wPDI & CSW: Residuals

Introduction

This is the third article in my series – wPDI & CSW. You can catch up by reading the first two articles – on called strikes and whiffs – found here and here.

Here is a quick recap of what we have covered so far:

In this series, we are looking at the PitcherList metric, CSW and how it relates to my plate discipline framework, wPDI. Last year’s FSWA Research Article of the Year by Alex Fast featured CSW, which is defined as:

Called Strikes + Whiffs
Total Pitches

With the Weighted Plate Discipline Index (wPDI) framework, all pitches are classified into six different outcomes as follows:

Each outcome is then assigned a weight, or an index. A% through F% are the percent of pitches thrown in each outcome. The general formula for wPDI, the Weighted Plate Discipline Index is given as:

wPDI = Index_A * A% + Index_B * B% + Index_C * C% + Index_D * D% + Index_E * E% + Index_F * F%

In the first two articles of my series, we tackled various topics surrounding pitcher whiffs and called strikes. We touched on the effectiveness of umpires, on the correlation of foul balls to swinging strikes, and dealt with a few other assorted nuggets of understanding. Our primary objective was to express CSW within the wPDI framework. We landed on the following equation:

wPDI_CSW = 105% * A% +  10% * C% +110% * D% + 90% * F%

Our wPDI approximation of Fast’s CSW formula yielded a 92% correlation coefficient, which can be seen graphically here:

In today’s article, we will look at the pitchers who exhibited the largest differences between the PitcherList version of deception, and my own. As always, our in-depth look will tangentially lead to some interesting discussions and other findings.

2019 Leaderboard

Let’s start out by looking at the 2019 wPDI_CSW leaderboard. Below are the top CSW plate discipline pitchers with a minimum of 250 pitches thrown last season.

Almost all of the elite members of the original wPDI formula appear at the top of the CSW version of the framework. That is hardly a coincidence, as four of the indexes are almost identical.

Andrew Kittredge and Sergio Romo are two relievers that seem to fall from the original list. The key difference can be found in their Outcome B component (out of zone, swung on, contact made). The aforementioned pair excel in generating swings out of the zone, albeit with contact. However, the quality of contact that they induce is superb. Kittredge and Romo sported hard hit contact rates of a 29% and 32% (respectively) in 2019. CSW does not look at any contact scenarios, whereas the original wPDI formula does.

For completion, as far as starting pitchers go – Stephen Strasberg, Zac Gallen and Mike Clevinger are the notable starting pitchers who fall from the original wPDI formula. The reasoning for their departure is in their Component E (in zone, swung on, contact made). The trio excelled at limiting contact in the zone (exceptionally low values of Component E), which was represented as a larger factor in the original formula. CSW does not distinguish between contact in and out of the zone.

Let’s now take a look at the residuals between CSW and its equation approximation, which is the main subject of today’s discussion.

Over-Fitters

Below are the pitchers who exhibit with the largest difference between wPDI_CSW to pure CSW, i.e. the over-fitters.

The first item that jumps out to me is that these are primarily relievers, or more notably – those with few innings pitched on the season. Jace Fry, Luis Garcia, Joshua James, and Francisco Liriano have the largest differences of players with more than 50 innings, but all still have 70 innings or fewer. The top two pitchers with at least 100 innings were Dakota Hudson (+.009) and Blake Snell (+.007).

We will talk a bit about Blake Snell in a moment, but at first glance, it seems that errors in the approximation formula has more to do with sample size than with anything else. That of course, is an excellent result.

As for Blake Snell, let’s dive a bit further into his 2019 season as far as plate discipline goes.

We have seen that Snell’s Outcome A was the largest of any starting pitcher in baseball in 2019. It was the 3^rd highest if we include relievers. His swinging strike rate was 17.7%, first in the majors for pitchers with at least 80 accumulated innings.

But was that the source of the residual?

My first thought was to check the foul tips for any abnormality. The league average ratio in 2019 of foul tips to pitches is 0.89%. Snell’s ratio was 1.00% – which squashes that thought. CSW (which includes foul tips) would be higher in this case.

Let’s now divide CSW into its underlying elements – called strikes and whiffs. We see that the majority of the discrepancy lies with his called strikes. The difference in CS% is +.005, but the W% is only +.002.

The key is in where Snell’s pitches are called strikes.

Snell generated far fewer called strikes out of the zone as a percentage of his called strikes in 2019. In fact, among starting pitchers – he had the lowest percentage. A few weeks back, I wrote about the debate of technology versus humanity. We discussed whether deceiving the umpire is a wanted sign of skill or not. Snell is an outlier case in this larger debate.

For Blake Snell in 2019, even with his incredible swinging strike rate, he was not getting a normal amount of in-zone called strikes. Was this related? Were umpires not giving him borderline calls because he was generating too many whiffs? Add that to the ever-growing list of future items to look at. For now, it seems (and we will assume) that Snell’s 2019 was somewhat random. Snell’s in-zone strikes comprised 82% of all called strikes from 2017 to 2018, which was close to an average figure. It is highly likely that this 2019 deviation was random noise caused by … umpires.

Under-Fitters

Below are the pitchers with the largest observed negative residuals between wPDI_CSW to pure CSW, i.e. the under-fitters.

What jumps out to me is the following:

• The magnitude of the top under-fitters is far larger than that of the over-fitters.
• The players with the largest difference from empirical to formula tend to be low valued players.
• We still mainly see low innings present, but a few high-volume starting pitchers now begin to creep in.

Let’s do a deeper dive into the first starting pitcher of note on this list, David Price.

What a stark difference to Snell’s 2019 season – especially in the swings and misses. Price’s Outcome A is quite poor, at 5.8%. That seems to be the primary source of his below average wPDI_CSW. If we look into the CSW underlying elements – we see once again that the difference lies within the called strikes. His W% residual is -.003, whereas the CS% is a much more significant -.027.

Taking a look into the zone location detail of Price’s called strikes yields:

The zone data depicts the mirror image story that we saw in Blake Snell. Where the strikes are called matters. There is also a compounding effect going on. Pitchers who tend throw more of their called strikes in the zone will certainly have a better CSW, but formulaically, they also get to take even more advantage of the wPDI index structure.

Conclusion

The differences between CSW and our wPDI_CSW regression formula are mostly random noise, the effects of which diminish with large sample sizes.

However, we can detect that there is a relationship between the distribution of called strikes (in zone vs. out of the zone) and the over/under fitting of our equation. Pitchers who have larger than average in-zone called strikes will tend to see a better wPDI result, and those with a poor in-zone percentage will have worse results.

The zone relationship is mostly negated for larger sample sizes (and thus for starting pitchers), but is more pronounced for smaller CSW values.

Bottom line – the wPDI_CSW regression is a good one, and may even be more indicative of what it takes to be a good pitcher, as it removes some of the umpire bias that pure CSW displays.