Tuesday, November 07, 2006

Repeatable Pitching Skills

Earlier in the week, I looked at repeatable hitting skills and it was shown that walks, strikeouts and power were substantially more repeatable than batting average. Today, I’ll talk about pitchers. I used the same method of correlation between two years for pitchers as I did for batters. In this study, I selected 320 pitchers with 15 or more starts in two consecutive years: 108 pitchers in 2000-2001, 99 in 2002-2003 and 113 in 2004-2005. The results are shown in Table 1 below.

Not surprisingly, pitching stats tended to be a little less consistent than hitting stats and relief pitcher stats (not shown here) were even more unreliable. Similar to batters, strikeouts (.77) and walks (.67) were the most repeatable stats. Other stats such as hits (.44) and home runs (.38) were less reliable. Two of the most traditional stats – ERA (.33) and winning percentage (.25) – were the most unreliable stats in the study. FIP ERA (which is essentially the ERA of a pitcher based on fielding independent skills) was more consistent having a correlation of .54. This is not surprising since it is heavily dependent on strikeouts and walks (FIP = (13*HR+3*BB-2*K)/IP ).

I conclude from this that strikeouts and walks for pitchers are better indicators of future performance (on the same stats) than any other stats. While you may not know from year to year how many hits per inning pitched a pitcher will allow, you can feel pretty confident that his strikeout and walk rates will stay relatively stable. Whereas a big increase in hits might be due to bad fielding or bad luck, a substantial increase in walks or decrease in strike outs is an indicator that the pitcher did not pitch as well as he did previously.

Table 1: Repeatability of Pitching Skills
















Winning Percentage



  1. Right on, Lee. There are really only three repeatable skills that I can reasonably identify, and you've nailed two: K/IP and BB/IP. The other is batted ball type (GB/FB). This is what is going to drive HR/9 to a great degree, and H/9 to a lesser degree.

    What would be really fun is to segment pitchers into type (GB pitchers versus FB pitchers) and see what their year-to-year r-squares are for H/9 and HR/9. I bet HR/9 is way more stable for GB pitchers than FB pitchers.

  2. Chuck, I had seen before that ground ball percentage was a reliable stat but I did not have the data to check it. If I can find time to pull together multiple years of ground ball from Hardball Times or some place else, I will look at this further.



Blog Archive


My Sabermetrics Book

My Sabermetrics Book
One of Baseball America's top ten books of 2010

Other Sabermetrics Books

Stat Counter