top of page
Kai Franke

PREDICTING PITCHER PERFORMANCE BASED ON STUFF ALONE; PART 3: SLIDERS

MY POSTS WILL NOW MAINLY SHOW UP ON PROSPECTS WORLDWIDE.

LINK FOR POST ON THAT SITE IS HERE.


To see the last post from the Pitch Grades series (Changeups), go to this link here.

To see part one of this series, go to this link here.


Now that I have gone through the fastball, curveball, and changeup pitch performance predictions, it’s time to get to my final article of this series: the Slider. The slider is a very different pitch than the previous ones, as it typically has more gyro spin to it. This, I thought, would make it more difficult to predict how well each pitch would do, but it actually wasn’t.


Once again, I went through the same process as the changeup as I had the same metrics in my linear regression model. As a reminder, I used velocity, true spin rate, horizontal break, vertical break, velocity difference between the fastball and slider, and true spin bauer units. The reason I felt that making this wouldn’t be as hard as I thought was because the linear model would be able to estimate how much true spin makes the slider a good one. So if less true spin gives better results, then the coefficient with it will be negative, the opposite is true for high spin, that will be positive.


Without further ado, here are the formulas for slider pxwOBA and pWhiff% (with ‘p’ meaning predicted): 


Slider pxwOBA = 0.7494 – (Velo. Difference+ * 0.001245) – (Slider True Spin+ * 0.002258) – (Avg. Velo+ * 0.003332) + (True Bauer Units+ * 0.002068) – (Z Break+ (Vertical Break) * 0.0001005) + (X Break+ (Horizontal Break) * 0.00005001)


Slider pWhiff% =  -61.537 + (Velo. Difference+ * 0.191) – (Slider True Spin+ * 0.081) + (Avg. Velo+ * 0.745) – (True Bauer Units+ * 0.097) + (Z Break+ (Vertical Break) * 0.052) – (X Break+ (Horizontal Break) * 0.004)


When we take a look here, you’ll see that for pxwOBA more true spin is bad for your projection as it is negative in the equation, however, take a look at the pWhiff% formula and more true spin is positive and that means that higher spin will get you a higher prediction. This was surprising to me, when you think of a slider, you think of low spin efficiency, but it makes a little bit of sense as you’ll get more total movement throughout the pitch instead of just some late break. That little late break could make it so that a hitter still hits the ball, but either tops it or pops it up, producing a lower xwOBA. 


Velocity still is the most influential factor in predicting performance here, while both horizontal and vertical break seem to not affect it too much, this could be because of the small movement that pitchers get out of the pitch due to, again, low spin efficiency. The difference between fastball and slider velocity is also a big factor for pWhiff%, as it has the second highest coefficient. 


Now let’s take a peak for what you all have been waiting for, the best and worst pitchers ranked by pxwOBA:



* pWhiff% average is 35.6, pxwOBA average is .268

* Whiff% average is 35.7, xwOBA average is .267


A name that stands out as having a nice sweeping slider is Adam Ottavino. People rave about how much movement it gets and it makes sense that he ends up at number 10. Here’s a video of that pitch: 



However, whiffs like this don’t happen as often as you may think for Ottavino, as he actually had a below average whiff% for sliders at 35.3%. My model was only off by 1.1% on that prediction and he’s one of the guys that it was able to predict well. Former teammates Mike Clevinger and Trevor Bauer also end up high on this list at 3 and 4, respectively. They performed similarly in my model as the had pxwOBAs of .201 for Clevinger and .207 for Bauer, and pWhiff%’s of 42.4 and 41.0. Bauer had a rough year last season, but looked fantastic in his first start with 6.1 innings pitched, 13 strikeouts, and 1 earned run allowed. He didn’t throw the slider a ton in that start, only 12 times, but had a 50% whiff% and a .013 xwOBA. Here is Bauer creating an ugly swing when Nicholas Castellanos was a Chicago Cub:



Kyle Freeland wasn’t at his best last year either, and his slider follows that same trend. He had a .311 xwOBA on it and batters only swung and missed 28% of the time, I was able to predict this very well as he had a pxwOBA of .309 and a pWhiff% of 29.8. Coors Field doesn’t fare well to pitchers, but yikes was he bad last year, his best batted ball statistic was his hard hit rate, but even that was only in the 25th percentile. In 2019, he used that slider 30% of the time, his second most used pitch, which could’ve been the reason he struggled, however, it was his best pitch in terms of wOBA at .340, which still isn’t very good. Like Bauer, he looked much better in his first start of 2020, as he went 6 innings and allowed 2 earned runs, he also dropped his slider pitch percentage down to 21.3, hopefully that’s the case for the whole season, as mixing his pitches more may help him succeed.


I was surprised to see that Max Scherzer, Patrick Corbin, and Justin Verlander all weren’t even in the top 50 for my predictions as they all out performed their by miles. Their sliders are revered for being great and the model wrongly didn’t agree, as it was off by .082 in xwOBA and 12.1 in whiff% for Scherzer, .066 and 14.7 for Corbin, and .100 and 5.9 for Verlander. All of their sliders are so good that CBS Sports ranked them all in the top three for their best slider rankings back in May. If you go by xwOBA, they’re all in the top 10, and for whiff%, Corbin is 5th, Scherzer is 6th, and Verlander drops down to 42nd. 


I wanted to investigate why that may be, so I created graphs for each of the stats compared to pxwOBA. Most of them were very scrambled and didn’t really seem like they would’ve had a huge effect on the three guys, but then I saw the Velocity Difference and pxwOBA graph:



All of those data points are very tight and have a definite downward trend, their points are also pretty close to that line, as all of the points are:



*Verlander left, Scherzer middle, Corbin right


This shows that the linear regression possibly brought the final pxwOBAs closer to that line as they’re all in proximity with it. I also wanted to see if the relationship above held true with non-predicted xwOBA as well, here are the findings for that:



It’s not as strong of a correlation, but it still does hold up! So, if a pitcher doesn’t have much velocity disparity between his fastball and slider, it’s likely that it will get crushed. Two examples of that are Jake Arrieta and Clayton Kershaw who only have velocity differences of 2.2 and 3.4, respectively. That lead to Arrieta having a .423 xwOBA and a 21.7 whiff%, Kershaw had a .284 xwOBA and a 34.1 whiff%. 


Once again, we need to see how predictive this model is for estimating performance. So, I figured out the R^2 and made graphs for pWhiff% and pxwOBA. Here are the final results:



The R^2’s are the highest they’ve been out of any of my pitch grades, none of the others even got past 0.15! They aren’t extremely high to where one could say that they’re very precise, but there is somewhat of a predictive value here. I consider this pitch modeling successful, but there is definitely more that could be done.


As stated in previous articles, the slider has to be utilized well, they could get plunked if a righty would throw it to a lefty, or if it’s just left down the middle. But, if you go back and take a look at the two videos of Ottavino and Bauer, you’ll see that both of those whiffs were way out of the strike zone, possibly meaning that the slider is a pitch that doesn’t need to be controlled all of the time, but has to have great movement in order to be pitched that way. That is possibly why the R^2’s are much higher with these models! This concludes my stuff based prediction series, however, improvements could be made in the future to the models that I may end up writing another article to this series in the future to present to you guys my new findings! 

To see the last post from the Pitch Grades series (Changeups), go to this link here.

To see part one of this series (Fastballs and Curveballs), go to this link here.




ALL STATS FROM BASEBALL SAVANT

34 views0 comments

Recent Posts

See All

Ohio State Sports Analytics Conference

I attended the Ohio State Sports Analytics Conference on April 1, 2023. I presented about Predicting Run Value to Evaluate MLB Team and...

Stuff+ RShiny Dashboard

I built a Stuff+ RShiny dashboard intended to help with pitch design based on 2022 MLB pitch data. It is based on a Stuff+ model that I...

Comments


bottom of page