This previous analysis I have done (here & here), I have shown that shift is working against LHH but not against RHH. Also, I have shown that what is going on with the shift, for both RHH and LHH, is something way more than just moving some infielders about to stop some singles…
It suggests that batters and/or pitchers and/or catchers are changing their mentality and approach to what is going on in front of them. If everyone was playing the game the same, you wouldn’t see these drastic changes in categories outside BABIP.
So, let’s start from the position that the fielding team doesn’t want to make any other changes in their approach bar the infield shift. They want the hitter to behave normally and possibly hit into their shift. This means the pitcher pitches the same with or without the shift. With this as the start point we ask, is the hitter doing anything different?
To determine this, we look at the swing rate of the batters in various areas using the breakdown of the strike zone used by Statcast and then calculating the hitters’ swing rates and miss rates for the four areas: Ozone, Chase Zone, Shadow Zone and Heart of Plate (as detailed below).
These are then compared as with the previous results. Adjustments made to account for pitchers, the impact of batters on bases and any selection bias of the hitters. Differences are highlighted green for good (for the shifting team) and red for bad.
For LHH they swing more at pitches which are on the edge and outside of the zone, but for RHH they swing less at these pitches. Also, the RHH the swing and miss rate is lower for pitches in the heart of the zone or on the shadow of the plate. It therefore looks like right handed batters are being more patient than normal when they are shifted against, and left handers are being less patient.
Let’s look at where they are swinging in more detail, to see if the data backs this theory up.
To do this we fitted a binomial generalized additive model, in R, to all base empty pitches for the same shifted batters. Where the outcome is 1 (swing) or 0 (no swing) and the logit of the probability of a swing is a smooth function of the pitch location. This is modelled for each of the 8 ‘handedness’ and ‘shifting’ scenarios. The data for the shifted and non-shifted batters is from Statcast, there has been no weighting adjustments for number of shifted and non-shifted at bats of these hitters (selection bias).
This generates a difference map showing areas where the modelled probability of swing is different for shifted and non-shifted at bats. For the difference graphs, a positive difference indicates that the non-shifted model had a higher probability of swing.
LHH v LHP – (No Shift: 11075 pitches, Shift: 8915 pitches)
Looking at the two swing models, the difference isn’t too noticeable. But looking at the difference graph, you can see that the non-shifted batters swung more at pitches inside and low. And when shifted, they swing more at low and outside pitches. The difference is greater for the shifted low and away swings which matches up to the Heart/Chase/Shadow/Ozone findings.
LHH v RHP – (No Shift: 34194 pitches, Shift: 45590 pitches)
The difference graph here models an increased swing rate for pitches high in the zone and outside the zone to the left handers when they are shifted. This model shows almost no areas where the batter swings more often when they are not shifted. This matches up with the Heart/Chase/Shadow/Ozone findings that these batters have swung more frequently in all locations.
RHH v LHP – (No Shift: 18485 pitches, Shift: 7819 pitches)
The difference model above, for the right-handed batter is completely different than those for the left handers. It models a shifted batter to be swinging less in lots of areas around the outside of the zone as well as an increase in part of the zone. This matches up with the Heart/Chase/Shadow/Ozone findings that RHH are swinging less around the edge of the zone when they are shifted.
RHH v RHP – (No Shift: 24398 pitches, Shift: 15098 pitches)
As with v LHP, RHH v RHP models the batter swinging to be less likely in lots of areas around the outside of the zone as well as an increase in the centre part of the zone. Which once again matches up with the Heart/Chase/Shadow/Ozone findings.
Both the zone breakdown, and swing probability models show the same, but what impact does that have on the at bats? We look at a few other factors.
Namely, how often to the players get pitches in the 4 Statcast ‘zones’ and how many pitches these at bats see? These values are calculated via the same method as before to account for pitchers, the impact of batters on bases and any selection bias of the hitters.
As you can see, the patience which RHH, when shifted, are showing has led to them seeing more balls in the heart of the zone and less in the shadow, chase zone and ozone. When we look at the average number of pitches (the second chart here), for these at bats we see that shifted at bats for LHH are on average longer, and for RHH they are shorter.
Do these have an impact on what we would expect from an at bat?
Thanks to Statcast we can find the estimated BA and wOBA for batted ball events (BBE) and if we added in values for strikeouts, walks and hit by pitch we can calculate an xBA and xwOBA to see if these have made a difference. The Statcast estimated BA and wOBA is based of launch speed and angle only.
This shows there is minimal change to the BA and wOBA on BBEs, the highest being 3.1% difference. But if we add the strikeouts, walks and hit by pitches the calculated xBA and xwOBA it shows, for LHH a decrease and an increase for RHH. This means that independent of any BABIP or fielding we would be expecting LHH to do worse against the shift, and RHH to do better, owing to the changes in walk rates and strikeout rates.
Does this transfer across on the teams which are doing well shifting?
Teams Shifting v Left Handed Hitters (LHH) (Ordered by wOBA change against expectation)
For 3 of the 4 the fielding teams which had the greatest success, most of that came from getting weaker contact on hits and striking out more and not from the infield actually being shifted. For Houston, Minnesota and Baltimore their wOBA – xwOBA differences are close to zero. Also, these teams are throwing less pitches into the heart of the plate when shifted.
There are a few teams in the mid-range who have small xBA and xwOBA change due to shifting but have larger BA – xBA and wOBA – xwOBA differences, these are the teams that are more likely being successful in the traditional sense of how we imagine the shift to support outcomes – reducing BABIP.
Most of the teams who are doing poorly were expected to do worse; the worst of these being the White Sox. They throw more into the heart of the plate, which leads to higher expectation from BBEs and a higher xwOBA.
Teams Shifting v Right Handed Hitters (RHH) (Ordered by shift percentage)
If you look at the estimated BA/BBE and wOBA/BBE you might think that the shift is was working by reducing the quality of contact but the difference in walk rate and strikeout rate means that xBA and xwOBA are higher that expectation for almost all of the teams that shift more than 15%. Most of these teams lower the BA and wOBA from expectation, but not enough for shifting be an overall benefit for fielding teams.
Almost all of the team’s pitch in the heart of the zone more often when shifting; this doesn’t seem to have a positive impact on the outcome of batted balls.
The bottom teams have not had enough shifted at bats to provide reliable comparisons.
Conclusion
The hitter and team metrics both have the same outcome. Overall shifting does, on average, decrease a hitter’s BABIP, regardless of handedness but, owing to the difference in approach of RHH v LHH, the overall outcome is completely different.
RHH hitters have taken a more patient approach, when shifted, swinging less on pitches that are on the edge or outside of the zone. This has led to fewer strikeouts and more walks, which in turn has driven up the wOBA of shifted at bats, even though they make weaker contact.
On the other hand, when shifted, LHH have been chasing more pitches that are on the edge or outside of the zone, which has led to both weaker batted balls and increased strikeouts. These have both driven down the wOBA of the shifted at bats, regardless of the shifted infield.
Variance of use and performance across teams shows that there is no ‘one and done’ approach to using the shift and further emphasises that where teams are with analytics is vastly different.
Why these RHH are more patient then the LHH is unclear; a we’d need to take up with the players themselves.
Appendix: Colorado Rockies
As we all know already, whenever the Rockies, or their players, appear to ‘look different’ in a stat you should check their home/road splits. In 2018 the Rockies, shifted similarly home and away, but the impact of the shift was dramatically different. The chart below is for LHH only.
The outcome analysis suggests that they were successful when shifting outside of Coors Field and when playing at home they were not successful when shifting. This looks largely down to the changes to the walk rate and strikeout rate but there is a large BABIP difference as well.
Looking at the further detail we see that the eBA/BBE and ewOBA/BBE drops when they are both home and away. The pitch location data suggest that they are pitching more than expected out of the zone when shifting when at home but not when away.
When at home, taking into account the changes to the walk rate and strikeout rate, the xBA and xwOBA are in line for the shifted and non-shifted at bats which is vastly different to the expected increase. This could suggest that shifting isn’t working for Colorado at Coors Field.
But if you look at how the batters doing against expectation you can see that the batters aren’t performing better than expected. The 8% and 12% increases are actually in line with what is expected for playing at Coors in comparison to their season average. What is happening is that the results batters were getting was a lot lower than expected when not being shifted, which makes it look like the shift isn’t working.
We are looking at a very small sample size so I cannot infer with much confidence, it would it would be worth monitoring this going forward so see if this trend continues but this might just be randomness.