The Effects of Major League Baseball's Ban on Infield Shifts: A Quasi-Experimental Analysis

Lee Kennedy-Shaffer

REVIEW 3 major objections 6 minor 43 references

The Effects of Major League Baseball's Ban on Infield Shifts: A Quasi-Experimental Analysis

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read MLB's 2023 ban on infield shifts raised left-handed batters' BABIP and on-base percentage by nine points each, and synthetic-control analysis attributes far larger offensive gains to individual high-shift players such as Corey Seager.

desk verdict A credible league-wide DID estimate of the shift ban plus a genuinely new per-player synthetic control application; the player-level results are suggestive, and the in-time placebo concern is real but the sign argument in the stress-test note is backwards. read the letter →

arxiv 2411.15075 v1 pith:2GDEUEXC submitted 2024-11-22 stat.AP stat.ME

Lee Kennedy-Shaffer This is my paper · ORCID

classification stat.APstat.ME MSC 62D20

keywords difference-in-differencesnaturalexperimentspaneldataquasi-experimentssportsanalyticssyntheticcontrol

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish whether Major League Baseball's 2023 ban on infield shifts achieved its aim of generating more offense, and it does so with two observational designs that borrow causal-inference methods from policy evaluation. It claims the ban raised left-handed batters' batting average on balls in play (BABIP) and on-base percentage by nine points each, a modest league-wide effect that persisted into 2024, while player-level synthetic control estimates attribute far larger gains to the most-shifted hitters, including a 271-point on-base-plus-slugging (OPS) gain for Corey Seager. A careful reader should care because the same tools apply to any sports rule change, injury, or tactical experiment in which a league provides no randomized control group.

What carries the argument

The argument runs on two quasi-experimental machines. Difference-in-differences compares left-handed batters, the group the shift primarily targeted, with right-handed batters, and subtracts the right-handers' year-to-year change in BABIP and on-base percentage from the left-handers' change across the 2022-2023 boundary, so that shared time trends cancel; its validity depends on the parallel-trends assumption. The synthetic control method builds a counterfactual for each of 30 high-shift target players as a weighted average of low-shift donor players, with weights between zero and one summing to one, chosen so that the donor composite matches the target's pre-2023 trajectory in the outcome, age, plate appearances, hits, singles, home runs, walk rate, and strikeout rate; the composite's actual 2023 value is the counterfactual, and the gap between it and the target's real outcome is the estimated effect of the ban. Placebo procedures that re-fit the same machinery for the donor players themselves and for a mid-shift group of players provide a null distribution, while an in-time placebo using 2022 as a fake intervention year checks for systematic overestimation.

What would settle it

Two checks would settle the player-level claims. First, re-run Seager's synthetic control with his most heavily weighted donor player (60% of the weight in the OPS fit) removed from the donor pool: if the estimated 271-point OPS gain collapses toward zero, the headline player result is an artifact of a single comparison player. Second, examine public batted-ball location data for left-handed ground balls: if the nine-point BABIP gain is really the ban working, the new hits should cluster in the vacated middle-infield zone, whereas if the gain appears instead in exit velocity or launch-angle distributions, it belongs to a different rule change or to luck.

Watch

Extended reading notes

Core claim

The central claim is that the 2023 shift ban raised offensive output for the batters it most affected, and that the size of the effect varied predictably with how often a batter had been shifted. The league-wide difference-in-differences estimate puts the effect on left-handed batters' batting average on balls in play and on-base percentage at nine points each, with near-zero estimates in prior no-policy years and a 2024 estimate of -0.001, which indicates the gain persisted. The player-level synthetic control estimates attribute the largest gains to the most-shifted hitters: Corey Seager, whose 2022 shift rate was 92.8%, is estimated to have gained 85 points of on-base percentage, 271 points of on-base-plus-slugging (OPS), and 115 points of weighted on-base average (wOBA) in 2023, and four players (Seager, Matt Olson, Yordan Alvarez, and Shohei Ohtani) show estimated wOBA gains above 80 points. The paper also finds a dose-response pattern: each additional 10 percentage points of 2022 shift rate maps to roughly 11 points of on-base percentage, 31 points of OPS, and 17 points of wOBA in estimated gain.

Load-bearing premise

The player-level results stand on the assumption that the weighted mix of low-shift players who tracked a target's stats before 2023 would have kept tracking his stats in 2023 and 2024 if the shift ban had never happened, so a changed role, injury, or aging divergence would masquerade as an effect of the ban.

Editorial extensions

If this is right

The league-wide effect, nine points of BABIP and on-base percentage for left-handed batters in bases-empty situations, amounts to roughly one additional on-base event per 500 plate appearances, so the ban was a modest success on its own terms.
The near-zero 2024 difference-in-differences estimate indicates the 2023 gain did not revert, meaning teams' counter-strategies such as 'strategic shades' did not erase the effect.
Because the paper finds a dose-response relationship between 2022 shift rate and 2023 gain, future rule changes intended to help a specific subset of players can be evaluated at both league and player levels with the same design.
The synthetic control estimates are average treatment effects on the treated for the 2023 season specifically, so they are not directly generalizable to other seasons, leagues, or rule environments without further assumptions.
The player-level estimates are transparent in a way batted-ball models are not: the weighted donor players are listed explicitly, so analysts can judge whether each comparison is plausible.
Most target players (over 75%) show positive estimated effects across all three outcomes, with target-player estimates averaging four to eight times the size of placebo-player estimates.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

An unstated implication is a valuation problem: if Seager's 271-point OPS gain is real, the shift was suppressing roughly a quarter of his offensive production, and teams' pre-2023 willingness to accept that suppression implies they judged the shift's run-prevention value to exceed the batter's adjustment cost, a tradeoff the new rule has now settled.
The league-wide design cannot separate the shift ban from the other 2023 rule changes, such as the pitch clock and larger bases; a clean extension would replicate the lefty-versus-righty contrast in minor-league seasons where the shift ban was introduced in isolation.
The same synthetic-control machinery could be turned on injuries or suspensions, since a player's pre-injury donor composite is already constructed here and could estimate the counterfactual season a team lost.
A batted-ball test follows directly: if the ban caused the effect, left-handed hitters' gains should concentrate in ground balls through the vacated middle-infield gaps, and if the gains instead appear in exit velocity or launch-angle changes, the attribution to the shift ban weakens.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The league-wide difference-in-differences result is the most solid part of this paper: a 9-point increase in BABIP and OBP for left-handed batters, supported by placebo pre-trends and a near-zero 2024 estimate that suggests persistence. The simple LHB/RHB comparison appeared in public analyst work before, but the formal panel design, the explicit statement of assumptions, and the full 2015-2023 series are a genuine step up. The synthetic control application is also new in this setting and is transparently presented, with code, data, and placebo checks. A reader who wants to know how quasi-experimental methods can be used in sports will get real value here.

The soft spots are what you would expect. Table 1 reports no confidence intervals or standard errors for the DID estimates, which is easy to fix and should be fixed. The player-level SCM results involve 30 targets times three outcomes, the p-values are not multiplicity-adjusted, and the donor pools are small. The paper is honest about this, explicitly framing the placebo p-values as screening indicators rather than strict tests. That is the right framing, and the reader should keep it.

On the stress-test note: the concern that negative in-time placebo estimates imply the player-level gains are inflated has the sign backwards. If the synthetic control over-predicts for high-shift players in a pre-ban year (which is what a negative placebo means), and that bias persists into 2023, then the estimated effect (observed minus synthetic) would be understated, not overstated. The author's interpretation that the negative placebo reflects a pre-existing decline in high-shift batters' performance is plausible, and it is consistent with the negative DID pre-trends for 2021-2022. So that specific worry does not land as stated. What does land is the more general point that the stable-weights assumption is untestable and the in-time placebo provides some evidence against it. The paper acknowledges this directly in Appendix B and the Discussion.

Worth a serious referee. The right outcome is likely minor-to-moderate revision: add uncertainty quantification to Table 1, keep the screening language for the individual estimates, and maybe show the sensitivity of the Seager result to dropping Turner or Correa from the donor pool. I would not desk-reject this.

Referee Report

3 major / 6 minor

Summary. The paper uses difference-in-differences (DID) and synthetic control methods (SCM) to estimate the effect of Major League Baseball's 2023 ban on infield shifts. The league-wide DID compares left-handed batters to right-handed batters in bases-empty plate appearances before and after the ban, reporting a 9-point increase in both BABIP and OBP for left-handed batters in 2023, with the effect persisting in 2024. The player-level SCM analysis focuses on 30 high-shift batters, using 58 low-shift batters as donors, and reports substantial estimated gains for several players, including Corey Seager (OBP +85 points, OPS +271 points, wOBA +115 points). The paper also discusses robustness checks, including placebo pre-trends, an in-unit placebo analysis, and an in-time placebo analysis using 2022 as a dummy intervention year, and provides all data and code in a GitHub repository.

Significance. If the league-wide estimate is credible, it provides a rigorous, reproducible quantification of a real policy change in professional sports and demonstrates the utility of quasi-experimental methods for sports analytics. The analysis is genuinely transparent: the synthetic control weights are shown in detail, placebo distributions are constructed, and all code and data are provided. The league-wide result is plausible and consistent with prior batted-ball analyses, and the 2024 persistence check is a valuable addition. However, the player-specific claims are substantially weaker than the abstract and discussion imply: the in-time placebo indicates a systematic divergence between targets and synthetic controls in a pre-intervention year, and the permutation p-values are not adjusted for the 90 tests performed. These issues do not invalidate the league-wide finding, but they do require a significant reframing and additional analysis of the player-level results.

major comments (3)

[Results, Analysis 1; Table 1] The league-wide DID estimates of 0.009 for BABIP and OBP are presented without any measure of statistical uncertainty. No standard errors, confidence intervals, or p-values are reported for the 2023 comparison, so the reader cannot judge whether the 9-point effect is distinguishable from year-to-year noise. The placebo pre-trends and the 2024 persistence check are helpful, but they do not provide a formal test of the 2023 estimate. Please report a variance estimate (e.g., a cluster-robust or bootstrap standard error) and a confidence interval for the main DID estimates, and ideally for the placebo-year estimates as well.
[Results, Analysis 2; Appendix A, Figure A.4; Appendix B, Assumptions] The in-time placebo analysis using 2022 as a dummy intervention year produces 'moderate negative effect estimates for many players.' This is a diagnostic failure of the synthetic control: under the stable-weights assumption stated in Appendix B, the synthetic control should track the target player's outcome in the absence of the intervention, but it systematically over-predicts the 2022 outcomes for high-shift players. The paper interprets this as a real pre-existing decline in high-shift batters' performance, but that interpretation is not independently verified. If the same over-prediction continues in 2023, the 2023 effect estimates are contaminated, although the direction of the bias is unclear (over-prediction would understate the positive effects, not overstate them). Please provide a sensitivity analysis that quantifies how the 2023 player-specific estimates change when adjusted for the mean (or player-specific) 2022 placebo effect, or otherwise demonstrate that the in-time placebo does not affect the substantive conclusions.
[Results, Analysis 2; Table 3] The player-level p-values in Table 3 are unadjusted for multiple testing. With 30 target players and 3 outcomes, 90 hypotheses are effectively tested, and the smallest possible permutation p-value with 58 placebo players is 1/59 ≈ 0.017, which is exactly the value reported for Seager, Olson, Alvarez, and Ohtani. After any standard multiple-comparison correction (e.g., Bonferroni or Benjamini-Hochberg), none of the individual player effects would be statistically significant. The authors acknowledge that the p-values are 'not strict hypothesis tests,' but the abstract and discussion highlight individual players with 'substantial' gains as if they were individually confirmed. Please re-analyze the player-level results using a family-wise error control procedure, or clearly reframe the player-specific estimates as exploratory findings that are not individually confirmatory.

minor comments (6)

[Table 1] Table 1 does not report the number of plate appearances behind each rate, so the reader cannot assess the precision of the group-specific averages. Please add PA counts for each cell.
[Figure 1] Figure 1 would benefit from confidence bands or at least a statement about the variability of the annual DID estimates. Currently the pre-trend panel (C,D) shows point estimates only, making it hard to judge how 'centered around 0' the placebo estimates are.
[Materials and Methods, Analysis 2] The description of the placebo p-value in the text could be more precise: it should state that the p-value is the proportion of placebo estimates, including the target estimate, that are at least as extreme in absolute value, and that the minimum attainable value is 1/(number of placebos+1).
[Table 2] The 'Weight Ranking' column in Table 2 is inconsistent across outcomes: for OBP only two weights are shown, while for OPS and wOBA five weights are shown. Use the same format for all three columns.
[Discussion] The term 'average treatment effect on the treated' (ATT) is applied to the player-specific SCM estimates, but the estimand is a unit-specific effect for a single player, not an average over treated units. Consider using 'unit-specific treatment effect' or 'synthetic control effect' to avoid confusion.
[Materials and Methods, Analysis 2] The paper states that the donor pool consists of 58 low-shift players, but Appendix B notes that the donor pool size changes for each target player because of the 250-PA requirement. Please report the distribution of donor pool sizes across the 30 target players, since this affects the resolution of the placebo p-values and the stability of the synthetic controls.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the DID and SCM estimates use held-out post-intervention outcomes and no parameter fitted to the target effect.

full rationale

The paper's central estimates are not circular. Analysis 1 computes the shift-ban effect as the 2023-vs-2022 change for left-handed batters minus the same change for right-handed batters; the 2023 post-period outcome enters only as the held-out response, and the pre-2023 placebo comparisons are diagnostics, not fitting inputs. Analysis 2 constructs synthetic controls from low-shift donor players using only 2015-2022 data and 2021-2022 covariates to minimize pre-intervention prediction error; the 2023 and 2024 target outcomes are not used in weight selection, so player-level gains such as Seager's are out-of-sample predictions rather than refitted values. The in-time placebo (2022 dummy intervention) and in-unit placebo (15-30% shift-rate players) analyses are robustness checks; even if the negative in-time placebo estimates raise a bias concern, the paper reports and interprets them as possible underestimation rather than using them to manufacture the effect. The two author self-citations (Kennedy-Shaffer 2022, 2024) are methodological references and are not load-bearing premises, fitted parameters, or uniqueness arguments. No equation defines the estimand in terms of the estimate or fits a parameter to the post-period effect, so no circular step is present.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

Causal identification rests on standard quasi-experimental assumptions; the most fragile are stable SCM weights and no spillover to comparison groups. The analysis introduces no invented entities, and the hand-chosen thresholds and scope restrictions are not tuned to the outcome.

free parameters (4)

Minimum plate appearance threshold = 250 PAs per season
Hand-chosen inclusion rule for player-level analyses; changes donor pools and pre-period availability for each target player.
Shift rate categorization cutoffs = 15% low, 75% high, 15-30% in-unit placebo
Hand-chosen boundaries assigning players to donor, target, and placebo groups; central to the SCM comparison.
Bases-empty restriction for DID = bases empty only
Limits the league-wide analysis to the plate-appearance subset where shifts were most common; effects are rescaled by the 23.3% share of such PAs.
Start year 2015 = 2015
Chosen as the beginning of full Statcast positioning data and the modern shift era; determines the length of the pre-period.

assumptions (6)

domain assumption Consistency and no anticipation (Appendix B, Assumption 1)
Assumes pre-2023 outcomes equal potential outcomes without the ban and no behavior changes before implementation.
domain assumption Parallel trends for DID (Appendix B, Assumption 3)
Assumes LHB and RHB bases-empty outcomes would have moved in parallel absent the ban; placebo pre-trends support but cannot prove it.
domain assumption No spillover to comparison units (Appendix B, Assumption 2)
Assumes the ban did not affect RHB PAs or low-shift donor players; acknowledged as only approximately true, with bias likely toward the null.
domain assumption Stable synthetic control weights (Appendix B, Analysis 2 Assumptions)
Assumes donor weights fit to pre-2023 data remain valid in 2023-2024; the central fragility for player-specific estimates.
standard math Convex hull / donor coverage condition (Appendix B, Analysis 2 Assumptions)
Standard SCM requirement that target covariates lie within the convex hull of donor covariates.
domain assumption Leaderboard data accuracy
Assumes FanGraphs and Baseball Savant correctly record PAs, outcome stats, and shift rates.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The Effects of Major League Baseball's Ban on Infield Shifts: A Quasi-Experimental Analysis." pith.science (2026). https://pith.science/paper/2GDEUEXC

@misc{pith2026241115075,
  author       = {Pith},
  title        = {Pith review of: The Effects of Major League Baseball's Ban on Infield Shifts: A Quasi-Experimental Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2GDEUEXC}},
  note         = {Machine review of arXiv:2411.15075}
}

read the original abstract

From 2020 to 2023, Major League Baseball changed rules affecting team composition, player positioning, and game time. Understanding the effects of these rules is crucial for leagues, teams, players, and other relevant parties to assess their impact and to advocate either for further changes or undoing previous ones. Panel data and quasi-experimental methods provide useful tools for causal inference in these settings. I demonstrate this potential by analyzing the effect of the 2023 shift ban at both the league-wide and player-specific levels. Using difference-in-differences analysis, I show that the policy increased batting average on balls in play and on-base percentage for left-handed batters by a modest amount (nine points). For individual players, synthetic control analyses identify several players whose offensive performance (on-base percentage, on-base plus slugging percentage, and weighted on-base average) improved substantially (over 70 points in several cases) because of the rule change, and other players with previously high shift rates for whom it had little effect. This article both estimates the impact of this specific rule change and demonstrates how these methods for causal inference are potentially valuable for sports analytics -- at the player, team, and league levels -- more broadly.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 41 canonical work pages

[1]

The Effects of Major League Baseball’s Ban on Infield Shifts: A Quasi- Experimental Analysis Lee Kennedy-Shaffer, PhD Department of Biostatistics, Yale School of Public Health, New Haven, CT, USA Correspondence: Lee.kennedy-shaffer@yale.edu Abstract From 2020 to 2023, Major League Baseball changed rules affecting team composition, player positioning, and ...

work page 2020
[2]

Players with weights below 0.1% are excluded

Top-weighted donor players in the synthetic control fits for Corey Seager, for the three outcomes. Players with weights below 0.1% are excluded. Weight Ranking OBP OPS wOBA 1 Starling Marte 63% Trea Turner 60% Trea Turner 47% 2 Carlos Correa 37% Carlos Correa 20% Jean Segura 25% 3 - Jean Segura 10% Carlos Correa 21% 4 - Yan Gomes 6% José Abreu 8% 5 - Paul...

work page 2023
[3]

Overall, for each of the three outcomes, most target players (over 75%) had an estimated increase. The placebo test estimates, as expected, had estimates greater than 0 near 50% of the time (although it was somewhat higher for OPS and wOBA, indicating possible systemic underestimation of the 2023 outcomes). The mean and median value of the estimates acros...

work page 2023
[4]

exciting

and in business analyses of revenue and attendance effects of decisions (Brook 2022; Cardazzi and Rodriguez 2024; Cisyk 2020). A recent study also looked at the effect of MLB’s automatic baserunner rule in extra innings on game times using DID (Kennedy-Shaffer 2022). Another study used both DID and SCM to analyze the effect of the 1981 change in match res...

work page 2024
[6]

See Appendix B and Abadie (2021) for further technical details of the synthetic control method; it is implemented using the tidysynth R package (Dunford 2023)

for the control players that minimize the squared distance from a weighted average of the covariates, where the covariate weights are chosen to best predict the outcome trajectory in the pre-intervention period. See Appendix B and Abadie (2021) for further technical details of the synthetic control method; it is implemented using the tidysynth R package (...

work page 2021
[7]

The synthetic control is re-fit for these groups and estimated effects obtained for both 2023 and

This yields a subset of the above player list (27 target players and 42 control players). The synthetic control is re-fit for these groups and estimated effects obtained for both 2023 and

work page 2023
[8]

Additional robustness checks are conducted by performing two similar analyses where zero or near-zero effects are expected. The first—an “in-unit” placebo analysis—conducts the same analysis for 2023 effects, using as the target players the 25 players who otherwise fit the criteria for inclusion with a shift rate in 2022 between 15 and 30%. We would expec...

work page 2023
[9]

Both indicate a modest increase in the outcome for left-handed batters in 2023 compared to what would have been expected in the absence of a rule change, with both estimates (coincidentally) equal to 0.009, or 9 points. To situate this magnitude in terms of yearly fluctuations and assess the parallel trends assumption, Figure 1 plots the trends in the two...

work page 2023

Show all 43 references

[10]

Two-by-two DID analysis for the effect of the shift ban in 2023, comparing left-handed and right-handed batters’ BABIP and OBP with the bases empty for 2023 vs

2023
[11]

Splits Leaderboard

Data source: FanGraphs “Splits Leaderboard” (2024). Batter Handedness BABIP OBP 2022 2023 Difference 2022 2023 Difference LHB 0.275 0.287 0.012 0.299 0.315 0.015 RHB 0.291 0.294 0.003 0.303 0.309 0.006 Difference –0.016 –0.007 0.009 –0.004 0.006 0.009 For years where no policy...

2024
[12]

If anything, this may indicate that LHBs would have continued to worsen compared to RHBs, causing the 2023 estimates to underestimate the positive effect of the change

This may reflect a divergence in the trends of RHBs and LHBs, possibly due to increasing shifts and more effective fielder positioning or other rule changes with a differential effect (like the universal designated hitter). If anything, this may indicate that LHBs would have c...

2023
[13]

Splits Leaderboard

Trends by batter handedness (A,B) and DID estimates (C,D) for BABIP (A,C) and OBP (B,D) by year, 2015–2023, excluding 2020, for bases empty plate appearances. The counterfactual 2023 value is also shown (A,B) assuming the increase from 2022 would be the same for LHBs as for RH...

2024
[14]

That analysis relied on modelling his batted-ball locations in 2022 and identifying balls in play that would have likely been hits in the absence of the shift

He was also specifically noted as a player likely to benefit from the new rule (Petriello 2023b). That analysis relied on modelling his batted-ball locations in 2022 and identifying balls in play that would have likely been hits in the absence of the shift. While valuable, esp...

2022
[15]

Corey Seager

The SCM accounts for these changes but relies on the identified weighted average being a good comparison for the target player, so there is value in conducting both analyses and triangulating results as a causal inference strategy (Matthay et al. 2020). The players with non-ne...

2020
[18]

Statcast Custom Leaderboard

Constructed using full (non-2020) seasons from 2016–2023 excluding seasons where Seager had fewer than 250 PAs (i.e., 2018). Donor players had at least 250 PAs in the same seasons. Data sources: Baseball Savant’s “Statcast Custom Leaderboard” (2024) and “Statcast Batter Positi...

2024
[19]

Statcast Custom Leaderboard

Shift rates in 2022 and estimated effects of the shift ban in 2023 on OBP, OPS, and wOBA, with associated placebo test p-values, for players with at least an 75% shift rate in 2022 and at least 250 PAs in each of the 2021–2023 seasons. Data sources: Baseball Savant’s “Statcast...

2024
[20]

Plots of estimated effects of the shift ban rule change in 2023 with pre-intervention trends in the difference between synthetic control estimated and observed statistic for OBP (A), OPS (B), and wOBA (C). All included players had at least 250 PAs in each of the 2021–2023 seas...

2023
[21]

Statcast Custom Leaderboard

Data sources: Baseball Savant’s “Statcast Custom Leaderboard” (2024) and “Statcast Batter Positioning Leaderboard” (2024). Figure 4 plots the effect estimates for the target players versus the player’s 2022 shift rate, indicating some dose response in the effect of the shift b...

2024
[22]

This relationship was modest, however, and the players’ responses exhibit high variability. On average, a 10 percentage point increase in 2022 shift rate corresponds to an increase in estimated effect for the player by 11 points in OBP, 31 points in OPS, and 17 points in wOBA....

2022
[23]

the player’s 2022 shift rate for OBP (A), OPS (B), and wOBA (C)

Plots of estimated effects of the shift ban rule change in 2023 for target players vs. the player’s 2022 shift rate for OBP (A), OPS (B), and wOBA (C). All included players had at least 250 PAs in each of the 2021–2023 seasons; target players had at least a 75% shift rate in 2...

2023
[24]

Statcast Custom Leaderboard

Dashed lines indicate ordinary least squares best fit lines. Data sources: Baseball Savant’s “Statcast Custom Leaderboard” (2024) and “Statcast Batter Positioning Leaderboard” (2024). To extend these results to the 2024 season, the SCM results were re-fit, only using the 27 ta...

2024
[25]

Using synthetic controls: Feasibility, data requirements, and methodological aspects,

Still other players (e.g., Joey Gallo, Seth Brown, and Rowdy Tellez) showed negative effect estimates in both years. Noticeable changes in the effect between years for a player could indicate one, or a combination, of three things: (1) a true change in the effect, due to chang...

2021
[27]

Statcast Custom Leaderboard

Constructed using full (non-2020) seasons from 2016–2024 excluding seasons where Seager had fewer than 250 PAs (i.e., 2018). Donor players had at least 250 PAs in the same seasons. Data sources: Baseball Savant’s “Statcast Custom Leaderboard” and “Statcast Batter Positioning L...

2024
[28]

Statcast Custom Leaderboard

Data sources: Baseball Savant’s “Statcast Custom Leaderboard” (2024) and “Statcast Batter Positioning Leaderboard” (2024). 28 Figure A.3. Plots of in-unit placebo estimated effects of the shift ban rule change in 2023 with pre-intervention trends in the difference between synt...

2024
[29]

Statcast Custom Leaderboard

Data sources: Baseball Savant’s “Statcast Custom Leaderboard” (2024) and “Statcast Batter Positioning Leaderboard” (2024). 29 Figure A.4. Plots of in-time placebo estimated effects of the shift ban rule change in the pre- intervention year of 2022 with pre-intervention trends ...

2024
[30]

Statcast Custom Leaderboard

Data sources: Baseball Savant’s “Statcast Custom Leaderboard” (2024) and “Statcast Batter Positioning Leaderboard” (2024). 30 Appendix B: Technical Appendix Analysis 1: League-Wide Effects (Difference-in-Differences) Notation Let 𝑌𝑗,𝑖,𝑡 be the value of outcome 𝑗 (i.e., BABIP o...

2024
[32]

The implementation can be found at the GitHub repository (https://bit.ly/QE-Baseball)

31 Estimation The estimators are calculated for all 𝑗, 𝑡 ≥ 2016 by: 𝜃̂𝑗,𝑡 = (𝑌𝑗,1,𝑡 − 𝑌𝑗,1,𝑡−1) − (𝑌𝑗,0,𝑡 − 𝑌𝑗,0,𝑡−1). The implementation can be found at the GitHub repository (https://bit.ly/QE-Baseball). Under Assumptions (1)–(3) above, 𝜃̂𝑗,2023 is unbiased for 𝜃𝑗,1,2023 for...

2016
[33]

= 𝐸[𝑌𝑗,1,2023 1 − 𝑌𝑗,1,2022 0 ] − 𝐸[𝑌𝑗,0,2023 0 − 𝑌𝑗,0,2022 0 ] (Assumption

2023
[34]

= 𝐸[𝑌𝑗,1,2023 1 − 𝑌𝑗,1,2022 0 ] − 𝐸[𝑌𝑗,1,2023 0 − 𝑌𝑗,1,2022 0 ] +𝐸[𝑌𝑗,1,2023 0 − 𝑌𝑗,1,2022 0 ] − 𝐸[𝑌𝑗,0,2023 0 − 𝑌𝑗,0,2022 0 ] = 𝐸[𝑌𝑗,1,2023 1 − 𝑌𝑗,1,2022 0 ] − 𝐸[𝑌𝑗,1,2023 0 − 𝑌𝑗,1,2022 0 ] +𝐸[𝑌𝑗,0,2023 0 − 𝑌𝑗,0,2022 0 ] − 𝐸[𝑌𝑗,0,2023 0 − 𝑌𝑗,0,2022 0 ] (Assumption

2023
[35]

= 𝐸[𝑌𝑗,1,2023 1 − 𝑌𝑗,1,2022 0 − 𝑌𝑗,1,2023 0 + 𝑌𝑗,1,2022 0 ] = 𝐸[𝑌𝑗,1,2023 1 − 𝑌𝑗,1,2023 0 ] = 𝜃𝑗,2023 Similarly, 𝜃̂𝑗,2024 is unbiased for 𝜃𝑗,2024 − 𝜃𝑗,2023 for all 𝑗: 𝐸[𝜃̂𝑗,2024] = 𝐸[(𝑌𝑗,1,2024 − 𝑌𝑗,1,2023) − (𝑌𝑗,0,2024 − 𝑌𝑗,0,2023)] = 𝐸[𝑌𝑗,1,2024 1 − 𝑌𝑗,1,2023 1 ] − 𝐸[𝑌𝑗,0,20...

2023
[36]

= 𝐸[𝑌𝑗,1,2024 1 − 𝑌𝑗,1,2024 0 ] − 𝐸[𝑌𝑗,1,2023 1 − 𝑌𝑗,1,2023 0 ] = 𝜃𝑗,2024 − 𝜃𝑗,2023 Placebo Test and Bias Analysis If Assumption (3) is replaced with the equivalent parallel trends assumptions for 𝑡 < 2023, then under the three assumptions, 𝜃̂𝑗,𝑡 would be an unbiased estimator...

2024
[37]

If the pre-2023 estimates are scattered around 0 and relatively small in magnitude, that suggests that Assumption 3 is reasonable

So we 32 calculate these estimators to use as placebo tests of this assumption. If the pre-2023 estimates are scattered around 0 and relatively small in magnitude, that suggests that Assumption 3 is reasonable. If a longer time series were available, a p-value could be calcula...

2022
[38]

For each player, we also record a set of covariates labelled 𝑋𝑘,1,𝑛 and 𝑋𝑘,0,𝑚 for target and control players, respectively, with 𝑘 = 1, … , 𝐾 indexing the covariates. For the main analysis, the covariates are: (1) Player age for 2022 season; (2) Outcome 𝑗 value; (3) Plate app...

2021
[39]

and can be found at the GitHub repository (https://bit.ly/QE-Baseball). For the main analysis, the estimate for outcome 𝑗 for target player 𝑛 in 2023 is given by: 𝜃̂𝑗,𝑛,2023 = 𝑌𝑗,1,𝑛,2023 − 𝑌̂𝑗,1,𝑛,2023 0 , where 𝑌̂𝑗,1,𝑛,2023 0 = ∑ 𝑤𝑗,𝑛,𝑚 58 𝑚=1 𝑌𝑗,0,𝑚,2023 is the synthetic co...

2023
[40]

weights that minimize the mean squared prediction error of the corresponding SC results in some set of the pre-intervention periods. That is, if we write 𝑤𝑗,𝑛,𝑚(𝒗) as the SC weight for outcome 𝑗, target player 𝑛, donor player 𝑚, if the vector 𝒗 of importance weights is used fo...

2015
[41]

Alternate specifications could validate on a smaller set of time periods and/or exclude individual seasons’ values of the outcome

where the target player had at least 250 PAs. Alternate specifications could validate on a smaller set of time periods and/or exclude individual seasons’ values of the outcome. This might be appropriate if data with less variability (potentially including bat-tracking data) we...

2024
[42]

This could also be done to estimate 𝜃̂𝑗,𝑛,2024 alone by including all target and donor players with at least 250 PAs in 2024 regardless of their number of PAs in

Note that because of the slightly different donor pool, the estimate 𝜃̂𝑗,𝑛,2023 using this method may not exactly match that in the main analysis. This could also be done to estimate 𝜃̂𝑗,𝑛,2024 alone by including all target and donor players with at least 250 PAs in 2024 regar...

2023
[43]

Only one additional target player is included with this weaker restriction

This is done as an additional sensitivity analysis and the results included in the GitHub repository. Only one additional target player is included with this weaker restriction. Placebo Tests To estimate the null distribution of estimates, we compute the SCM estimator for each...

2010
[793]

Demand for offense: Designated hitters and MLB attendance,

Cardazzi, A., and Rodriguez, Z. (2024), “Demand for offense: Designated hitters and MLB attendance,” Applied Economics, 1–14. https://doi.org/10.1080/00036846.2024.2302929. Carleton, R. A. (2018a), The Shift: The Next Evolution in Baseball Thinking, Chicago, Illinois: Triumph ...

2024
[2020]

Let 𝜃𝑗,𝑖,𝑡 = 𝐸[𝑌𝑗,𝑖,𝑡 1 − 𝑌𝑗,𝑖,𝑡 0 ] be the effect (on the additive scale) of the infield shift ban on outcome 𝑗 for population 𝑖 in season 𝑡

Let 𝑌𝑗,𝑖,𝑡 0 and 𝑌𝑗,𝑖,𝑡 1 be the potential outcomes for outcome 𝑗, population 𝑖, season 𝑡, in the absence and presence, respectively, of the infield shift ban. Let 𝜃𝑗,𝑖,𝑡 = 𝐸[𝑌𝑗,𝑖,𝑡 1 − 𝑌𝑗,𝑖,𝑡 0 ] be the effect (on the additive scale) of the infield shift ban on outcome 𝑗 for ...

2023
[2022]

and vaccines (Kennedy-Shaffer 2024), among many others. More detailed discussions of the requirements and implementations of these 4 methods are available in textbooks and academic articles such as Cunningham (2021), Huntington-Klein (2022), Caniglia and Murray (2020), Abadie ...

2021
[2023]

The outcome, player’s age, PAs, hits, singles, home runs, base-on-balls percentage, and strikeout percentage are used as the covariates. All covariates except the outcome itself are used individually for 2021 and 2022 and averaged over all included pre-2020 seasons; the outcom...

2021
[2024]

and the National Basketball Association has changed player rest rules (Marks 2023). Major League Baseball (MLB) recently implemented several changes around team makeup, pitcher hitting, pitch timing, baserunning, and fielding play, largely designed to reduce game times and/or ...

2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

The Effects of Major League Baseball’s Ban on Infield Shifts: A Quasi- Experimental Analysis Lee Kennedy-Shaffer, PhD Department of Biostatistics, Yale School of Public Health, New Haven, CT, USA Correspondence: Lee.kennedy-shaffer@yale.edu Abstract From 2020 to 2023, Major League Baseball changed rules affecting team composition, player positioning, and ...

work page 2020

[2] [2]

Players with weights below 0.1% are excluded

Top-weighted donor players in the synthetic control fits for Corey Seager, for the three outcomes. Players with weights below 0.1% are excluded. Weight Ranking OBP OPS wOBA 1 Starling Marte 63% Trea Turner 60% Trea Turner 47% 2 Carlos Correa 37% Carlos Correa 20% Jean Segura 25% 3 - Jean Segura 10% Carlos Correa 21% 4 - Yan Gomes 6% José Abreu 8% 5 - Paul...

work page 2023

[3] [3]

Overall, for each of the three outcomes, most target players (over 75%) had an estimated increase. The placebo test estimates, as expected, had estimates greater than 0 near 50% of the time (although it was somewhat higher for OPS and wOBA, indicating possible systemic underestimation of the 2023 outcomes). The mean and median value of the estimates acros...

work page 2023

[4] [4]

exciting

and in business analyses of revenue and attendance effects of decisions (Brook 2022; Cardazzi and Rodriguez 2024; Cisyk 2020). A recent study also looked at the effect of MLB’s automatic baserunner rule in extra innings on game times using DID (Kennedy-Shaffer 2022). Another study used both DID and SCM to analyze the effect of the 1981 change in match res...

work page 2024

[5] [6]

See Appendix B and Abadie (2021) for further technical details of the synthetic control method; it is implemented using the tidysynth R package (Dunford 2023)

for the control players that minimize the squared distance from a weighted average of the covariates, where the covariate weights are chosen to best predict the outcome trajectory in the pre-intervention period. See Appendix B and Abadie (2021) for further technical details of the synthetic control method; it is implemented using the tidysynth R package (...

work page 2021

[6] [7]

The synthetic control is re-fit for these groups and estimated effects obtained for both 2023 and

This yields a subset of the above player list (27 target players and 42 control players). The synthetic control is re-fit for these groups and estimated effects obtained for both 2023 and

work page 2023

[7] [8]

Additional robustness checks are conducted by performing two similar analyses where zero or near-zero effects are expected. The first—an “in-unit” placebo analysis—conducts the same analysis for 2023 effects, using as the target players the 25 players who otherwise fit the criteria for inclusion with a shift rate in 2022 between 15 and 30%. We would expec...

work page 2023

[8] [9]

Both indicate a modest increase in the outcome for left-handed batters in 2023 compared to what would have been expected in the absence of a rule change, with both estimates (coincidentally) equal to 0.009, or 9 points. To situate this magnitude in terms of yearly fluctuations and assess the parallel trends assumption, Figure 1 plots the trends in the two...

work page 2023