Ahead in the Count: Pitcher BABIP by Count

It is well established that pitchers who have high strikeout rates one year tend to have high strikeout rates the following year, and that walk rates and ground-ball rates have similar year-to-year correlations as well. However, pitchers do not exhibit the same kind of consistency in their Batting Average on Balls In Play (BABIP). Although pitchers’ BABIP depends on the defense behind them, they do not appear to control this very much. As I have pointed out before, that does not mean they do not control BABIP, but rather that they make adjustments constantly in such a way that BABIP does not exhibit any significant year-to-year correlation. The lack of year-to-year correlation in BABIP was originally discovered by Voros McCracken and discussed here at Baseball Prospectus. Even McCracken later realized that pitchers do exhibit some persistence in their BABIP, even net of team effects, but the year-to-year correlation is approximately .12, or hardly anything to trust. Given the small variations in pitcher skill at preventing BABIP as compared with the larger variations in BABIP (only two-thirds of starters will end up with a BABIP that year that is within .020 points of their true skill level), there is not much information contained in a pitcher’s BABIP one year. Most of the variation is going to be luck and defense.

However, there are definitely some differences in pitcher’s impacts on BABIP. In fact, there is a correlation between pitcher BABIP and some defense-independent pitching statistics: there is a correlation of -.138 between BABIP and K/PA among pitchers who faced more than 400 hitters in 2005-2008, and there is a positive correlation of .127 between BABIP and ground-ball percentage. However, there is no significant correlation between BABIP and BB/PA (-.002). However, that is only because BB/PA is negatively correlated with ground-ball percentage (GB%), which is positively correlated with BABIP. Using regression to isolate the effects of each skill on BABIP, one can see that between two pitchers with equivalent K/PA and GB% skills, the one with higher BB/PA will have a higher BABIP on average.

Upon realizing these correlations, I was curious as to why this was true. One thing that should be noted is that BABIP depends strongly on the count. Although BABIP overall is .299 this year, BABIP on full counts is .312, on other three-ball counts it is .306, and on other two-strike counts it is just .287. Earlier in the count (before three balls or two strikes), BABIP is .303. With this knowledge, I became curious how much of the effect that strikeout pitchers have lower BABIPs was merely a function of their being ahead in the count. We have all seen hitters shorten their swings with two strikes, and perhaps power pitchers only have slightly lower BABIPs because hitters are reacting this way. The results actually showed that something else must be going on:

Above-Average K/PA
Count Type                 BABIP
Full Count                 .300
Other Three-Ball Counts    .293
Other Two-Strike Counts    .284
Earlier Counts             .294
Overall                    .292

Below Average K/PA
Count Type                 BABIP
Full Count                 .308
Other Three-Ball Counts    .316
Other Two-Strike Counts    .291
Earlier Counts             .303
Overall                    .301

Instead, strikeout pitchers seem to do better in all counts. The effect is not simply that they are ahead in the count, or even that hitters shorten their swings with two strikes-instead, it seems to be something that these pitchers do to affect BABIP on all counts. Hitters may simply be shortening their swings early in the count as well, but the effect is that they are consistently lowering BABIPs by about .010 points in all counts. Pitchers who are at least one standard deviation above average in strikeouts enjoy the benefit of even lower BABIPs than moderately high strikeout pitchers. These extreme power pitchers have overall BABIP of about .285, and this drop in BABIP is consistent among all four count types. Pitchers more than a standard deviation below average in BABIP have a BABIP of .304, and this rise in BABIP is consistent among all four count types.

As I mentioned earlier, pitchers with high walk rates are not more likely to have higher BABIP, unless you adjust for the fact that they tend to be fly-ball pitchers. Breaking down BABIP by count type for pitchers with above- and below-average walk rates, we get the following tables:

Higher than Average BB/PA
Count Type                 BABIP
Full Count                 .308
Other Three-Ball Counts    .301
Other Two-Strike Counts    .293
Earlier Counts             .299
Overall                    .292

Lower than Average BB/PA
Count Type                 BABIP
Full Count                 .308
Other Three-Ball Counts    .313
Other Two-Strike Counts    .288
Earlier Counts             .299
Overall                    .301

There does not seem to be much here, but it seems that pitchers who walk a lot of hitters tend to have lower BABIPs with three balls. This could hint at some questionable decision-making by hitters, or it could simply be some form of selection bias.

Ground-ball pitchers’ high BABIP is not very surprising given the fact that ground balls have higher BABIP outcomes than fly balls, and line drives do not exhibit much year-to-year correlation. However, looking at this breakdown for ground-ball pitchers, we can see some interesting facts emerge:

Higher than Average GB%
Count Type                 BABIP
Full Count                 .315
Other Three-Ball Counts    .324
Other Two-Strike Counts    .293
Earlier Counts             .303
Overall                    .302

Lower than Average GB%
Count Type                 BABIP
Full Count                 .301
Other Three-Ball Counts    .293
Other Two-Strike Counts    .288
Earlier Counts             .296
Overall                    .294

The most interesting discrepancy is the difference in BABIP on three-ball counts (excluding full counts). Ground-ball pitchers surrender hits on balls in play at a .324 clip, but non-ground ball pitchers do so at a .293 clip. Since pitchers frequently throw fastballs in 3-0 and 3-1 counts, perhaps this suggests that the reason for this discrepancy is that variations in ground-ball rates are variations in ground-ball rates on fastballs. It would be helpful to know the ground-ball rates for difference pitch types.

Lastly, it is important to remember that popups are the batted balls that are most likely to lead to outs, and these occur on about 7.8 percent of all balls in play. Eric Seidman informs me that popups have an intra-class correlation of .55, and there is -.30 correlation between pop-up rate and BABIP. Interestingly, popups occur more often when the pitcher is behind in the count, as the pop-up rate on three-ball counts (excluding full counts) is 9.5 percent, but is only 7.7 percent on two-strike counts (excluding full counts). On full counts, the pop-up rate is 8.3 percent, and on early counts it clocks in at 8.1 percent.

Many were skeptical when McCracken initially released his theory, and although BABIP differences between pitchers are small, there are some differences. Given that there is so much noise already incorporated in yearly BABIP, it is more useful to estimate a pitcher’s future BABIP by looking at what type of pitcher he is rather than what his recent BABIPs have been. High-strikeout pitchers have lower BABIPs than low-strikeout pitchers, and this seems to hold true in all counts. Pitchers who have high ground-ball rates tend to have higher BABIPs than those with low rates of ground balls, but this is particularly because of their higher BABIP in fastball counts. Walk rate is not positively correlated with BABIP, but only because walk rate is negatively correlated with ground-ball rate; otherwise, it would have a positive correlation as seen in regression analysis. Knowing these facts, we can better predict pitcher’s BABIP without being tricked by less-reliable information.