Forecasting Pitcher Strikeout Rates

It’s no secret that strikeouts are important parts of a baseball game. While there’s plenty of evidence to show that it’s just another out (no more or less valuable than a fielder’s choice, by some measures) strikeouts are undeniably important to pitchers because they are the pitcher’s only means of earning an out alone. Where all other outs rely on some kind of defensive assistance, strikeouts are all about the pitcher – and have an added bonus of not allowing the opportunity for runners to advance.

It must be important to know what influences a starting pitcher’s ability to gather strikeouts over the course of a season, then. Starting pitchers who strike out more batters should, on average and all else equal, give up fewer runs and contribute more wins to their respective teams. If it’s possible to identify what affects strikeout rates on a nine-inning basis, or K/9, it’s also possible to identify pitchers who should be expected to improve over the balance of the season.

In an attempt to forecast strikeout rates for pitchers, I gathered pitcher-controlled stats from 2007-2015, the years for which PITCHf/x data is available, and only from starters. The 16(!) variables I considered are shown below against K/9 rates, along with a linear trendline that indicates whether a variable is associated with improved or worsened K/9 rates.

Discuss Orioles pitching and possible additions (or subtractions) to the team on the BSL Forums.

The Features

Right away we can see that pitchers who throw harder are more likely to strike out batters. Interestingly, pitchers who throw in the zone less are more likely to strike out batters, presumably because they have better stuff and get batters to swing and miss at pitches outside the zone, where batters wouldn’t be able to do damage even if they connected anyway.

In the second row, we find that pace of play appears to affect pitchers’ strikeout rates pretty significantly. As pitchers take more time between pitches, maybe they let batters get in their own heads. Maybe kicking dirt and asjusting batting gloves isn’t as helpful as players originally thought! This could be a case of correlation and not causation, though: K/9 rates have risen dramatically over time, and so have the length of games (up until 2015 sped them up ever so slightly).

 

Games take a long time in part because pitchers and batters take (or took, before the new rules were instituted) time between pitches. If high-strikeout pitchers just happened to be in vogue during that same period, well, then, so be it. Then again, some players felt that they performed better with more time between pitches. It could be a mix of coincidence and substance, but it’ll take a long time to figure that out.

Contrary to what I expected, the difference in speed between a pitcher’s fastball and his changeup doesn’t significantly affect his ability to strike batters out. I thought that getting a batter in front of a changeup would be a regular strategy for strikeouts. Instead, getting off on the right foot is more important: K/9 increases dramatically as a pitcher throws more first-pitch strikes. It’s a lot easier to get into a two-strike count if the batter is already in a 0-1 hole.

Sliders and curveballs – both high movement pitches – were included, but because not every pitcher throws both, I combined them to make the fifth row: Strikeout Pitch, or K_pitch information. If these pitches stay up, they tend not to be as effective. They don’t seem to be better or worse for strikeouts with horizontal movement. Instead, frequency and speed (along with dropping from their release point) are key. The harder strikeout pitches are thrown, the less time the batter has to react to a pitch that doesn’t end up where he expects it, and presumably, the more they move. The more often these pitches are thrown, the more often the batter is swinging at something that isn’t there. Pretty simple math.

The Model

In an attempt to predict strikeout rates based on the features listed above (I skipped the third and fourth rows in favor of the fifth row because it allowed for more observations), I chose to work with a linear regression. It’s simple, fast, and none of the plots showed much of a shape, much less some kind of curve. If you’re the type of person who’s interested in looking at the residual plots to see if there’s a pattern that would indicate a need for a special model specification, I’ve got you covered.

The resulting equation for predicting K/9 is a nice long equation, but it’s not particularly effective; the R² value, or the percent of variation in K/9 rates that can be described by the features used, is a paltry 36.5%. Yuck. At the very least, it can give us an outside sense of who might improve over the rest of 2015, and who might not perform as well.

Plenty of factors aren’t included here. It’s arguable that pitchers aren’t the only people responsible for their K/9 rate. Batterymates can help out with superb framing, for example, and a pitcher with a stud receiver behind the plate can carry higher K/9 than an identical pitcher without a framing artist to work with.

The Orioles

Unfortunately, on Wei-Yin Chen, Ubaldo Jimenez, and Chris Tillman have all of the data necessary to be run through the model. Kevin Gausman, for example, doesn’t have any sliders or curveballs listed on Baseball Prospectus’ PITCHf/x leaderboard, which is where I got my data. So what to expect from Chen, Jimenez, and Tillman?

Despite this model not scoring high as a predictor, it’s pretty much nailed Chen and Tillman’s K/9 rates thus far in 2015. Meanwhile, this model predicts Jimenez’ K/9 rate to be more than 2 strikeouts less than it currently is in 2015. I’m inclined to say that the model is wrong in this case; Jimenez’ strikeout rate is pretty much in line with his career 8.3 K/9 (as are Chen and Tillman’s, for that matter).

It looks like what we see with the Orioles is what we’re going to get.

The Rest of Baseball

The pitchers who should be expected to improve their strikeout rates (based on our admittedly iffy model) most dramatically down the stretch are:

Interestingly, we see a name on the market here: Yovani Gallardo is a potential plan B for the Dodgers, apparently, and the Rangers are listening to offers on him. He pitched out of his shorts for a stretch and is widely expected to come back to earth a bit, but an improved strikeout rate in the last third of the season could be a boon for whatever team employs Gallardo.

And the pitchers expected to strike out fewer batters as playoff races heat up are:

Ubaldo shows up here, but I’ve already covered how his 2015 K/9 rate is pretty much in line with career norms. Aces for a number of playoff hopefuls do show up here though: Dallas Keuchel, Carlos Carrasco, and Chris Archer might all decline as the season continues. Missteps by those three could help the Orioles in a pennant or wild card battle at the end of the season.

Part of me thinks that Kershaw, Sale, and Scherzer show up here because they have – and have always sustained – incredibly high K/9 rates relative to the rest of the league. At least, I hope it’s nothing more than an illusion: Kershaw and Scherzer are both on my fantasy team.