Hans Hassell points me to this article by Baseball Bob on what happens to pitchers after they show extreme differences in ERA and FIP. He studied 2009 versus 2010, and found that pitchers with FIP << ERA in 2009, their FIP is a good predictor of their 2010 ERA. For pitchers with a FIP >> ERA in 2009, the average of the two is a better indicator of the following season’s ERA. In other words, pitchers with low ERA regress less toward their FIP than pitchers with high ERAs.
This was a very small study, however, and sample size issues could easily skew the results. I’d like to see how this holds up over many years.
left wall probably. and it probably understates the problem on the top end since these things are usually done as binomial, the “group 4” isn’t a nominal issue as regards FIP (bad FIP result attrite out at a high rate giving the impression of more regression than actually happens if you just segregate the single season results).