Pith. sign in

REVIEW 2 major objections 5 minor 100 references

This paper claims that EDBCD and EDBCD2, a new family of multi-treatment response-adaptive designs, simultaneously attain the Cramér-Rao lower bound on allocation variance and the theoretical optima for selection bias and entropy.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 18:47 UTC pith:WABKY3EK

load-bearing objection This paper essentially closes the multi-treatment ERADE open problem with a new continuous allocation family, and the stress-test objection about EDBCD2 being ill-defined does not hold up under arithmetic check. the 2 major comments →

arxiv 2607.17176 v1 pith:WABKY3EK submitted 2026-07-19 math.ST math.PRstat.MEstat.TH

Efficient Doubly Adaptive Biased Coin Designs for Multiple Treatments

Zhang Li-Xin This is my paper
classification math.ST math.PRstat.MEstat.TH MSC 60F1562G1060F0560F10
keywords response-adaptive designsclinical trialsCramér-Rao lower boundselection biasentropydoubly adaptive biased coin designmultiple treatmentsasymptotic efficiency
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper aims to dissolve a long-standing three-way conflict in response-adaptive clinical trials: a design that adapts to accumulating outcomes typically becomes predictable (high selection bias, low entropy), while a design that stays random loses statistical efficiency because the allocation proportions fluctuate more. The author proposes a general framework for K-treatment trials in which the next patient's treatment probability is a continuous function of the current allocation proportions and a sequential estimate of the target proportions. Within this framework, two concrete procedures — the efficient doubly adaptive biased coin design (EDBCD) and a variant EDBCD2 — are shown to target any desired allocation proportion, have asymptotic allocation variance equal to the Cramér-Rao lower bound, and simultaneously achieve the theoretical best values of selection bias and entropy. The paper also proves strong consistency, asymptotic normality, and functional central limit theorems for the allocation proportions and parameter estimators, including at random stopping times. The paper is explicit that these are large-sample guarantees; for fixed small samples the original trade-offs may still be felt.

Core claim

The central claim is that efficiency and randomness are not in conflict once the allocation rule is built from a weight function with an infinite cusp at the target ratio. Concretely, the allocation probability for treatment k is g_k(x,ρ)=ρ_k ψ(ρ_k/x_k)/Σ_j ρ_j ψ(ρ_j/x_j), with ψ continuous, ψ(1)=1, and ψ'_+(1)=∞. This infinite right derivative makes the design overcorrect any tiny overrepresentation of a treatment, driving the asymptotic covariance of √n(N_n/n - v) down to Σ, the Cramér-Rao lower bound, while the continuity of g makes the allocation probabilities converge pointwise to v, so the selection bias reaches max_k v_k and the entropy reaches H(v). The same qualitative result holds

What carries the argument

The load-bearing object is the weight function ψ in the allocation function g_k(x,ρ)=ρ_k ψ(ρ_k/x_k)/Σ_j ρ_j ψ(ρ_j/x_j), together with Condition C/C′: whenever treatment k is overrepresented, the allocation probability must be bounded above by ρ_k + λ_{m,k}(x_k-ρ_k)+..., with λ_{m,k} eventually diverging to -∞ (not just being negative). That divergence is exactly what the cusp ψ'_+(1)=∞ produces, and it is what cancels the extra variance term that makes the ordinary DBCD inefficient. The proofs of the asymptotic results use a recursive stochastic approximation representation, martingale strong approximation by Brownian motion, and a Gaussian comparison theorem to show that the allocation proc

Load-bearing premise

The efficiency claim rests on the allocation weight ψ having an infinitely steep cusp at the target ratio (ψ'_+(1)=∞) — with a smooth weight of finite slope, the design degenerates to the ordinary non-efficient DBCD and the variance stays strictly above the Cramér-Rao lower bound.

What would settle it

Compute the asymptotic covariance of √n(N_n/n - v) for EDBCD when ψ is replaced by a smooth weight with finite derivative at 1, e.g. ψ(x)=x^γ for x≥1 and ψ(x)=x^γ for x≤1 (the ordinary DBCD weight). The paper's own Theorem A.4 predicts the limit becomes Λ_γ = (1/(1+2γ))diag(v) + ... + (2(1+γ)/(1+2γ))Σ, which is strictly larger than Σ for every γ finite; observing exactly that inflation (rather than Σ) would confirm that the infinite cusp, not mere continuity, carries the efficiency claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the claims hold, a multi-treatment trial can be run with any chosen target allocation (for example one that minimizes expected treatment failures) without paying an asymptotic penalty in the variance of the final allocation proportions.
  • The optimality of selection bias and entropy means an experimenter trying to guess the next assignment can do no better asymptotically than guessing the largest target proportion, and the per-patient Shannon entropy reaches its theoretical ceiling for that target; no adaptive design converging to the same proportions can be less predictable.
  • The strong consistency, functional central limit theorem, and random-time central limit theorem make the proposed designs suitable for sequential monitoring and for trials stopped by interim analysis rules.
  • Because Condition C/C′ is general, the paper's framework can generate other efficient designs besides EDBCD and EDBCD2; several examples are given, including step-up, step-down, and monotone-weight procedures.
  • The theoretical results cover both continuous and discrete allocation functions, subsuming earlier efficient designs for two treatments and extending them to any K.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A practical caveat follows from the mathematics: the infinite cusp must be realized in computation. If an implementation smooths or discretizes the weight function so that the right derivative at 1 is finite, the variance may revert toward the inflated value of the ordinary DBCD; an editor-level simulation varying the grid resolution near the cusp would show how much of the theoretical gain surviv
  • The result reframes the efficiency-randomness trade-off as a rate condition rather than a structural conflict: most-random behaviour only requires allocation probabilities to converge to v pointwise, while efficiency additionally requires that convergence be fast enough (λ_{m,k}→-∞). This suggests one could build a one-parameter family interpolating between DBCD and EDBCD by tuning the derivative
  • The entropy optimality proof suggests that for any design converging to v, 'most random' is equivalent to the allocation probabilities themselves converging to v; if so, randomness optimality is a topological property of the allocation function, not a property of the randomization mechanism's tuning parameters.
  • The paper's own suggestion to substitute a delayed-response estimator into the framework indicates the results should transfer to survival trials with delayed and censored outcomes, giving a candidate efficient-adaptive design in that setting; testing this on simulated survival data would be a natural extension.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper develops a general framework for efficient response-adaptive randomization in K-arm clinical trials. It introduces Condition C on allocation functions and proves that, under Conditions A and B, a design satisfying Condition C is asymptotically normal with the minimal covariance matrix Sigma and hence attains the Cramer-Rao lower bound for allocation variability. It also proves lower/upper bounds for asymptotic selection bias and entropy, defining 'most random' designs, and proposes two concrete procedures: EDBCD (Procedure B), based on a weight function with an infinite right derivative at the target ratio, and EDBCD2 (Procedure C), based on a linear imbalance adjustment. The main theorem (Theorem 5.1) claims that both procedures are simultaneously efficient and most random. Proofs of strong consistency, rates, functional central limit theorems, and Gaussian approximations are deferred to a supplementary file, and simulation comparisons with DBCD, ERADE, and GERADE are reported.

Significance. If the central claims are correct, the paper would provide a substantial resolution of the multi-treatment version of the open problem raised by Hu and Rosenberger (2006): a fully randomized procedure that targets arbitrary allocation proportions and is asymptotically best in both variability and randomness. The Condition C framework is broader than the existing two-treatment ERADE and gives a flexible template for constructing new efficient designs. The entropy and selection-bias optimality results, though elementary, supply useful theoretical baselines. The EDBCD construction is elegant: the cusp in the weight function is exactly what forces the allocation probability to track the estimated target closely enough to achieve efficiency while remaining continuous enough to attain optimal randomness. However, as written, one of the two flagship procedures (EDBCD2) is not a well-defined allocation rule, and the main theorem overclaims for it. The general framework and the EDBCD procedure appear salvageable, but the current version requires nontrivial repair before the central claim is fully supported.

major comments (2)
  1. [Section 3, Procedure C, Eq. (3.4); Theorem 5.1] Procedure C is not a well-defined allocation rule because g_k in (3.4) is not guaranteed to lie in [0,1] for valid simplex states. The components sum to 1 by construction, but individual components can be negative or exceed 1. For example, take K=3, rho=(0.5,0.25,0.25), x=(0.5001,0.5,0), and alpha=0.5. Then c=0.25/(0.0001)^0.5=25, so g_2=0.25-25(0.5-0.25)=-6.0, an invalid probability. The paper never truncates, clamps, or renormalizes, and never proves g(x,rho) is a probability vector. The subsequent verification of Condition C and the proofs of asymptotic normality and randomness silently treat g as a probability vector; a negative component breaks the boundedness of the martingale differences and the simplex invariance argument. Consequently, Theorem 5.1's explicit claim that EDBCD2 is efficient and most random is not established as written. The design must either be modified (e.g., by
  2. [Section 3, Procedure B, Eq. (3.3); Procedure A] Even for EDBCD (Procedure B), the algorithm is not fully specified for small m. Since Procedure A only prescribes complete randomization for the first patient, for m<K the vector N_m/m has some zero coordinates. For any such state, g_k = rho_k psi(rho_k/x_k) / sum_j rho_j psi(rho_j/x_j) is undefined whenever x_k=0, because psi(rho_k/0) = psi(infinity) = infinity and the denominator also involves infinite terms (possibly more than one). Thus the allocation probabilities at steps 2,...,K cannot be computed from (3.3). This affects the EDBCD part of Theorem 5.1, not only the secondary design. The authors need to add a standard initialization convention (e.g., assigning one patient to each treatment before adaptation) or provide an extended definition for x_k=0 and verify that the asymptotics are unchanged.
minor comments (5)
  1. [Section 3, Eq. (3.4)] The remark '0/0 is defined to be 0' is confusing and inconsistent with the formula: the numerator rho_j ^ (1-rho_j) is always positive for rho in (0,1), so when x_j=rho_j the ratio is infinite, not 0/0. Please restate the intended convention clearly.
  2. [Supplementary Materials, proof of Theorem A.3] After the display for E[Z_n], the condition is written as 'lambda0 > 0', but throughout the proof lambda0 is negative. This is a sign typo and should be corrected to lambda0 < 0.
  3. [References] The reference to Wei and Durham (1978) is listed as 'randomized pay-the-winner'; the standard name is 'play-the-winner'.
  4. [Section 4.2, Theorem 4.3] The assumption that rho_1(hat_Theta_m) has no mass at rational points is used to assert P(N_m,1 = m rho_hat_m,1) = 0 for all m. This is fine, but the proof of (4.10) appears to assume v1 >= 1/2 to identify max(alpha v1, 1-alpha v1); the statement should be made symmetric for the case v1 < 1/2, or a sentence should clarify that the labeling is without loss of generality.
  5. [Section 6, Theorem 6.2 and 6.3] Equation (6.4) reads 'n(hat_Theta_n - Theta) = B_t V^{1/2} + o(n^{1/2})'. This is dimensionally correct if B_t is Brownian motion at time t, but the use of both n and t in the same equation is confusing. Clarify that the Brownian motion is evaluated at time n (so the right-hand side is B(n)V^{1/2} + o(n^{1/2})).

Circularity Check

0 steps flagged

No significant circularity: the main derivations verify Condition C directly and use external theorems as benchmarks.

full rationale

The paper's efficiency claim is a conditional theorem: if an allocation function satisfies Condition C and the estimator is efficient, then n^{1/2}(N_n/n - v) converges to N(0, Sigma), so the design attains the Hu et al. lower bound. The verification of Condition C for EDBCD is an explicit calculation using the cusp of psi at 1: the text shows g_k <= rho_k when x_k > rho_k and then derives g_k - rho_k <= -lambda (x_k - rho_k) for arbitrarily large lambda near v, which gives lambda_{m,k} -> -infinity. For EDBCD2, Condition C is checked by algebra from the definition g_k = rhohat_k - c(x_k - rhohat_k). Neither check assumes the conclusion; both are direct sufficient-condition verifications. The randomness/optimality part is also not circular: Theorems 4.1 and 4.2 prove lower/upper bounds on selection bias and entropy and establish attainment iff p_{m,k} converges to rho_k(Theta); Theorem 5.1 then proves that convergence from continuity and the strong consistency theorems, rather than assuming it. The lower-bound result of Hu et al. (2006) and the technical lemmas from Zhang (2004, 2024) are external mathematical results with stated assumptions, not the target theorem, and are not fitted to any data, so they do not create circularity. One non-circular concern: Procedure C's allocation function (3.4) is not proved to lie in [0,1] and can take negative values for valid simplex states, so Theorem 5.1's claim for EDBCD2 may be unsupported. That is a correctness/well-definedness gap, not a circularity, and does not affect the circularity score.

Axiom & Free-Parameter Ledger

2 free parameters · 6 axioms · 0 invented entities

The central claims rest on standard conditions (A, B) for the target proportion and response distributions, on the Hu et al. (2006) lower bound, on martingale strong-approximation lemmas from the author's earlier work (Zhang 2004) and Hu-Zhang (2004), and on the Gaussian comparison (Slepian) lemma. No new entities are invented; the only free design constants are the tuning parameters gamma (EDBCD) and alpha (EDBCD2/ERADE), which are recommended heuristically, not fitted.

free parameters (2)
  • gamma (EDBCD weight exponent) = gamma > 0, recommended 2-4
    Tuning parameter in psi(x)=1+sqrt(x^{2 gamma}-1). Efficiency holds for any gamma>0; the value affects finite-sample behavior but is not fitted to data.
  • alpha (EDBCD2 / ERADE randomization level) = 0 < alpha < 1
    Controls the degree of randomization; simulations use alpha=2/3 or 1/2. Not fitted.
axioms (6)
  • domain assumption Condition A: target proportion function rho(y) is continuous and differentiable at Theta with rho(Theta)=v.
    Required for the delta method and the definition of the variance matrix Sigma in (2.7).
  • domain assumption Condition B: Bahadur representation theta_hat_{m,k} = N_{m,k}^{-1} sum X_{j,k} xi_{j,k} + o(N_{m,k}^{-1/2}) with E||xi_{1,k}||^{2+epsilon} < infinity.
    Standard for asymptotic normality of parameter estimators; used throughout the proofs.
  • standard math Hu et al. (2006) lower bound: the asymptotic variance of N_n/n is at least Sigma for any adaptive design with target rho(Theta).
    External benchmark used to define efficiency; cited as prior result.
  • standard math Lemma A.2 (Hu and Zhang 2004): law of iterated logarithm and consistency of parameter estimators.
    Used to derive rates for rho_hat_n and N_n.
  • standard math Lemma A.3 (Zhang 2004): strong approximation of martingale vectors by Brownian motion.
    Key technical tool for Gaussian approximation; proof referenced to earlier work by the author.
  • standard math Gaussian comparison / Slepian's lemma.
    Used in the proof of Theorem A.3 to bound maxima of Gaussian processes.

pith-pipeline@v1.3.0-alltime-deepseek · 3289 in / 3366 out tokens · 159357 ms · 2026-08-01T18:47:55.513679+00:00 · methodology

0 comments
read the original abstract

The randomness, efficiency (power and variability), and desirable allocation proportions are important components for evaluating a response-adaptive design in clinical trials and conflicted demands in applications. The aim of this paper is to provide designs dealing with these dilemmas. We first give a general framework for efficient response-adaptive randomization procedures that attain the Cram\'er-Rao lower bounds of the allocation variances for any desired allocation proportions. The general framework is flexible for us to define new families of efficient designs with good properties for both two and multiple-treatment clinical trials. We also prove that, among all response-adaptive randomization procedures with the same limit allocation proportions, the selection biases and entropies as measures of the randomness of the designs have their optimal values. Basing on the theory on efficiency and randomness, we propose a new family of doubly adaptive biased coin designs for multi-treatment clinical trials that can target any allocation proportion and are asymptotically best in terms both the randomness and efficiency so that their randomness is asymptotic optimal and asymptotic allocation variance attains the Cram\'er-Rao lower bound. Theoretical properties, including the strong consistency, the asymptotic normality, and the functional central limit theorem for both the sample allocation proportions and the estimators of the distribution parameters, are developed by using the technique of Gaussian approximation and Gaussian comparing theorems.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

100 extracted references · 2 linked inside Pith

  1. [1]

    and Hu, F

    Alkhnefr, N. and Hu, F. and Zhai, G. , title =. Statistical Methods in Medical Research , year =

  2. [2]

    , title =

    Baldi Antognini, A. , title =. J. Statist. Plann. Inference , year =

  3. [3]

    and Giovagnoli, A

    Baldi Antognini, A. and Giovagnoli, A. , title =. J. Royal Statist. Soc. Series C , year =

  4. [4]

    , year =

    Billingsley, P. , year =. Convergence of Probability Measures , publisher =

  5. [5]

    , year =

    Duflo, M. , year =. Random iterative models,

  6. [6]

    , year =

    Eberlein, E. , year =. On strong invariance principles under dependence , journal =

  7. [7]

    , year =

    Eisele, J. , year =. The doubly adaptive biased coin design for sequential clinical trials , journal =

  8. [8]

    and Woodroofe, M

    Eisele, J. and Woodroofe, M. , year =. Central limit theorems for doubly adaptive biased coin designs , journal =

  9. [9]

    , year =

    Efron, B. , year =. Forcing a sequential experiment to be balanced , journal =

  10. [10]

    Hayre, L. S. , year =. Two-population sequential tests with three hypotheses , journal =

  11. [11]

    and Rosenberger, W

    Hu, F. and Rosenberger, W. F. , year =. Optimality, variability, power: Evaluating responseadaptive randomization procedures for treatment comparisons , journal =

  12. [12]

    and Rosenberger, W

    Hu, F. and Rosenberger, W. F. , year =. The Theory of Response-Adaptive Randomization in Clinical Trials , publisher =

  13. [13]

    and Rosenberger, W

    Hu, F. and Rosenberger, W. F. and Zhang, L.-X. , year =. Asymptotically best response-adaptive randomization procedures , journal =

  14. [14]

    and Zhang, L.-X

    Hu, F. and Zhang, L.-X. , year =. Asymptotic properties of doubly adaptive biased coin designs for multi-treatment clinical trials , journal =

  15. [15]

    and Zhang, L.-X

    Hu, F. and Zhang, L.-X. and Cheung, S. H. and Chan, W. S. , year =. Doubly adaptive biased coin designs with delayed responses , journal =

  16. [16]

    and Zhang, L.-X

    Hu, F. and Zhang, L.-X. and He, X. , year =. Efficient randomized adaptive designs , journal =

  17. [17]

    , year =

    Ivanova, A. , year =. A play-the-winner type urn model with reduced variability , journal =

  18. [18]

    , year =

    Ivanova, A. , year =. Urn designs with immigration: Useful connection with continuous time stochastic processes , journal =

  19. [19]

    and Turnbull, B

    Jennison, C. and Turnbull, B. W. , year =. Group Sequential Methods with Applications to Clinical Trials , publisher =

  20. [20]

    Kushner, H. J. and Yin, G.G. , year =. Stochastic approximation and recursive algorithms and applications,

  21. [21]

    and Simon, R

    Pocock, S.J. and Simon, R. , year =. Sequential treatment assignment with balancing for prognostic factors in the controlled clinical , journal =

  22. [22]

    and Page, C

    Melfi, V. and Page, C. and Geraldes, M. , year =. An adaptive randomized design with application to estimation , journal =

  23. [23]

    , year =

    Robbins, H. , year =. Some aspects of the sequential design of experiments , journal =

  24. [24]

    Rosenberger, W. F. and Lachin, J. M. , year =. Randomization in Clinical Trials Theory and Practice , publisher =

  25. [25]

    Rosenberger, W. F. and Stallard, N. and Ivanova, A. and Harper, C. and Ricks, M. , year =. Optimal adaptive designs for binary response trials , journal =

  26. [26]

    Smith, R. L. , year =. Properties of Biased coin designs in sequential clinical trials , journal =

  27. [27]

    Smith, R. L. , year =. Sequential treatment allocation using biased coin designs , journal =

  28. [28]

    Thompson, W. R. , year =. On the likelihood that one unknown probability exceeds another in view of the evidence of the two samples , journal =

  29. [29]

    and Rosenberger, W

    Tymofyeyev, Y. and Rosenberger, W. F. and Hu, F. , year =. Implementing optimal allocation in sequential binary response experiments , journal =

  30. [30]

    Journal of the American Statistical Association , Year =

    An application of an urn model to the design of sequential controlled clinical trials , Author =. Journal of the American Statistical Association , Year =

  31. [31]

    Wei, L. J. , year =. The adaptive biased coin design for sequential experiments , journal =

  32. [32]

    , year =

    Wei, L.J. , year =. The generalized P\'olya's urn design for sequential medical trials , journal =

  33. [33]

    and Durham, S

    Wei, L.J. and Durham, S. , year =. The randomized pay-the-winner rule in medical trials , journal =

  34. [34]

    J., Smythe, R.T

    Wei, L. J., Smythe, R.T. and Smith, R. L. , year =. K -treatment comparisons with restricted randomization rules in clinical trials , journal =

  35. [35]

    Zhang, L. J. and Rosenberger, W. F. , year =. Response-adaptive randomization for clinical trials with continuous outcomes , journal =

  36. [36]

    Zhang, L. J. and Rosenberger, W. F. , year =. Response-adaptive randomization for survival trials: the parametric approach , journal =

  37. [37]

    , year =

    Zhang, L.-X. , year =. Strong approximations of martingale vectors and their applications in Markov-chain adaptive designs , journal =

  38. [38]

    Handbook of Statistics-- Probability Models,

    The recursive stochastic algorithm, randomized urn models, and response-adaptive randomization in clinical trials , author=. Handbook of Statistics-- Probability Models,. 2024 , publisher=

    2024

  39. [39]

    and Chan, W.S

    Zhang, L.-X. and Chan, W.S. and Cheung. S.H. and Hu, F. , year =. A generalized urn model for clinical trials with delayed responses , journal =

  40. [40]

    and Hu, F

    Zhang, L.-X. and Hu, F. and Cheung, S.H. and Chan, W.S. , year =. Immigrated urn models -- theoretical properties and applications , journal =

  41. [41]

    Zhu, H. J. and Hu, F. , year =. Sequential monitoring of response-adaptive randomized clinical trials , journal =

  42. [42]

    1991 , author =

    Probability in Banach Spaces, Isoperimetry and Processes , publisher =. 1991 , author =

    1991

  43. [43]

    1980 , publisher=

    Martingale Limit Theory and its Application , author=. 1980 , publisher=

    1980

  44. [44]

    Antognini, A. B. and Zagoraiou, M. , title =. Biometrika , year =

  45. [45]

    Annales de l'IHP Probabilit

    Relation de domination entre deux processus , author=. Annales de l'IHP Probabilit

  46. [46]

    The risks and rewards of covariate adjustment in randomized trials: An assessment of 12 outcomes from 8 studies , journal =

    Kahan, Brennan C and Jairath, Vipul and Dor. The risks and rewards of covariate adjustment in randomized trials: An assessment of 12 outcomes from 8 studies , journal =. 2014 , volume =

    2014

  47. [47]

    2015 , author =

    Modern Adaptive Randomized Clinical Trials: Statistical and Practical Aspects , publisher =. 2015 , author =

    2015

  48. [48]

    The Lancet , volume=

    Clinical assessment incorporating a personal genome , author=. The Lancet , volume=. 2010 , publisher=

    2010

  49. [49]

    Journal of the American Statistical Association , volume=

    Statistical inference for covariate-adaptive randomization procedures , author=. Journal of the American Statistical Association , volume=. 2020 , publisher=

    2020

  50. [50]

    Journal of the American Statistical Association , volume=

    Inference under covariate-adaptive randomization , author=. Journal of the American Statistical Association , volume=. 2018 , publisher=

    2018

  51. [51]

    Contemporary clinical trials , volume=

    The use of minimization in clinical trials , author=. Contemporary clinical trials , volume=. 2010 , publisher=

    2010

  52. [52]

    Controlled Clinical Trials , volume=

    Statistical comparison of random allocation methods in cancer clinical trials , author=. Controlled Clinical Trials , volume=. 2004 , publisher=

    2004

  53. [53]

    Statistics in Medicine , volume=

    Comparison of stratification and adaptive methods for treatment allocation in an acute stroke clinical trial , author=. Statistics in Medicine , volume=. 2003 , publisher=

    2003

  54. [54]

    American Economic Journal: Applied Economics , volume=

    In pursuit of balance: Randomization in practice in development field experiments , author=. American Economic Journal: Applied Economics , volume=

  55. [55]

    and Wei, L

    Cheng, S.C. and Wei, L. J. and Ying, Z. , title=. Biometrika , volume=

  56. [56]

    Handbook of Development Economics , volume=

    Using randomization in development economics research: A toolkit , author=. Handbook of Development Economics , volume=. 2007 , publisher=

    2007

  57. [57]

    2015 , publisher=

    Randomization in Clinical Trials: Theory and Practice , author=. 2015 , publisher=

    2015

  58. [58]

    Censoring and stochastic integrals,

    Gill, Richard D , volume=. Censoring and stochastic integrals,. 1980 , publisher=

    1980

  59. [59]

    The Annals of Statistics , volume=

    Large sample behavior of the product-limit estimator on the whole line , author=. The Annals of Statistics , volume=

  60. [60]

    arXiv preprint arXiv:2004.02994 , year=

    On the theory of covariate-adaptive designs , author=. arXiv preprint arXiv:2004.02994 , year=

    Pith/arXiv arXiv 2004

  61. [61]

    The Journal of Alternative & Complementary Medicine , Year =

    Beyond randomization , Author =. The Journal of Alternative & Complementary Medicine , Year =

  62. [62]

    Statistica Neerlandica , Year =

    A covariate-adjusted adaptive design for two-stage clinical trials with survival data , Author =. Statistica Neerlandica , Year =

  63. [63]

    Controlled Clinical Trials , Year =

    Adaptive allocation in randomized controlled trials , Author =. Controlled Clinical Trials , Year =

  64. [64]

    Biometrika , Year =

    Linear regression with censored data , Author =. Biometrika , Year =

  65. [65]

    Analysis of Survival Data , Author =

  66. [66]

    Draft guidance , Year =

    Guidance for industry: Adaptive design clinical trials for drugs and biologics , Author =. Draft guidance , Year =

  67. [67]

    Computational Statistics & Data Analysis , Year =

    Validity and power of tests when groups have been balanced for prognostic factors , Author =. Computational Statistics & Data Analysis , Year =

  68. [68]

    Statistics in Medicine , Year =

    Dynamic balancing randomization in controlled clinical trials , Author =. Statistics in Medicine , Year =

  69. [69]

    Clinical Investigation , Year =

    Statistical inference of adaptive randomized clinical trials for personalized medicine , Author =. Clinical Investigation , Year =

  70. [70]

    The Annals of Statistics , Year =

    Asymptotic properties of covariate-adaptive randomization , Author =. The Annals of Statistics , Year =

  71. [71]

    Biometrika , Year =

    Rank-based inference for the accelerated failure time model , Author =. Biometrika , Year =

  72. [72]

    The Statistical Analysis of Failure Time Data , Author =

  73. [73]

    The Annals of Statistics , Year =

    Regression analysis with randomly right-censored data , Author =. The Annals of Statistics , Year =

  74. [74]

    The Annals of Statistics , Year =

    Large sample theory of a modified Buckley--James estimator for regression analysis with censored data , Author =. The Annals of Statistics , Year =

  75. [75]

    The Lancet Neurology , Year =

    Restenosis after carotid artery stenting and endarterectomy: A secondary analysis of CREST, a randomised controlled trial , Author =. The Lancet Neurology , Year =

  76. [76]

    2004 , publisher=

    Elements of large-sample theory , author=. 2004 , publisher=

    2004

  77. [77]

    Statistics in Medicine , Year =

    Generalized multidimensional dynamic allocation method , Author =. Statistics in Medicine , Year =

  78. [78]

    Statistics in Biosciences , Year =

    Robust inferences for covariate effects on survival time with censored linear regression models , Author =. Statistics in Biosciences , Year =

  79. [79]

    Journal of Statistical Planning and Inference , Year =

    Semiparametric inference for the accelerated life model with time-dependent covariates , Author =. Journal of Statistical Planning and Inference , Year =

  80. [80]

    Journal of the American Statistical Association , volume=

    Testing hypotheses of covariate-adaptive randomized clinical trials , author=. Journal of the American Statistical Association , volume=. 2015 , publisher=

    2015

Showing first 80 references.