Using Batted Ball Data to Evaluate and Predict Pitcher Performance

[Editor's Note: Please welcome Ryan Potter to Beyond the Box Score! Ryan will be providing link reactions several times a week.]

At last week’s MIT Sloan Sports Analytics Conference, Dan Rosenheck of The Economist and the New York Times gave a presentation on predicting BABIP that included a section on the ability of IFFB% and contact rates on strikes provided an explanation for a 15% variance in a the future BABIP of pitchers. Rosenheck’s presentation reignited the sabermetric community’s discussion of how best to use batted ball data when evaluating and predicting the performance of pitchers.

Using batted ball data to analyze and predict pitcher performance isn't exactly a new concept; ERA Estimators like SIERA and tERA have been incorporating batted ball data in their calculations for years. However, the renewed discussion of the ability of IFFB% to shed some light on a portion of the variance in a pitcher’s future BABIP inspired Dave Cameron of Fangraphs to expand upon FIP using IFFB%. Dave explains:

Maybe we should consider that the IFFB is essentially not that different from a HR, in that it measures the results of specific batted balls that have a distinct run value and that aren’t influenced by the defenders behind any given pitcher. Just like we penalize pitchers for giving up home runs, logic would suggest that we should be giving them credit for infield flies.

Last year, Bill Petti analyzed a number of pitching metrics for starting pitchers from seven seasons (2004-2011) to find the year to year correlation for each metric. Bill found that the year to year correlation for IFFB% was .37. Dave mentions Bill’s work with year to year correlations for pitching metrics and the year to year correlation for IFFB% in particular, but doesn’t dwell much on it. The relatively low year to year correlation indicates that IFFB% is not a repeatable skill for pitchers. So Dave’s ERA estimator, which he named "INNIP" doesn’t seem to be a reliable tool for making predictions of a pitcher’s future performance.

FIP was developed to as a way to determine a pitcher’s skill independent of factors they have no actual control over. Because the year-to-year correlation for IFFB% is not strong enough to indicate that it is a skill, adding it to FIP would cloud the perception of a pitcher’s actual ability.

It is also nearly impossible given the data we currently have available to determine how much of IFFB% is the result of the batter’s actions. Further development of PitchF/X may help us to better understand batted ball data in the near future. At that point we will be better able to evaluate the usefulness of IFFB% in determining a pitcher’s true skill level.

Should we incorporate IFFB% into ERA Estimators?