Filed under:

# K% and BB% standard deviations: Which affects performance more?

If we can control for one variable: strikeout percentage or walk percentage, can we try and discover which metric most affects performance?

Continuing my line of research into standard deviations of K%, BB%, and GB%, I will attempt to inform the relationship between K%, BB%, and performance. By grouping pitchers according to their standard deviation categories of K% and BB%, we can observe pitcher performance while controlling for one or the other variable. For example, we can look at how the range of K% affects performance within the group of pitchers whose BB% is at least two standard deviations below the middle.

The methodology for this analysis can be found here; basically, pitchers must have faced at least 170 batters in a season during the years of 1995-2013 to be included in this analysis. I have chosen the median of RA9-WAR as the performance metric in each group; I didn't want to choose a performance metric whose calculation was based on strikeouts and walks, like FIP or fWAR, because that would confound the results. Observe the table below.

 BB% -2 -1 0-1 0+1 1 2 RA9-WAR K% -2 N/A -0.6 0.3 -0.7 0.2 N/A -0.7 -1 1.7 0.5 0.2 -0.1 -0.5 -0.9 -0.1 0-1 2.6 1.5 1 0.5 -0.1 -0.1 0.7 0+1 4.5 3.1 2.3 1.2 0.5 -0.2 1.7 1 7.1 5 3.6 2.9 1.2 0.7 3 2 8 7.3 5.8 3.7 2.7 0.4 4.8 RA9-WAR 3.2 2.6 1.7 1 0.3 0.0

Across the top of the table are the BB% standard deviation groups, and going down the table are the K% standard deviation groups. -2 means at least two standard deviations below the middle, -1 means at least one standard deviation below the middle, 0-1 means between 0 and 1 standard deviations below the middle, etc. The RA9-WAR values shown in the far right column are the median RA9-WAR values for each K% group regardless of BB%, and the RA9-WAR values shown in the bottom row are the median RA9-WAR values for each BB% group regardless of K%. For example, the median RA9-WAR for all pitchers who were at least 2 standard deviations above the middle K% is 4.8. N/A means there were no pitchers in that group.

Given what we know about strikeouts (they're good) and walks (they're bad), the trends within the table make sense. In general, while holding the other variable steady, increased strikeouts and fewer walks leads to better performance.

However, there is one counter-intuitive group present. The bottom right group with an extremely high K% and BB% did not follow the general trend with only 0.4 RA9-WAR. This group contained 3 pitchers: 1998 Dennys Reyes, 1998 Hideo Nomo, and 2006 Daniel Cabrera. Cabrera accrued 1.9 RA9-WAR in 148 innings, which would be well within the general trend of the table. Reyes accrued the same RA9-WAR as Nomo in only 1/3 the innings, however. Had Reyes thrown similar innings to the other two and his performance remained the same, his RA9-WAR likely would have brought the median up to be in line with the general trend. This is an example of how a small sample size can affect the interpretation of a trend.

There are some general takeaways from this table. If you divide the table into quadrants, you can compare "analogues" of pitcher groups. For example, the top left quadrant of the table represents the pitchers whose K% and BB% are both below the middle. The analogue for this group is the bottom right quadrant, whose K% and BB% are both above the middle. Looking at the RA9-WAR numbers in each quadrant, it appears as though the high K% and high BB% group performs better than the low K% and low BB% group.

Let's look at a specific example. The RA9-WAR of the group whose K% is between 1 and 2 standard deviations above the middle and whose BB% is between 0 and 1 standard deviations above the middle, abbreviated as group (1, 0+1), was 2.9. The +1 K% group had an overall RA9-WAR value of 3, which means that the slightly elevated walk rate had little effect on performance. The analogue of this group, which is group (0-1, -1), had an RA9-WAR of 1.5. The -1 BB% group overall had an RA9-WAR value of 2.6, which means that the slightly decreased strikeout rate had a significant effect on performance.

The idea being presented here is that in general it is better to have a high K%/high BB% pitcher than a low K%/low BB% pitcher. I suspect this is partially related to the conclusion I drew from my previous analysis, in which I stated that the distributions of K% and BB% are changing. I believe that high walk pitchers are being selected out of the pitching population, which means that the walk rate is becoming more homogeneous. It appears as though having an elevated walk rate is not as harmful as having a lower strikeout rate.

Despite the general trend, there is still considerable variation within groups. For example, 2005 Carlos Silva was in group (-1, -2), whose RA9-WAR value was 1.7. Silva accrued 4 RA9-WAR in that season; his GB% was 49.2%, which put him in the upper end of the 0+1 group for GB%. In future performance analyses, I will be looking at K% and GB% together and BB% and GB% together to see how GB% can affect performance.

. . .

All statistics courtesy of FanGraphs.

Kevin Ruprecht is a contributor to Beyond the Box Score. He also writes at Royal Stats for Everyone. You can follow him on Twitter at @KevinRuprecht..

More from Beyond the Box Score