Pith. sign in

REVIEW 2 major objections 5 minor 119 references

Balancing Interference and Correlation in Spatial Experimental Designs: A Causal Graph Cut Approach

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A graph-cut surrogate for the MSE of a doubly robust ATE estimator turns spatial experiment design into a spectral clustering problem.

desk verdict A genuinely useful graph-cut design for spatial experiments, with strong empirical support and a real but addressable gap between the theory (known outcome regressions) and the deployed algorithm (estimated regressions). read the letter →

arxiv 2505.20130 v3 pith:6IU5O2TD submitted 2025-05-26 cs.LG stat.COstat.ML

Jin Zhu , Jingyi Li , Hongyi Zhou , Yinan Lin , Zhenhua Lin , Chengchun Shi This is my paper
classification cs.LGstat.COstat.ML MSC 62K0562D2062H30
keywords spatialA/Btestingcluster-randomizeddesigninterferencecorrelationgraphcutdoublyrobustestimationaveragetreatmenteffectspectralclustering
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 claims that the optimal way to assign treatment clusters in a spatial A/B test—one where a policy in one region spills over to its neighbors and residuals are spatially correlated—is the solution to a graph-cut problem. It proposes a surrogate for the mean squared error of a doubly robust average-treatment-effect estimator that has two opposing terms: interference favors giving neighboring regions the same policy, while positive spatial correlation favors giving them different policies. The paper proves that in the two extreme cases the optimal design is the global design (all regions treated alike) and the individual design (each region randomized separately), and that the surrogate interpolates between them. If the claim is correct, experimenters can use classical spectral clustering to pick clusters that substantially lower estimation error, and the paper reports a 3.5-fold MSE reduction over existing designs in a city-scale ridesharing simulator.

What carries the argument

The load-bearing object is the surrogate function in Equation (2): for two clusters C1, C2 it is (8R/N) sum_{i in C1, i' in C2} W_{ii'} Sigma^+_{ii'} - (8/N) sum_{i in C1, i' in C2} Sigma_{ii'}. The first term is an upper bound on the first-order interference term I1 (valid when the covariance decays with distance), and the second term equals the between-cluster covariance whose maximization minimizes the within-cluster correlation term SC. Written as sum of weights omega_{ii'} = R W_{ii'} Sigma^+_{ii'} - Sigma_{ii'}, minimizing the surrogate is a minimum-cut problem on a graph whose vertices are regions; the paper solves it by spectral relaxation, taking the Fiedler eigenvector (or first eigenvector when the smallest eigenvalue is nonzero) of the Laplacian and running k-means, and searches over cluster counts m with mmax = $R^{{2/3}}$, selecting m by plugging each candidate design into the Theorem 1 MSE formula.

What would settle it

Simulate a spatial grid with a known nonlinear outcome regression gi and known covariance Sigma, then compare the CGC design (which estimates gi and Sigma) with OCGC (which knows Sigma) and with the true MSE-optimal cluster design computed by exhaustive search; if CGC's selected design consistently differs from the oracle or its empirical MSE does not approach OCGC's as the number of repeated experiments grows, the assumption that estimated regression functions can stand in for known ones is the failing point.

Watch

Extended reading notes

Core claim

The paper's central claim is that the MSE of the doubly robust estimator under a cluster-randomized design decomposes (Theorem 1) into a design-agnostic term, a within-cluster correlation term SC, and two boundary interference terms I1, I2; the design that minimizes this MSE is driven by two opposing forces. With weak non-neighbor correlation and positive neighbor covariance, minimizing SC+I1+I2 makes the global design optimal (Proposition 1). Under SUTVA and non-negative covariance, the interference terms vanish and minimizing SC makes the individual design optimal (Proposition 2). To get a tractable objective, the paper replaces I1 by an upper bound that is tight under decaying covariance (Propositions 3 and 4), omits I2 as higher-order, and obtains a quadratic surrogate (Equation (2), general form Equation (3)) whose weights blend boundary covariance with between-cluster covariance. The same paper reports that in an 85-region ridesharing simulator driven by real data, the resulting design attains roughly 3.5 times smaller estimator MSE than competing designs, and that in synthetic tests it closely tracks the oracle version that knows the covariance.

Load-bearing premise

The MSE decomposition and the optimality results assume the outcome regression functions gi are known; in the algorithm they are estimated from data, and the theory does not quantify how estimation error in gi or in the residual covariance propagates into the chosen design.

Editorial extensions

If this is right

  • With weak correlation among non-neighboring regions and positive neighbor covariance, the global design (all regions receiving the same policy) is MSE-optimal, so cluster boundaries should be avoided entirely.
  • When there is no interference and all residual covariances are non-negative, the individual design (independent randomization per region) is MSE-optimal.
  • For moderate interference and correlation, the optimal design lies between these extremes and is found by spectral graph cut, with the omitted second-order boundary term I2 treated as negligible.
  • Because the covariance enters the surrogate directly, the design adapts to the correlation structure rather than optimizing worst-case MSE, and the iterative algorithm can update the design as data accumulate.
  • In the real-data ridesharing simulator the CGC estimator's MSE is about 3.5 times smaller than the benchmark designs, and it tracks the oracle design in synthetic settings.

Reading between the lines

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

  • A scalar weight could be introduced in front of the interference upper-bound term to trace a continuous family of designs from individual to global, which may be useful when the relative cost of boundary bias versus correlation variance is a policy choice rather than fixed by the estimated covariance.
  • The theory treats the regression functions gi as known; a natural stress test is to compare CGC with OCGC under misspecified gi, since random-forest estimation error in gi enters the residual covariance and could mis-calibrate the design.
  • The same surrogate applies to any experiment with an exposure graph and an estimated outcome covariance, such as marketplace or social-network experiments, if one can estimate the covariance from historical data; the paper only demonstrates this in single-experiment settings with a noisy proxy covariance.
  • When the covariance has negative entries, Proposition 2 no longer holds and the optimal design may mix clusters not captured by the current analysis; the sign structure of the covariance is a candidate driver of richer designs.
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.

Referee Report

2 major / 5 minor

Summary. This paper studies the design of cluster-randomized spatial experiments under neighborhood interference and spatially correlated residuals. The authors propose a surrogate objective for the MSE of a doubly robust ATE estimator, express it as a graph-cut problem, and develop a spectral clustering algorithm (CGC) with an iterative procedure that estimates outcome regressions and residual covariance from repeated experiments. Theoretical results decompose the MSE into design-agnostic, within-cluster correlation, and boundary interference terms; two propositions identify the global design and the individual design as optimal in limiting cases; Proposition 3 bounds an interference term by the surrogate. Simulations on four spatial layouts and three covariance structures and a ridesharing simulator show that CGC substantially reduces MSE relative to benchmarks, with a 3.5x improvement in the simulator and near-oracle performance in synthetic settings.

Significance. If the results hold, the paper offers a practical and computationally efficient alternative to existing design methods: it accommodates moderate-to-large interference, adapts to the covariance structure, and scales via graph cuts. The empirical portion is unusually thorough: four urban morphologies, three covariance families, an oracle comparison, a realistic city-scale simulator, and ablation-style comparisons (OCGC vs CGC, local vs pooled regression) that support the algorithm's iterative design. The theoretical analysis gives intuition and formal statements for limiting cases. The main caveats are that the theory is developed under known outcome regressions and known covariance, and that the omitted second-order interference term is not controlled; these gaps are acknowledged in the manuscript but not closed.

major comments (2)
  1. [Appendix B.1, Theorem 1; Algorithm 1 (Steps 3–4)] The MSE decomposition in Theorem 1 is derived under the assumption that the outcome regressions g_i are known: the proof in Appendix B.1 begins 'assuming functions gi are known,' and Propositions 1–4 inherit this assumption. Algorithm 1, however, estimates g_i by random forest and plugs the residuals into bSigma, and the model-selection step (Step 8) selects clusters based on bSigma. Double robustness of the ATE estimator does not protect bSigma: a biased bg contributes Cov(g_i - bg_i, g_i' - bg_i') and treatment-dependent terms to E[bSigma_ii'], which can shift the graph-cut weights and the selected number of clusters. No theorem bounds ||bSigma - Sigma|| or shows that the design learned from bSigma is near-oracle; the OCGC-vs-CGC comparison is reassuring but is limited to the synthetic family tested. I would ask the authors either to provide finite-sample or asymptotic guarantees for the plug-in design, or to state clearly that the theoretical optimality results apply to the oracle version and that the deployed algorithm is a heuristic whose guarantees are empirical.
  2. [Section 3.2 (Eq. (2)) and Section D] The surrogate (2) optimizes an upper bound of SC + I1 only; the second-order boundary term I2 from Theorem 1 is omitted. The text justifies this by arguing that 'only a small fraction of the regions lie at the boundary,' and Section D acknowledges that I2 'may be non-negligible.' No formal condition on the graph, the covariance decay, or the number of clusters ensures that I2 is small relative to I1; Proposition 4 bounds the surrogate's interference term by I1, not by I2. Consequently, unless I2 is provably negligible, minimizing (2) need not minimize the MSE of the DR estimator. I recommend adding a quantitative bound on I2 (for instance under a decaying-covariance condition with boundary-size control) or explicitly reframing the method as a heuristic approximation of the MSE with the I2 caveat prominently stated in the main text.
minor comments (5)
  1. [Section 4, real-data simulator paragraph] The sentence 'the global design is significantly less effective than the individual design, further supporting Proposition 1' appears to refer to Proposition 2, since strong spatial correlation is the stated reason; please correct the reference.
  2. [Appendix C.3 and Section 4] The main text states 'numerical results above were obtained without cross-fitting,' whereas Appendix C.3 says 'we adopt the data-splitting and cross-fitting method.' Please clarify which estimator is used in the reported experiments and in Algorithm 1/2.
  3. [Algorithm 1, Step 4] The covariance estimate is written as a sum over t = 1, ..., B, but Step 3 uses datasets D(1), ..., D(l); the sum should be over the pooled sample of size lB (or the indices should be defined accordingly).
  4. [Section 3.3] The maximum number of clusters is described as 'set to R2/3'; this should be R^(2/3). The word 'sightly' should be 'slightly'.
  5. [Appendix B.2, Proposition 1] The proof replaces 'delta sufficiently small' by setting non-neighboring covariances exactly to zero; a continuity argument or an explicit threshold on delta is needed to establish the stated small-delta result.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central derivation is a new surrogate-to-graph-cut argument; the overlapping-author citation supplies a prior MSE decomposition but does not presuppose the paper's conclusions.

full rationale

The paper's derivation chain is self-contained in the relevant sense. Theorem 1's MSE decomposition is imported from Theorem 2 of Yang et al. (2024), a preprint with overlapping authorship, but that citation is a dependency rather than a circle: the quoted decomposition applies to the DR estimator under known gi and neighborhood interference, and it does not assume the surrogate (2), the graph-cut optimization, Propositions 1-4, or any experimental outcome. The paper's actual contribution—the surrogate (2), its upper-bound justification (Propositions 3-4), the graph-cut relaxation, and the iterative bSigma plug-in—is derived from, not assumed by, that decomposition. Propositions 1 and 2 follow algebraically from Theorem 1 (e.g., under SUTVA, SC = (4/N)[sum Sigma - between-cluster Sigma], so nonnegativity implies the individual design maximizes the between-cluster term). Algorithm 1 optimizes the surrogate and selects cluster count by the Theorem-1 MSE formula with estimated bSigma; this is a plug-in estimation gap (a correctness risk, since Theorem 1 assumes known gi), not a circular reduction. The empirical evaluation is external: synthetic MSEs are Monte Carlo squared errors against a known ATE, and the ridesharing simulator uses a separate Monte Carlo approximation of the ATE, so the results are not forced by fitting the claimed objective. The acknowledged omission of I2 (Appendix D) is an approximation caveat, not circularity.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The method does not introduce new physical or conceptual entities. It relies on standard modeling assumptions about interference and covariance, an external theorem for the estimator's MSE, and a heuristic omission of the second-order boundary term. The free parameters are the maximum cluster count and the fixed treatment probability.

free parameters (2)
  • m_max (maximum number of clusters) = R^(2/3)
    The search over cluster counts is truncated at R^(2/3), following a recommendation from Leung (2022). This is a heuristic choice not derived in the paper.
  • p (treatment assignment probability) = 0.5
    Fixed to ensure a balanced design, asserted optimal via Yang et al. (2024). This affects the MSE scaling but is not fitted to data here.
assumptions (6)
  • domain assumption Outcome model Yi = gi(A,O) + ei with independent, mean-zero errors ei.
    The entire MSE analysis and covariance estimation rely on additive errors independent of actions and covariates, stated at the start of Section 2.
  • domain assumption Assumption 1: Neighborhood interference, gi depends only on treatments within the neighbor set Ni.
    Restricts interference to neighbors and is used to define the DR estimator and the MSE decomposition.
  • domain assumption Assumption 2: Decaying covariance, neighbor covariance is at least as large as non-neighbor covariance for disjoint regions.
    Used in Proposition 3 to prove that the first surrogate term upper-bounds I1.
  • standard math Theorem 2 of Yang et al. (2024) giving the MSE of the DR estimator with known gi.
    The proof of Theorem 1 starts from this external theorem; the present paper does not reprove it.
  • ad hoc to paper Known gi for the theoretical MSE; estimation error of gi is ignored.
    The proof of Theorem 1 assumes gi are known, while the algorithm estimates gi from data. This gap is not addressed theoretically.
  • ad hoc to paper The omitted I2 term is negligible relative to I1.
    Section 3.2 asserts that only a small fraction of regions lie at boundaries, but no theorem bounds the ratio I2/I1; the paper later acknowledges this limitation in the Discussion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Balancing Interference and Correlation in Spatial Experimental Designs: A Causal Graph Cut Approach." pith.science (2026). https://pith.science/paper/6IU5O2TD

@misc{pith2026250520130,
  author       = {Pith},
  title        = {Pith review of: Balancing Interference and Correlation in Spatial Experimental Designs: A Causal Graph Cut Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6IU5O2TD}},
  note         = {Machine review of arXiv:2505.20130}
}
read the original abstract

This paper focuses on the design of spatial experiments to optimize the amount of information derived from the experimental data and enhance the accuracy of the resulting causal effect estimator. We propose a surrogate function for the mean squared error (MSE) of the estimator, which facilitates the use of classical graph cut algorithms to learn the optimal design. Our proposal offers three key advances: (1) it accommodates moderate to large spatial interference effects; (2) it adapts to different spatial covariance functions; (3) it is computationally efficient. Theoretical results and numerical experiments based on synthetic environments and a dispatch simulator that models a city-scale ridesharing market, further validate the effectiveness of our design. A python implementation of our method is available at https://github.com/Mamba413/CausalGraphCut.

Figures

Figures reproduced from arXiv: 2505.20130 by the authors.

Figure 1
Figure 1. Illustrations of the optimal cluster-randomized designs in three experiments, where different colors represent different clusters. (a) With weakly correlated residuals, the optimal design simplifies to the global design. (b) Without interference, the op￾timal design reduces to the individual design. (c) With moderate to large levels of interference and correlation, the optimal design falls between the two extremes, … view at source ↗
Figure 2
Figure 2. An illustration of each term in the MSE decomposition from Theorem 1 with m = 2 clusters distinguished by different colors. (a) SC captures the residual covariance between any i, i′ within the same cluster. (b) I1 captures the residual covariance between any i, i′ that belong to different clusters, with one region lying on the boundary between the two clusters. (c) I2 captures residual covariance between any i, i′ f… view at source ↗
Figure 4
Figure 4. (a) Spatial layout of 4 regions to be partitioned. (b) Each region is represented as a vertex, with edges determined by the weight ω. (c) The cut highlighted in red minimizes the total loss weight. (d) The spatial layout obtained after applying the 2-cut, illustrating how the regions are partitioned based on the graph cut. Proposition 3. Under Assumption 2, the first term in (2) upper bounds I1. Meanwhile, when the … view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: (a) Four types of spatial grids in synthetic environments: square (top-left), rectangle (top-right), circle (bottom-left), fans (bottom-right). (b) MSEs with different numbers of repetitions N in the real-data-based simulator. The shaded area visualizes the confidence …
Figure 6
Figure 6. Figure 6: The plot of ρ versus MSE, with each panel representing a different spatial arrangement setting. Real-data-based simulator. We develop a simulator using a five-day historical dataset from a ridesharing company to evaluate the proposed approach. This simulator repli￾cate…
Figure 8
Figure 8. Figure 8: The plot of number of repeated experiments versus MSE on square spatial setting. Each panel corresponds to a spatial corre￾lation setting. For clearer representation, we adopt a logarithmic scale for the values on the y-axis. Constant Truncated Exponential Square 20 40…
Figure 9
Figure 9. Figure 9: We use the logarithmic scale on values of y-axis for a clearer presentation. From [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

119 extracted references · 52 canonical work pages

  1. [1]

    Multi-armed bandits with network interference

    Agarwal, A., Agarwal, A., Masoero, L., and Whitehouse, J. Multi-armed bandits with network interference. arXiv preprint arXiv:2405.18621, 2024

    arXiv 2024

  2. [2]

    Aronow, P. M. and Samii, C. Estimating average causal effects under general interference, with application to a social network experiment. The Annals of Applied Statistics, 11 0 (4), December 2017. ISSN 1932-6157. doi:10.1214/16-aoas1005. URL http://dx.doi.org/10.1214/16-AOAS1005

    doi:10.1214/16-aoas1005 2017

  3. [3]

    A., McIntosh, C., and Özler, B

    Baird, S., Bohren, J. A., McIntosh, C., and Özler, B. Optimal design of experiments in the presence of interference. The Review of Economics and Statistics, 100 0 (5): 0 844--860, 12 2018. ISSN 0034-6535. doi:10.1162/rest_a_00716. URL https://doi.org/10.1162/rest\_a\_00716

    doi:10.1162/rest_a_00716 2018

  4. [4]

    W., Masoero, L., McQueen, J., Richardson, T., and Rosen, I

    Bajari, P., Burdick, B., Imbens, G. W., Masoero, L., McQueen, J., Richardson, T., and Rosen, I. M. Multiple randomization designs. arXiv preprint arXiv:2112.13495, 2021

    arXiv 2021

  5. [5]

    Causal inference under interference and network uncertainty

    Bhattacharya, R., Malinsky, D., and Shpitser, I. Causal inference under interference and network uncertainty. In Uncertainty in Artificial Intelligence, pp.\ 1028--1038. PMLR, 2020

    2020

  6. [6]

    Off-policy evaluation in doubly inhomogeneous environments

    Bian, Z., Shi, C., Qi, Z., and Wang, L. Off-policy evaluation in doubly inhomogeneous environments. Journal of the American Statistical Association, pp.\ 1--27, 2024

    2024

  7. [7]

    V., Chades, I., and Dezfouli, A

    Blau, T., Bonilla, E. V., Chades, I., and Dezfouli, A. Optimizing sequential experimental design with deep reinforcement learning. In International Conference on Machine Learning, pp.\ 2107--2128. PMLR, 2022

    2022

  8. [8]

    Design and analysis of switchback experiments

    Bojinov, I., Simchi-Levi, D., and Zhao, J. Design and analysis of switchback experiments. Management Science, 69 0 (7): 0 3759--3777, 2023

    2023

Show all 119 references
  1. [9]

    Optimized covariance design for a/b test on social network under interference

    Chen, Q., Li, B., Deng, L., and Wang, Y. Optimized covariance design for a/b test on social network under interference. Advances in Neural Information Processing Systems, 36, 2024

    2024

  2. [10]

    and Qi, Z

    Chen, X. and Qi, Z. On well-posedness and minimax optimal rates of nonparametric q-function estimation in off-policy evaluation. In International Conference on Machine Learning, pp.\ 3558--3582. PMLR, 2022

    2022

  3. [11]

    Double/debiased machine learning for treatment and structural parameters

    Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., Newey, W., and Robins, J. Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal, 21 0 (1): 0 C1--C68, 01 2018. ISSN 1368-4221. doi:10.1111/ectj.12097. URL https:...

    2018 doi

  4. [12]

    K., and Singh, R

    Chernozhukov, V., Newey, W. K., and Singh, R. Debiased machine learning of global and local parameters using regularized riesz representers. The Econometrics Journal, 25 0 (3): 0 576--601, 2022

    2022

  5. [13]

    Do labor market policies have displacement effects? evidence from a clustered randomized experiment

    Cr \'e pon, B., Duflo, E., Gurgand, M., Rathelot, R., and Zamora, P. Do labor market policies have displacement effects? evidence from a clustered randomized experiment. The quarterly journal of economics, 128 0 (2): 0 531--580, 2013

    2013

  6. [14]

    Statistics for spatial data

    Cressie, N. Statistics for spatial data. John Wiley & Sons, 2015

    2015

  7. [15]

    Coindice: Off-policy confidence interval estimation

    Dai, B., Nachum, O., Chow, Y., Li, L., Szepesv \'a ri, C., and Schuurmans, D. Coindice: Off-policy confidence interval estimation. Advances in Neural Information Processing Systems, 33: 0 9398--9411, 2020

    2020

  8. [16]

    Causal deepsets for off-policy evaluation under spatial or spatio-temporal interferences

    Dai, R., Wang, J., Zhou, F., Luo, S., Qin, Z., Shi, C., and Zhu, H. Causal deepsets for off-policy evaluation under spatial or spatio-temporal interferences. arXiv preprint arXiv:2407.17910, 2024

    2024 arXiv

  9. [17]

    Doubly robust policy evaluation and learning

    Dud \' k, M., Langford, J., and Li, L. Doubly robust policy evaluation and learning. In Proceedings of the 28th International Conference on International Conference on Machine Learning, pp.\ 1097--1104, 2011

    2011

  10. [18]

    Eichhorn, M., Khan, S., Ugander, J., and Yu, C. L. Low-order outcomes and clustered designs: combining design and analysis for causal inference under network interference. arXiv preprint arXiv:2405.07979, 2024

    2024

  11. [19]

    Markovian interference in experiments

    Farias, V., Li, A., Peng, T., and Zheng, A. Markovian interference in experiments. Advances in Neural Information Processing Systems, 35: 0 535--549, 2022

    2022

  12. [20]

    and Zheleva, E

    Fatemi, Z. and Zheleva, E. Minimizing interference and selection bias in network experiment design. In Proceedings of the International AAAI Conference on Web and Social Media, volume 14, pp.\ 176--186, 2020

    2020

  13. [21]

    Accountable off-policy evaluation with kernel bellman statistics

    Feng, Y., Ren, T., Tang, Z., and Liu, Q. Accountable off-policy evaluation with kernel bellman statistics. In International Conference on Machine Learning, pp.\ 3102--3111. PMLR, 2020

    2020

  14. [22]

    Algebraic connectivity of graphs

    Fiedler, M. Algebraic connectivity of graphs. Czechoslovak mathematical journal, 23 0 (2): 0 298--305, 1973

    1973

  15. [23]

    Laplacian of graphs and algebraic connectivity

    Fiedler, M. Laplacian of graphs and algebraic connectivity. Banach Center Publications, 1 0 (25): 0 57--70, 1989

    1989

  16. [24]

    Best of three worlds: Adaptive experimentation for digital marketing in practice

    Fiez, T., Nassif, H., Chen, Y.-C., Gamez, S., and Jain, L. Best of three worlds: Adaptive experimentation for digital marketing in practice. In Proceedings of the ACM on Web Conference 2024, pp.\ 3586--3597, 2024

    2024

  17. [25]

    A., Fisher, R

    Fisher, R. A., Fisher, R. A., Genetiker, S., Fisher, R. A., Genetician, S., Britain, G., Fisher, R. A., and G \'e n \'e ticien, S. The design of experiments, volume 21. Springer, 1966

    1966

  18. [26]

    M., and Mealli, F

    Forastiere, L., Airoldi, E. M., and Mealli, F. Identification and estimation of treatment and interference effects in observational studies on networks. Journal of the American Statistical Association, 116 0 (534): 0 901--918, 2021

    2021

  19. [27]

    R., Malik, I., and Rainforth, T

    Foster, A., Ivanova, D. R., Malik, I., and Rainforth, T. Deep adaptive design: Amortizing sequential bayesian experimental design. In International Conference on Machine Learning, pp.\ 3384--3395. PMLR, 2021

    2021

  20. [28]

    and Ding, P

    Gao, M. and Ding, P. Causal inference in network experiments: regression-based analysis and design-based properties. arXiv preprint arXiv:2309.07476, 2023

    2023 arXiv

  21. [29]

    and Hochbaum, D

    Goldschmidt, O. and Hochbaum, D. S. A polynomial algorithm for the k-cut problem for fixed k. Mathematics of operations research, 19 0 (1): 0 24--37, 1994

    1994

  22. [30]

    Network a/b testing: From sampling to estimation

    Gui, H., Xu, Y., Bhasin, A., and Han, J. Network a/b testing: From sampling to estimation. In Proceedings of the 24th International Conference on World Wide Web, pp.\ 399--409, 2015

    2015

  23. [31]

    and Kahng, A

    Hagen, L. and Kahng, A. B. New spectral methods for ratio cut partitioning and clustering. IEEE transactions on computer-aided design of integrated circuits and systems, 11 0 (9): 0 1074--1085, 1992

    1992

  24. [32]

    On the role of the propensity score in efficient semiparametric estimation of average treatment effects

    Hahn, J. On the role of the propensity score in efficient semiparametric estimation of average treatment effects. Econometrica, pp.\ 315--331, 1998

    1998

  25. [33]

    Halloran, M. E. and Struchiner, C. J. Causal inference in infectious diseases. Epidemiology, 6: 0 142--151, 1995

    1995

  26. [34]

    P., Thomas, P

    Hanna, J. P., Thomas, P. S., Stone, P., and Niekum, S. Data-efficient policy evaluation through behavior policy search. In International Conference on Machine Learning, pp.\ 1394--1403. PMLR, 2017

    2017

  27. [35]

    Bootstrapping fitted q-evaluation for off-policy inference

    Hao, B., Ji, X., Duan, Y., Lu, H., Szepesv \'a ri, C., and Wang, M. Bootstrapping fitted q-evaluation for off-policy inference. In International Conference on Machine Learning, pp.\ 4074--4084. PMLR, 2021

    2021

  28. [36]

    A., and Zhang, P

    Harshaw, C., S \"a vje, F., Spielman, D. A., and Zhang, P. Balancing covariates in randomized experiments with the gram--schmidt walk design. Journal of the American Statistical Association, 119 0 (548): 0 2934--2946, 2024

    2024

  29. [37]

    J., Ichimura, H., and Todd, P

    Heckman, J. J., Ichimura, H., and Todd, P. Matching as an econometric evaluation estimator. The review of economic studies, 65 0 (2): 0 261--294, 1998

    1998

  30. [38]

    and Wager, S

    Hu, Y. and Wager, S. Switchback experiments under geometric mixing. arXiv preprint arXiv:2209.00197, 2022

    2022

  31. [39]

    and Wager, S

    Hu, Y. and Wager, S. Off-policy evaluation in partially observed markov decision processes under sequential ignorability. The Annals of Statistics, 51 0 (4): 0 1561--1585, 2023

    2023

  32. [40]

    Average direct and indirect causal effects under interference

    Hu, Y., Li, S., and Wager, S. Average direct and indirect causal effects under interference. Biometrika, 109 0 (4): 0 1165--1172, 2022

    2022

  33. [41]

    Hudgens, M. G. and Halloran, M. E. Toward causal inference with interference. Journal of the American Statistical Association, 103 0 (482): 0 832--842, 2008

    2008

  34. [42]

    Imbens, G. W. and Rubin, D. B. Causal inference in statistics, social, and biomedical sciences. Cambridge university press, 2015

    2015

  35. [43]

    Jia, S., Kallus, N., and Yu, C. L. Clustered switchback experiments: Near-optimal rates under spatiotemporal interference. arXiv preprint arXiv:2312.15574, 2023

    2023 arXiv

  36. [44]

    Multi-armed bandits with interference

    Jia, S., Frazier, P., and Kallus, N. Multi-armed bandits with interference. arXiv preprint arXiv:2402.01845, 2024

    2024 arXiv

  37. [45]

    and Li, L

    Jiang, N. and Li, L. Doubly robust off-policy value evaluation for reinforcement learning. In International conference on machine learning, pp.\ 652--661. PMLR, 2016

    2016

  38. [46]

    and Wang, H

    Jiang, Y. and Wang, H. Causal inference under network interference using a mixture of randomized experiments. arXiv preprint arXiv:2309.00141, 2023

    2023 arXiv

  39. [47]

    Causal inference in social platforms under approximate interference networks

    Jiang, Y., Deng, L., Wang, Y., and Wang, H. Causal inference in social platforms under approximate interference networks. arXiv preprint arXiv:2408.04441, 2024

    2024 arXiv

  40. [48]

    To appear in ACM Economics and Computation 2020

    Johari, R., Li, H., and Weintraub, G. Experimental design in two-sided platforms: An analysis of bias. arXiv preprint arXiv:2002.05670, 2020. "To appear in ACM Economics and Computation 2020"

    2002 arXiv

  41. [49]

    Always valid inference: Continuous monitoring of a/b tests

    Johari, R., Koomen, P., Pekelis, L., and Walsh, D. Always valid inference: Continuous monitoring of a/b tests. Operations Research, 70 0 (3): 0 1806--1821, 2022

    2022

  42. [50]

    Optimal a priori balance in the design of controlled experiments

    Kallus, N. Optimal a priori balance in the design of controlled experiments. Journal of the Royal Statistical Society Series B: Statistical Methodology, 80 0 (1): 0 85--112, 2018

    2018

  43. [51]

    and Uehara, M

    Kallus, N. and Uehara, M. Efficiently breaking the curse of horizon in off-policy evaluation with double reinforcement learning. Operations Research, 70 0 (6): 0 3282--3302, 2022

    2022

  44. [52]

    Network experimentation at scale

    Karrer, B., Shi, L., Bhole, M., Goldman, M., Palmer, T., Gelman, C., Konutgan, M., and Sun, F. Network experimentation at scale. In Proceedings of the 27th acm sigkdd conference on knowledge discovery & data mining, pp.\ 3106--3116, 2021

    2021

  45. [53]

    Active adaptive experimental design for treatment effect estimation with covariate choices

    Kato, M., Oga, A., Komatsubara, W., and Inokuchi, R. Active adaptive experimental design for treatment effect estimation with covariate choices. arXiv preprint arXiv:2403.03589, 2024

    2024 arXiv

  46. [54]

    The handbook of urban morphology

    Kropf, K. The handbook of urban morphology. John Wiley & Sons, 2017

    2017

  47. [55]

    Confident off-policy evaluation and selection through self-normalized importance weighting

    Kuzborskij, I., Vernade, C., Gyorgy, A., and Szepesv \'a ri, C. Confident off-policy evaluation and selection through self-normalized importance weighting. In International Conference on Artificial Intelligence and Statistics, pp.\ 640--648. PMLR, 2021

    2021

  48. [56]

    Batch policy learning under constraints

    Le, H., Voloshin, C., and Yue, Y. Batch policy learning under constraints. In International Conference on Machine Learning, pp.\ 3703--3712. PMLR, 2019

    2019

  49. [57]

    Leung, M. P. Rate-optimal cluster-randomized designs for spatial interference. The Annals of Statistics, 50 0 (5): 0 3064--3087, 2022

    2022

  50. [58]

    Leung, M. P. and Loupos, P. Graph neural networks for causal inference under network confounding. arXiv preprint arXiv:2211.07823, 2022

    2022

  51. [59]

    Li, H., Zhao, G., Johari, R., and Weintraub, G. Y. Interference, bias, and variance in two-sided marketplace experimentation: Guidance for platforms. In Proceedings of the ACM Web Conference 2022, pp.\ 182--192, 2022

    2022

  52. [60]

    Optimal treatment allocation for efficient policy evaluation in sequential decision making

    Li, T., Shi, C., Wang, J., Zhou, F., et al. Optimal treatment allocation for efficient policy evaluation in sequential decision making. Advances in Neural Information Processing Systems, 36, 2023

    2023

  53. [61]

    Evaluating dynamic conditional quantile treatment effects with applications in ridesharing

    Li, T., Shi, C., Lu, Z., Li, Y., and Zhu, H. Evaluating dynamic conditional quantile treatment effects with applications in ridesharing. Journal of the American Statistical Association, pp.\ 1--15, 2024

    2024

  54. [62]

    Liao, P., Qi, Z., Wan, R., Klasnja, P., and Murphy, S. A. Batch policy learning in average reward markov decision processes. Annals of statistics, 50 0 (6): 0 3364, 2022

    2022

  55. [63]

    G., and Becker-Dreps, S

    Liu, L., Hudgens, M. G., and Becker-Dreps, S. On inverse probability-weighted estimators in the presence of interference. Biometrika, 103 0 (4): 0 829--842, 2016

    2016

  56. [64]

    G., Saul, B., Clemens, J

    Liu, L., Hudgens, M. G., Saul, B., Clemens, J. D., Ali, M., and Emch, M. E. Doubly robust estimation in observational studies with partial interference. Stat, 8 0 (1): 0 e214, 2019

    2019

  57. [65]

    Breaking the curse of horizon: Infinite-horizon off-policy estimation

    Liu, Q., Li, L., Tang, Z., and Zhou, D. Breaking the curse of horizon: Infinite-horizon off-policy estimation. Advances in Neural Information Processing Systems, 31, 2018

    2018

  58. [66]

    and Zhang, S

    Liu, S. and Zhang, S. Efficient policy evaluation with offline data informed behavior policy design. In International Conference on Machine Learning, pp.\ 32345--32368. PMLR, 2024

    2024

  59. [67]

    Cluster-adaptive network a/b testing: From randomization to estimation

    Liu, Y., Zhou, Y., Li, P., and Hu, F. Cluster-adaptive network a/b testing: From randomization to estimation. Journal of Machine Learning Research, 25 0 (170): 0 1--48, 2024

    2024

  60. [68]

    J., Laber, E

    Luckett, D. J., Laber, E. B., Kahkoska, A. R., Maahs, D. M., Mayer-Davis, E., and Kosorok, M. R. Estimating dynamic treatment regimes in mobile health using v-learning. Journal of the American Statistical Association, 115 0 (530): 0 692--706, 2020

    2020

  61. [69]

    Policy evaluation for temporal and/or spatial dependent experiments

    Luo, S., Yang, Y., Shi, C., Yao, F., Ye, J., and Zhu, H. Policy evaluation for temporal and/or spatial dependent experiments. Journal of the Royal Statistical Society Series B: Statistical Methodology, pp.\ qkad136, 2024

    2024

  62. [70]

    and Tresp, V

    Ma, Y. and Tresp, V. Causal inference under networked interference and intervention policy enhancement. In International Conference on Artificial Intelligence and Statistics, pp.\ 3700--3708. PMLR, 2021

    2021

  63. [71]

    Treatment effects in market equilibrium

    Munro, E., Wager, S., and Xu, K. Treatment effects in market equilibrium. arXiv preprint arXiv:2109.11647, 2021

    2021 arXiv

  64. [72]

    Design of panel experiments with spatial and temporal interference

    Ni, T., Bojinov, I., and Zhao, J. Design of panel experiments with spatial and temporal interference. Available at SSRN 4466598, 2023

    2023

  65. [73]

    G., Halloran, M

    Perez-Heydrich, C., Hudgens, M. G., Halloran, M. E., Clemens, J. D., Ali, M., and Emch, M. E. Assessing effects of cholera vaccination in the presence of interference. Biometrics, 70 0 (3): 0 731--741, 2014

    2014

  66. [74]

    A graph-theoretic approach to randomization tests of causal effects under general interference

    Puelz, D., Basse, G., Feller, A., and Toulis, P. A graph-theoretic approach to randomization tests of causal effects under general interference. Journal of the Royal Statistical Society Series B: Statistical Methodology, 84 0 (1): 0 174--204, 2022

    2022

  67. [75]

    J., Yang, S., Guan, Y., Giffin, A

    Reich, B. J., Yang, S., Guan, Y., Giffin, A. B., Miller, M. J., and Rappold, A. A review of spatial causal inference methods for environmental and epidemiological applications. International Statistical Review, 89 0 (3): 0 605--634, 2021

    2021

  68. [76]

    Randomized experimental design via geographic clustering

    Rolnick, D., Aydin, K., Pouget-Abadie, J., Kamali, S., Mirrokni, V., and Najmi, A. Randomized experimental design via geographic clustering. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp.\ 2745--2753, 2019

    2019

  69. [77]

    Saveski, M., Pouget-Abadie, J., Saint-Jacques, G., Duan, W., Ghosh, S., Xu, Y., and Airoldi, E. M. Detecting network effects: Randomizing over randomized experiments. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, pp.\ 10...

    2017

  70. [78]

    Deeply-debiased off-policy interval estimation

    Shi, C., Wan, R., Chernozhukov, V., and Song, R. Deeply-debiased off-policy interval estimation. In International conference on machine learning, pp.\ 9580--9591. PMLR, 2021

    2021

  71. [79]

    Statistical inference of the value function for reinforcement learning in infinite-horizon settings

    Shi, C., Zhang, S., Lu, W., and Song, R. Statistical inference of the value function for reinforcement learning in infinite-horizon settings. Journal of the Royal Statistical Society Series B: Statistical Methodology, 84 0 (3): 0 765--793, 2022

    2022

  72. [80]

    A multiagent reinforcement learning framework for off-policy evaluation in two-sided markets

    Shi, C., Wan, R., Song, G., Luo, S., Zhu, H., and Song, R. A multiagent reinforcement learning framework for off-policy evaluation in two-sided markets. The Annals of Applied Statistics, 17 0 (4): 0 2701--2722, 2023 a

    2023

  73. [81]

    Dynamic causal effects evaluation in a/b testing with a reinforcement learning framework

    Shi, C., Wang, X., Luo, S., Zhu, H., Ye, J., and Song, R. Dynamic causal effects evaluation in a/b testing with a reinforcement learning framework. Journal of the American Statistical Association, 118 0 (543): 0 2059--2071, 2023 b

    2023

  74. [82]

    Off-policy confidence interval estimation with confounded markov decision process

    Shi, C., Zhu, J., Shen, Y., Luo, S., Zhu, H., and Song, R. Off-policy confidence interval estimation with confounded markov decision process. Journal of the American Statistical Association, 119 0 (545): 0 273--284, 2024

    2024

  75. [83]

    Sobel, M. E. What do randomized studies of housing mobility demonstrate? causal inference in the face of interference. Journal of the American Statistical Association, 101 0 (476): 0 1398--1407, 2006

    2006

  76. [84]

    and Papadogeorgou, G

    Song, Z. and Papadogeorgou, G. Bipartite causal inference with interference, time series data, and a random network. arXiv preprint arXiv:2404.04775, 2024

    2024 arXiv

  77. [85]

    and Wagner, F

    Stoer, M. and Wagner, F. A simple min-cut algorithm. Journal of the ACM (JACM), 44 0 (4): 0 585--591, 1997

    1997

  78. [86]

    Arma-design: Optimal treatment allocation strategies for a/b testing in partially observable time series experiments

    Sun, K., Kong, L., Zhu, H., and Shi, C. Arma-design: Optimal treatment allocation strategies for a/b testing in partially observable time series experiments. arXiv preprint arXiv:2408.05342, 2024

    2024 arXiv

  79. [87]

    and Joachims, T

    Swaminathan, A. and Joachims, T. The self-normalized estimator for counterfactual learning. advances in neural information processing systems, 28, 2015

    2015

  80. [88]

    Bounded, efficient and doubly robust estimation with inverse weighting

    Tan, Z. Bounded, efficient and doubly robust estimation with inverse weighting. Biometrika, 97 0 (3): 0 661--682, 2010

    2010

  81. [89]

    A deep value-network based approach for multi-driver order dispatching

    Tang, X., Qin, Z., Zhang, F., Wang, Z., Xu, Z., Ma, Y., Zhu, H., and Ye, J. A deep value-network based approach for multi-driver order dispatching. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pp.\ 1780--1790, 2019

    2019

  82. [90]

    Tchetgen, E. J. T. and VanderWeele, T. J. On causal inference in the presence of interference. Statistical methods in medical research, 21 0 (1): 0 55--75, 2012

    2012

  83. [91]

    and Brunskill, E

    Thomas, P. and Brunskill, E. Data-efficient off-policy policy evaluation for reinforcement learning. In International Conference on Machine Learning, pp.\ 2139--2148. PMLR, 2016

    2016

  84. [92]

    High-confidence off-policy evaluation

    Thomas, P., Theocharous, G., and Ghavamzadeh, M. High-confidence off-policy evaluation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 29, 2015

    2015

  85. [93]

    Graph cluster randomization: Network exposure to multiple universes

    Ugander, J., Karrer, B., Backstrom, L., and Kleinberg, J. Graph cluster randomization: Network exposure to multiple universes. In Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining, pp.\ 329--337, 2013

    2013

  86. [94]

    J., Vandenbroucke, J

    VanderWeele, T. J., Vandenbroucke, J. P., Tchetgen, E. J. T., and Robins, J. M. A mapping between interactions and interference: implications for vaccine trials. Epidemiology, 23 0 (2): 0 285--292, 2012

    2012

  87. [95]

    and Raudenbush, S

    Verbitsky-Savitz, N. and Raudenbush, S. W. Causal inference under interference in spatial settings: a case study evaluating community policing program in chicago. Epidemiologic Methods, 1 0 (1): 0 107--130, 2012

    2012

  88. [96]

    Causal clustering: design of cluster experiments under network interference

    Viviano, D., Lei, L., Imbens, G., Karrer, B., Schrijvers, O., and Shi, L. Causal clustering: design of cluster experiments under network interference. arXiv preprint arXiv:2310.14983, 2023

    2023 arXiv

  89. [97]

    and Xu, K

    Wager, S. and Xu, K. Experimenting in equilibrium. arXiv preprint arXiv:1903.02124, 2019

    1903 arXiv

  90. [98]

    Safe exploration for efficient policy evaluation and comparison

    Wan, R., Kveton, B., and Song, R. Safe exploration for efficient policy evaluation and comparison. In International Conference on Machine Learning, pp.\ 22491--22511. PMLR, 2022

    2022

  91. [99]

    Wang, J., Qi, Z., and Wong, R. K. Projected state-action balancing weights for offline reinforcement learning. The Annals of Statistics, 51 0 (4): 0 1639--1665, 2023

    2023

  92. [100]

    Graph neural networks: Theory for estimation with application on network heterogeneity

    Wang, Y., Gu, C., and Otsu, T. Graph neural networks: Theory for estimation with application on network heterogeneity. arXiv preprint arXiv:2401.16275, 2024

    2024 arXiv

  93. [101]

    Experimental designs for heteroskedastic variance

    Weltz, J., Fiez, T., Volfovsky, A., Laber, E., Mason, B., Jain, L., et al. Experimental designs for heteroskedastic variance. Advances in Neural Information Processing Systems, 36, 2023

    2023

  94. [102]

    Unraveling the interplay between carryover effects and reward autocorrelations in switchback experiments

    Wen, Q., Shi, C., Yang, Y., Tang, N., and Zhu, H. Unraveling the interplay between carryover effects and reward autocorrelations in switchback experiments. In International Conference on Machine Learning. PMLR, 2025

    2025

  95. [103]

    Xiong, R., Chin, A., and Taylor, S. J. Data-driven switchback experiments: Theoretical tradeoffs and empirical bayes designs. arXiv preprint arXiv:2406.06768, 2024

    2024 arXiv

  96. [104]

    An instrumental variable approach to confounded off-policy evaluation

    Xu, Y., Zhu, J., Shi, C., Luo, S., and Song, R. An instrumental variable approach to confounded off-policy evaluation. In International Conference on Machine Learning, pp.\ 38848--38880. PMLR, 2023

    2023

  97. [105]

    Linear contextual bandits with interference

    Xu, Y., Lu, W., and Song, R. Linear contextual bandits with interference. arXiv preprint arXiv:2409.15682, 2024

    2024 arXiv

  98. [106]

    Spatially randomized designs can enhance policy evaluation

    Yang, Y., Shi, C., Yao, F., Wang, S., and Zhu, H. Spatially randomized designs can enhance policy evaluation. arXiv preprint arXiv:2403.11400, 2024

    2024

  99. [107]

    Estimating treatment effects under recommender interference: A structured neural networks approach

    Zhan, R., Han, S., Hu, Y., and Jiang, Z. Estimating treatment effects under recommender interference: A structured neural networks approach. arXiv preprint arXiv:2406.14380, 2024

    2024

  100. [108]

    A., Laber, E

    Zhang, B., Tsiatis, A. A., Laber, E. B., and Davidian, M. A robust method for estimating optimal treatment regimes. Biometrics, 68 0 (4): 0 1010--1018, 2012

    2012

  101. [109]

    A., Laber, E

    Zhang, B., Tsiatis, A. A., Laber, E. B., and Davidian, M. Robust estimation of optimal dynamic treatment regimes for sequential treatment decisions. Biometrika, 100 0 (3): 0 10--1093, 2013

    2013

  102. [110]

    Spatial interference detection in treatment effect model

    Zhang, W., Yang, Y., and Yao, F. Spatial interference detection in treatment effect model. arXiv preprint arXiv:2409.04836, 2024

    2024 arXiv

  103. [111]

    and Wang, Z

    Zhang, Z. and Wang, Z. Online experimental design with estimation-regret trade-off under network interference. arXiv preprint arXiv:2412.03727, 2024

    2024 arXiv

  104. [112]

    Experimental design for causal inference through an optimization lens

    Zhao, J. Experimental design for causal inference through an optimization lens. In Tutorials in Operations Research: Smarter Decisions for a Better World, pp.\ 146--188. INFORMS, 2024 a

    2024

  105. [113]

    A simple formulation for causal clustering

    Zhao, J. A simple formulation for causal clustering. Available at SSRN 5008213, 2024 b

    2024

  106. [114]

    J., and Kosorok, M

    Zhao, Y., Zeng, D., Rush, A. J., and Kosorok, M. R. Estimating individualized treatment rules using outcome weighted learning. Journal of the American Statistical Association, 107 0 (499): 0 1106--1118, 2012

    2012

  107. [115]

    Robust on-policy sampling for data-efficient policy evaluation in reinforcement learning

    Zhong, R., Zhang, D., Sch \"a fer, L., Albrecht, S., and Hanna, J. Robust on-policy sampling for data-efficient policy evaluation in reinforcement learning. Advances in Neural Information Processing Systems, 35: 0 37376--37388, 2022

    2022

  108. [116]

    P., Zhu, J., Yang, Y., and Shi, C

    Zhou, H., Hanna, J. P., Zhu, J., Yang, Y., and Shi, C. Demystifying the paradox of importance sampling with an estimated history-dependent behavior policy in off-policy evaluation. In International conference on machine learning. PMLR, 2025

    2025

  109. [117]

    Seller-side experiments under interference induced by feedback loops in two-sided platforms

    Zhu, Z., Cai, Z., Zheng, L., and Si, N. Seller-side experiments under interference induced by feedback loops in two-sided platforms. arXiv preprint arXiv:2401.15811, 2024

    2024 arXiv

  110. [118]

    Zigler, C. M. and Papadogeorgou, G. Bipartite causal inference with interference. Statistical science: a review journal of the Institute of Mathematical Statistics, 36 0 (1): 0 109, 2021

    2021

  111. [119]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.