People routinely fail to grasp that the significance of any performance is not simply a function of the sample size, but that of the sample size and the divergence of the performance from the norm or expectation. (Bill James pointed this out long ago, citing Roger Clemens’ 15-K, 0 BB game early in his career as something that an ordinary pitcher was very unlikely to ever accomplish).

So it’s not actually true that anything at all can happen randomly in a 27 PA big league sample. If that sample has the hitter making contact 18 times and recording 3 line drive singles, 5 doubles, 2 triples, and 2 homers, you could calculate 95% and 99% confidence intervals on the hitting skill that could produce that, and the lower limits would, I think, be interesting.

]]>The article is an interview with a baseball player about his work. Just take it for what it is.

]]>Isn’t that even beyond small sample sizes. Aren’t there typically dozens of such starts that end up meaning nothing every year [plus the occurrences where is doesn’t happen to start the season, I’m sure Omar Infante will keep up the .672 wOBA he put up in the last week]

]]>