February 24, 2012
The Stats Go Marching In
The Art of Handling the Pitching Staff
While evaluating Jose Molina’s defensive skills a couple weeks ago, we were able to assign a run value to four aspects of catcher defense: blocking errant pitches, preventing the opposing team from stealing bases, fielding short batted balls, and inducing the home plate umpire to call a few extra borderline pitches.
However, we acknowledged that something had been left out. We often hear about how some catchers can improve their pitching staffs. Think about the praise Ivan Rodriguez received for handling a young crop of Marlins arms back in 2003, and how he was subsequently considered the perfect batterymate for Stephen Strasburg as the highly-regarded rookie first took to a major-league mound.
Here’s today question: Do catchers actually have the ability to improve the performance of their pitchers? Or is this a baseball myth?
Keith Woolner explored this issue over a decade ago. His seminal “Field General or Backstop?” appeared in Baseball Prospectus 1999 and was later made available on our site. More recently, Sean Smith wrote “Do Catchers have an ERA?” in The Hardball Times Baseball Annual 2011. The analysis that follows is heavily influenced by Sean’s article.
The forces influencing the outcome
Thus, if we want to assess the effect a catcher has on his batterymates, we need to adjust for all the above factors.
Since batted-ball types are not available for all the seasons in the Retrosheet play-by-play files, I devised an alternate method that simply gives all the balls in play the same value.
Then I assigned every plate appearance an expected run value. This was estimated with a With-Or-Without-You approach in order to take care of the other factors I’ve already mentioned. The statistical method I used for this analysis is called cross-classified mixed model. Introducing an advanced statistical technique is way beyond the scope of this article, but you shouldn’t have to take my word for it that the model works just because it sounds sophisticated. Fortunately, it has performed well when I’ve used it to analyze baseball before.
The first time I used cross-classified mixed models, it was to evaluate how effective catchers throughout history were at blocking pitches and controlling the running game. The resulting article contains several files that you’re free to examine. However, I’ll highlight a few of the results to show that what the model says jibes with our expectations.
After adjusting pitch-blocking ratings to account for the pitchers on the mound, known finesse pitchers all emerged as the ones giving their catchers the easiest time. The lowest rate of pitches passing by the receiver occurred when Greg Maddux was on the mound. Brad Radke, Tom Glavine, and Jamie Moyer all place high on this list. At the other end of the spectrum are knuckleballers Hoyt Wilhelm, Charlie Hough, the Niekro brothers, and Tim Wakefield.
I also used the technique to find biases in PITCHf/x location measurements. After adjusting for pitcher, pitch type, batter handedness, and ball/strike count, I came up with an estimate of the PITCHf/x offset for every game. In order to evaluate whether the results made sense, I looked at how umpires called the games. Using the corrected locations I came up with, the umpires appeared to be much more consistent in their judging of the strike zone than they did when I used the data as provided by PITCHf/x.
Finally, when performing calculations for my first article here at Baseball Prospectus, I found out that with Carlos Pena and Mark Teixeira at first base, the number of successful plays on short batted balls is higher than average; conversely, when Prince Fielder is manning the cold corner, we should expect fewer short grounders to be converted into outs.
Summing up, whenever I have used this method, the results it produced were closely aligned with what one would expect. Thus, I feel comfortable going back to the well to investigate game-calling.
The method (continued)
The expected run value of a plate appearance featuring Ryan Howard is quite different depending on the handedness of the pitcher. The same is true from the pitcher’s perspective: a so-called LOOGY should have different expected run values depending on the handedness of the batter. Thus, when Howard faces Marc Rzepczynski, my model calculates the expected run value of the event, given that Howard is facing a lefty and Rzepczynski is throwing to a lefty.
After some tinkering with the model, I decided to incorporate batter handedness into the park factor as well. When making that decision, Yankee Stadium, with its short porch favoring left-handed batters and its “Death Valley” in left-center, was looming large in my mind. The results, coming in the next section, showed that to be a sensible choice.
I’ve mentioned the umpire as another factor that might influence the outcome of a plate appearance. However, I found the men in blue to have a rather small effect, so I decided to leave them out in order to reduce the computational time.
The expected run value estimated in light of the batter, the pitcher, their handedness, and the ballpark, was compared to the observed run value of the plate appearance. The difference was assigned to the catcher. (Note: obviously, the difference is attributable to numerous factors aside from the catcher, including chance.)
The top closers are ranked as the best at preventing runs. This doesn’t mean that they are the best pitchers, but that on a per-plate-appearance basis, they allow the lowest number of runs. Mariano Rivera tops the list. Actually, Rivera tops the list only against right-handed batters, though he does fare very well (within the top 15) against lefties, too. Following him are Rafael Soriano (facing righties) and Craig Kimbrel and Jonathan Papelbon (both versus lefties).
I would have immediately discarded my model if Albert Pujols had not come out on top as the best batter. Luckily, he does. Albert versus lefties is head-and-shoulders above the rest of the league, while he’s tied for second against righties with Adrian Gonzalez. Following them are all the other big boppers one would expect: Prince Fielder, Ryan Howard, Joey Votto, and David Ortiz, all when facing righties. Jose Bautista is a bit behind because his pre-superman seasons are included, and also because he obtains very good results both against righties and against southpaws. In fact, when combining results obtained when facing pitchers of either handedness, the Jays outfielder is ranked right behind Pujols. Another sanity check for the model is looking at the worst hitters. The pitchers dominate that list: smell test passed.
Finally, I checked out the ballparks. Finding Petco Park and Dodgers Stadium to be the most pitcher friendly and Coors Field and the Ballpark in Arlington to be the most hitter friendly is another point in favor of the model. When I switched to separate park factors by batter handedness, I found the expected Yankee Stadium behavior: the ballpark in the Bronx scores as a left-handed hitter’s heaven while playing close to neutral when righties are at the plate.
These results confirm that Jose Molina is a defensive wizard, as he was by far the best despite playing half as many games as the catchers listed below him.
The numbers above measure the cumulative effect a catcher has on plate appearances. That effect surely includes framing, so we shouldn’t add the framing runs we calculated last time to these totals. Fielding is more or less removed from this analysis, however, so the runs prevented by catchers on short batted balls can be added. Stealing occurs on the basepaths, and blocking pitches also has an effect on baserunners, so both of those components can be added as well.
There might be an argument that catchers who are good at blocking pitches (and to some extent at preventing steals) also add some value at the plate, as pitchers might be more confident in delivering dropping pitches when throwing to them. However, I believe those effects are captured separately by the metric proposed here, while what we showed last time measured only the effect on the events happening on the basepaths.
Let’s also look at the bottom 10:
Since the analysis is based on just four years, it’s possible that some catchers had a limited set of comps. That could be the case with Carlos Santana and Lou Marson, who wound up in the top-10 and bottom-10 lists, respectively, and would be particularly problematic for catchers whose pitching staffs had not changed much in the four-year span. In the Santana-Marson case, when one considers the trades involving Indians pitchers, Santana was actually compared to 48 other catchers, while Marson was compared to 87. Among the catchers featured on both lists, only Butera has a lower number of paired colleagues (60) than Marson.
Another problem always looming in this kind of analysis is the aging of all the actors in the scene. In the model, Jose Bautista circa 2008 contributes to the expected run value as much as Jose Bautista circa 2010. Bautista’s transformation notwithstanding, players should experience at least gradual improvements or declines over the years, and the four-year window was chosen in order to avoid equating, say, Ken Griffey Jr. in 1996 with Ken Griffey Jr. in 2006.
The ideal model would consider data from the entire Retrosheet era and apply proper age adjustments to pitchers and batters (and possibly parks as well). However, that model would be quite CPU-consuming.
The following plot depicts the comparison of ratings compiled by all the catchers in even days versus their ratings on odd days.
The weighted correlation between the two ratings is 0.51. Now let’s compare the above number with those of other, better-known statistics. For each of the measures reported in the table below, I repeated the even/odd split described above and used the same time span.
Removing batted-ball-type information
One is that batted-ball type are reported by stringers, who can be biased. As our own Colin Wyers has shown, ballparks make use of different locations for stringers to follow the game, and the different angles from which they keep track of the action lead to biased scoring of batte- ball types.
The second reason for removing batted-ball type is that that information is not available beyond a certain point. Thus, a model not requiring those data would allow us to rate catchers going back to the earliest Retrosheet-tracked games in 1948.
The model without batted-ball types simply assigns specific run values to strikeouts, walks, home runs (the Three True Outcomes), and balls in play. Yes, every ball put into play by the batter that does not leave the park is assigned the same run value.
As you can see from the chart below, the alternate model yields results very similar to the ones produced by the model incorporating batted-ball-type information. The correlation for the 2008-2011 period is 0.95.
Mark Buehrle left the White Sox for the Marlins this offseason. While in Chicago, he teamed with A.J. Pierzynski who, according to our model, improved his pitchers by roughly 80 runs in 18,000 PAs. That translates into nearly four runs for the 858 batters faced by Buehrle in 2011, or more than 0.1 runs per nine innings.
The above estimate should be mitigated a bit, since Pierzynski did not catch every Buehrle pitch in the past (75 percent from 2008 to 2011, 61 percent last year) and Marlins catcher John Buck is also a tick above average. Plus, it might turn out that Ozzie Guillen was the guy responsible for Pierzynski’s great rating, and since Guillen accompanied Buehrle to Miami, the left-hander could still benefit from that. Still, that’s one example of how to do a back-of-the-envelope calculation, and here are the complete rankings, so you can utilize that additional knowledge when choosing your 2012 fantasy rotation.
Coming up next