Is Contact Management Consistent In-Season?

Last week, I took a look at Statcast data from 2016 and 2017 and attempted to find contact-management skills among pitchers. The basic conclusion of that study? Pitchers might well have skills to manage contact once the ball hits the bat; if they do, however, neither xwOBA nor Statcast classifications seem to reveal it. Quality of contact didn’t hold up from year to year — i.e. last year’s results on contact aren’t likely to inform much of this year’s results on contact.

In the comments section, however, one reader wondered if in-season results might create a different result. That’s what I’d like to examine in this post. Here we go.

To answer this new question, I looked at the pitchers in 2016 who recorded at least 1,000 pitches in each half. Before looking at the correlations between two halves for contact-management numbers, let’s first get a sense of how other metrics correlate between the first and second half of a season.

For example, here are strikeout rates from the first half and the second half of 2016.

We see a pretty strong relationship here for the 86 pitchers who met the requisite sample in both halves. Pitchers who struck out a lot of batters in the first half tended to keep doing so in the second, Meanwhile, pitchers who struck out fewer hitters tended to keep doing that, as well.

The same is true for other metrics. Just to provide a decent idea about a few different stats, the table below features the correlation coefficient (r) for several stats from the first half to the second half in 2016.

As we can see, pitchers do have a decent amount control of contact type; the presence of ground-ball rate at the top of the above table indicates as much. We see that FIP stays fairly true from half to half, which would be expected given the strong relationship with strikeouts and walks. ERA, meanwhile, doesn’t easily replicate itself from first half to second half. Also, the number of fly balls that fly out of the park versus the ones that stay in tends not to match up between halves.

Now let’s look at BABIP between the halves.

That’s pretty noisy. What if we factor in some measure of contact quality by looking at wOBA on contact?

Same deal. Noisy again.

Now let’s see if Statcast can do better. Here’s xwOBA between the two halves.

If you squint, you can maybe see a slightly more dense grouping in the middle there, but there are a lot of pitchers who are all over the board. The correlation coefficient is larger than it was when comparing 2016 to 2017, but it’s difficult to tell if that’s just noise, particularly when we aren’t dealing with a strong relationship to begin with and squaring those numbers makes them quite similar and small.

Owing to the relative youth of Statcast and the improvements being made to it, it seems possible that the numbers from this season might be better than those from last. We’re dealing with only three-quarters of a season at this point, so we have a smaller sample, but looking at pitchers with at least 750 pitches before and after June 9 might shed some light on things.

What do we find? For BABIP and wOBA on contact, the correlation coefficient is essentially zero, with no relationship between the first two months of the season and the past two months. Here is what xwOBA on contact yielded.

So we have something that seems to be better than BABIP and wOBA early in the season, but not really any better when we had the larger samples in 2016, and there is still a ton of noise. Does it mean xwOBA on contact might offer more than BABIP does in terms of identifying pitcher skill? We can hope. But does it mean we’ve come any closer to isolating a contact-management skill? Probably not, as the relationship still isn’t strong and isn’t stronger over the course of two seasons than in the middle of a season.

If the numbers were to get better at the end of the season, it’s possible we could get closer. It’s possible more in-depth studies are already getting there. In any event, it’s possible xwOBA on contact reaches it’s highest level of reliability faster than BABIP or xwOBA on contact, but it doesn’t appear that it’s any better in-season than it is across seasons.

The reader (wobatus) who spurred this exercise mentioned two articles in his comment. The first, by Dan Szymborski, references the stickiness of in-season results as it relates to rest-of-season projections. While the section of that piece referenced BABIP specifically regarding hitters, it’s possible there might be some application for pitchers as well. In the second piece, Rob Arthur at FiveThirtyEight finds that pitcher performance fluctuated over the course of the season based on the velocity of pitches. This piece is a good piggyback off of yet another piece mentioned in the comments section, one in which Russell Carleton found that exit velocity for pitchers becomes reliable pretty quickly, but that reliability did not increase over time with more results.

Those latter two articles seem to intimate that pitcher skill likely varies over time, even over short intervals. Given the fluctuation of those skills, it seems reasonable to believe that, if there is a skill in managing contact, it’ll be difficult to find. That theory tends to support my last post and much of my research. There’s considerable difficulty in isolating a specific skill in managing contact and or attempting to predict it in any meaningful way.

This isn’t to say anything against a specific skill existing or providing credit for the skill. Both Carleton and Arthur found that pitcher performance likely has meaningful fluctuations in short spurts. It very well could be that managing contact is a skill that fluctuates, making broader estimates of that skill very difficult. It’s also possible we just don’t have enough information. Nothing above disproves pitcher contact as a skill, but nothing above likely helps predict pitcher performance in a positive manner, either.

Likewise, giving credit for this skill or lack thereof is a matter of interpretation. FIP has long taken most of the potential skill/randomness of batted balls out of the equation, only including home runs and total outs in terms of contact. Runs allowed and ERA include all results on contact, but they also do a poor job of predicting future runs allowed, where FIP and others like it have done a much better job with both. The search continues.