP + SLG is pretty high. So, if I use 1.7 or 1.9 for OBP, no one is really going to care.
4 - I created this equation for 2006 a few weeks ago:
wins per PA = .025 * (1.7 * OBP + SLG - 1)
It is very nice to see that the historical record, which uses .0246, is extremely close to my quick equation. I derived my equation based completely on LWTS, and a runs to win converter.
Now, back to Cy’s clutchiness. All you need to do is attach his “predicted” and “actual” with PA, figure out how many standard deviations that is for each player (z-score), and then take the standard deviation of the z-scores. If it’s 1.00, then clutchiness is not shown to exist here. If it’s above 1.00, clutchiness may exist. Figure out the uncertainty level around that SD, and you can say how much clutch exists.
In The Book, Andy was able to show that Clutch hitting does indeed exist.