What did Jack Zduriencik know?

Last week, I posted an article looking at the differences between AJ Burnett’s 10 best and 10 worst starts of 2009. I considered a lot of variables, including aspects of his “stuff”, location and approach, and found that there was virtually no difference between how he pitched in each of the two groups of starts. In my conclusion I offered that the reason his results in those two groups of starts were so different, despite a similar process, was primarily luck. I theorized that a lot of the differences we see between pitchers’ best and worst starts are factors out of the pitchers’ control, although I did note that Burnett was only one example.

The article generated a lot of feedback. In addition to the two pages worth of comments here at THT, there was some discussion at other blogs and I received several emails as well. Mainly, the comments offered suggestions of ways to dig deeper into the data to see if it could reveal something about Burnett that I missed in the original article. I did take a look at those suggestions (at least the ones easily doable), and didn’t find much. If anyone else wants to take a look at the data for Burnett, I put it in a Google Docs spreadsheet:

Good Burnett, Bad Burnett

If anyone needs help deciphering any of the columns, feel free to email me or comment on this article.

Anyway, with one case study giving me some pretty interesting results, I wanted to check out some more pitchers to see if the same phenomenon held true. This time, I chose Jarrod Washburn. Washburn is basically the opposite of Burnett, making him a nice complement in the study. He’s a lefty, throws in the mid to high 80s, has like 12 different pitches, and is a pretty extreme flyball pitcher.

The other reason I chose Washburn was that he had a really good first half of the season with Seattle, and then was traded to Detroit where he was awful. Many Seattle fans had felt that Washburn was simply getting lucky during his brilliant first half of 2009 and was in line for a lot of regression.

Mariner’s GM Jack Zduriencik apparently thought the same thing and pulled the trigger on the trade, netting a couple of decent prospects. Now let me just make things clear. Washburn had a 2.66 ERA with the Mariners last year and was consistently great throughout the season. So trading him when the Mariners were still somewhat in contention was a risky move and hinged on the idea that he was simply getting lucky and would return to his usual “meh” performance pitching in front of a less stellar defense. Based on the results of his time in Detroit, Zduriencik appeared to have been right…. or was he?

Given the storyline behind the trade, this seemed like a nice opportunity to combine my research on good and bad starts by pitchers with a real-life situation in which this might be useful. I wanted to see if Washburn had actually pitched differently during his time in Seattle in comparison to his time in Detroit, or if the drastic difference in results was simply due to factors outside of his control. So I looked at Washburn’s gamelogs this year, and took his five best starts in Seattle, and his five worst starts in Detroit as sorted by game score*.

*Using five starts instead of 10 was inspired by a comment from obssesivecomplusivegiant in the Burnett article. He thought that using 10 starts for each group would mix in too many average-ish starts with the extremes. That made a great deal of sense, so I decided to limit it to five. Thanks OCG

Let’s check out just how much the two groupings of starts differed in terms of results:

As you can, there was a HUGE difference. He allowed twice as many hits in his bad starts, despite pitching 15 fewer innings. I know that hits allowed aren’t really under the pitcher’s control, but that’s a still a pretty amazing stat. He struck out twice as many batters in his good starts while walking fewer. He also allowed zero home runs in his good starts compared to seven in his bad starts. Overall, his RA was 10 points higher in his bad starts, and his FIP was nearly five runs higher.

Now, let’s take a look under the hood of Washburn’s results to try to identify the differences in the way he actually pitched. I used Pitch f/x data, which, by means of laser cameras, captures many key aspects of each pitch thrown in baseball. Like with Burnett, I chose to look at stuff, location, and approach.

Stuff

First, I reclassified each of Washburn’s pitches, going game-by-game so I could get the most accurate classifications*

*Side anecdote: I was stumped on one of his games, July 6, 2009 against the Orioles, regarding whether or not he was throwing a changeup. Looking at the pitch data, it appeared that he only threw one changeup, which seemed rather weird, and I thought that I might be missing something. So I looked at ESPN’s online recap for that game, and read… “Washburn threw mostly fastballs against the Orioles, mixing in a handful of breaking pitches and one changeup.” Excellent.

Even then it was hard to get a feel for what Washburn was throwing. Some changeups looked awfully like two-seamers, and some of his cutters looked a lot like four-seamers; however, I think I did a pretty good job. Anyway, here is a pretty graph of Washburn’s spin deflection by pitch type for the two groups of starts. Spin deflection is the technical term for “movement” and it basically represents, in inches, how far that ball moved from a theoretical pitch thrown without spin. It’s from the catcher’s point of view, so negative numbers mean the pitch moved inside to a righty and vice versa:

As you can see, Washburn has a virtual rainbow of pitches. He throws at least six pitches (I say at least, because there were a few changeups that were a little too fast to be changeups and a little too slow to be two-seamers, so they might in fact be a different pitch), and basically covers all white space on the movement spectrum. This chart doesn’t tell you much, but you should be able to notice that Washburn’s pitchers were slightly more condensed (meaning there was less overlap in spin deflection by pitch type) in his good starts compared to his bad starts. Honestly, I have no idea whether or not that is a good thing. Now let’s check out some numbers…

Washburn throws more pitches than Burnett, so it’s a tougher chore making a good comparison. Still, the major differences here are that he threw fewer four-seam fastballs in his good starts, and compensated for that by throwing more twoseamers, changeups and sliders. The percentage of curveballs and cutters thrown remained roughly the same.

All of his pitches had more movement into righties and more sink in his bad starts, interestingly enough; however, the relative movement of all of his pitches remained similar. The pitch speeds all looked very similar, with the biggest difference coming from the changeup, which was about 2 MPH slower in his bad starts. That gives it more separation from the fastball.

So, unlike with Burnett, there are some significant differences in Washburn’s pitch selection/attributes in his good starts vs. his bad. The problem is that we really don’t know about what makes pitches good or bad based on their physical attributes – especially in very specific individual cases like this. The fact that his pitches had more sink in his bad starts probably bodes well for the two-seamers, curveballs, sliders and changeups, but maybe not as well for the four-seamers and cutters. His changeup having more speed separation from the fastball(s) could also go either way. So we will leave the stuff section with a question mark.

Approach and location

For Washburn, I decided to combine the approach and location parts into one section. The reason for that is that, unlike Burnett, Washburn is a pitcher who will presumably have to rely more on sequencing his pitches and pitching to the count than just trying to blow the batter away due to lesser stuff.

However, with Washburn, looking at his approach and location can get tricky. He’s got so many pitches that it becomes very hard to pick up distinctive patterns. One thing we can look at for starters is how often he gets himself into hitters vs. pitchers vs. neutral counts. So, I broke down the count into those three states based on the run expectancy for each count, and measured the percentage of pitches that Washburn threw in each count:

	Good starts	Bad starts
Pitcher	30.1%	30.0%
Neutral	57.5%	55.7%
Hitter	11.7%	14.1%

As you can see, he was in the same amount of pitcher’s counts in his good starts in compared to his bad ones; however, he was also in more hitter’s counts and fewer neutral counts in his bad starts. In fact, when I weigh the frequency of batters faced in each count by the run expectancy, and average the results, I get a run expectancy of -0.11 per 100 pitches for his good starts and 0.33 per 100 pitches for his bad starts.

In other words, you would expect him to allow about .4 runs more per game in his bad starts simply due to the count situation he was in. That’s significant, however, it doesn’t nearly explain the massive difference in results over his good and bad starts. The answer to that, if there is one, probably depends on his pitch selection and location, as well as other possible variables like sequencing. First, let’s take a look at his pitch selection by count type:

The biggest difference comes in hitter’s counts, where he threw more four-seam fastballs in his good starts compared to his bad starts – to the tune of about 10 percent. He supplemented that by throwing more cutters. In pitcher counts, his pitch distribution was similar in both groups of starts, however, he threw more changeups and fewer cutters in his bad starts. In neutral counts the pitch distributions were nearly identical.

Again, unlike with Burnett, there are some differences to be found here. However, it’s hard to tell whether they are significant and if they favor his good starts or his bad ones. Baring that in mind, we’ll move on to his pitch location. This time, in addition to splitting it up by pitch type and batter-handedness, I also split it up by count type (pitcher, neutral or hitter). That gave me a total of 36 bins for each group of starts. Let’s take a look at a couple of plots here, starting with first with four-seam fastballs:

The top row is his good starts while the bottom row is his bad ones. These are from the catcher’s point of view.

For the most part, his four-seam fastball location looks very similar in the two groups of starts, even when separated by count type. The biggest difference appears to be that, in hitter’s counts, Washburn threw more pitches over the heart of the plate, especially to righties in his good starts. This is obviously surprising given that he allowed so many more hits and home runs during his bad starts. He also threw more four-seam fastballs in neutral counts inside on righties in his good starts. Now, let’s take a look at his two-seam fastballs:

Here the advantage looks to go to the bad starts. He rarely threw two-seamers in hitter’s counts in either group of starts, and the locations were very similar in pitcher’s counts. However, in neutral counts, the aggregate pitch locations appear to be lower in the strikezone in the bad starts. When you combine that with the fact that his two-seamer had more sink, his GB rate was nearly 15 percent higher overall in his bad starts. Again, very surprising given the amount of home runs he gave up.

Let’s check out his cutters now. I won’t bother posting a chart in the article to save room, but you can see it here. To be honest, I can’t make much conclusions from the chart – there are simply too few cutters thrown, especially in his bad starts. The main thing is that he attacked the strike zone pretty well in pitcher’s counts in his good starts.

Now, let’s move onto his changeups. In both groups of starts, he really only threw changeups to righties and in pitcher or neutral counts. In his bad starts, he pounded the low outer half, throwing fewer pitches in the strikezone than in his good starts, but also more quality pitches and fewer pitches over the heart of the plate.

For his curveballs, his control looked pretty miserable. Like with the changeups, he mainly threw it to righties and in pitcher or neutral counts. He hung a bunch of big slow curveballs out and over the plate, moreso in neutral counts in his good starts. However, he threw a lot of pitches in the strikezone with them, especially in neutral counts in his bad starts.

Finally, we’ll look at his sliders. In both groups of starts, he mainly threw them in pitcher or neutral counts and to left-handed batters. In pitcher counts, the locations were almost identical. In his neutral counts there is a fairly big difference. In his good starts, he threw many more pitches in the strikezone, but also a lot more hangers. In his bad starts, he was consistently right around the low outside corner, but also threw more pitches outside of the strikezone. In fact, in his bad starts, his slider location in neutral counts mirrored that in hitter’s counts. It’s also worth noting that the actual called left-handed strikezone is pushed a couple inches more to the third-base side.

What went wrong?

So, by comparing how Washburn actually pitched in his five best starts with the Mariners vs. his five worst starts with the Tigers, the conclusions are… inconclusive. Looking at stuff, and approach and location, we can definitely identify some significant differences between good Washburn and bad Washburn. Some appear to favor his good starts, while others appear to favor his bad starts. Still, it’s hard to tell for sure.

So it looks we need another tactic to figure out why his results were so much better in his good starts. One way to do that is to look at where Washburn had the most problems in his bad starts. The easiest place to start is with home runs. Washburn allowed seven home runs in his bad starts compared to zero in his good starts. So on which types of pitches did those home runs occur?

If you want to see the location of the home runs, look here. We see that all of his home runs came on pitches up and over the plate, and against his fastballs in either pitcher or neutral counts. As I showed in the “Approach and location” section, his four-seam fastball location was virtually identical in such counts in both groups of starts, his two-seam fastball location was similar in pitcher’s counts and most likely better (farther down in the zone and with more sink) in his bad starts. So barring any other factor that I’m missing, the difference in home runs allowed appears to be simply due to luck and batter’s squaring up a couple more pitches.

So what else hurt Washburn in his bad starts? Well, he walked four more batters despite pitching 15 fewer innings, and his strike percentage was 3 percent higher in his good starts. According to my calculations, he threw a little more than 5 percent more pitches inside the strikezone (I adjusted for batter stance and height) in his good starts, so there appears to be no bad luck in his bad starts in terms of walks. Unfortunately, I’m not really sure how to weed out the luck involved in strikeouts, at least without another thorough look through the location data. At first glance, it appears that Washburn threw more pitcher’s pitches in his bad starts, but I’m not sure… so I’ll end here.

Conclusion

Unlike with Burnett, Washburn showed some significant differences between his best five starts with Seattle and his worst five with Detroit. Almost all aspects of his pitching – pitch selection, stuff, approach and location – were different in some way, big or small. The problem is there is really no way to tell whether or not the differences in the way he pitched caused the differences in results. The fact that he threw fewer pitches in the strikezone implies causation toward the extra walks. The massive difference in home runs, hits and strikeouts don’t appear to jibe with the in-depth look at pitch location by count type and stuff.

The takeaway from this is that we need to know a lot more about what makes pitches good. I think that with both Burnett and Washburn, I’ve demonstrated that a pitcher’s results will often be reliant upon things outside of his control – much more so that DIPS has suggested. When you think about it, that makes sense. The outcome of a pitch relies on at least five factors – the pitcher, the hitter, the umpire, the fielders and the ballpark. It’s not unreasonable to suggest that a pitcher might have less than 50 percent control over what happens, even if he makes a perfect pitch.

The next step is to look at just how much variance is associated with certain pitches, and try to isolate how much control pitchers actually have over the outcomes of those pitches. Keep an eye out for the next few weeks as I hope to be able to offer some solid insight into that. The eventual endpoint, I hope, will be a halfway decent way of evaluating pitchers based on what they do, and not based on the actual outcomes. However, we’re not there yet.

References & Resources
Run expectancy by count taken from this article by John Walsh.

Yes I realize this article really had nothing to do with “what Jack Zduriencik knew”, but I thought that was a catchy title and it made for a good intro. Sue me.