In Michael Wolverton's great research article "Support-Neutral" Statistics -- A Method of Evaluating the True Quality of a Pitcher's Start, he makes the statement that
"Looking at each start's contribution to winning, rather than cumulative run-prevention over the course of a year (ERA or Pitching Runs), can help us answer questions like: Given equal ERAs, do some pitchers pitch in a way that will tend to win more games than other pitchers? In particular, is it better for a starter to be flaky -- either very good or very bad on a given day -- or consistently average?"
I don't think that either of the questions that Wolverton asserts as answerable using his support-neutral game-by-game statistics actually were satisfactorily answered in the research article or anywhere else, unless it has been part of BP's premium content.
Flaky vs Steady Pitchers
As a reminder, 'flaky' and 'steady' are relative terms describing the amount of variance in SNVA between a pitcher's starts. Wolverton makes a reasonable, logical argument that flaky pitchers are undervalued using seasonal statisics such as ERA or APW, and conversely that steady pitchers are overvalued. In addition to the logical argument, he also points out that a large majority of the flakiest and steadiest pitchers were under- and overvalued respectively in 1992. OK, so maybe Wolverton does answer the flaky question reasonable well. However,the data he presents is just for 1992.
Using the historical reports linked from Baseball Prospectus' Statistics I compiled the 10 flakiest and steadiest pitchers each season from 1998-2003 (a total of 120). A graph of [(SNVA-APW) / (GS/37)] vs (SNVA Var) is included below.
A linear regression provided the following equation:
[(SNVA-APW) / (GS/37)] = 10.816 * (SNVA Var) - 0.4818
R^2 = 0.19
Unfortunately the variance of SNVA is only given for the top 10 flaky and steady pitchers each year. So until Wolverton/BP includes the historical variance of SNVA (or we calculate it ourselves), we only know the relationship at the extremes.
Run-Shaping Aability (Pitching to the Score?)
Clearly there are many pitchers each year who have a discrepancy between APW and SNVA of more than half a game - you can quickly find many examples in any of the yearly reports at BP. In fact, the standard deviation of the difference is just under 0.5 wins over the course of a full season. But is this an ability (i.e. reproducible)? Is there a correlation between [(SNVA-APW) / (GS/37)] from one year to the next? ... a correlation betwen odd and even years?
To answer that question I transferred all the data from the Support-Neutral pitching reports from 1998 through 2003 to Excel. I combined the data in two ways: by odd vs even years with a minimum of 30 starts in each, and consecutive seasons ('98 vs '99, '00 vs '01, '02 vs '03, and then taking the remaining non-matched starts of '99 vs '00 and '01 vs '02) with a minimum of 20 starts in each season. This effort resulted in a sample size of 157 for odd vs even seasons, and 253 for consecutive seasons. For each of these two data sets I compared Diff=[(SNVA-APW) / (GS/37)] in each sample. In other words, I looked at Diff_Even vs Diff_Odd and Diff_Succ vs Diff_Prec. Performing a linear regression on each resulted in the following equations:
Diff_Even = 0.1094 * Diff_Odd + 0.0837
R^2 = 0.0125
Diff_Succ = 0.1009 * Diff_Prec + 0.1187
R^2 = 0.0088
The two graphs are included below:
As you can see from the linear regressions and the graphs, there is little relationship between the difference of a starting pitcher's contribution according to game-by-game analysis and end-of-season analysis from one year to the next. This is getting pretty close to adressing "pitching to the score". One who pitches to the score would have the ability to shape the distribution of runs against in such a manner as to minimize the L/RA conversion. But that's not what is seen from the 1998-2003 data. At best, whatever "clutch pitching ability" that exists among MLB starting pitchers is dwarfed by factors not included in the model.
Go to Kevin Harlow's home page.