April 4, 2017

Baseball Therapy

Stop Shifting (So Much)

by Russell A. Carleton

4,576
6,881
13,298
17,737
28,072

I promise we haven’t taken to just creating articles entirely out of numbers here at BP. That’s a list of the number of balls hit into play against a shifted defense in each of the past five seasons (2012-2016, starting at the top). Those numbers aren’t even the whole story. If a batter faces a shift, but strikes out, walks, gets hit by the pitch, or hits a home run, there is no tally made in the record. If the trend line holds, we could easily see 35,000 of these in 2017. Then again, David Ortiz has retired.

MLB has become over-run with shifting defenses. As some point of comparison, there were 123,246 balls hit into play (in the BABIP sense) in 2016, meaning that 22.8 percent of all balls hit into play were hit against The Shift. It’s reasonable to say that 20-25 percent of all plate appearances now feature The Shift.

And hey ... it works! Right? Here is BABIP on ground balls hit into The Shift vs. BABIP on ground balls not hit into The Shift.

Year	GB BABIP vs. The Shift	GB BABIP vs. no shift
2012	.208	.234
2013	.196	.234
2014	.210	.241
2015	.206	.238
2016	.224	.240

Great success! The Shift really does make ground balls easier to grab!

Well, let’s dial back our enthusiasm a tiny bit. Here is overall BABIP against The Shift vs. BABIP when The Shift isn’t on, this time for all batted ball types.

Year	BABIP vs. The Shift	BABIP vs. no shift
2012	.293	.296
2013	.281	.298
2014	.290	.299
2015	.291	.299
2016	.298	.298

In 2016, the two were equal, and in years past, the overall edge that The Shift bequeathed defenses wasn’t earth-shattering. I also think that whatever advantage that the strategy seemed to grant was mostly a mirage and that there are hidden side effects to the practice that no one has noticed.

Warning! Gory Mathematical Details Ahead!

Let’s start with the obvious. The Shift does exactly what it says on the label. If a hitter is prone to pulling his ground balls, putting more fielders on the pull side is going to increase the chances of generating an out—when he hits a ground ball. The first of the BABIP charts above tells us that, and if that’s as deep as you look, The Shift seems like a rollicking success. You would also get a "D" in my research methods class.

First off, we need to account for who gets shifted against. Last year, I looked at split stats of players who got at least 100 plate appearances without The Shift on, and their BABIP and slugging percentage not against The Shift. I used that as a baseline for comparison. For example, if we had a hitter with a .300 BABIP when not facing The Shift, and he had 10 PA in which he hit a ball in play against The Shift, we would expect three hits from him if The Shift made no difference. If the Shift was a better defense against him, we might expect one or two hits.

In small samples like that, there’s going to be random variation, but if we sum across the league, we can get a better idea of The Shift as a strategy. Last year, I produced this table, which I have updated with the 2016 numbers.

Year	Hits (expected)	Hits (actual)	Total bases (expected)	Total bases (actual)
2012	1,018	1,055	1,334	1,362
2013	1,653	1,556	2,159	1,984
2014	2,632	2,767	3,423	3,490
2015	3,874	3,844	5,024	4,895
2016	5,246	5,472	6,788	7,009
Total	14,423	14,694	18,728	18,740

We can see that over the past five years, The Shift has been a slight net negative for the defense, once we control for the talent level of the players who are shifted against. We see a few more hits than we might expect from shifted hitters, compared to how they performed in non-shift situations.

After I published the first article, there was a reasonable critique of my method by Ben Jedlovic of Sports Info Solutions (formerly Baseball Info Solutions). He pointed out that someone like David Ortiz would not have been in my original sample, because in almost all of his plate appearances, he saw The Shift, and so he wouldn’t have notched 100 PA without The Shift. He’s also exactly the sort of player that The Shift would, in theory, be the best play against. In 2016, there were 18 hitters who had fewer than 100 PA against The Shift, but more than 200 with The Shift in place, so those guys are out there (and Ortiz is on that list near the top).

So, I ran the analyses from the previous paragraph backwards. What happens when teams take The Shift off? If The Shift is a better defense overall, we should see that once we get rid of it, there should be more base hits for those players. I found players who had been shifted at least 50 times in the season and used that as the basis for my expectations when they weren’t shifted. The table is now updated with the 2016 numbers.

Year	No-Shift Hits (expected)	No-Shift Hits (actual)	No-Shift Total Bases (expected)	No-Shift Total Bases (actual)
2012	1,714	1,696	2,206	2,239
2013	3,091	3,290	3,966	4,274
2014	4,310	4,157	5,464	5,422
2015	5,264	5,321	6,772	6,888
2016	7,733	7,391	9,992	9,594
Total	22,112	21,855	28,400	28,417

We see that once teams take The Shift off, they actually gave up fewer hits than we would have expected. This table again suggests that The Shift, against this group of hitters, is actually a slight net negative for the defense.

The method above is still going to have a problem with a guy like Ortiz. The sample in the table immediately above is weighted by how many non-shift PA a hitter gets, so Ortiz isn’t in there very much either. The extreme guys are harder to pin down and I’m willing to entertain the idea of that we have to maintain a healthy skepticism about how well these findings apply to those extreme guys. The thing is that The Shift has expanded well beyond those extreme guys. It’s all over the league now (again, 20-25 percent of PA). I think the fairest statement that we can make on the matter is that The Shift—in the aggregate—isn’t doing all that much, but what it is doing is actually counter-productive.

In my previous work, I found that against The Shift, hitters tended to pull the ball less often than they usually do. That makes sense. I also found that batters hit more grounders than normal against The Shift. Again, here are the updated numbers.

Year	Grounders (expected)	Grounders (actual)	Pulled balls (expected)	Pulled balls (actual)
2012	1,607	1,498	1,495	1,385
2013	2,253	2,371	2,309	2,123
2014	3,802	4,200	4,023	3,914
2015	5,558	6,107	5,382	5,120
2016	7,714	8,191	7,403	6,882
Total	20,934	22,367	20,612	19,424

At first, I interpreted the findings about ground balls as something of a double-edged sword. Pitchers knew that they had a defense behind them that was created specifically to help them field ground balls, so why not pitch to try to induce a grounder? For the purposes of BABIP, that can cut the other way though. The reason is contained in this chart. It’s BABIP, based on batted ball type, 2016 numbers.

Batted Ball Type	BABIP
Fly Ball	.127
Ground Ball	.239
Line Drive	.682

If you put more ground balls into circulation (.239 BABIP in general, and .224 against The Shift), that’s great, but you’re taking more fly balls out of circulation (which also happens), and the fly balls have a .127 BABIP. When you replace something with a .127 BABIP with something that has a .224 BABIP, that’s going to provide upward pressure on BABIP.

Inducing more ground balls is not a bad thing. There’s no SLG or OPS on the ground and fewer fly balls probably means fewer home runs. Then I realized that something else was happening. Similar to what I did above, I took split data for pitchers, both when they are pitching in front of The Shift and when they are not. Because shifting tends to be a function of the batter’s tendencies, and not the pitcher’s (although we’ll see in a moment that’s not entirely true), just about all pitchers have experience in both columns.

One reason that batters might be hitting more ground balls than we would otherwise expect is that they are facing pitchers who tend to give up ground balls at a greater than average rate. We know that ground-ball rate is a quickly reliable stat for pitchers and that the pitcher and batter share the credit for the type of batted ball hit. That turns out to be correct. Here’s a weighted average of ground-ball rate for pitchers, weighted by the number of shifted PA they participated in. (Someone’s going to ask, so I put the batters in there as well.) The league average for the year is next to it as a point of comparison

Year	Pitcher weighted GB rate	Batter weighted GB rate	League GB rate	League GB rate (shift)
2012	47.1%	42.6%	45.1%	45.7%
2013	46.5%	41.3%	44.5%	43.6%
2014	46.6%	40.7%	44.8%	44.6%
2015	47.3%	42.8%	45.3%	45.9%
2016	47.7%	43.4%	44.7%	44.8%

We can see that the types of pitchers teams play The Shift behind are somewhat more likely to give up a ground ball. It’s not that teams don’t shift in front of fly-ball pitchers. They just shift more often in front of grounder guys. So, when we see that grounders go up for hitters, it’s that they are more likely to be facing a ground-ball pitcher. This kinda makes sense on the surface. The Shift is a ground-ball trap. It’s best used when a team thinks that a ground ball is coming.

Let’s reason this one out. If the kinds of pitchers that are shifted behind are ground-ball pitchers to begin with, that could explain why we see higher ground-ball rates for batters against The Shift than we expect. But are our ground-ball generators getting all the ground balls that we would normally expect out of them?

Year	Grounders (expected)	Grounders (actual)	Fly balls (expected)	Fly balls (actual)
2012	1,947	1,904	1,298	1,304
2013	2,913	2,747	1,988	2,071
2014	5,474	5,293	3,783	3,921
2015	6,964	6,764	4,568	4,734
2016	10,332	9,689	6,738	7,143
Total	27,630	26,397	18,375	19,173

Pitchers, pitching in front of The Shift, tend to get fewer ground balls than we would expect of them (based on their non-shift performance) and allow more fly balls. They still generate a lot of ground balls, but they were already high-grounder guys to begin with. The Shift has taken them to around merely average. The answer for why that happens might be the reverse of what I argued a few paragraphs above. Hitters who are shifted against tend to be lower ground-ball rate types.

So, hitters are hitting more grounders than they normally do (but hey, they’re facing a bunch of ground ball-pitchers!) and pitchers are giving up more grounders than they normally do (but hey, they’re facing a bunch of fly-ball hitters!). We’re probably seeing the same GB/FB ratio that we would see if the two sides were squaring off without The Shift. So, shift or no shift, they’d probably be hitting the same types of balls, and The Shift makes those ground balls more likely to turn into outs.

Well, there’s another problem ...

Year	Batter line drives (expected)	Batter line drives (actual)	Pitcher line drives (expected)	Pitcher line drives (actual)
2012	773	786	890	924
2013	1,213	1,258	1,359	1,443
2014	1,922	2,031	2,483	2,577
2015	2,823	2,818	3,163	3,197
2016	3,862	3,906	4,583	4,822
Total	10,593	10,799	12,478	12,963

Uh oh. Both batters and pitchers are hitting/giving up more line drives against The Shift than we would have otherwise expected of them. It’s not a huge increase, but it doesn’t have to be to make a difference. Line drives have an expected BABIP (using those 2016 numbers) that’s 443 points higher than a ground ball. If we assume that The Shift makes grounders worth 30 fewer points of BABIP (seems reasonable from what we saw on the first chart in the article), then replacing one ground ball with a line drive undoes the benefit that 15 ground balls got from The Shift.

That’s why The Shift doesn’t seem to have a very big effect on BABIP. While the ground balls that are hit are easier to field, something about The Shift also makes it easier for batters to square one up.

There’s another red flag, one that’s visible because of some extra data lurking in the splits. I have previously lamented (and will again) the fact that the data with which we are working only relates to balls hit into play. Strikeouts against The Shift aren’t noted, nor are walks. The evidence that we’ve seen above suggests that once we correct for the selective sampling bias involved in who gets a shift, The Shift actually has a fairly minimal impact on BABIP. The fact that it gets back to the origin point through a couple of effects that essentially cancel each other out is important, though.

If we’re consistently seeing a small tick up in line drives, is that evidence that pitchers are changing the way that they pitch? If so, is it possible that this will also affect things like walks or strikeouts? In the splits, there is data concerning the numbers of balls and strikes (and by extension, the number of pitches) that a batter saw in the plate appearances covered by the split. So, I know how many balls and strikes Ortiz saw ... in plate appearances in which he eventually didn’t walk or strikeout. However, we can still squeeze some information out of this.

Using the same method above, where I looked at plate appearances in which a batter was not shifted against, I got a baseline rate of how many balls he saw per plate appearance. Again, if The Shift doesn’t change the way that pitchers pitch, we should see roughly the same number of balls per PA when he’s being shifted against. We can prorate that, and sum across the league (again, to correct for low sample sizes for individual batters). We can do the same thing for pitchers. We can put them in this lovely table.

Year	Batter called balls (expected)	Batter called balls (actual)	Pitcher called balls (expected)	Pitcher called balls (actual)
2012	4,104	4,331	4,543	5,060
2013	6,269	6,677	6,816	7,680
2014	10,316	11,086	12,657	14,076
2015	14,627	15,445	16,024	17,426
2016	20,035	20,811	23,822	25,387
Total	55,351	58,350	63,862	69,629

And now, strikes.

Year	Batter strikes (expected)	Batter strikes (actual)	Pitcher strikes (expected)	Pitcher strikes (actual)
2012	7,966	8,151	9,408	9,473
2013	12,298	12,645	14,200	14,381
2014	20,745	21,445	26,726	27,118
2015	29,434	29,954	33,838	33,582
2016	40,600	40,712	50,274	49,286
Total	111,043	112,907	134,446	133,840

Batters are seeing more called balls than they usually do, and pitchers are throwing a lot more balls than they usually do. Strikes don’t have the same bump. In fact, pitchers are throwing fewer strikes than we would otherwise expect from them. I can’t directly say that The Shift increases walk totals, but I can come to its doorstep. If nothing else, more balls in the count in plate appearances where the ball goes into play generally means higher quality batted balls. BABIP is much higher on a ball struck in a hitter’s count than it is in a pitcher’s count.

On top of that, it means that pitchers are throwing more pitches per plate appearances in front of The Shift. That might not have an effect on this particular plate appearance, but we know that pitchers tend to fade as their pitch count climbs, and so the effect of a couple of extra pitches against this shifted hitter might not be felt until he is facing someone else in the sixth inning. It’s something that we can’t directly price into the cost of The Shift, but it’s there.

If I may summarize:

1. Plate appearances involving The Shift suffer from a sample selection problem. They generally feature a high ground-ball/low fly-ball pitcher facing off against a low ground-ball/high fly-ball hitter. We know that the natural outcome for this sort of matchup is that the probability of GBs and FBs tend to meet in the middle, and that’s what happens. The Shift doesn’t seem to change the GB/FB ratio that we see, but the grounders that are hit are easier to field and turn into outs. This is the primary benefit of The Shift. I would argue that we are guilty of looking only at this benefit, and not the whole system.
2. There are also side effects that are not immediately apparent. There is some force inherent in The Shift that is giving a slight boost to the number of line drives that batters hit above and beyond what we would expect. Yes, plenty of balls just became a little bit easier to field, but there are a few new balls in the system that are a lot harder to field. As a result, the effect on BABIP appears to be largely null, if not a little bit negative for the defense.
3. We have evidence that pitchers are throwing more balls. We can’t directly say that this means “more walks” but in the absence of public data to directly test that hypothesis, it at least shifts the preponderance of evidence to “yes, it does.” There’s not much of a corresponding increase in strikes either, so it’s not like it’s going to be balanced out by a bunch of strikeouts.
4. Whatever is happening with those balls, we know that The Shift tends to increase pitch count and we know that all else equal, increasing a pitcher’s pitch count is a bad thing.
5. This one I can only classify as informed speculation, but if we’re seeing that batters are having an easier time squaring up a line drive, then it’s reasonable to suspect (and let’s just leave it at a “suspicion”) that batters might have an easier time clobbering a ball out of the park. Because we don’t have data on whether the ball leaves the park, we don’t know. But it doesn’t have to be a big effect to erase whatever benefit is left from The Shift.

It Does What It Says on the Label ... and So Much More

The Shift was sold as a miracle drug. Like most miracle drugs, it does what it says on the label and everyone is happy with it, until they realize what the side effects are. Sometimes the treatment is worse than the disease.

In my past work on the topic, I have made an argument that I will make again, and it starts with the clause “... as The Shift is currently practiced and in the aggregate.” It’s reasonable to think that The Shift really does “work” for someone who—to take an extreme strawman example—pulls 100 percent of his ground balls, even if we accept that there’s going to be some value bled away in the form of a few more line drives and a few more walks.

There comes a point where that bargain no longer makes sense, and it seems that there are plenty of hitters being shifted who are probably on the wrong side of that break-even point. I don’t think that The Shift should go away. I think that we need a more complete accounting of what exactly it does, both in its obvious effects and in its subtler effects.

One thing is abundantly clear. The idea that The Shift is having a profound effect on the game, at least statistically, is over-played. The evidence we have available is that—compared to what that hitter and pitcher would have done otherwise—we’re seeing mostly the same outcomes.

I don’t think The Shift is unsalvageable from a strategic standpoint. For example, there’s probably nothing about the fact that there are three infielders on the right side that increases line-drive rate. The active ingredient there could be the way in which a pitcher changes his approach when pitching in front of The Shift. Maybe that’s also the reason for the increase in called balls. Maybe there’s a way to work with pitchers to neutralize some of that.

I do think that The Shift needs to be curtailed, not for moral reasons, but for practical ones. When you evaluate a strategy, you have to look for the obvious impacts and the effects on the rest of the system. In this case, there are some serious side effects that I think are threatening to cancel out whatever progress The Shift thinks that it’s making.

Russell A. Carleton is an author of Baseball Prospectus. Follow @pizzacutter4
Click here to see Russell's other articles. You can contact Russell by clicking here