In baseball, there are two ways for pitchers to be able to improve their run prevention skill. The first is by limiting the amount of chances the opposing lineup has to hit the ball by increasing strikeouts and lowering walks. The second is by improving the outcome of balls in play, by coercing hitters to hit the ball to spots where the seven to eight other defenders on the field can convert into outs. Teams are continuously searching for pitchers who fit into either category, whether it’s a pitcher with nasty stuff that can generate whiffs almost at will or a pitcher who aims for weak fly ball or ground ball contact to get outs. Teams aren’t necessarily looking for one type over another, although the first group gets paid more, as it makes sense to have a diverse set of arms to throw at opposing lineups.
When it comes down to measuring run prevention on an actual performance basis, we’re going to use ERA and FIP. ERA is a fairly simple statistic that most baseball fans understand, as it’s taking a runs allowed stat and averaging out how many earned runs they’ve allowed per 9 innings pitched. FIP on the other hand estimates the pitcher’s performance based on the number of home runs, walks, hit batters, and strikeouts a pitcher has before adding a constant that allows FIP to be comparable to ERA. That constant usually sits around 3.15 in most years, but you can look up the values here for every season on Fangraphs. By using two different dependent variables with the same independent variables, we should see if there is a similar trend or more noise.
With the dependent variables established, I will now introduce what I’ll be using as the two independent variables. The first will be strikeout minus walk percent (K-BB%), which is a simple calculation of the pitcher’s strikeout percentage minus his walk percentage. I prefer percentage stats over 9-inning averages for the two metrics because most pitchers aren’t throwing nine innings in an appearance any more. The second variable will be wOBACON, a stat that measures a batter outcome on every ball put into play, assigns a linear weight to them, then divides by the number of plate appearances a batter is not intentionally walked or drops a sacrifice bunt. Like with the FIP constant, you can look up the linear weights for a single (~0.9), double (~1.25), triple (~1.6), or home run (~2) on Fangraphs and I can tell you the weight for an out is 0. The reason for including that statistic is more extra base hit contact generally tends to lead to more runs scoring.
Typically the best way to avoid extra base hit contact is by either getting soft fly ball contact or ground ball contact. Both types of contact typically lead to a high percentage of those balls being converted into outs, no matter the skill of the defenders in the area. Ground ball outs tend to be more reliable than a soft contact fly ball pitcher, as ground balls yield an average of .240 with a slugging percentage of .265. If you are looking for a pitcher who may have an ability to reduce the impact of fly ball contact, you can look at pitchers with a high fly ball rate but a low barrel rate on Statcast. Barrels are a specific subset of batted balls based on a specific range of exit velocity and launch angle, that typically lead to extra base hits more often than not with an expected average of at least .500 and expected slugging percentage of at least 1.500.
In this exercise, I chose 358 pitchers who faced 200 or more hitters in the 2021 season. If this sounds familiar to you, it’s the same data I used to see what the limits for the Called Strike + Whiff% statistic were. I concluded that CSW% only correlates well with strikeout rate (K%) and any other statistic that is driven by strikeout rates such as FIP, xFIP, or K-BB%. In those three relationships, a linear estimation model produces an R-squared value of 0.39, 0.42, and 0.53 respectively. Finding out what more we can do with those particular relationships when projecting future performances is a different article for a different day. With that same data, I did the same linear estimation model and wanted to see how ERA and FIP would respond to each variable.
ERA/FIP vs. K-BB% and wOBACON
|ERA vs. K-BB%||0.26|
|ERA vs. wOBACON||0.45|
|FIP vs. K-BB%||0.44|
|FIP vs. wOBACON||0.34|
Unsurprisingly, K-BB% had a better correlation with the run prevention model that uses strikeouts and walks in their input. wOBACON had a stronger correlation with ERA than FIP, as the latter model tries to minimize the effects of balls in play to estimate a pitcher’s true run prevention skill on outcomes the pitcher has the most control over. The big takeaway here is that both are necessary to try to improve run prevention.
So now that we’ve established that eliminating the amount of contact as well as the quality of contact are essential to run prevention. With that in mind, I decided to put together a table of ERA and FIP vs. both K-BB% and wOBACON and see how they change in relation to both variables.
The data is a bit skewed due to a couple outliers in specific demographics that have a count of 1-2 players. So we’re going to to localize the area to a K-BB% of 5-35% and a wOBACON of .275 to .450. We’ll keep the ranges for K-BB% but the ranges for wOBACON will be intervals of 0.035.
Localizing the data to that range, you still see a similar pattern with ERA and FIP relative to K-BB% and wOBACON. There are some examples where the change in ERA or FIP at certain spots seem unusual, but can be attributed to a small player pool in that specific demographic. From both charts, we can see that solid run prevention is possible with an elevated wOBACON or a low K-BB%, but having both issues would be too problematic to roster for much longer.
The final question to answer will be what’s the trade off between K-BB% and quality of contact? Any pitcher would have to be average or better in one, if not both, areas in order to have sustainable above-average run production in both ERA and FIP. ERA has a much stronger correlation to wOBACON than K-BB%, although in the case of FIP it’s close to even between the two variables. Since both variables have a decent correlation, they do have some predictive possibilities if there is a significant improvement in either area. An improvement of one, you might see a small improvement but an improvement in both may suggest a future breakout candidate on the mound.