Exploring the Battery Effect by Max Weinstein August 19, 2013 Today’s article will concern the “battery effect” and its far reaching influences on passed balls and wild pitches. However, before we delve in, I will fill you in on the details of my previous research as a reference point for today’s research. The “Battery Effect” The “battery effect” is most easily explained as the relationship between the pitcher and the catcher and how they affect each other. The effect is often subtle, but still significant in the big picture. Let’s dive into the details. My previous study on battery combinations included investigating which of the two battery mates — the pitcher or the catcher — deserved the credit for catching a runner. The basic take away from this research was, surprisingly, that the pitcher had more of a profound effect on the caught stealing percent of the battery. To measure this effect I ran a regression of the pitcher’s CS% — caught stealing percentage — on the battery’s CS%, and vice versa for catchers. Now, I will kindly spare you the guts and gore of the math involved, but in the end, the analysis yielded that a pitcher’s CS% correlates about 80% with how the battery performed — opposed to a 40% correlation for catchers. The said findings suggested that a pitcher’s effect on the running game is underemphasized. Simply put, it would be negligible on our parts, as students of the game, to completely disregard the influence the pitcher has on the running game — in a positive or negative way. In the big picture, that goes for anything in the scope of a battery. We cannot simply ignore the influence that these two highly intertwined positions have on each other. That brings us to our topic for today. How Does the Battery Effect the Passed Ball and Wild Pitch? Enter Pitcher X throwing out of the stretch with a man on first base, Catcher Y puts the first sign down, the ball is released from the windup and makes its way to home plate. Strike one, two-seam fastball on the edge of the plate. Again as the catcher puts the same sign down, the pitcher nods. Strike two, two-seam fastball inside. The third pitch, the catcher puts the sign down for a four-seamer up and away. The pitcher, for whatever reason, throws a curveball down and in the dirt. The catcher is quick to realize the cross-up and is able to put his body in front of the pitch, but to no avail as the ball squirts away and allows the runner to advance to second base. There are thousands of different scenarios that are not captured by our illustration. Pitcher X could have a propensity for his control to run amok. Catcher Y could be old with bad knees and has trouble with his mobility. It goes on and on, and we certainly can’t cover them all. The point is, battery mates are highly, but subtly, dependent on each other — as we’ve concluded in previous battery studies. If you noticed, I did not classify our play as a wild pitch nor a passed ball for a reason. That distinction would imply one of the battery mates was to blame for the hypothetical pitch gone awry. I am not ready to make that distinction until I have done my due diligence with the appropriate analyses. Analyzing the Battery Effect on Passed Balls and Wild Pitches To effectively analyze the battery effect, I have pulled all 301 battery combinations since 2002 with at least 200 innings together — and at least 150 innings without. From the battery combination data, we have pulled their Passed Ball (PB) and Wild Pitch (WP) numbers in the scope of the battery. We have also isolated the individual battery mates’ WP and PB data isolated from the battery totals, so that the regression is dependent of their performance within and outside of the battery. Simply put, if the battery of Pitcher X and Catcher Y had a 6 WP and a 2 PB in 450 innings together, Pitcher X’s isolated numbers would be dependent from the battery’s totals — meaning we have subtracted his battery numbers from his career numbers, since 2002, and similarly for Catcher Y to isolate the battery effect. To begin, lets look at a nice visualization of the range of WP% data for pitchers — “WP%_P” — mapped against the WP% for the batteries — “WP%_B”: Chart #1 This chart is telling us what we would expect – with a correlation of 0.52 between WP%_B and WP%_P. Essentially, the pitcher has a strong influence over wild pitches in the scope of the battery. Being that a WP is something that is designated solely to the pitcher, and not the catcher, the strong relationship is not unwarranted. Similarly, we have a visualization of a pitcher’s PB% — “PB%_P” — compared to the battery’s PB% — “PB%_B”: Chart#2 Here is where things become interesting. Traditionally, the PB is a stat that we assign solely to the catcher. What I expected is that the pitcher would have some sort of weak association with the PB% of the battery, yet the regression yielded a correlation of 66%. This means between the pitcher’s isolated WP% and PB%, the pitcher has more of a relationship with the PB% of the battery —which seems counter-intuitive. Before we get ahead of ourselves, let’s take a look at how this phenomenon relates to the catcher by analyzing their relationship with both the WP and the PB, in the environment of the battery: To begin with, we have a chart like the ones above, relating the WP% of a catcher — “WP%_C” — with the WP% of the battery — “WP%_B”: Chart #3 If you compare this chart with “Chart #1”, you can visualize that the relationship of the red line in “Chart #1” fits the blue series better. By running a regression on these two series, the evidence supports our observation — with a measly 20% correlation on the battery WP%. Once again, this weak relationship is expected, as traditionally a WP is not a stat that we blame on the catching position. And lastly, we will illustrate the correlation between a catcher’s propensity to “allow” a ball roll to the backstop —”PB%_C”— and the battery’s passed ball percentage —”PB%_B”: Chart #4 Generally, the PB is classified as the fault of the catcher so conventional baseball wisdom would suggest that this chart would have a similarly strong correlation as in “Chart #1” — pitcher’s WP% on battery’s WP%. However, this graph gives us the lowest correlation out of all the regressions we have performed today with a disappointing 16%. This clearly demonstrates that the catcher has little control over the passed ball within the scope of the battery. Summary of Results Regression Chart # r = WP%_B~WP%_P 1 0.52 PB%_B~PB%_P 2 0.66 WP%_B~WP%_C 3 0.2 PB%_B~PB%_C 4 0.16 Above are the correlations — “r =” — for each of the graphs that we have broken down. Essentially the closer to “1” the more correlated the regression. Given the summary of results, there are a few basic conclusions that we can make from the above analyses. It’s important to note, that correlation does not always imply causation; however, it is difficult to find causation without correlation. We can conclude that in the environment of the battery, the pitcher has more of an influence than the catcher over both WP% and PB%. I am not concluding that the pitcher has all the control over the outcomes, rather I am saying that it is some combination of both the pitcher and catcher who influence the PB% and WP% of the battery — with the pitcher being the more influential of the two. However, the pitcher had a higher correlation with the PB% of the battery than the WP%. For me this seems counter-intuitive. Conventional wisdom suggests that a PB is the fault of the catcher, the WP the fault of the pitcher. The regression tells us that the pitcher is more of a factor, than perhaps they get credit for, on the outcome of a PB. I also find it strange that in our research, catchers have no significant correlation with any of the battery stats — CS% and now PB% and WP%. This has left me curious as to whether multi-regression analysis will give us further insight into why the void between the pitcher and catcher is so profound. The multi-regressions, simply put, are measuring the correlation of two variables — here the WP% and PB% of the pitcher and catcher— on one variable — the WP% and PB% of the battery. These should give us a bigger picture of how the past performance of the battery mates affects the battery’s performance: Multi-Regressions Chart # r = PB%_B~WP%_P+PB%_P N/A 0.83 WP%_B~WP%_P+PB%_P N/A 0.78 PB%_B~WP%_C+PB%_C N/A 0.6 WP%_B~WP%_C+PB%_C N/A 0.51 The above chart suggests that the gap in influence between the pitcher and catcher is less than what the previous regression charts suggest. Meanwhile, there is still a significant gap and therefore I would argue that our conclusions still hold true. The pitcher has more correlation to the PB than the catcher and remains the battery mate with the stronger association to the battery’s statistics. For me, these findings are not conclusive — a strong correlation does not dictate definitive blame but rather provides additional insight into the battery effect. I do think the findings are talking points and the beginning of the conversation to start looking at the battery differently — as an environment in which every outcome is subject to influences by the both the pitcher and the catcher. Nonetheless, our initial research into the battery effect dynamic indicates that the pitcher is the “wild card” in the equation.