Previous Entry Share Next Entry
Where Jacoby Ellsbury's Power Came From (Part 1 of 2)
In June of 2009 I was still receiving severance pay from the Boston Red Sox after being laid off as a baseball operations consultant unexpectedly at the start of the year—a move dictated by upper management as a response to the recession. I was grateful for this unexpected generosity, and when we parted ways I told them I’d send them anything I found truly interesting and important, as a freebie.

The previous year I had noticed that Jacoby Ellsbury had astounding splits depending on the identity of the following hitter, and they were opposite from the conventional wisdom: he was amazingly better with a weak hitter behind them.

In this pair of long e-mails to the team, I explain that phenomenon, and advise them on a simple change Ellsbury could make in his approach—one that would unlock his power potential.

Note that I am not claiming responsibility for Ellsbury’s blossoming as a hitter. In fact, it's quite possible that this argument was ignored in the short run, and that his hitting coaches eventually noticed what I did, and gave him the same advice. But I believe it completely explains where Ellsbury’s 2011 power spike (usually regarded as a bizarre fluke) came from. More about that in Part 2.

June 14: A Freebie (Ellsbury Solved)

I’ve got 5 different lines of evidence, all independent, that point to the same conclusion:

When Jacoby Ellsbury thinks it’s his job to get on base (which is most of the time), he does no guessing, no sitting on pitches in hitter’s counts, and just reacts to what’s thrown. And he’s not very good. When he doesn’t think that’s his job, he does guess and does look for pitches to drive—and is a vastly better hitter as a result.

Here’s the direct evidence that he usually is just reacting. It’s from this year’s pitch/fx data and is about a week old. I’ve been looking all year at pitches thrown right down the middle (center third of the 20” of the plate plus the black, center third of the zone top to bottom as defined by pitch/fx). Here’s Jacoby’s rank among the 9 regulars:

1st in percentage seen (of all pitches)
7th in % taken (only Ortiz and Bay have swung at more)
2nd in lowest % of swings and misses (only Lowell is better)
2nd in lowest % of fouls per swing (barely trailing Lowell)
and hence a huge #1 in % of pitches down the middle hit fair

He is 5th in BA on those balls hit fair, but next to last (to Green) in Iso and SA.

This was puzzling. He swings at way more pitches down the middle than average and rarely misses them, but doesn’t hit them hard at all.

Well, why would a player take a pitch right down the middle? Because he was expecting something else. Why would he swing and miss at one? Because he was expecting something else and didn’t recognize the difference.

That Ellsbury rarely takes or misses a pitch down the middle suggests that, on the whole, he’s not expecting anything. He’s just up there reacting. Not that big league hitters guess on every pitch, but I think the good ones all do it, to a degree, especially when ahead in the count.

Is there more evidence to support this? Take his career splits by count (as of last week). He actually has a higher BA and SA after falling behind 0-1 (.305 / .401) than getting ahead 1-0 (.278 / .398). The reverse split is just random, but the lack of the expected positive split (last year the league was .240 / .363 after getting behind 0-1, .280 / .452 after getting ahead) indicates that he’s usually not using the count to help predict what he might see. The entire reason hitters hit better when ahead in the count is that the opposing pitching repertoire is narrowed, allowing more effective guessing / sitting on pitches.

Here’s his SA after various counts compared to AL ‘08 average, sorted by the difference:



AL ‘08

After 0-2 .326 .274 .052
After 0-1 .401 .363 .038
After 1-2 .309 .289 .020
After 2-1 .431 .426 .005
After 1-1 .393 .389 .004
After 2-2 .302 .319 -.017
After 1-0 .398 .452 -.054
After 2-0 .433 .487 -.054
Full .288 .376 -.088
First pitch .441 .544 -.103
After 3-0 .385 .508 -.123
After 3-1 .256 .488 -.232

Just a massive failure to take advantage of being ahead in the count, despite better-than-average numbers while being behind.

This insight allows us to reinterpret my last year’s analysis. And that gives us two more lines of evidence. As you recall, he had a split where he was immensely better with a weak hitter behind him than with an elite one. And he has been much better in his career with 2 outs than with 0.

I thought that both of these were about the pitcher’s different approach (pitch around him vs. challenge him). But now we see that it’s his own approach that changes. Hitting 8th with a Lugo type behind him, he’s not putting a premium on setting the table and starts to think about driving the ball. So he starts to guess, sit on pitches in hitter’s counts. Same thing when he’s up with 2 outs versus 0.

I would have sent this a week ago, but I wanted to test the theory with another line of evidence, another set of numbers compiled after I came up with the idea. I was going to break down his performance by count depending on following hitter, but that’s an enormous amount of work and I’m busy.

Then tonight he comes up with a four-run lead in the 9th, is obviously sitting on a pitch, and hits it out. And I realize, there’s another split to look at. He should be better with the Sox leading comfortably, when getting on base is not a premium, than with them trailing, when it’s everything. And here are the numbers from last year:

Sox trailing, 181 PA, .272 / .320 / .361 (and 4 GDP)
Up 3 or more runs, 121 PA, .297 / .355 / .495 (and 1 GDP).

Just tell him that he should always be up there with an idea of what he’s going to see, always guessing and looking for a pitch to drive when he gets ahead, regardless of score, outs, men on base, or the hitter on the on-deck circle. And he should be terrific.


June 18: RE: Ellsbury Solved

Thanks much for [an opportunity to buy tickets at face value]. And as reward (or punishment), the final word on Ellsbury, including, for the first time ever, why he has a 1000+ OPS hitting in front Lugo, Lowrie, Green, Crisp, and Cash (not just combined as a group, in front of each and every one of them!). I can’t believe it took me so long to figure out…
Numbers are through Tuesday night.

I. Admonition

Tell Ellsbury to stop trying to move the runners over!

He has 28 career PA with runners on 1st or 1st and 2nd, 0 outs, and a good hitter on deck. He’s hit .111 / .143 / .111 with 6 GDP. That’s 31 outs in 28 PA. And a 21.4% GDP rate. And he’s not improving at it—in fact, he’s 1 for his last 15, HBP, 5 GDP going back to last July 11.

He has 10 GDP in his other career 207 GDP opportunities—a 4.8% rate. The difference in rates has a 1 in 937 chance of happening randomly. He also has a .036 K+BB% in these PA vs. .182 otherwise, which is also statistically significant, so he’s exhibiting no patience at all.

II. Facts

Next, here’s what I think are his relevant career splits: [HRC = HR / Contact, XIP = XBH / Balls in Play, 1B% = 1B / (1B + Outs in Play)]












Move 'em Over 28 .111 .143 .111 .254 -0.24 .036 .000 .000 .000 .115
Other gd next, 0 outs or less than 2 run lead 634 .262 .309 .353 .662 4.00 .125 .057 .014 .053 .251
Gd next, 1 or 2 outs and 2+ run lead 170 .340 .398 .431 .829 7.41 .112 .082 .000 .088 .323
Bad next, 0 outs or trail 95 .384 .442 .500 .942 8.51 .095 .074 .026 .053 .380
Bad next, 1 or 2 outs, tied or leading 73 .464 .513 .791 1.310 20.34 .123 .055 .083 .109 .429

You can see that his strike zone command is better in rows 3 and 4 than row 2. In the last row, his K% and BB% revert to baseline (quite possibly a SSS fluke) but he has a massive spike in HR power; 36% of his career HR have been hit in these 7% of his career PA. He also has his best splits for XBH / BIP and for 1B / (1B + Outs in Play).

His overall good next / bad next split is now .272 / .322 / .360 in 832 PA (4.15 RC/27) vs. .419 / .474 / .632 in 168 PA (12.78 RC/27).

The good hitters are almost all Pedroia: 739 PA vs. 39 for Youkilis, Ortiz, Drew, and Lowell. In case you think the Bad next splits are driven by one guy, here’s all the guys with 10 or more PA:

Next Batter






Lugo 61 .382 .452 .564 1.015 11.16
Lowrie 39 .486 .538 .743 1.281 14.95
Green 20 .500 .545 .500 1.045 16.42
Crisp 11 .500 .545 .900 1.445 27.64
Cash 10 .500 .500 1.100 1.600 24.30

III. Interpretation

There are two related things going on here.

  • He has two different approaches depending on game situation, but the one he uses most of the time is very counter-productive.

  • As a result, his overall numbers and reputation do not reflect his actual ability as a hitter. And when he bats in front of a weak hitter, the opposing pitchers have no fear of him or the next guy and are not working to get him out at all. They’re just throwing it right over the plate, and he’s killing it.

Two approaches. The first we’ll call “just get on base” [GOB] and the second “try to do some damage” [DD]. He just tries to get on base whenever there’s 0 outs or whenever the team is trailing. He tries to do some damage when there’s 1 or 2 outs and the team is up by 2 or more runs.

When there’s 1 or 2 outs and the score is tied or we’re up by a run, his approach depends on the next hitter. He tries to just get on if the hitter is good, do some damage if he’s bad.

All of this makes absolutely perfect sense. The problem is, of course, that his GOB approach sucks. Just trying to get on knocks 90 points off his OBP.

I earlier identified one aspect of the two approaches: that he’s just reacting when he’s GOB, while he’s guessing and sitting on pitches when he’s DD. There may of course be other differences. The hypothesis is backed up by the pitch/fx data: this year he’s taken 22% of pitches down the middle in GOB situations but 38% in DD. The odds against that being random are just 5 to 1 but it is absolutely in the expected direction.

No respect. How a pitcher attacks a hitter is clearly more complex than simply challenging them vs. pitching around them, which really only applies to good hitters. Bad hitters are essentially always challenged, but when they’re protected by a good hitter, they are pitched much more carefully. When Ellsbury hits in front of Pedroia they take him very seriously and work to get him out. When Julio Lugo or Nick Green is up next, they essentially get way too overconfident and treat him like the punch-and-Judy hitter that he isn’t.

IV. Advice.

You don’t really need to know the components of the two approaches. Just verify that he does change his approach according to the game situation, and tell him to abandon the GOB approach and just try to do damage all the time. If he does that, he should go on a tremendous tear hitting 8th.

In terms of batting order, you paradoxically want to leave him there as long as they’re disrespecting him and he’s killing the ball. It’ll probably take a few weeks before teams notice that he’s not a weak and easy out and stop feeding him meatballs. But from that point onward he should still be a .380 OBP, .430 SA guy at a minimum if he sticks with the DD approach. (And his walk rate will go up—it’s true that he sees more 3-ball pitches in the strike zone than almost anyone on the team.) So you would move him back to the top of the order when he appears to “cool down” to normal.

  • 1
I wanted to leave my original e-mails as they were, without comment, but I want to clarify two points.

First, the set of approach splits I presented were indeed created by evil data manipulation, and were an attempt to capture the best possible approximation of Ellsbury's rationale for deciding whether to just get on base or try to make harder contact. He would of course have no conscious algorithm, just a sense of when each approach was warranted, and he would doubtless be inconsistent in their application. So the logic I present for choosing his approach is an after-the-fact reverse-engineering that attempts to codify his instincts or gut feelings. The brain does work that way.

Second, my comments about weak hitters being affected by protection (and in fact suffering from it) were not meant to describe a universal phenomenon, since hitters respond in a wide variety of ways to altered pitching approaches. I hope to have much more to say about the entire subject of protection sometime within the next few years, perhaps at a SABR conference.

Edited at 2013-12-06 10:01 am (UTC)

  • 1

Log in