Gerrit Cole is the Pittsburgh Pirates ace. He posted the best fWAR of the seven starters for the Pirates this year at 3.1 wins, despite struggles with the homerun.
However, this was a far cry from Cole’s 2015 campaign when he posed 5.5 WAR and a 2.60 ERA while leading the Pittsburgh Pirates to 98 wins. What is interesting is that so many of Cole’s 2017 peripheral stats are similar to his 2015 campaign with two notable exceptions. Those being that his walk rate and his homerun rates skyrocketing.
This suggests that, in part, batters were less afraid of Cole’s stuff this year, due to a reduced ability to get strikes. This in turn allowed them to be more selective on the pitches they swung at, which ended in more balls in the stands.
So let’s take a look at Cole’s stuff.
According to Statcast, Gerrit Cole featured five types of pitches this season, a four seam fastball, a two seam fastball, a changeup, a knuckle curve and a slider. Here’s the list of the frequency that he threw each of these pitches and number of homeruns hit off of them this year.
|Four Seam Fastball||49.24%||15|
|Two Seam Fastball||10.53%||3|
You can see that the four seam fastball was far and away Cole’s favorite pitch, throwing it nearly half the time. So that’s the Pitch we will begin with.
Cole threw his four seam fastball about half the time in 2017 and, predictably, gave up about half of his homeruns while using it. It took 10.1 pitches per out using this four seamer, which is reasonable for a pitcher’s primary pitch, although lowering that ratio will help Cole return to the sort of dominance that we saw in the 2015 campaign.
The question then becomes, how he does that.
Building the Model
In order to figure that out we have to find the probability of Gerrit Cole’s four seamer being thrown for a strike, given certain input factors. To do this we will be utilizing a logit regression; we’ve previously used one in this article if you’re interested in the mathematics behind it. Essentially all this regression does is give us the probability of an event, like a strikeout, occurring.
We will be running a model of Strikes as a function of pitch speed, pitch location (both side to side, denoted as “x”, and up and down, denoted as “z”), the spin rate of the pitch, the release point of the ball from the pitcher’s hand (also in x and z dimensions), which pitch in the at bat the pitch comes, and the extension home of the pitcher. Additionally, to control for tiredness of the player, we will be noting the number of at bats the player has faced to that point.
We will be modeling for strikes, both called and swinging, as that seems to be the desired outcome for most pitchers. There are obviously other outcomes, like weak contact that may also be preferred, but pitchers that miss bats tend to give up fewer runs than those who don’t.
We will be running two separate models, one for right handed batters and one for left handed batters, in order to control for the obvious difference in pitching to each kind of batter.
The Four Seam Fastball (FSFB)
For Cole’s FSFB against righties we get this model:
And against Lefties we get:
The numbers themselves are somewhat meaningless without a calculator, but the signs on these numbers are quite informative. Any number with a positive next to it means an increased probability of a strike, and numbers with a negative imply a decreased probability of a strike.
Starting at the top of both models, Cole’s release speed is negatively correlated with his strike probability. Initially this seems a bit counterintuitive; Cole is known as a power pitcher, who can touch 100mph on the radar gun and should be able to blow balls by batters.
You can see that the four seam fastball was far and away Cole’s favorite pitch
That is by and large true, however, it is also somewhat intuitive that the faster any pitcher throws, the less control they have, meaning the fewer strikes they throw. This trend plays out in the data as Cole’s pitch speed negatively correlates to his strikes thrown.
Plate_x is the distance from the middle of the strike zone, where anything toward the right-handed batter’s box is negative, and towards the left handed batter’s box is positive. Plate_z is similar to plate_x, only it measures above (positive) or below (negative) of the middle of the zone.
What this model states is that Cole improves his FSFB strike probability when he goes up and in on righties and down and away from lefties. Obviously for this pitch type and each of the other pitches we talk about, this ball placement has diminishing effects as moving too far to either of these extremes means that Cole would be lowering his strike probability because he would be outside of the strike zone. This just means that going up and in or down and away, within or at least very near to the strike zone is more effective than elsewhere.\n
Additionally, I would not suggest only working balls in this direction as hitters can obviously adjust. What I am suggesting is that more utilization of these areas would likely help Cole to miss bats more often, resulting in fewer balls in play and fewer runs scored.
What’s interesting is that this up an in location on righties does not change for Cole across all five pitches we will be looking at. This makes sense, a well-placed pitch on the hands of a batter is significantly more difficult to hit than one that is in the middle or even outer third of the strike zone. The being said, the pitch placement does vary for lefties depending on the pitch.
Spin rate is a proxy variable for movement on the ball. The greater the spin rate the more “life” the ball has. For both lefties and righties, the more RPMs Cole gives his FSFB the more likely it is to go for a strike.
Pitch number, or the number of pitches into the at bat that this pitch came, is interesting. Righties were more susceptible to get a strike on Gerrit Cole’s FSFB the later into the at bat that this pitch type came, whereas lefties were more likely to get a strike on the same pitch the earlier into the at bat they were.
The data on Gerrit Cole’s release was largely statistically insignificant to both righties and lefties. This means that Cole is either very consistent in his delivery of the FSFB and so we don’t get enough variance to have significant findings, or whatever variance there is, has such a small impact on the probability of Cole getting a strike that we can’t tell given our sample size. In either, case we’ll just move on.
In total Cole should work a controllable four seam fastball, with movement, up and in on righties, and down and away on lefties in order to maximize his chances of getting strikes.
The Two Seam Fastball (TSFB)
The two seamer is a pitch that hasn’t been particularly popular with Cole in the Statcast era, although he did use it more this season than in the previous two by about two percent. This year Cole’s two seamer accounted for 10.5 percent of his pitches, and just 9.7 percent of his homeruns, meaning it was relatively effective at keeping the ball in the park.
When we run our same basic model on Cole’s two seam pitches we get somewhat similar results to those of his FSFB. Pitch speed again is negatively correlated to strikes due to lack of control and increased spin rates meant a greater probability of strikes. This time, however, Cole had improved probability of a strike by going up and in on both righties and lefties. Cole also saw improvements in strike probability with both righties and lefties the deeper into the at bat that he used it.
Cole got an out with his TSFB every 5.4 times he used it, or a little less than twice as effective as his FSFB. To what degree this jump in effectiveness is because of how little Cole uses it is hard to tell, but if he can sustain it over a larger portion of his pitches, using the two seamer could be part of the fix for Cole for 2018.
One area of concern is that he gets the TSFB in for strikes just 44 percent of the time, relative to the 50 percent of the time that his four seamer does. Even still, finding how to throw it for strikes and better utilization of the two seam fastball may be an option for Cole going forward.
Cole’s change up was effective this year by getting opposing players out at an average of 5.9 pitches per out, second only to his two seamer. That being said, it was also pretty effective at getting knocked around posting the worst difference between pitch utilization (10.7 percent) and percent of homeruns hit off of it (12.7 percent).
It isn’t immediately clear why this is, but from anecdotal evidence of my watching Cole pitch, it seemed his problem came largely when he hung changeups right in the wheelhouse of hitters.
When we run our logit analysis on both lefties and righties, we get some pretty clear results. The obvious ones are generally things we expect from a changeup; we want it to be a slower pitch with relatively little spin to it.
Now for the more technical stuff.
For Cole to use his changeup effectively, he has to place it up and inside on batters. When working changeups inside on hitters, Cole increases his likelihood of a strike by an average of 0.55 percent per additional inch inside. When working balls up, Cole improves his strike probability by an average of 1.4 percent per inch.
His release point also plays an important role in determining the strike probability. Increasing his release height just 1 additional inch would improve his strike probability by about 3 percentage points.
Cole’s changeup has been a good pitch for him in the past. Setting aside the home runs given up on it this season, the changeup was pretty effective at generating outs this season. Finding way to shave off a few homeruns from it next season is a must for Cole to return to dominance.
The story of Cole’s slider is similar to that of his changeup this year in terms of getting knocked around. Cole threw it 17.4 percent of the time but gave up 19.3 percent of his homeruns with it, making it the second worst pitch for the homerun, relatively speaking. He threw the slider for a strike 44.7 percent of the time and took 9.3 sliders thrown to record an out.
The up and in placement was again effective on both lefties and righties. One interesting take-away from these models is that Cole’s slider was more effective against lefties when he added some velocity and more effective against righties when he reduced the velocity. The other interesting bit is that Cole’s slider was more effective the earlier in the count he used it.
Given that the slider was Cole’s secondary pitch after the FSFB, the slider is actually the more effective of the two at getting early strikes.
The other interesting fact is that Cole’s slider tended to be more effective with a bit less spin on it. Why this is, is a bit unclear. Ostensibly, more movement on a slider the more effective it is. It could be the case that Cole is already operating at a good level of movement on his slider so any more movement means loss on control. Alternatively, it could just be a factor of how Statcast categorizes and measures what a slider is. In either case it is a bit outside of the scope of this article to identify the solution.
Cole improving the strike rate of his slider is an important factor in him returning to form. Increasing the effect of his second most utilized pitch would go a long way towards making batters afraid of Cole’s pitches again.
The Knuckle Curve
Gerrit Cole’s Curveball is an effective strike pitch, getting thrown for a strike 47.5 percent of the time. Additionally it was his most effective pitch at keeping balls in the park, accounting for just 9.7 percent of home runs despite being 12.2 percent of Cole’s pitches thrown. It is not as effective as an out pitch, needing to be thrown 11.4 times per out recorded on it, which is the least efficient rate of Cole’s 5 pitches. Cole has also been throwing the curveball more and more since 2015 when he threw it just 7.8 percent of the time; though, throwing more curves has been a league wide trend over this time period.
Our analysis of Cole’s curveball gives us a few interesting results. The first is that to lefties and righties alike, the faster the curveball is thrown, the better the likelihood of a strike.
Fans and observers alike have long asked why Gerrit Cole refuses to use his curveball more often.
The next interesting thing is that the spin rate had a negative impact on the likelihood of a strike, similar to that of the slider. Again this is counterintuitive, however it lends a bit of evidence to the scenario where Cole is actually operating at the appropriate level of spin, so any additional spin gives the ball too much break and goes for a ball.
The final interesting point is the location of the curve to lefties. Cole’s curve was most effective up and away from lefties. Since this is the same spot as the most effective place for righties, only up and in on them, this tells us that Cole was most effective at generating strikes when his curveball hit the same spot for both types of batters.
Much of this information seems rather novel on its face; slight increases of a few percentages in strike likelihood seem almost too small to matter. However, if Cole had thrown just 5 percent more strikes this season, meaning 5 percent fewer balls and hits, his WHIP would likely have been in the to 15 of NL starters, rather than the top 25.
Additionally, Cole has the stuff to be a Cy Young contender, as he showed in 2015. 2016 saw a regression in Cole’s abilities, but 2017 was actually a bounce-back year when we discount his inflated HR/FB rate. The key to Cole’s return to form will be finding ways to better attack the strike zone to prevent batters from sitting back and waiting for a mistake.
For our sake it is difficult to separate out why exactly these things would make each pitch more effective. Perhaps, for instance with location, Cole would be more effective in those spots because he rarely goes there so batters aren’t expecting it, but batters could adjust to that change. However, it is exactly those kinds of changes that Cole has to continually make, to attack zones in ways that he hasn’t before, that will keep hitters off balance and allow Cole to return to dominant form.