from Amazon
Tuesday, December 04, 2007
Giving the finger to correlation coefficient
I get mighty p.o.ed when talk comes to correlation coefficient (r), and how the r-squared “explains” something. What it’s really saying is that, given that size of sample, the variance of the parameter you are studying explains a certain percentage of the total variance (including the variance from luck which is completely based on the size of the sample). You can, in effect, control your r-squared to your liking. Wanna prove clutch exists? Use a sample size of 5000 PA. Wanna prove clutch doesn’t exist? Use a sample size of 50.
Anyway, while I get mighty p.o.ed, Phil takes it to another level. Here are his archives on the matter.
If you see a researcher (baseball or otherwise) give a correlation coefficient, without giving you a regression equation (like Phil does), or a correlation coefficient equation (like r=PA/[PA+200]), give him the finger.