Predictive Value of Pitch Combinations vs. Single Pitches
Pitches are the weapons of a pitcher’s arsenal, and most analysis looks at the effectiveness of a single pitch type on its own. I think there’s a critical blind spot in that type of analysis and this analysis shows why that is true.
Our hypothesis is that
If that hypothesis were true, then we should see higher predictive value in pitch combinations or sequences of pitches than in single, isolated pitch evaluations. That is, the outcomes of pitch combos should be more consistent than the outcomes of single pitches.
Testing Predictive Value
To test this, I looked at the standard deviations for wOBA Added for both discrete
pitch types by
For each of these data sets, I was focused on
wOBA Addedoutcomes observed. A larger standard deviation would indicate lower predictive value of the input variable (pitch type and pitch combo), and a smaller standard deviation of outcomes would indicate more predictive value.
Larger Standard Deviation of wOBA Added Outcomes = Less Predictive Value
Smaller Standard Deviation of wOBA Added Outcomes = More Predictive Value
The chart below shows the results from the 2018 season. I filtered to only include pitches (as defined by pitch type and location) or pitch combinations with at least 40 observed samples to try to remove as much noise as possible. I then charted every sample by average wOBA Added with reference bands showing the Standard Deviation for wOBA Added.
What The Data Shows
The results are subtle but telling. As we look at pitch types and pitch combos with a minimum sample size of 30 observations to eliminate noise from the analysis, we see that standard deviation of wOBA Added outcomes can be 10-33% lower for pitch combos compared to single pitches. Let’s dive in.
If we set the minimum sample size to 40, then the typical Standard Deviation size for wOBA Added on Single Pitches is 0.300. However, the same measure for Pitch Combos is 0.270. This 0.030 represents a 10% improvement in predictive value for the effectiveness of pitch selection.
I even think this 10% improvement is underestimating the potential advantage of Pitch Combos over Single Pitches. As we increase the minimum sample size threshold, we see the Pitch Combo advantage increase, getting as high as a
33% advantage at `n = 50`. This analysis DOES come with a serious qualification as there are only 3 Pitch Combos that have at least 50 observations, and they are all Justin Verlander Four Seam Fastball / Four Seam Fastball combos. However, the trend holds and intuitively it makes sense.
Any analysis or pitch decision that only looks at a Single Pitches effectiveness is only looking at half of the picture; not all Four Seam Fastballs low and away are created equal. If we want to look at effective pitch selection, we have to look at the compounding effects of sequencing to get better predictions about what each incremental pitch will add to the pitcher’s chance of success.