For years sabermetrics has theorized that pitchers can’t control what happens after a batter strikes a ball. Whether it bloops in for a hit, rockets its way to an outfielder’s glove or lands just inches outside the foul line, it’s a consequence of the batter and the defense, but not the pitcher. That argument is what led to Fielding Independent Pitching (FIP), a statistic that attempts to quantify a pitcher’s performance only for the events for which we know the pitcher has definite responsibility (walks, strikeouts and home runs — in sabermetrics parlance, the “Three True Outcomes”). FIP throws out base hits completely on the basis that they are partially a product of the defense.

But now we have Statcast. The new technology that collects data on the position and velocity of the ball and the players on the field is beginning to change what we know (or thought we knew) about all sorts of things — pitching included. And that’s raising new questions about how much effect a pitcher can have on a ball once it’s put in play. The answer might be: a lot.

But first, let’s talk batters. Last week, I used Statcast to show that harder-struck pitches tend to fall for hits more often. That, generally, makes hard contact good for the hitter and bad for the pitcher. Obviously, some hitters are better at hitting the ball hard. What Statcast tells us is that some pitchers are better at *making *batters hit the ball softly.

That’s not to say pitchers hold the upper hand. In my models of batted ball velocity that incorporate the pitcher, batter and ballpark, the batter’s effect dominates the pitcher’s. A ball’s exit velocity after a bat strikes it is about five times more the batter’s doing than the pitcher’s. This fact seems to partially vindicate FIP — batters really are the ones in control.

At the same time, the pitcher’s effect is not negligible. While the best batters increase batted ball velocity by as much as 7-8 mph, the best pitchers suppress it by 1.5 mph compared with the average pitcher.

That has real significance: Such a decrease roughly equates to a 13-point decrease in batting average on balls in play (BABIP) for a given batted ball. Over the course of a game, the pitchers who can best decrease exit velocity save about a quarter of a run (on average). A quarter of a run doesn’t sound like much? Multiplied over a season, all those quarters of a run add up to about one win of value.

So that means FIP is flawed as an overall value metric, at least for some pitchers. Who are those pitchers? Here’s a table of all 485 pitchers with batted ball data this season as of the writing of this article. Search for your favorite pitcher and see how many miles per hour he takes away from or adds to the average batted ball.

The five best pitchers in the league: the Baltimore Orioles’ Wei-Yin Chen (balls leave the bat 1.63 mph slower than average when Chen pitches); the Chicago White Sox’s Chris Sale (1.56 mph); the Los Angeles Angels’ Garrett Richards (1.53 mph); the St. Louis Cardinals’ Adam Wainwright (1.46 mph, before he experienced a season-ending injury April 25); and the Houston Astros’ Dallas Keuchel (1.40 mph). Many of these pitchers are bona fide aces, most obviously Sale, Richards and Wainwright (Clayton Kershaw also lurks in ninth place). These players are not only adept at managing contact, they are also skilled strikeout artists. FIP accurately tabs them as great pitchers even without information about their contact-controlling abilities.

These pitchers control their opposition’s quality of contact partly by driving the hitters into bad counts. In pitcher’s counts, hitters tend to put weak, defensive swings on the ball, resulting in glancing contact. About 15 percent of pitchers’ exit velocity suppression comes from controlling the count. Richards, for example, has reached two-strike counts in 92 of his opponents’ plate appearances, compared with three-ball counts in only 38; in the former, hitters have a .271 BABIP, whereas in the latter, they have a .333 BABIP.

Chen is an intriguing case. The best at suppressing batted ball velocity, Chen also has the largest gap between ERA and FIP among qualified starters. In fact, Chen has put up a sizable gap between his ERA and FIP in three of the four years in which he’s pitched in MLB. Lacking batted ball velocity in years prior, we cannot say that his skill is consistent, but his results appear to be.

FIP doesn’t only fail to credit the pitchers who manage their opponent’s batted ball velocity, it also fails to blame bad ones who consistently get hit hard.

The league’s bottom five in that respect: the Pittsburgh Pirates’ Vance Worley (1.43 mph added to a ball’s exit velocity, compared with average); the Tampa Bay Rays’ Nate Karns (1.39 mph); the San Francisco Giants’ Tim Lincecum (1.16 mph); the Kansas City Royals’ Yordano Ventura (1.11 mph); and the San Diego Padres’ James Shields (1.06 mph). Just as the best pitchers tended to be better than average even by FIP, these five are worse. And while the aces use the count to their advantage, these pitchers are liable to find themselves in hitter’s counts, which causes some of their problems in the first place.

Their exit velocity stats are also worse because they serve up pitches down the middle of the plate. Batters crave these meatballs and can punish them for extra-base hits and home runs. When I took pitch location out of the model, the pitchers’ effects on batted ball velocities fell by 20 percent on average.

Ventura, the Royals’ young flamethrower, is a conspicuous member of this worst-in-the-league list. Like Worley and Karns, he tends to throw his four-seam fastball more often than the league average. That’s significant because fastballs tend to get hammered the hardest (even adjusting for count and location). Ventura, and other fastball-heavy starters, run the risk of allowing harder contact and more hits.

The idea that pitchers can, in fact, influence their BABIP is not new. Shortly after the initial publication of DIPS, Tom Tippett (currently employed by the Red Sox) wrote about how the best pitchers seemed to be able to control the probability that their struck pitches would fall for hits. Tippett had only anecdotal examples such as Pedro Martinez and Greg Maddux, so the sabermetric community coalesced on the idea that they could be exceptions to a very reasonable rule.

Now that we have Statcast’s data, we can see otherwise. No one will mistake Wei-Yin Chen for Pedro Martinez, but it appears that Chen can repeatably depress his opponent’s batted ball velocity, and a statistic such as FIP will fail to credit him for that skill. Statcast’s data is beginning to challenge not only our views of specific players, but also some of the fundamental precepts of sabermetrics.

*Special thanks to **Baseball Savant** for the batted ball data; Pitch Info and Harry Pavlidis for the use of pitch tag data; and Jonathan Judge, Greg Matthews, Harry Pavlidis and Dan Turkenkopf for helpful comments and feedback.*

**CORRECTION (May 22, 11:38 a.m.): **An earlier version of this article stated that Wei-Yin Chen had a sizable gap between his ERA and FIP all four years in which he pitched in the league. That wasn’t true in 2013, when his ERA was slightly higher than his FIP.

Specifically, I used a linear random effects model with the R package lme4. The model was specified as follows:

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

In total, batter explains 3 percent of the variance in batted ball velocity, while pitcher explains 0.6 percent and park 0.03 percent.

Specifically, I used a linear random effects model with the R package lme4. The model was specified as follows:

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

In total, batter explains 3 percent of the variance in batted ball velocity, while pitcher explains 0.6 percent and park 0.03 percent.

Deriving a p-value in a random effects model is tricky. However, the random effect for pitchers significantly improves the model by AIC, improves out-of-sample prediction accuracy, and is larger than expected under a null distribution derived from permutations.

Specifically, I used a linear random effects model with the R package lme4. The model was specified as follows:

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

In total, batter explains 3 percent of the variance in batted ball velocity, while pitcher explains 0.6 percent and park 0.03 percent.

Deriving a p-value in a random effects model is tricky. However, the random effect for pitchers significantly improves the model by AIC, improves out-of-sample prediction accuracy, and is larger than expected under a null distribution derived from permutations.

This number comes from a logistic regression of exit velocity for each batted ball.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

Deriving a p-value in a random effects model is tricky. However, the random effect for pitchers significantly improves the model by AIC, improves out-of-sample prediction accuracy, and is larger than expected under a null distribution derived from permutations.

This number comes from a logistic regression of exit velocity for each batted ball.

Run expectancy numbers are derived from a linear regression of linear weights value per pitch on batted ball velocity. Linear weights values come from Pitch Info.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

This number comes from a logistic regression of exit velocity for each batted ball.

Run expectancy numbers are derived from a linear regression of linear weights value per pitch on batted ball velocity. Linear weights values come from Pitch Info.

This amounts to 17,768 batted balls. You may notice that most relievers are close to zero on this list. That’s because the model does not have enough data per reliever to be certain that they are altering velocity heavily, so it regresses their readings to the mean.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

This number comes from a logistic regression of exit velocity for each batted ball.

Run expectancy numbers are derived from a linear regression of linear weights value per pitch on batted ball velocity. Linear weights values come from Pitch Info.

This amounts to 17,768 batted balls. You may notice that most relievers are close to zero on this list. That’s because the model does not have enough data per reliever to be certain that they are altering velocity heavily, so it regresses their readings to the mean.

This estimate is derived by incorporating count into the model as a fixed effect and then determining how much the pitcher’s estimated random effect decreased.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

This number comes from a logistic regression of exit velocity for each batted ball.

This amounts to 17,768 batted balls. You may notice that most relievers are close to zero on this list. That’s because the model does not have enough data per reliever to be certain that they are altering velocity heavily, so it regresses their readings to the mean.

This estimate is derived by incorporating count into the model as a fixed effect and then determining how much the pitcher’s estimated random effect decreased.

The league difference is a bit smaller, but still significant: .291 BABIP with three balls, .281 with two strikes.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

This number comes from a logistic regression of exit velocity for each batted ball.

This estimate is derived by incorporating count into the model as a fixed effect and then determining how much the pitcher’s estimated random effect decreased.

The league difference is a bit smaller, but still significant: .291 BABIP with three balls, .281 with two strikes.

Location was added into the model with a quadratic term for horizontal and vertical coordinates after adjusting for the count.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

This number comes from a logistic regression of exit velocity for each batted ball.

The league difference is a bit smaller, but still significant: .291 BABIP with three balls, .281 with two strikes.

Location was added into the model with a quadratic term for horizontal and vertical coordinates after adjusting for the count.

Using pitch tags generously supplied to me by Pitch Info.

Batted ball velocity ~ (1|hitter) + (1|pitcher) + (1|ballpark).

This number comes from a logistic regression of exit velocity for each batted ball.

Location was added into the model with a quadratic term for horizontal and vertical coordinates after adjusting for the count.

Using pitch tags generously supplied to me by Pitch Info.

Fortunately, new statistics such as Baseball Prospectus’ DRA (Deserved Run Average) do take into account all the events in which a pitcher plays a part. In fact, individual pitchers’ DRAs better correlate with the velocity suppression effects I calculated than their FIPs do, indicating that DRA is capturing some of this skill.