Some pitching stats, like Called Strikes plus Whiffs (CSW) and Strikeouts minus Walks per inning (K-BB)/IP, are useful to get an initial estimate of pitcher’s performances due to their descriptive capacities, but also because of their predictability qualities, confirmed through their correlation with other stats. As an example, in a constrained data set, CSW and (K-BB)/IP for pitchers with more than six games started in the 2017 regular season, correlated 61.4% and 66.3% to the 2018 regular season SIERA for pitchers with more than 13 GS (this number of games started was the sweet point where correlation began to be higher but even at just 6 GS, a high correlation is noticeable):
Why to correlate with SIERA? And, specifically in this case, why with the following year’s version of it? Because it provides a good understanding of what a pitcher’s proficiency really is now and how that relates to ERA. Also, SIERA has a good relationship with fWAR, e.g., in 2018 it had a 73.43% correlation with it (R2=0.5392).
This doesn’t mean that having a good SIERA will de facto increase your value in terms of fWAR, but it is a good stepping stone so, predicting (I hate to use that word) next year’s SIERA is a good first step into finding a pitcher’s future value.
2017 (K-BB)/IP does a better job to estimate 2018 SIERA than 2017 CSW, even better than 2017 SIERA itself when evaluating for different GS thresholds. This is really nice because (K-BB)/IP is very simple to calculate and its widely used variant, Strikeouts percentage minus Base on Balls percentage (K%-BB%), is too. The latter did just marginally worse (not shown in the graph).
One caveat with (K-BB)/IP or K%-BB% is that, because of their nature, they could make us ignore guys like the 2017’s version of Aaron Nola, who after posting a K%-BB% of 19.5%, 16th among qualifying pitchers, had a 4.5 fWAR in 2018, good for 9th overall. His 2017 CSW was, at 32.3%, a better estimator in this case.
That’s one of the reasons we shouldn’t get fixated with just one stat even if it proves to be useful in a lot of situations because there is not a single catch-it-all approach. What could be done, and this is something I like to try, is to identify some stats that can provide a high floor and help us separate as much of the noise from the signal as possible, that way our estimation floor is still high but it’s also wider, sort of speaking.
Using four stats, CSW, (K-BB)/IP, SIERA, and real F-Strike%, I’ve been able to balance them in a meaningful manner into a single indicator, with minor adjustments for scale and weight. Initial tests for this Simple Pitching Estimation indeX, speX, show a correlation with the following year’s SIERA as good as (K-BB)/IP or K%-BB% but also doing a better job at not overlooking those cases in which the pitcher can succeed without having high K%-BB% rates.
I’ve used 2017-2018 as an example but data for other years behave similarly.
The actual formula for speX is:
speX = [2.12766 * CSW% + 60.606 * (K-BB)/IP + 10.1523 * (10-SIERA) + 1.16279 * Real F-Strike%] * 0.3125
It’s very straightforward, basically a sum of weighted terms, which allows getting a proportional and scaled index.
To exemplify its use, the following chart ranks every pitcher in 2020 with more than 40 IP during regular season:
It is important to remember we are using a smaller than desirable sample here due to the shortened season so proceed with caution when interpreting these stats. For a better analysis, I am partial to adding the final 60 games from 2019 but for the moment let’s look at the available results.
The following graph allows us to look at the list from a broader perspective; better pitchers will have a darker green upper left location while the opposite will be darker red in the lower right of the graph (gliding over any bubble will display the complete info for that player).
We can think of this as the graphical representation of the dominance capability by means of power as the axes are representing two stats, CSW and (K-BB)/IP which are in their core a reflection of the striking out proficiency of the pitcher.
If we graph speX in a SIERA vs real F-Strike% relationship, we’ll get a similar representation:
This relationship can be seen as dominance by control and other factors (although it does not exclude those showed in the previous graph).
Doing a crossover among the two relationships, we find that there are pitchers integrally good in both of them like Bieber and deGrom, others that achieve their speX rating more by means of power like Glasnow or by control like Kershaw, and others that do terrible by all accounts, like Castellani, Dunn, Fedde, and Cease.
In the middle, there are tons of pitchers that are graded regardless of their primary ability, so it is easier to start the process of comparing them; this is how using a previous version of speX, very early in the season I could estimate the possibilities from Kevin Gausman and Framber Valdez to achieve good results in 2020.
I like to think of this index as something related to what Tom Tango calls a naive model when referring to his forecast systems, the Marcels. By no means speX is a model, of course, it is just a simple stat but in its simplicity and naivety, it works very well as a soundness check before/during using more complex ways of estimation and forecasting.
EE, Data geek, Baseball fan. Twitter: @camarcano