Pith. sign in

REVIEW 3 major objections 5 minor 46 references

Simulation-Based Inference for Adaptive Experiments

T0 review · 3 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 Simulating adaptive experiments with positively biased nuisance means yields hypothesis tests and confidence intervals with asymptotic type I error control for designs where standard reweighting fails.

desk verdict Genuinely new simulation-with-optimism method with solid proofs for three designs, but the 'wide variety' claim overreaches and the paper admits it. read the letter →

arxiv 2506.02881 v1 pith:DZPMHFDN submitted 2025-06-03 stat.ME cs.LGstat.ML

Brian M Cho , Aurélien Bibaut , Nathan Kallus This is my paper
classification stat.MEcs.LGstat.ML MSC 62L0562F4062F12
keywords simulation-basedinferenceadaptiveexperimentsmulti-armedbanditstypeIerrorcontrolsimulationwithoptimismconfidenceintervalspost-experimentexplore-then-commit
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

After an adaptive experiment, the sample mean of an arm is a natural test statistic, but its distribution is shaped by the data-dependent sampling rule and is generally not normal; standard Wald intervals and reweighting schemes either require conditional positivity or are conservative. This paper proposes to learn the null distribution by resimulating the whole experiment many times, and proves that adding a carefully sized positive bias to the estimated means of all non-target arms—'simulation with optimism'—makes the simulated distribution conservative enough for valid tests. Theorem 1 establishes asymptotic type I error control for explore-then-commit, UCB, and clipped reward-maximizing designs, with no conditional-positivity assumption. Inverting the test gives confidence intervals and consistent point estimates, and the experiments show intervals up to 50% narrower than existing approaches, with the largest gains for arms the design samples least. If the guarantees hold, adaptive experiments can be analyzed with a simple simulation procedure instead of specialized asymptotics.

What carries the argument

The mechanism is 'simulation with optimism': Algorithm 1 resimulates the experiment under the null by following the known adaptive policy, drawing each arm's outcomes as Gaussian with the target arm's mean fixed at $\theta_0$ and every other arm's mean shifted upward by $\epsilon_a$, and with variances estimated from the observed data. The bias $\epsilon_a$ is chosen to dominate the law-of-iterated-logarithm scale $\sqrt{\log\log N_T(a)/N_T(a)}$, which guarantees the optimistic nuisance eventually lies above the true mean almost surely. In the designs considered, that upward shift makes the simulated sample-mean distribution of the target arm wider than the true distribution, so comparing the observed statistic to simulated quantiles errs on the side of not rejecting. The proofs verify this quantile-widening case by case, using stability results for UCB, almost-sure convergence lemmas for random pull counts, and Glivenko-Cantelli convergence as the number of simulations $B$ grows.

What would settle it

Take an adaptive design not covered by Examples 1-3, for example a two-armed rule that commits to the target arm when the optimistic estimate of the other arm is high, so that raising the nuisance mean increases target-arm sampling. Resimulate under the null with the Theorem 1 bias and compare the empirical rejection rate at $\theta^*$ over many independent replications; if the rate exceeds $\alpha$ as $T$ and $B$ grow, the optimism-widens-quantiles principle fails for that design.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 1. In Algorithm 1, set the nuisance mean of each non-target arm $a$ to $\hat\mu_a = \hat\mu_T(a) + \epsilon_a$, where $\epsilon_a > 0$ and $\sqrt{\log\log N_T(a)/N_T(a)}/\epsilon_a \to 0$, and set $\hat\sigma_a^2$ to the sample variance. Then the two-sided resimulation test in Algorithm 2 satisfies $\limsup_{T\to\infty}\lim_{B\to\infty} P(\xi(\theta^*, \alpha, H_T) = 1) \le \alpha$ for the ETC, UCB, and clipped reward-maximizing designs, for every $\alpha \in [0,1]$. The force of the result is that it covers designs where the probability of selecting an arm can vanish, so asymptotic-normality reweighting is unavailable, and it bypasses the plug-in failure documented in Remark 2, where even $\sqrt{T}$-consistent nuisance estimates do not make the simulated and observed statistics share a limiting distribution. Lemma 2 adds that the inverted confidence set collapses almost surely to $\{\theta^*\}$ and the point estimate is strongly consistent.

Load-bearing premise

The guarantee hinges on the principle that adding positive bias to the other arms' means makes the simulated null distribution of the target arm's sample mean at least as wide as the true distribution; the paper proves this only for its three example designs and states that no unifying theory for a broader class is known.

Editorial extensions

If this is right

  • Tests and confidence intervals for arm means and their differences become valid after explore-then-commit, UCB, and clipped reward-maximizing designs, even when the design violates conditional positivity.
  • In the paper's experiments, confidence intervals are up to 50% narrower than the best baseline, with the largest gains for arms the adaptive design samples least.
  • The sample-mean test has asymptotic power 1: any false null is rejected almost surely as the horizon grows, so the confidence set collapses to the true mean.
  • The procedure is computationally practical: constructing a confidence interval costs $O(G B)$ simulations of length $T$, where $G$ is the grid of nulls and $B$ the number of trajectories per null.

Reading between the lines

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

  • The optimism principle suggests a general recipe: any adaptive design for which raising the other arms' means weakly increases the spread of the target arm's sample-mean distribution should inherit error control, and a unifying monotonicity condition would extend the theorem well beyond the three verified designs.
  • Because the bias term only needs to dominate the law-of-iterated-logarithm scale, one could tune $\epsilon_a$ arm-by-arm, using a smaller bias for well-sampled arms to recover power while preserving the theorem's rate condition; the paper's appendix already shows that smaller bias improves power.
  • The same resimulation idea could apply to other test statistics or off-policy estimates whenever the adaptive policy is known, but the paper proves guarantees only for sample means and their differences, so such extensions would need fresh proofs.
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

3 major / 5 minor

Summary. The paper proposes a simulation-based inference method for adaptive experiments. Given an observed trajectory and a point null for a target arm, Algorithm 1 resimulates trajectories under the null using Gaussian outcomes, setting non-target arm means to positively biased estimates plus a vanishing optimism term and using sample variances. Algorithm 2 rejects the null when the observed sample mean falls outside the empirical quantiles of the simulated statistics, and Algorithm 3 inverts these tests to build confidence intervals and a heuristic point estimate. Theorem 1 claims asymptotic type I error control for an explore-then-commit design (Example 1), a UCB design (Example 2), and a clipped reward-maximizing design (Example 3) as the number of simulations B grows. Lemmas 1 and 2 provide power-one consistency and convergence of the confidence set/point estimate. Empirical comparisons on synthetic data and an MTurk adaptive experiment report tighter intervals than baseline anytime-valid and reweighting approaches.

Significance. Conditional on Theorem 1, the paper offers a genuinely different route to post-adaptive inference: it avoids conditional positivity, handles non-normal limiting distributions, and is computationally more attractive than nuisance-grid scanning. The case-by-case proofs for ETC, UCB, and the clipped reward-maximizing design are detailed, and the empirical study includes both synthetic and real-world adaptive data, with runtime scaling reported. The main weakness is that the announced scope substantially exceeds what is proved: the 'optimism widens quantiles' principle is verified only for three designs, and the paper's own limitations section concedes that no unifying theory is provided. This scope mismatch, rather than the core construction, is the principal reason the manuscript needs revision.

major comments (3)
  1. [Abstract, §1, §E] The abstract and Introduction claim guarantees 'over a wide variety of common bandit designs' and over a 'wide class of commonly used designs,' but Theorem 1 proves type I error control only for Examples 1, 2, and 3. Section E explicitly states that there is no unifying theory identifying which designs enjoy the guarantee, and the epsilon-greedy experiments in Appendix C fall outside Theorem 1 and therefore cannot repair the theoretical gap. The stated scope should be narrowed to the verified designs, or a general sufficient condition on the allocation rule should be proved.
  2. [Lemma 7 and §D.2] Lemma 7's condition (i) is the load-bearing 'optimism widens quantiles' property: for every sample path, the simulated null quantiles must be at least as extreme as the observed quantiles. The manuscript verifies this only through separate limiting calculations in Sections D.2.1-D.2.3 and gives no design-level condition that implies it. The property is not automatic: for an allocation rule that increases target-arm pulls when non-target arms look better, adding positive bias to non-target means could shrink the simulated target-arm sample size and make simulated quantiles less extreme, inflating type I error. In addition, condition (i) is stated with a sup/inf over the entire sample space of random quantiles, and the proof's passage to F(sup_omega ...) needs a uniformity or measurability argument that is not supplied. The theorem would be on solid ground if the monotonicity property were stated as a formal condition on the design and verified, rather than checked case by case.
  3. [Lemma 2, §D.3, Remark 5] The proof of Lemma 2 in Section D.3 invokes Remark 5, which says that Algorithm 3 should be modified so that the confidence set always contains the empirical mean estimate. Algorithm 3 as printed already initializes the set with rho(H_T), so the gap is repairable, but the statement, proof, and remark are internally inconsistent, and Remark 5 is phrased as a future edit rather than as a property of the presented algorithm. Since Lemma 2 is the formal basis for the paper's confidence-interval consistency claim, this part of the manuscript needs to be brought into alignment before the result can be accepted as stated.
minor comments (5)
  1. [Abstract] The sentence 'Using these simulations, we characterize the distribution potentially non-normal sample mean test statistic to conduct inference' is grammatically incomplete and should be rewritten.
  2. [Algorithm 1, line 2] The displayed nuisance vector ends with sigma_hat_2^2 where sigma_hat_K^2 is clearly intended.
  3. [Lemma 5] Both conditions in Lemma 5 are labelled (i); the second should be labelled (ii).
  4. [Table 1] The text says that as G doubles the runtime 'doubles exactly,' but the reported values 11.50 to 20.21 and 20.66 to 39.17 are only approximately proportional; the wording should be softened.
  5. [§E and Remark 5] The phrase 'there exists minimal unifying theory' in the Limitations section should read 'there is little or no unifying theory,' and Remark 5 should be removed or converted into a formal part of the algorithm statement rather than a note about a future edit.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Theorem 1's optimistic-bias condition is proven via external limiting-distribution calculations rather than assumed or fit, and the admitted scope gaps concern generality, not circularity.

full rationale

We walked the derivation chain from Algorithms 1-2 to Theorem 1. The load-bearing step is Lemma 7's condition (i), which requires the simulated null quantiles to be at least as extreme as the observed statistic's quantiles. The paper proves this condition separately for Examples 1-3 in Appendix D by deriving the limiting distributions of both the observed and simulated test statistics (e.g., Eqs. 39, 40, 49, 50) and checking the resulting type I error bound. These calculations rely on external mathematical results (the strong law of large numbers, the law of iterated logarithm, and Khamaru-Zhang's UCB stability theorem), not on the paper's own conclusions. The positive bias epsilon_a is chosen to satisfy a stated sufficient condition (dominating the LIL fluctuation rate) and to make the simulated distribution provably more dispersed; it is not calibrated to match the observed data's coverage or p-values. The self-citations in the introduction and related work are contextual and do not carry the proof. We flag two non-circular limitations that the manuscript itself acknowledges: Section E states there is 'minimal unifying theory' covering which designs are valid, so the abstract's 'wide variety' claim exceeds the proven Examples 1-3; and Remark 5 plus Section D.3 admit that Lemma 2's confidence set could be empty and needs a fix. These are correctness and scope concerns, not circularity, because the three proven cases stand on their own external verifications.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a small set of domain assumptions and one ad hoc monotonicity principle that is verified only for the three example designs; no new physical entities are postulated.

free parameters (1)
  • epsilon_a (optimism bias) = log log N_T(a) / sqrt(N_T(a))
    Chosen by hand (not fitted to data) to satisfy Theorem 1's condition sqrt(log log N_T(a)/N_T(a))/epsilon_a -> 0. Appendix C.3 shows alternative biases also control type I error but reduce power.
assumptions (5)
  • domain assumption Assumption 1 (Infinite Sampling): for each arm a, N_T(a) diverges to infinity almost surely for any arm distributions.
    Used throughout to invoke SLLN and LIL; excludes designs that stop sampling some arm.
  • domain assumption Arm distributions have finite variance (and subgaussianity for the UCB example).
    Required for CLT/LIL and for the Khamaru-Zhang stability result used in Example 2.
  • ad hoc to paper 'Optimism widens quantiles' principle: increasing non-target arm means makes the target arm's sample mean distribution more dispersed.
    Verified case-by-case in Appendix D for ETC, UCB, and clipped reward-maximizing designs, but not proven for the general class claimed in the abstract. This is the key structural premise.
  • domain assumption The true parameter theta* lies in the tested grid Theta0 (needed for Lemma 2).
    Remark 4 offers a grid approximation; without it, the confidence interval convergence result is not exact.
  • standard math Variance estimators are strongly consistent (Lemma 6).
    Proved in Appendix D under Assumption 1 and finite variance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Simulation-Based Inference for Adaptive Experiments." pith.science (2026). https://pith.science/paper/DZPMHFDN

@misc{pith2026250602881,
  author       = {Pith},
  title        = {Pith review of: Simulation-Based Inference for Adaptive Experiments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DZPMHFDN}},
  note         = {Machine review of arXiv:2506.02881}
}
read the original abstract

Multi-arm bandit experimental designs are increasingly being adopted over standard randomized trials due to their potential to improve outcomes for study participants, enable faster identification of the best-performing options, and/or enhance the precision of estimating key parameters. Current approaches for inference after adaptive sampling either rely on asymptotic normality under restricted experiment designs or underpowered martingale concentration inequalities that lead to weak power in practice. To bypass these limitations, we propose a simulation-based approach for conducting hypothesis tests and constructing confidence intervals for arm specific means and their differences. Our simulation-based approach uses positively biased nuisances to generate additional trajectories of the experiment, which we call \textit{simulation with optimism}. Using these simulations, we characterize the distribution potentially non-normal sample mean test statistic to conduct inference. We provide guarantees for (i) asymptotic type I error control, (ii) convergence of our confidence intervals, and (iii) asymptotic strong consistency of our estimator over a wide variety of common bandit designs. Our empirical results show that our approach achieves the desired coverage while reducing confidence interval widths by up to 50%, with drastic improvements for arms not targeted by the design.

Figures

Figures reproduced from arXiv: 2506.02881 by the authors.

Figure 1
Figure 1. Plot of Realized Type I Error Rate with Plug-in and Optimistic Nuisances for ETC adaptive [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Coverage probabilities, average CI widths, and MSE for synthetic setup, with [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. 90% Confidence Intervals for real-world data. The difference in CI widths between our [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Top: Type I error rates based on magnitude of bias term [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Coverage probabilities, average CI widths, and MSE for synthetic setup using the clipped [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 25 canonical work pages

  1. [1]

    P. Auer, N. Cesa-Bianchi, Y . Freund, and R. E. Schapire. The nonstochastic multiarmed bandit problem.SIAM Journal on Computing, 32(1):48–77, 2002. doi: 10.1137/S0097539701398375. URLhttps://doi.org/10.1137/S0097539701398375

    doi:10.1137/s0097539701398375 2002

  2. [2]

    Demistifying Inference after Adaptive Experiments

    A. Bibaut and N. Kallus. Demystifying inference after adaptive experiments, 2024. URL https://arxiv.org/abs/2405.01281

    work page Pith review arXiv 2024

  3. [3]

    Bibaut, A

    A. Bibaut, A. Chambaz, M. Dimakopoulou, N. Kallus, and M. van der Laan. Post-contextual- bandit inference, 2021

    work page 2021

  4. [4]

    Bibaut, A

    A. Bibaut, A. Chambaz, M. Dimakopoulou, N. Kallus, and M. van der Laan. Post-contextual- bandit inference, 2021. URLhttps://arxiv.org/abs/2106.00418

    arXiv 2021

  5. [5]

    Bibaut, N

    A. Bibaut, N. Kallus, and M. Lindon. Near-optimal non-parametric sequential tests and confidence sequences with possibly dependent observations, 2024. URL https://arxiv. org/abs/2212.14411

    arXiv 2024

  6. [6]

    Chernozhukov, D

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

    doi:10.1111/ectj.12097 2018

  7. [7]

    B. Cho, K. Gan, and N. Kallus. Peeking with peak: Sequential, nonparametric composite hypothesis tests for means of multiple data streams, 2024. URL https://arxiv.org/abs/ 2402.06122

    work page Pith review arXiv 2024

  8. [8]

    B. Cho, D. Meier, K. Gan, and N. Kallus. Reward maximization for pure exploration: Minimax optimal good arm identification for nonparametric multi-armed bandits, 2024. URL https: //arxiv.org/abs/2410.15564

    arXiv 2024

Show all 46 references
  1. [9]

    T. J. DiCiccio and B. Efron. Bootstrap confidence intervals.Statistical Science, 11(3):189 – 228,

  2. [10]

    Gabillon, M

    V . Gabillon, M. Ghavamzadeh, A. Lazaric, and S. Bubeck. Multi-bandit best arm identifica- tion. InProceedings of the 25th International Conference on Neural Information Processing Systems, NIPS’11, page 2222–2230, Red Hook, NY , USA, 2011. Curran Associates Inc. ISBN 9781618395993

    2011

  3. [11]

    Hadad, D

    V . Hadad, D. A. Hirshberg, R. Zhan, S. Wager, and S. Athey. Confidence intervals for policy evaluation in adaptive experiments.Proceedings of the National Academy of Sciences, 118(15): e2014602118, 2021. doi: 10.1073/pnas.2014602118. URL https://www.pnas.org/doi/ abs/10.1073/...

    2021 doi

  4. [12]

    P. Hall, C. Heyde, Z. Birnbaum, and E. Lukacs.Martingale Limit Theory and Its Application. Communication and Behavior. Academic Press, 2014. ISBN 9781483263229. URL https: //books.google.com/books?id=gqriBQAAQBAJ

    2014

  5. [13]

    S. R. Howard, A. Ramdas, J. McAuliffe, and J. Sekhon. Time-uniform, nonparamet- ric, nonasymptotic confidence sequences.The Annals of Statistics, 49(2), apr 2021. doi: 10.1214/20-aos1991. URLhttps://doi.org/10.1214%2F20-aos1991

    2021 doi

  6. [14]

    Hu and W

    F. Hu and W. F. Rosenberger.The theory of response-adaptive randomization in clinical trials. John Wiley & Sons, 2006

    2006

  7. [15]

    Khamaru and C.-H

    K. Khamaru and C.-H. Zhang. Inference with the upper confidence bound algorithm, 2024. URLhttps://arxiv.org/abs/2408.04595

    2024 arXiv

  8. [16]

    Lai and H

    T. Lai and H. Robbins. Asymptotically efficient adaptive allocation rules.Advances in Applied Mathematics, 6(1):4–22, 1985. ISSN 0196-8858. doi: https://doi.org/10.1016/ 0196-8858(85)90002-8. URL https://www.sciencedirect.com/science/article/ pii/0196885885900028. 10

    1985

  9. [17]

    Lai and H

    T. Lai and H. Robbins. Asymptotically efficient adaptive allocation rules.Adv. Appl. Math., 6 (1):4–22, Mar. 1985. ISSN 0196-8858. doi: 10.1016/0196-8858(85)90002-8. URL https: //doi.org/10.1016/0196-8858(85)90002-8

    1985 doi

  10. [18]

    D. L. McLeish. Dependent Central Limit Theorems and Invariance Principles.The Annals of Probability, 2(4):620 – 628, 1974. doi: 10.1214/aop/1176996608. URL https://doi.org/ 10.1214/aop/1176996608

    1974

  11. [19]

    X. Nie, X. Tian, J. Taylor, and J. Zou. Why adaptively collected data have negative bias and how to correct for it. In A. Storkey and F. Perez-Cruz, editors,Proceedings of the Twenty-First International Conference on Artificial Intelligence and Statistics, volume 84 ofProceedi...

    2018

  12. [20]

    Offer-Westort, A

    M. Offer-Westort, A. Coppock, and D. Green. Adaptive experimental design: Prospects and applications in political science.American Journal of Political Science, 65, 02 2021. doi: 10.1111/ajps.12597

    2021 doi

  13. [21]

    Ramdas, P

    A. Ramdas, P. Grünwald, V . V ovk, and G. Shafer. Game-Theoretic Statistics and Safe Anytime- Valid Inference.Statistical Science, 38(4):576 – 601, 2023. doi: 10.1214/23-STS894. URL https://doi.org/10.1214/23-STS894

    2023 doi

  14. [22]

    H. Robbins. Some aspects of the sequential design of experiments.Bulletin of the American Mathematical Society, 58(5):527 – 535, 1952

    1952

  15. [23]

    W. F. Rosenberger and F. Hu. Bootstrap methods for adaptive designs.Statistics in Medicine, 18 (14):1757–1767, 1999. doi: https://doi.org/10.1002/(SICI)1097-0258(19990730)18:14<1757:: AID-SIM212>3.0.CO;2-R. URL https://onlinelibrary.wiley.com/doi/abs/10. 1002/%28SICI%291097-02...

    1999 doi

  16. [24]

    J. Shin, A. Ramdas, and A. Rinaldo. On the bias, risk and consistency of sample means in multi-armed bandits, 2021. URLhttps://arxiv.org/abs/1902.00746

    2021 arXiv

  17. [25]

    J. Trommer. Resampling methods for dependent data.Biometrics, 62(2):633–634, 06 2006. ISSN 0006-341X. doi: 10.1111/j.1541-0420.2006.00589_12.x. URL https://doi.org/10. 1111/j.1541-0420.2006.00589_12.x

    2006

  18. [26]

    A. W. v. d. Vaart.Asymptotic Statistics. Number 9780521784504 in Cambridge Books. Cam- bridge University Press, Enero 2000. URL https://ideas.repec.org/b/cup/cbooks/ 9780521784504.html

    2000

  19. [27]

    Ville.Étude critique de la notion de collectif

    J. Ville.Étude critique de la notion de collectif. 1939. URL http://eudml.org/doc/192893

    1939

  20. [28]

    Waudby-Smith and A

    I. Waudby-Smith and A. Ramdas. Estimating means of bounded random variables by betting,

  21. [29]

    Waudby-Smith, D

    I. Waudby-Smith, D. Arbour, R. Sinha, E. H. Kennedy, and A. Ramdas. Time-uniform central limit theory and asymptotic confidence sequences, 2024. URL https://arxiv.org/abs/ 2103.06476

    2024 arXiv

  22. [30]

    Waudby-Smith, L

    I. Waudby-Smith, L. Wu, A. Ramdas, N. Karampatziakis, and P. Mineiro. Anytime-valid off- policy inference for contextual bandits, 2024. URL https://arxiv.org/abs/2210.10768

    2024 arXiv

  23. [31]

    L. J. Wei, R. T. Smythe, D. Y . Lin, and T. S. Park. Statistical inference with data-dependent treatment allocation rules.Journal of the American Statistical Association, 85(409):156–162,

  24. [32]

    R. Zhan, V . Hadad, D. A. Hirshberg, and S. Athey. Off-policy evaluation via adaptive weighting with data from contextual bandits, 2021. URLhttps://arxiv.org/abs/2106.02029

    2021

  25. [33]

    K. W. Zhang, L. Janson, and S. A. Murphy. Statistical inference with m-estimators on adaptively collected data, 2021. URLhttps://arxiv.org/abs/2104.14074. 11

    2021 arXiv

  26. [34]

    J. Zhao. Adaptive neyman allocation. InProceedings of the 25th ACM Conference on Economics and Computation, EC ’24, page 776, New York, NY , USA, 2024. Association for Computing Machinery. ISBN 9798400707049. doi: 10.1145/3670865.3673535. URL https://doi.org/ 10.1145/3670865.3...

    2024

  27. [38]

    normal random variables

    is the true standard deviation of arm 1 and Z1, Z2, Z3 are i.i.d. normal random variables. If we set the nuisance value ˆη= ˆµ2 equal to the maximum possible value of 1, then, for any null value θ0, the distribution of the test statistic using simulated trajectories H (i) T is...

  28. [39]

    is unchanged for values ofθ0 < µ∗ 2, a vanishing bias term results in improved power for all values ofθ0 > µ∗

  29. [40]

    When θ0 > µ2, the simulated distribution of the test statistic ρ(H (i) T ) is more tightly centered around the value θ0 compared to the choice of ˆµ2 = 1 by a factor of 1/ √

  30. [41]

    A.2 Confidence Intervals for Unbounded Parameter Spaces

    As a result, the probability that our observed test statistic ρ(HT ) lies beyond the lower and upper quantiles of simulated distribution is larger under nuisances ˆηthat converge toη ∗. A.2 Confidence Intervals for Unbounded Parameter Spaces. While the main body of our paper f...

    2018

  31. [42]

    Let σ∗ 1, σ∗ 2 denote the true standard deviations of arm 1 and 2 respectively. Analysis of Case (i)We first characterize the distribution of the observed sample mean test statistic under the ETC design for Case (i), keeping the true arm distributions P1, P2 fixed. The observe...

  32. [43]

    If ϵ2(ω)≥µ ∗ 2 −µ ∗ 1, then the distribution of the observed test statistic ρ(HT ) and sample-path dependent simulated distribution ρ(H (i) T )(ω) match in distribution due to1[µ∗ 1 ≥µ ∗ 2+ϵ2(ω)] =1[µ ∗ 1 ≥µ ∗ 2]. In the case where µ∗ 1 > µ∗ 2, if ϵ2(ω)≥µ ∗ 1−µ∗ 2, the distrib...

  33. [44]

    Because we assume σ∗ 2 <∞ , term (a) is upper bounded by the constant σ∗ 2 √

  34. [45]

    Because NT (2)(ω)→ ∞for all ω∈Ω , term (b)→0 for all ω∈Ω

    For term (b), note that ˆσ1(ω)→σ ∗ 1 and ˆσ2(ω)→σ ∗ 2 for all ω∈Ω , meaning that the scalar values in the numerator are finite. Because NT (2)(ω)→ ∞for all ω∈Ω , term (b)→0 for all ω∈Ω . Lastly, for term (c), we use the condition that ϵ2(ω)/ q log logN T (2)(ω) NT (2)(ω) −1 →0...

  35. [46]

    (74)   (Φ(3c/2)−Φ(3c ′/2))| {z } ≥0 (1 + Φ(σ∗ 2x/σ∗ 1)) + 2 Φ(3c′/2)2 −Φ(3c/2) 2 | {z } ≤0 Φ(3c/2)Φ(3c′/2)   d f(x)(75) ≤0. (76) Now, note that by Equation 50, we know that the following holds: lim T→∞ sup ω∈Ω ˆF −1(α/2)(ω) = Φ−1(α/2),lim T→∞ inf ω∈Ω ˆF −1(1−α/2)(ω) ...

  36. [1990]

    URLhttp://www.jstor.org/stable/2289538

    ISSN 01621459, 1537274X. URLhttp://www.jstor.org/stable/2289538

  37. [1996]

    URLhttps://doi.org/10.1214/ss/1032280214

    doi: 10.1214/ss/1032280214. URLhttps://doi.org/10.1214/ss/1032280214

  38. [2022]

    URLhttps://arxiv.org/abs/2010.09686

    2010 arXiv

Pith tools

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