Last week, MLBAM delivered a lovely surprise to those of us in the baseball research community. You (yes, you!) can now get data on infield and outfield positioning, on a pitch-by-pitch basis, from the Baseball Savant website, specifically whether The Shift (infield or outfield) was being played. Up to this point, the only shift data that had been publicly available was provided by Sports Info Solutions (formerly Baseball Info Solutions), and it was only available at the plate appearance level (and only in the aggregate), and then only for balls that had been put into play (i.e., no home runs, no walks, no strikeouts, no hit batters).
I’ve written extensively on that SIS data, and about two years ago I found myself asking the question: “How many home runs were hit in front of The Shift?” It seems like a silly question because if the batter hit the ball over the fence, there aren’t any defenders back there (and certainly no infielders) so The Shift shouldn’t make a difference. But I thought it was an important question. While the infielders aren’t going to catch a fly ball hit over the fence, does the infield shifting lead to the pitcher throwing differently? Maybe The Shift kinda freaks him out?
Even the data that we did have told a different story than the usual narrative about The Shift. Sure, there are those plays where the batter hits the ball past the place where one of the fielders would otherwise have been, but there were supposedly more balls that The Shift snags than it lets through. But I didn’t find that. I found that in front of The Shift, batting average on balls in play (BABIP) actually went up slightly, once we accounted for the fact that the guys who tend to be shifted have lower BABIPs overall.
Or so I thought.
In March of this year, Mark Simon (formerly of ESPN, now with Sports Info Solutions) presented data at the SABR Analytics Conference that suggested the truth was a little more complicated. SIS–who provided the public data that I had been using–counted as a “shift” both the times when there was a third infielder on one side of the field (which is what most people generally picture when they hear “shift”) and times when there was an infielder (usually the shortstop) playing well out of position (mostly shaded very much up the middle) even though he hadn’t crossed the imaginary line in the middle of the field.
Simon suggested that when you broke shifts into these “full” and “partial” categories, the problem was that the “partial” shifts weren’t very effective, but the full shifts were. If teams were guilty of something in their shifting, it was soft-pedaling it. If you’re going to shift, lean all the way into it.
But there were other warning signs that The Shift might be a mirage. That same SIS data showed that for plate appearances in which the ball was hit into play, there was a spike in the number of balls (as in ball one, ball two, etc.) in the count. Because walks specifically weren’t counted in that data, it wasn’t possible to directly tie shifting to an increase in walks, but if there are more balls in the count, that’s a pretty good piece of evidence that something’s going on. And it’s probably going wrong.
Well … now we can answer a lot of those questions.
Warning! Gory Mathematical Details Ahead!
To start, the Statcast data seems to use the same formulation that SIS did in classifying two different types of shifts, and as we will see, they do perform differently. Data was available from 2015-2017. (The 2018 data is available on an ongoing basis, but I wanted to work with full-season data.) First, a little sanity check. Here’s a chart comparing the total number of partial (Statcast calls them “strategic”) shifts versus how many full shifts were reported in 2016 and 2017 by SIS and Statcast. Consistent with how SIS reports them, I selected only for plate appearances that ended with a ball in play (no walks, no strikeouts, no hit batters, no homers).
|Partial/Strategic Shifts||Full Shifts|
(Note: Data from SIS is from Simon’s presentation at SABR Analytics.)
We see that Statcast is a little more conservative in calling something a shift, particularly among partial shifts, so we are dealing with somewhat different definitions of what “counts.” Still, we’ll go with what we have.
First, let’s look at the shift as a defense against batted balls. In theory, that’s the entire point of what The Shift is supposed to do. In the past, using SIS data (but the version that didn’t differentiate between partial and full shifts), I found evidence that The Shift actually (slightly) increased BABIP over what we might expect, based on the hitters who get shifted. Part of the problem with doing any sort of analysis on The Shift is that it’s a very select group of hitters seeing the weird alignment. Obviously they are chosen because they have extreme pull tendencies, at least when they hit ground balls. But they also tend to be big, slow power hitters. It’s not a good idea to compare their performance to a league-average BABIP. They’re not likely to hit for a high BABIP because they tend to get the type of hits that aren’t counted by BABIP (homers) and fewer of the cheapies that inflate BABIP a bit (infield hits).
To correct for this, I looked at all hitters who had at least 100 plate appearance without The Shift. If a batter had a .300 BABIP without The Shift, then in the 200 plate appearance that he got in front of The Shift–if The Shift was exactly as good a defense against batted balls as a standard two-right/two-left alignment–we would expect 60 hits from him. If The Shift is a better defense, we’d see fewer than 60. If it’s worse, we’d see more. Now, we know that because of small-sample-size weirdness, you can’t always trust the results you’d get from this sort of analysis on an individual level. But if we sum across the league, we can get a good idea on how The Shift is doing in the aggregate.
|Shift Type (2015-2017)||Expected Hits (BIP)||Actual Hits (BIP)||Difference|
|Partial/Strategic (n = 24,446)||7,462||7,516||+54|
|Full (n = 26,616)||7,958||7,746||-212|
|Total (n = 51,062)||15,420||15,262||-158|
And if you want to see the real effect of this, here are singles per plate appearance in front of both types of shifts.
|Shift Type (2015-2017)||Expected Singles||Actual Singles||Difference|
|Partial/Strategic (n = 38,733)||6,123||6,129||+6|
|Full (n = 49,502)||7,103||6,610||-493|
|Total (n = 88,235)||13,326||12,739||-487|
In the Statcast data set, we see that BABIP (and singles) go down in front of The Shift overall, but there’s a difference between the Partial Shift and the Full Shift. The Full Shift seems to reduce BABIP by about eight points. This is consistent with what SIS found in their own data set, and their recommendation was that teams should dispense with half-measures and just do the Full Shift. It seems to be the better defense (and a better defense than the two-right/two-left variety), at least at defending batted balls.
It seems like a victory for The Shift, but that’s where things start to fall apart. BABIP has its uses, but it’s not the same thing as success. The point of the defensive team is to turn batters into outs. For that, we need (and now can calculate!) on-base percentage.
|Shift Type (2015-2017)||Expected On-Base Events||Actual On-base Events||Difference|
|Partial/Strategic (n = 38,733)||12,549||12,823||+274|
|Full (n = 49,502)||16,380||16,646||+266|
|Total (n = 88,235)||28,929||29,469||+540|
Uh oh. There are fewer outs recorded in front of The Shift (both kinds) than we might expect. It’s worth about seven points of on-base percentage for the Partial Shift and five points for the Full Shift. If you’re wondering what’s going on:
|Shift Type (2015-2017)||Expected Walks||Actual Walks||Difference|
|Partial/Strategic (n = 38,733)||2,951||3,216||+265|
|Full (n = 49,502)||4,310||4,884||+574|
|Total (n = 88,235)||7,261||8,100||+839|
We see that the Full Shift “took away” 493 singles, but it somehow gave back 574 walks. It seems that the primary effect of The Shift is to change the way that a batter reaches first base, and it seems that he is standing on first base more often. You can’t throw him out if he gets to walk there.
There’s more to it. In a previous piece, I suggested that one of the reasons The Shift might persist is evidence to suggest that hitters were pulling the ball less in front of The Shift, and since most power is pulled power, perhaps this would suppress home runs. (There was direct evidence that doubles and triples were slightly reduced.)
|Shift Type (2015-2017)||Expected HR||Actual HR||Difference|
|Partial/Strategic (n = 38,733)||1,148||1,104||– 44|
|Full (n = 49,502)||2,019||2,086||+67|
|Total (n = 88,235)||3,167||3,190||+23|
It seems that the Full Shift tends to increase homers (by a bit), while the Partial Shift suppresses them (by a bit). In reality, it’s mostly a wash, but that’s a story unto itself. I hypothesized that The Shift persisted because it was actually a trap that made hitters go away from their power. No such thing actually happens, according to this data. (The extra-base hit chart–not shown here–tells a similar story, with both types of shift showing a slight uptick in doubles and triples over expectations. The Shift is not a power-sapper.) If it were, it might claw back enough value to make the whole enterprise worth it, but it doesn’t do that.
Let’s roll this into a simple linear weights model and see what the final score is.
|Shift Type (2015-2017)||Expected LWTS runs||Actual LWTS runs||Difference|
|Partial/Strategic (n = 38,733)||9 (yes, 9)||127||+118|
|Full (n = 49,502)||657||856||+199|
|Total (n = 88,235)||666||983||+317|
In this set of plate appearances (players who had at least 100 non-shifted plate appearances, when they are facing The Shift), the average team has a net loss of about 3.5 runs a year through their shifty ways. The Shift ends up being a net negative for the defense (in the aggregate … hold on to that prepositional phrase).
When I started using this method, Ben Jedlovic (then of Sports Info Solutions), suggested a possible flaw in my methodology, which was that in selecting players who had received 100 plate appearances in a non-shifted situation, I was leaving out the guys who were prime candidates for The Shift. This is the part of the article where I drop in the name Chris Davis, who almost exclusively sees shifts when he comes to the plate. (see also: Ortiz, David).
So, what I did was to run the analyses “backwards.” I looked for hitters who had seen a Full Shift (three on the right side) at least 250 times during the season in question. These are going to be the guys who are the most pull-happy and the ones for whom The Shift was designed. I then looked at what happened to them when they did occasionally face the (presumably inferior?) two-right/two-left defense. If The Shift is superior, we should see their outcomes getting worse as teams go to the inferior defense. I used this group’s performance against The Shift as a baseline for expectations and then looked at their performance when no one shifted them.
|Outcome (n = 9,975 PA)||Expected Number||Actual Number||Difference|
|Outs in Play||4,526||4,502||-24|
|Linear Weights Runs||209||267||+58|
Here we see that for our particularly pull-happy players, when a team removes the third infielder from the pull side, singles go way up, which is what we would expect. So The Shift is effectively doing its job. We also see, though, that taking away that third infielder actually suppresses extra-base hits and, again, walks. But this time, when we look at linear weights, we see that for this group, going to an inferior defense makes them 58 runs better overall (in the slightly fewer than 10,000-plate appearance sample we have). For this group, The Shift is a perfectly reasonable strategy.
But what about the group that’s not quite as shift-able? Below is table of linear weights runs. Again, the “Expected Number” refers to how well we would expect the sample to perform, based on how they perform in front of a Full Shift. The “actual number” is their actual performance when they see what is supposed to be an inferior defense, the standard formation of two-right/two-left. If the actual number is higher than the expected number, that means the standard defense is allowing more runs (which is bad for the defense). If the actual number is lower, it means the standard defense is preventing more runs.
When we group batters by how often they see a Full Shift, we get:
|Group||Expected LWTS runs||Actual LWTS runs||Difference|
|250+ Full Shifts (n = 9,975 PA)||209||267||+58|
|200-249 Full Shifts (n = 9,194 PA)||176||116||-61|
|150-199 Full Shifts (n = 9,385 PA)||137||114||-23|
|100-149 Full Shifts (n = 13,623 PA)||518||269||-249|
|50-99 Full Shifts (n = 44,590 PA)||612||502||-110|
Even for the group that encounters a lot of full shifting (the 200-249 shifts group), they seem to perform worse when facing the standard defense than they do against a shifted defense. To give some sort of perspective on this, in the three years in the data set, there were just over 60 player seasons in which a batter saw more than 250 Full Shifts (and it was mostly the same names over and over). This means the standard defense is the better choice for all but the most severe pull hitters.
(A few addenda: To check myself, I ran the same types of analysis from the pitcher’s perspective and found the same story over and over. Pitchers in front of The Shift give up fewer singles than we would have expected, but they walk more hitters. I also ran these analyses using a log-odds ratio setup, which simultaneously models the batter’s and pitcher’s odds of a certain outcome happening. The same finding kept coming up over and over. The Shift reduces singles, doesn’t do all that much to power outcomes, and adds an avalanche of walks.)
The Power of Listening
Ever since The Shift became “a thing” there’s been a fairly universal reaction among pitchers. When asked about it, pitchers often talk about how The Shift made them uncomfortable. It was rarely put into words further than that, perhaps because pitchers didn’t want to give away any strategic edge through the media, or perhaps because they didn’t quite know how to put it into words. But the data shows that maybe we should have been listening to the pitchers.
It seems that hitters have already figured out the best way to beat The Shift. Let the pitcher walk you. The walk penalty isn’t something that’s been publicly discussed when it comes to The Shift. However, it turns out that it’s effectively swallowing all of the benefits that The Shift was supposedly granting. While The Full Shift is good at preventing singles, batters have just found another way to get to first base. Maybe we should have listened harder to the pitchers.
The data here doesn’t say that The Shift should be abandoned. It’s hard to make recommendations on an individual level here. Remember, all of these analyses are in the aggregate. But we do come away with one guiding principle. The threshold for when a shift should be used should be very high. It should be practiced only on those hitters with the most extreme pull tendencies. There should probably be a few thousand shifts per year, league wide, and they should all be against the same 20 or so guys. And those guys should be bunting against The Shift anyway.
For what it’s worth, I don’t think this means the end of The Shift. I’m no “play the game the right way” iconoclast when it comes to The Shift as a strategy, and even if I were, teams would still do it if it works. I’m just not so sure it actually works.
The fact that one of the major problems with The Shift is walking hitters suggests perhaps pitchers might be trained in some sort of counter-measure (maybe they’re nibbling too much?). Maybe pitchers whose game is to fill the strike zone are less prone to the walking penalty. And maybe they’re safer to shift in front of. There might be other markers of hitters that change the calculus enough to make them worth shifting. There’s a lot of ground still to cover, but going forward The Shift needs to be understood in the context of these findings.
The infield shift has a nasty side effect that makes the treatment worse than the disease in most cases, and until that side effect is better understood, The Shift should be used very cautiously.