Many people don’t realize that one of the (many) weaknesses of UZR, at least for the infield, is that it ignores any ground ball in which the infield was configured in some kind of a “shift” and it “influenced the play.” I believe that’s true of DRS as well.
What exactly constitutes “a shift” and how they determine whether or not it “influenced the play” I unfortunately don’t know. It’s up to the “stringers” (the people who watch the plays and input and categorize the data) and the powers that be at Baseball Info Solutions (BIS). When I get the data, there is merely a code, “1” or “0”, for whether there was a “relevant shift” or not.
How many GB are excluded from the UZR data? It varies by team, but in 2015 so far, about 21% of all GB are classified by BIS as “hit into a relevant shift.” The average team has had 332 shifts in which a GB was ignored by UZR (and presumably DRS) and 1268 GB that were included in the data that the UZR engine uses to calculate individual UZR’s. The number of shifts varies considerably from team to team, with the Nationals, somewhat surprisingly, employing the fewest, with only 181, and the Astros with a whopping 682 so far this season. Remember these are not the total number of PA in which the infield is in a shifted configuration. These are the number of ground balls in which the infield was shifted and the outcome was “relevant to the shift,” according to BIS. Presumably, the numbers track pretty well with the overall number of times that each team employs some kind of a shift. It appears that Washington disdains the shift, relatively speaking, and that Houston loves it.
In 2014, there were many fewer shifts than in this current season. Only 11% of ground balls involved a relevant shift, half the number than in 2015. The trailer was the Rockies, with only 92, and the leader, the Astros, with 666. The Nationals last year had the 4th fewest in baseball.
Here is the complete data set for 2014 and 2015 (as of August 30):
2014
| Team | GB | Shifted | Not shifted | % Shifted |
| ari | 2060 | 155 | 1905 | 8 |
| atl | 1887 | 115 | 1772 | 6 |
| chn | 1958 | 162 | 1796 | 8 |
| cin | 1938 | 125 | 1813 | 6 |
| col | 2239 | 92 | 2147 | 4 |
| hou | 2113 | 666 | 1447 | 32 |
| lan | 2056 | 129 | 1927 | 6 |
| mil | 2046 | 274 | 1772 | 13 |
| nyn | 2015 | 102 | 1913 | 5 |
| phi | 2105 | 177 | 1928 | 8 |
| pit | 2239 | 375 | 1864 | 17 |
| sdn | 1957 | 133 | 1824 | 7 |
| sln | 2002 | 193 | 1809 | 10 |
| sfn | 2007 | 194 | 1813 | 10 |
| was | 1985 | 116 | 1869 | 6 |
| mia | 2176 | 125 | 2051 | 6 |
| ala | 1817 | 170 | 1647 | 9 |
| bal | 1969 | 318 | 1651 | 16 |
| bos | 1998 | 247 | 1751 | 12 |
| cha | 2101 | 288 | 1813 | 14 |
| cle | 2003 | 265 | 1738 | 13 |
| det | 1995 | 122 | 1873 | 6 |
| kca | 1948 | 274 | 1674 | 14 |
| min | 2011 | 235 | 1776 | 12 |
| nya | 1902 | 394 | 1508 | 21 |
| oak | 1980 | 244 | 1736 | 12 |
| sea | 1910 | 201 | 1709 | 11 |
| tba | 1724 | 376 | 1348 | 22 |
| tex | 1811 | 203 | 1608 | 11 |
| tor | 1919 | 328 | 1591 | 17 |
2015
| Team | GB | Shifted | Not shifted | % Shifted |
| ari | 1709 | 355 | 1354 | 21 |
| atl | 1543 | 207 | 1336 | 13 |
| chn | 1553 | 239 | 1314 | 15 |
| cin | 1584 | 271 | 1313 | 17 |
| col | 1741 | 533 | 1208 | 31 |
| hou | 1667 | 682 | 985 | 41 |
| lan | 1630 | 220 | 1410 | 13 |
| mil | 1603 | 268 | 1335 | 17 |
| nyn | 1610 | 203 | 1407 | 13 |
| phi | 1673 | 237 | 1436 | 14 |
| pit | 1797 | 577 | 1220 | 32 |
| sdn | 1608 | 320 | 1288 | 20 |
| sln | 1680 | 266 | 1414 | 16 |
| sfn | 1610 | 333 | 1277 | 21 |
| was | 1530 | 181 | 1349 | 12 |
| mia | 1591 | 229 | 1362 | 14 |
| ala | 1493 | 244 | 1249 | 16 |
| bal | 1554 | 383 | 1171 | 25 |
| bos | 1616 | 273 | 1343 | 17 |
| cha | 1585 | 230 | 1355 | 15 |
| cle | 1445 | 335 | 1110 | 23 |
| det | 1576 | 349 | 1227 | 22 |
| kca | 1491 | 295 | 1196 | 20 |
| min | 1655 | 388 | 1267 | 23 |
| nya | 1619 | 478 | 1141 | 30 |
| oak | 1599 | 361 | 1238 | 23 |
| sea | 1663 | 229 | 1434 | 14 |
| tba | 1422 | 564 | 858 | 40 |
| tex | 1603 | 297 | 1306 | 19 |
| tor | 1539 | 398 | 1141 | 26 |
The individual fielding data (UZR) for infielders that you see on Fangraphs is based on non-shifted ground balls only, or on ground balls where there was a shift but it wasn’t relevant to the outcome. The reason that shifts are ignored in UZR (and DRS, I think) is because we don’t know where the individual fielders are located. It could be a full shift, a partial shift, the third baseman could be the left-most fielder as he usually is or he could be the man in short right field between the first baseman and the second baseman, etc. The way most of the PBP defensive metrics work, it would be useless to include this data.
But what we can do, with impunity, is to include all ground ball data in a team UZR. After all, if a hard ground ball is hit at the 23 degree vector, and we are only interested in team fielding, we don’t care who is the closest fielder or where he is playing. All we care about is whether the ball was turned into an out, relative to the league average out rate for a similar ground ball in a similar or adjusted for context. In other words, using the same UZR methodology, we can calculate a team UZR using all ground ball data, with no regard for the configuration of the IF on any particular play. And if it is true that the type, number and timing (for example, against which batters and/or with which pitchers) of shifts is relevant to a team’s overall defensive efficiency, team UZR in the infield should reflect not only the sum of individual fielding talent and performance, but also the quality of the shift in terms of hit prevention. In addition, if we subtract the sum of the individual infielders’ UZR on non-shift plays from the total team UZR on all plays, the difference should reflect, at least somewhat, the quality of the shifts.
I want to remind everyone that UZR accounts for several contexts. One, park factors. For infield purposes, although the dimensions of all infields are the same, the hardness and quality of the infield can differ from park to park. For example, in Coors Field in Colorado and Chase Field in Arizona, the infields are hard and quick, and thus more ground balls scoot through for hits even if they leave the bat with the same speed and trajectory.
Two, the speed of the batter. Obviously faster batters require the infielders to play a little closer to home plate and they beat out infield ground balls more often than slower batters. In some cases the third baseman and/or first baseman have to play in to protect against the bunt. This affects the average “caught percentage” for any given category of ground balls. The speed of the opposing batters tends to even out for fielders and especially for teams, but still, the UZR engine tries to account for this just in case it doesn’t, especially in small samples.
The third context is the position of the base runners and number of outs. This affects the positioning of the fielders, especially the first baseman (whether first base is occupied or not). The handedness of the batters is the next context. As with batter speed, these also tend to even out in the long run, but it is better to adjust for them just in case.
Finally, the overall GB propensity of the pitchers is used to adjust the average catch rates for all ground balls. The more GB oriented a pitcher is, the softer his ground balls are. While all ground balls are classified in the data as soft, medium, or hard, even within each category, the speed and consequently the catch rates, vary according to the GB tendencies of the pitcher. For example, for GB pitchers, their medium ground balls will be caught at a higher rate than the medium ground balls allowed by fly ball pitchers.
So keep in mind that individual and team UZR adjust as best as it can for these contexts. In most cases, there is not a whole lot of difference between the context adjusted UZR numbers and the unadjusted ones. Also keep in mind that the team UZR numbers you see in this article are adjusted for park, batter hand and speed, and runners/outs, the same as the individual UZR’s you see on Fangraphs.
For this article, I am interested in team UZR including when the IF is shifted. Even though we are typically interested in individual defensive performance and talent, it is becoming more and more difficult to evaluate individual fielding for infielders, because of the prevalence of the shift, and because there is so much disparity in how often each team employs the shift (so that we might be getting a sample of only 60% of the available ground balls for one team and 85% for another).
One could speculate that teams that employ the most shifts would have the best team defense. To test that, we could look at each team’s UZR versus their relevant shift percentage. The problem, of course, is that the talent of the individual fielders is a huge component of team UZR, regardless of how often a team shifts. There may also be selective sampling going on. Maybe teams that don’t have good infield defense feel the need to shift more often such that lots of shifts get unfairly correlated with (but are not the cause of) bad defense.
One way we can separate out talent from shifting is to compare team UZR on all ground balls with the total of the individual UZR’s for all the infielders (on non-shifted ground balls). The difference may tell us something about the efficacy of the shifts and non-shifts. In other words, total team individual infield UZR, which is just the total of each infielder’s UZR as you would see on Fangraphs (range and ROE runs only), represents what we generally consider to be a sample of team talent. This is measured on non-shifted ground balls only, as explained above.
Team total UZR, which measures team runs saved or cost, with no regard for who caught each ball or not, and is based on every batted ball, shifted or not, represents how the team actually performed on defense and is a much better measure of team defense than totaling the individual UZR’s. The difference, then, to some degree, represents how efficient teams are at shifting or not shifting, regardless of how often they shift.
There are lots of issues that would have to be considered when evaluating whether shifts work or not. For example, maybe shifting too much with runners on base results in fewer DP because infielders are often out of position. Maybe stolen bases are affected for the same reason. Maybe the number and quality of hits to the outfield change as a result of the shift. For example, if a team shifts a lot, maybe they don’t appear to record more ground ball outs, but the shifted batters are forced to try and hit the ball to the opposite field more often and thus they lose some of their power.
Maybe it appears that more ground balls are caught, but because pitchers are pitching “to the shift” they become more predictable and batters are actually more successful overall (despite their ground balls being caught more often). Maybe shifts are spectacularly successful against some stubborn and pull-happy batters and not very successful against others who can adjust or even take advantage of a shift in order to produce more, not less, offense. Those issues are beyond the scope of UZR and this article.
Let’s now look at each team in 2014 and 2015, their shift percentage, their overall team UZR, their team UZR when shifting, when not shifting, and their total individual combined UZR when not shifting. Remember this is for the infield only.
2015
| Team | % Shifts | Shift Runs | Non-Shift Runs | Team Runs | Total Individual Runs | Team Minus Individual after prorating Ind Runs to 100% of plays |
| KCA | 20 | -2.2 | 10.5 | 10 | 26.3 | -19.6 |
| LAN | 13 | -5 | -7.3 | -13.3 | 0.8 | -14.2 |
| TOR | 26 | -2.5 | 13.9 | 11 | 22.6 | -15.6 |
| CHA | 15 | -7.7 | -12.3 | -21.8 | -11.9 | -8.9 |
| CLE | 23 | 0.6 | 3.3 | 3.3 | 12.8 | -11.4 |
| MIN | 23 | 3.5 | -11.6 | -7.6 | 1.8 | -9.7 |
| MIL | 17 | 0.3 | -7.1 | -6.7 | 2.5 | -9.5 |
| SEA | 14 | -2.6 | -8.7 | -13.8 | -5.1 | -8.3 |
| SFN | 21 | 2.3 | 12.6 | 15.8 | 24.4 | -11.8 |
| MIA | 14 | 0.5 | 2.7 | 2.4 | 8.4 | -6.7 |
| ARI | 21 | 3.4 | -1.5 | 2.1 | 8 | -7.0 |
| HOU | 41 | -7.6 | -3.2 | -11.3 | -6.1 | -3.1 |
| PHI | 14 | -6.4 | -16.4 | -23.5 | -19 | -3.0 |
| COL | 31 | -7.3 | 0 | -5.5 | -1.5 | -3.7 |
| ATL | 13 | 3.1 | 6.9 | 9.8 | 12.6 | -3.7 |
| SLN | 16 | -1.1 | -5.8 | -8.8 | -7 | -1.1 |
| DET | 22 | 1.8 | -16.2 | -17.8 | -16 | 0.5 |
| ALA | 16 | -2.4 | -0.4 | -3.6 | -2.8 | -0.5 |
| BOS | 17 | 0.3 | 4.8 | 3.5 | 2.7 | 0.5 |
| NYN | 13 | -3.8 | 3.1 | 0.8 | -2.7 | 3.7 |
| WAS | 12 | 1.1 | -9.4 | -8.4 | -12.6 | 5.1 |
| CIN | 17 | 5 | 9.8 | 16.2 | 11.2 | 3.9 |
| CHN | 15 | 0.2 | 18.7 | 17.4 | 10.5 | 6.0 |
| BAL | 25 | 10.6 | -0.5 | 14.4 | 5.8 | 7.6 |
| SDN | 20 | 7.5 | -6.8 | 1.5 | -7.8 | 10.3 |
| TEX | 19 | 4.1 | 12.8 | 19.6 | 10.1 | 8.3 |
| TBA | 40 | 0.1 | 4.5 | 7 | -9.2 | 19.3 |
| NYA | 30 | 0.1 | 11.8 | 12.2 | -6.6 | 20.2 |
| PIT | 32 | 0.3 | 0.3 | 0.1 | -21 | 26.0 |
| OAK | 23 | 3.9 | -8.8 | -5 | -31.4 | 31.1 |
The last column, as I explained above, represents the difference between how we think the infield defense performed based on individual UZR’s only (on non-shifted GB), prorated to 100% of the games (the proration is actually regressed so that we don’t have the “on pace for” problem), and how the team actually performed on all ground balls. If the difference is positive, then we might assume that the shifts and non-shifts are being done in an effective fashion regardless of how often shifts are occurring. If it is negative, then somehow the combination of shifts and non-shifts are costing some runs. Or the difference might not be meaningful at all – it could just be noise. At the very least, this is the first time that you are seeing real infield team defense being measured based on the characteristics of each and every ground ball and the context in which they were hit, regardless of where the infielders are playing.
First of all, if we look at all the teams that have a negative difference in the last column, the teams that presumably have the worst shift/no-shift efficiency, and compare them to those that are plus and presumably have the best shift/no-shift efficiency, we find that there is no difference in their average shift percentages. For example, TBA and HOU have the most shifts by far, and HOU “cost” their teams 5.2 runs and TBA benefited by 16.2 runs. LAN and WAS had the fewest shifts and one of them gained 4 runs and the other lost 14 runs. The other teams are all over the board with respect to number of shifts and the difference between the individual UZR’s and team UZR.
Let’s look at that last column for 2014 and compare it to 2015 to see if there is any consistency from year to year within teams. Do some teams consistently do better or worse with their shifting and non-shifting, at least for 2014 and 2015? Let’s also see if adding more data gives us any relationship between the last column (delta team and individual UZR) and shift percentage.
| Team | 2015 % Shift | 2014 % Shift | 2015 Team Minus Individual | 2014 Team Minus Individual | Combined 2014 and 2015 Team Minus Individual |
| HOU | 41 | 32 | -5.2 | 45.6 | 40.4 |
| TBA | 40 | 22 | 16.2 | 12.7 | 28.9 |
| PIT | 32 | 17 | 21.1 | 5.5 | 26.6 |
| TEX | 19 | 11 | 9.5 | 9.9 | 19.4 |
| WAS | 12 | 6 | 4.2 | 13.0 | 17.2 |
| OAK | 23 | 12 | 26.4 | -9.3 | 17.1 |
| BAL | 25 | 16 | 8.6 | 7.6 | 16.2 |
| NYN | 13 | 5 | 3.5 | 9.0 | 12.5 |
| NYA | 30 | 21 | 18.8 | -8.4 | 10.4 |
| CHA | 15 | 14 | -9.9 | 12.5 | 2.6 |
| CHN | 15 | 8 | 6.9 | -5.8 | 1.1 |
| TOR | 26 | 17 | -11.6 | 12.6 | 1.0 |
| DET | 22 | 6 | -1.8 | 2.4 | 0.6 |
| SFN | 21 | 10 | -8.6 | 6.0 | -2.6 |
| CIN | 17 | 6 | 5 | -8.2 | -3.2 |
| CLE | 23 | 13 | -9.5 | 5.2 | -4.3 |
| MIL | 17 | 13 | -9.2 | 3.1 | -6.1 |
| ARI | 21 | 8 | -5.9 | -0.2 | -6.1 |
| SDN | 20 | 7 | 9.3 | -15.7 | -6.4 |
| MIA | 14 | 6 | -6 | -0.9 | -6.9 |
| BOS | 17 | 12 | 0.8 | -10.6 | -9.8 |
| KCA | 20 | 14 | -16.3 | 6.3 | -10.0 |
| ATL | 13 | 6 | -2.8 | -7.5 | -10.3 |
| PHI | 14 | 8 | -4.5 | -6.2 | -10.7 |
| ALA | 16 | 9 | -0.8 | -11.6 | -12.4 |
| SLN | 16 | 10 | -1.8 | -12.2 | -14.0 |
| LAN | 13 | 6 | -14.1 | -2.5 | -16.6 |
| MIN | 23 | 12 | -9.4 | -9.3 | -18.7 |
| SEA | 14 | 11 | -8.7 | -11.3 | -20.0 |
| COL | 31 | 4 | -4 | -23.0 | -27.0 |
Although there appears to be little correlation from one year to the next for each of the teams, we do find that of the teams that had the least efficient shifts/non-shifts (negative values in the last column), they averaged 14% shifts per season in 2014 and 2015. Those that had the most effective (plus values in the last column) shifted an average of 19% in 2014 and 2015. As well, the two teams with the biggest gains, HOU and TB, had the most shifts, at 37% and 31% per season, respectively. The two worst teams, Colorado and Seattle, shifted 17% and 13% per season. On the other hand, the team with the least shifts in baseball in 2014 and 2015 combined, the Nationals, gained over 17 runs in team UZR on all ground balls compared to a total of the individual UZR’s on non-shifted balls only, suggesting that the few shifts they employed were very effective, which seems reasonable.
It is also interesting to note that the team that had the worst difference in team and individual UZR in 2014, the Rockies, only shifted 4% of the time, easily the worst in baseball. In 2015, they have been one of the most shifted teams and still their team UZR is 4 runs worse than their total individual UZR’s. Still, that’s a lot better than in 2014.
It also appears that many of the smarter teams are figuring out how to optimize their defense beyond the talent of the individual players. TB, PIT, HOU, WAS, and OAK are at the top of the list in plus value deltas (the last column). These teams are generally considered to have progressive front offices. Some of the teams with the most negative numbers in the last column, those teams which appear to be sub-optimal in their defensive alignment, are LAN, MIN, SEA, PHI, COL, ATL, SLN, and ALA, all with reputations for having less than progressive front offices and philosophies, to one degree or another. In fact, other than a few outliers, like Boston, Texas, and the White Sox, the order of the teams in the chart above looks like a reasonable order of teams from most to least progressive teams. Certainly the teams in the top half appear to be the most saber-savvy teams and those in bottom half, the least.
In conclusion, it is hard to look at this data and figure out whether and which teams are using their shifts and non-shifts effectively. There doesn’t appear to be a strong correlation between shift percentage and the difference between team and individual defense although there are a few anecdotes that suggest otherwise. As well, in the aggregate for 2014 and 2015 combined, teams that have been able to outperform on team defense the total of their individual UZR’s have shifted more often, 19% to 13%.
There also appears to the naked eye to be a strong correlation between the perceived sabermetric orientation of a team’s front office and the efficiency of their shift/non-shift strategy, at least as measured by the numbers in the last column, explained above.
I think the most important thing to take away from this discussion is that there can be a fairly large difference between team infield UZR which uses every GB, and the total of the individual UZR’s which uses only those plays in which no shift was relevant to the outcome of the play. As well, the more shifts employed by a team, the less we should trust that the total of the individual performances are representative of the entire team’s defense on the infield. I am also going to see if Fangraphs can start publishing team UZR for infielders and for outfielders, although in the outfield, the numbers should be similar if not the same.
MGL on Baseball 