What Actually Makes a Curveball Effective?

The other day I began pulling together Savant data to determine whether there was an ideal zone percentage for different types of curveballs (CUs) and sliders (SLs). I haven’t found much on that front yet. However, I did realize that I don’t really know what makes curveballs effective, both from a results standpoint (extra whiffs, weaker contact, etc.) or a trait standpoint (vertical break, horizontal break, velo). I took a look at all of these factors for the curveballs in the 2019 baseball season to see if anything stuck out.

I analyzed a sample of 214 pitchers, representing everyone from 2019 who threw at least 20 innings, a curveball at least 10% of the time, and qualified for Savant’s pitch movement leaderboard. From this sample I pulled info on every pitcher’s spin profile, wOBA, xwOBACON, zone percentage, SwStr %, and RHB/LHB splits. I even noted all that same info for the rest of their arsenal as well as just to have a full view. Then they were bucketed in every way imaginable with averages and standard deviations to see which ones stood out. I do want to preface all my findings by saying that the sample size is not ideal, as the buckets were mostly of decent size (roughly 100-plus players), but I did get granular at times (the smallest group was 48).

I am most focused on the following metrics: CU wOBA, CU xwOBACON, CU SwStr %, CU Drop & Tail (as a % difference vs. the average pitcher at similar velocity). Here are the averages across the entire sample:

Some quick takeaways: CU wOBA is skewed just because of the counts they’re generally thrown in, which is why xwOBACON and SwStr % are important for this analysis. CU Tail has a lot of variance because of a wide pitch type spectrum (12-6 CUs, Sweeping CUs, and possibly misclassified SLs). The thing that really surprises me here is the SwStr rate. I had always assumed that one of the strengths of the CU was in its swing & miss potential, but when you compare it to a similar sample for other pitch types (pitchers with more than 20 innings pitched and at least 10% usage), we find that the average SL (16.8%) and CH (15.7%) actually outperforms it. There is the chance that CUs produce more called strikes or a better pure whiff rate (SwStr per swing), but this made me wonder what defines a good hammer.

The next step was to split the sample into good vs. bad curveballs. Although the measure has its flaws when observing individual pitch types, wOBA is still a decent marker for all-round success.

The better curveballs get more whiffs but not nearly at the rate you would expect, only about 1.4% better than average. This SwStr gap between good and bad curves (3.1%) is actually smaller than the difference between good and bad sliders and changeups. The xwOBACON is interesting to me, you would think it’d be higher if you consider CUs to be more of a strikeout-or-homerun type of offering. When compared to other pitch types I’ve looked at so far, CUs have more variance across many metrics, so it’s harder to define what a perfect one looks like. This means it’s very likely that you can have just as much success by optimizing for weak contact as you would by optimizing for whiffs.

Not every curveball has to be a strikeout pitch to be successful, as we need to make sure we understand what a player’s individual CU produces and set expectations tailored to his profile. To do this, I split the sample into different spin profiles to see what kind of results we could expect. This will help us understand the value of 12-6 curveballs vs. sweeping curveballs, above- vs. below-average movement, and matchup implications.

We already know that more pitch movement is generally a good thing, but let’s see how it impacts this sample. Now this is where things start getting really weird, but here are our key metrics for groups split by whether they are above- or below-average in movement:

There could be some noise in these numbers, but each of these groups have at least 100 pitchers. Either way I would never have guessed negative-drop curveballs would get anywhere close to the same amount of SwStrs as positve-drop CUs, much less more. The +drop group does outperform the -drop one, but we can’t say for sure that it’s because they get more whiffs or weaker contact. It isn’t always the case, but most of the time in pitch-design sessions the emphasis is on trying to get more depth on the CU, not tail. However, based on these numbers you could make the case for doing exactly that. The +tail group was the best performer even with a negligible difference in SwStr rate. The only way I could think to explain this is that maybe these sweeping curveballs are only being used in RHP vs. RHB/LHP vs. LHB matchups, which would skew in their favor, but the usage is similar on average. The splits back this up as well:

Splits still matter, but apparently the shape of the curveball does not seem to dictate whether you can use it against lefties. This is a smaller sample size (60-80 pitchers in each group), so take it with a grain of salt, but it was important to note. There’s also some overlap in these groups since many guys are above or below average in both drop and tail. At the risk of analyzing a criminally small sample size (roughly 50 per group), let’s look at the guys who only have one or the other:

This article makes it seem like I’m trying to devalue vertical movement and the 12-6 curveball, but that’s not the point. My point is that curveballs can profile in many ways and be effective. There is also a lot of value in sweeping curveballs that I think has gone underappreciated. I wonder why they did so well in comparison to 12-6 CUs in this study. Maybe it’s because the movement in both planes challenges the batter’s posture/timing more. It may also be that the value added by increasing drop is not linear, and if so, then we might be able to identify the points where the value plateaus and adding tail instead becomes more beneficial. This may turn out to be unactionable info due to variance in the data caused by either small sample sizes or noise from the many ways that pitchers use these CUs within the context of their arsenal. Regardless, it deserves further research, and it’s something to consider the next time we design a breaking ball.

*Original version of this article was posted to my Reddit page here.

*I’m a private coach & analyst based out of Dallas, Texas, and am always looking to connect with other baseball minds on Twitter @jacob_foster17.

We hoped you liked reading What Actually Makes a Curveball Effective? by Jacob Foster!

Please support FanGraphs by becoming a member. We publish thousands of articles a year, host multiple podcasts, and have an ever growing database of baseball stats.

FanGraphs does not have a paywall. With your membership, we can continue to offer the content you've come to rely on and add to our unique baseball coverage.

Support FanGraphs

newest oldest most voted

Great stuff. I wonder if part of the reason swing strike rate is drop agnostic is how it pairs with your other pitches. For example, if you have a fastball with a lot of ride (and you are trying to change the eye level of the hitter with the fastball curveball pair) then you can afford to have less drop. The opposite I guess would be true (less ride, need more drop in the curveball)