Is "High Heat" A Myth?

I'm a big fan of strikeouts. Unlike Steve McCatty, I think they are vastly important to being a successful major league pitcher. I probably value strikeouts as much if not more than anyone reading this piece.

A pitcher's ability to get a batter to swing and miss ("whiff") has been shown to be pretty important in helping pitchers to rack up strikeouts.

By simple logic and the transitive property, If strikeouts are important to me and whiffs are important to strikeouts then, whiffs should be important to me as well. This conclusion then brought to me what I think is a very logical question.

If whiffs explain strikeouts then what explains whiffs?

I decided it may be a worthwhile venture to go pitch-by-pitch across baseball and try to find what characteristics may differentiate pitchers in terms of inducing a swing and miss from batters.

The best place to go when attempting to answer this sort of question is PITCHf/x data.

I started my journey by looking at Dan Brooks' PITCHf/x Leaderboards over at Baseball Prospectus and focused only on four-seam fastball.

Study #1

My hypothesis was that pitchers who had more "stuff" would induce more whiffs on average.

Luckily, the leaderboards provide an easy why to go about testing that hypothesis, as metrics that measure "stuff" are positioned right by whiff per swing rate.

I took this information and ran a multiple regression with every pitcher who threw at least 200 four-seam fastballs in 2012 (n=180). Horizontal movement, vertical movement and velocity were the three separate predictors and whiff per swing rate was the dependent variable.

The result was a fairly weak, but not horrible, correlation (r=.35). Velocity and vertical movement were both very significant and horizontal movement was trending towards significance at a 95 percent confidence level (p=.079), with this resulting equation:

Whiff/Swing = -.358 + .0048*Velo + .0026*H-Mov + .0067*V-Mov

This result supported my hypothesis, but in no way should it set the world on fire.

If you walked up to a random Joe on the street who knew anything about baseball and told him that pitchers who throw harder with more movement (have better "stuff") on average get batters to swing and miss more often, I think his response would be something along the lines of "Obviously?!".

"Pure stuff" only explained 12 percent of the variation in whiff rate, which isn't a whole a lot. However, there's obviously a lot of factors that go into a swing that are not being considered in this regression, such as the count, the hitter's ability, the location of the pitch, the situation in the game, the pitcher's deception, the sequence of pitches and many more.

Some of those can be factored in, others can't, but the more predictors you add into this model the more complex it would become which could cause some issues.

I personally think that based on the hypothesis and the results of the test, it's fairly safe to conclude that pitchers who have better fastballs (harder and move more) induce more whiffs on average.

Like I said earlier though, that should not shock anyone; thus, I decided to delve deeper.

Study #2

My next idea was to test this conclusion at the individual level. The first study compared pitchers against each other based on their average velocity, average horizontal movement, average vertical movement and average whiff per swing rate. The obvious theme there is average, my major issue with that is pitchers don't throw their average fastball on every pitch, there's variation.

I decided to test whether an individual pitcher could dial it up or throw a nastier fastball to induce a whiff on a per pitch basis. The only way to test this was to dust off my handy PITCHf/x database.

I pulled data on 50 randomly selected pitchers who threw at least 200 four seam fastballs in 2012. I only looked at pitches that were four seam fastballs and resulted in a swing. All swings that did not result in a whiff were classified as a 0 and in turn all swings that resulted in a whiff were classified as 1.

A separate multiple regression was run on each pitcher in an attempt to predict whether or not the result would be a 0 (non-whiff) or 1 (whiff). Based on the original study, I expected to find that pitchers were throwing their fastballs harder and with more movement when a whiff occurred.

Well, it turns out that I was wrong.

For the majority of pitchers, whiffs were not occurring on fastballs with more movement (as those predictors did not come back as significant) and it seemed that only a few pitchers were throwing harder when a whiff occurred.

Before I bowed my head and disappointedly returned to the drawing board, I decided to include both horizontal and vertical location as predictors.

This actually brought about a pretty interesting result.

In most cases horizontal location (if the pitch was on the edge or not) was not significant; however, the vertical location of fastballs were statistically significant.

I broke down the results into three separate columns in a table below.

• The first column shows the correlation between vertical location/velocity and whiffs,
• The second column shows whether or not vertical location was statistically significant at a 95 percent confidence level
• The third column shows whether or not velocity was statistically significant at a 95 percent confidence level

Pitcher	r	Vert. Location	Velocity
Zach McAllister	0.33	Yes	No
Chris Sale	0.32	Yes	No*
Danny Duffy	0.31	Yes	No
Drew Hutchison	0.3	Yes	Yes
Jordan Lyles	0.28	Yes	No
Roy Oswalt	0.28	Yes	Yes
Liam Hendriks	0.27	Yes	No
Franklin Morales	0.27	Yes	No
Jason Hammel	0.25	Yes	No
Wei-Yen Chen	0.25	Yes	Yes
Bud Norris	0.25	Yes	Yes
Hiroki Kuroda	0.25	Yes	No
Matt Harvey	0.25	Yes	Yes
Neftali Feliz	0.24	Yes	No*
C.J. Wilson	0.24	Yes	No
Yovani Gallardo	0.24	Yes	Yes
Justin Masterson	0.24	Yes	No
Brian Duensing	0.24	Yes	No
Chad Billingsley	0.23	Yes	Yes
Felix Hernandez	0.22	Yes	No
Juan Niascio	0.22	Yes	No
Francisco Liriano	0.22	No*	No
Jordan Zimmermann	0.21	Yes	No
Mike Minor	0.21	Yes	Yes
James McDonald	0.21	Yes	No
Gavin Floyd	0.2	Yes	No
Justin Verlander	0.19	Yes	Yes
Daniel Bard	0.17	Yes	No*
Mark Rogers	0.16	No*	No
Brad Lincoln	0.16	Yes	No
Johnny Cueto	0.16	Yes	No
Christian Friedrich	0.14	Yes	No
Stephen Strasburg	0.14	Yes	No
Jarrod Parker	0.14	Yes	No
Miguel Gonzalez	0.14	Yes	No
Jaime Garcia	0.12	No	No
Ivan Nova	0.11	No*	No
A.J. Burnett	0.11	Yes	No*
Henderson Alvarez	0.1	No	Yes
Zack Greinke	0.1	Yes	No
James Shields	0.09	No	No
Casey Kelly	0.09	No	No
Edwin Jackson	0.08	Yes	No
Jeff Locke	0.08	No	No
Josh Outman	0.07	No	No
Tommy Hunter	0.05	No	No
Josh Beckett	0.05	No	No
Ervin Santana	0.05	No	No
Garrett Richards	0.02	No	No
Clay Buchholz	0.02	No	No

*-Indicates that the predictor was significant at a 90 percent confidence level

All of these correlations are pretty weak, but that should be expected when attempting to predict a result on a pitch-by-pitch basis.

For the majority of pitchers (70%), vertical location was a statistically significant predictor of whiffs at the 95 percent level and that percentage was even higher (76%) at the 90 percent confidence interval.

It's also pretty clear from this table the velocity of the pitches that resulted in whiffs, in most cases, were not different from other pitches where a swing took place, as velocity was only a significant predictor of whiffs for 20% of the pitchers (28% at 90 percent confidence).

Also, I must note that for a few of the pitchers were velocity came back significant the relationship was actually negative, meaning that on the whiffs their pitches were not actually thrown harder.

Despite its weak predictive ability velocity was left in the model for a very specific reason. Velocity was used as an attempt to control for just how much vertical location was explaining whiffs regardless of velocity.

It seems to me, based on the results of this sample, that on average individual pitchers generate more whiffs on fastballs that are higher in the zone relative to their other fastballs.

This may seem like an obvious conclusion.

The idea that batters have to speed up their bats to catch up to higher fastballs is something that is generally accepted; the notion of "high heat".

At the same time, just because something is generally accepted across baseball does not always mean that statistical evidence will end backing that assumption. Many sabermetricians have done great work to debunk what was held for years as "generally accepted".

While I think that is all well and good, I also find it just as fascinating and fulfilling when the results actually back preconceived notions.

The PITCHf/x data used in the first regression came courtesy of Baseball Prospectus.

The PITCHf/x database used for the second regression came courtesy of our friend Jeff Zimmerman.

Also please note that there could be discrepancies between the two studies as BP's numbers use pitch classifications from Pitch Info LLC and Zimmerman's data uses MLBAM's classifications.

You can follow me on twitter @Glenn_DuPaul