Pith. sign in

REVIEW 2 major objections 5 minor 37 references

Impact of interim analyses on Bayesian and frequentist operating characteristics of Bayesian clinical trials

T0 review · 2 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read Bayesian clinical trials are no less harmed by repeated interim analyses than frequentist ones; multiplicity still matters.

desk verdict Solid, fully reproducible simulation paper that settles a live practical confusion: Bayesian sequential trials are not multiplicity-immune once you look at either Type I or Bayesian discovery metrics under prior disagreement. read the letter →

arxiv 2607.04976 v1 pith:NSDQ4CR7 submitted 2026-07-06 stat.AP stat.ME

Clément R Massonnaud , Lucas Manns , Cédric Laouénan , André Gillibert This is my paper
classification stat.APstat.ME MSC 62P1062L0562F15
keywords BayesianstatisticsclinicaltrialsequentialanalysisinterimTypeIerrorpriorprobabilityadaptivedesignmultiplicity
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 settles a common claim that Bayesian group-sequential designs escape the multiplicity problems of looking at data repeatedly. Using simulations of normal outcomes with and without futility rules and threshold adjustments, it shows that unadjusted interim analyses inflate Type I error and also degrade Bayesian metrics: the risk of wrongly concluding efficacy and the informative value of an efficacy conclusion (positive true discovery odds). Those Bayesian metrics worsen as the number of looks grows and as the investigator’s prior diverges from a reviewer’s prior; even perfect prior agreement, more looks reduce the value of a positive stop. Frequentist error rates of Bayesian designs can be controlled by adjusting decision thresholds, but Bayesian operating characteristics cannot be guaranteed for every stakeholder because they depend on the prior. The result matters for adaptive and platform trials that rely on Bayesian interim decisions and often assume no statistical penalty for continual analysis.

What carries the argument

Positive true discovery odds (pTDO = Pr(μ>0|V)/Pr(μ≤0|V)), equal to the Bayes factor of an efficacy conclusion when the prior is symmetric about zero, evaluated by Monte Carlo under a reviewer’s prior that may differ from the investigator’s analysis prior, together with the analogous Bayesian risk A and power 1−B.

What would settle it

Repeat the simulation suite with non-conjugate logistic or survival models, unknown variance, and decisions based on expected utility or a minimal clinically important difference; if Type I error and pTDO no longer degrade with the number of looks, the central claim is false.

Watch

Extended reading notes

Core claim

Repeated interim analyses alter both frequentist and Bayesian operating characteristics of group-sequential trials, regardless of the framework used. Without multiplicity adjustment the Type I error rate rises with the number of analyses. Bayesian alternatives—risk of erroneous efficacy conclusions and the informative value of an efficacy conclusion measured by positive true discovery odds—are meaningful but degrade with more analyses and with prior divergence between investigator and reviewer; even under full prior agreement that informative value falls. Frequentist operating characteristics of Bayesian trials can be controlled by known methods, but Bayesian operating characteristics cannot

Load-bearing premise

Binary stop-or-continue decisions under a normal–normal conjugate model with known unit variance and equally spaced looks adequately represent how interim analyses affect real trial conclusions.

Editorial extensions

If this is right

  • Multiplicity must be designed for in every trial with interim analyses, Bayesian or frequentist.
  • Reviewers of positive Bayesian sequential trials should prefer pTDR or pTDO under their own prior rather than the authors’ reported posterior probability.
  • Perpetual Bayesian designs that analyse continuously cannot keep erroneous-efficacy risk controlled indefinitely; the informative value of an efficacy conclusion approaches zero.
  • The classical Type I error under the null remains a useful upper-bound metric for Bayesian designs with many looks because it is the most sceptical prior.
  • Public, reproducible simulation code is required to assess operating characteristics of Bayesian adaptive designs.

Reading between the lines

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

  • Platform and master protocols that drop or add arms under Bayesian rules will inherit the same degradation of decision value unless they hard-code multiplicity-adjusted thresholds calibrated across a range of reviewer priors.
  • Even decision-theoretic designs that maximise expected utility rather than threshold a posterior probability can still suffer sequential-selection bias in the reported posterior once early stopping is allowed.
  • Guidance that treats Bayesian designs as free of multiplicity penalties risks systematically overstating the reliability of early-stopping claims in adaptive trials.
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. The manuscript evaluates how repeated interim analyses affect both frequentist and Bayesian operating characteristics of group-sequential clinical trials under a normal–normal conjugate model. Through large Monte Carlo simulations (100 000 trials; m up to 500 equally spaced looks), the authors show that unadjusted efficacy stopping inflates Type I error for Bayesian designs much as for frequentist ones (especially under weakly informative analysis priors), that Bayesian analogues A and 1−B and the proposed positive true discovery odds (pTDO) degrade with more looks and with investigator–reviewer prior divergence, and that even under exact prior agreement pTDO falls as the number of analyses rises. Frequentist Type I error of Bayesian designs can be controlled by boundary recalibration (Pocock-style), whereas Bayesian OCs cannot be guaranteed across divergent reviewer priors. The central claim is that Bayesian trials are not immune to multiplicity, contrary to some statements in the literature.

Significance. The paper addresses a persistent and practically important confusion in adaptive and Bayesian trial design. Its contribution is clear: it separates the likelihood-principle invariance of the posterior from the operating characteristics of binary stopping rules, defines transparent Bayesian decision metrics (including pTDO), and demonstrates with reproducible code and extensive simulations that both frequentist and Bayesian OCs are altered by repeated looks. The public GitHub repository and conjugate closed forms strengthen credibility. If the qualitative conclusions hold beyond the normal model—as the authors argue in §6.6—the work should influence how Bayesian group-sequential and “perpetual” designs are planned, reported, and reviewed, and it usefully reframes Type I error as a consensus upper bound under the most sceptical prior.

major comments (2)
  1. [§5.2, Figure 4, §6.5] §5.2 and Figure 4: After calibrating e_i and n so that A = 0.05 and 1−B = 0.80 under π_investigator = π_reviewer = N(0, 0.25²), the authors correctly show that more sceptical reviewer priors inflate A and collapse pTDO. The practical implication is left somewhat open. A short discussion of whether designers should (a) control frequentist Type I as a consensus bound (§6.5(iv)), (b) pre-specify a range of reviewer priors and control worst-case A, or (c) report pTDO under a grid of σ_reviewer would make the recommendation actionable without changing the central claim.
  2. [§6.4, Supplementary Figures S74–S78] §6.4 and Supplementary Figures S74–S78: The observation that, with many looks and a fixed posterior-probability threshold, the posterior probability at stopping concentrates at the threshold itself is load-bearing for the claim that the posterior loses quantitative interpretability. Elevating a concise version of this result (and one main-text figure) would strengthen the argument that binary stopping, not conjugacy per se, drives the loss of informativeness, and would better support the advice that external reviewers should rely on pTDO rather than the reported posterior.
minor comments (5)
  1. [Table 1, §2.2] Table 1 and §2.2: Clarify that pTDO equals the Bayes factor of V for μ > 0 only when the reviewer prior is symmetric about zero (as assumed throughout the simulations); the general odds form is already stated but easy to miss.
  2. [Figure 1] Figure 1B and analogous power-vs-α plots: The left-to-right ordering of points by number of analyses is described in the caption but hard to read; a small annotation or colour gradient for m would help.
  3. [§2.4] §2.4 / Table 3: State explicitly that the unit-variance assumption is without loss of generality (by rescaling) so that readers do not misread the design as restricted to σ = 1 endpoints.
  4. [References] References: The FDA Bayesian guidance link is dated “accessed 14 June 2026”; confirm the access date and citation format for the journal.
  5. [Supplementary tables] Supplementary tables S1–S6 are dense; consider moving a single representative pTDO table (e.g., under prior agreement) into the main text to illustrate the decline with m without forcing readers into the appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: operating characteristics are Monte-Carlo estimates under independently generated data and priors, not algebraic restatements of the decision rules or fitted inputs.

full rationale

The paper's central claims rest on simulation studies that generate data under fixed nulls (μ=0), alternatives (μ=0.1112), or reviewer priors π_reviewer=N(0,σ_reviewer²), apply investigator decision rules (posterior probability thresholds or p-value boundaries), and estimate Type I error, power, A, 1-B, pTDR, pFDR and pTDO as empirical frequencies (Methods §§2.3–2.4; Tables 4–7; Figs. 1–4). These quantities are not defined in terms of one another, nor are parameters fitted to a subset of the same outcomes and then re-presented as predictions. pTDO is obtained from Bayes' theorem applied to the simulated joint distribution of (μ,V) and equals (1-B)/A under the paper's symmetric priors; it is not forced by construction of the stopping boundary. Boundary recalibration (Pocock-style or numerical optimisation of e_i and n) is performed to hit pre-specified targets and then re-evaluated under mismatched priors; the degradation is an observed simulation result, not an identity. Self-citations (e.g., the authors' platform-trial review) supply only methodological background and are not load-bearing premises. The normal–normal conjugate model is an explicit modelling choice, not a uniqueness theorem imported from prior work by the same authors. Consequently the derivation chain contains no self-definitional steps, no fitted-input-as-prediction, and no circular self-citation.

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

The central claim rests on a standard sequential-testing setup plus a dual-prior evaluation device (investigator vs reviewer). No new physical entities are postulated. Free parameters are design knobs (prior SDs, thresholds, n, μ1) chosen for illustration, not fitted to real trial data. Domain assumptions are classical i.i.d. normal observations and conjugate updating; the main ad-hoc modeling choice is treating binary efficacy conclusions under fixed posterior-probability thresholds as the primary decision metric.

free parameters (5)
  • Maximum sample size n = 500 (unadjusted); re-derived for controlled scenarios
    Fixed at 500 for unadjusted scenarios so that a single frequentist analysis has ~80% power at μ1; later re-optimized for controlled designs.
  • Alternative effect μ1 = 0.1112
    Chosen so a single 5%-level test at n=500 has 80% power; anchors power comparisons.
  • Investigator prior SD σ_investigator = grid in {0.001, 0.01, 0.05, 0.1, 0.25, 1, 10, 100}
    Grid of values spanning strongly informative to essentially noninformative priors used for analysis rules.
  • Reviewer prior SD σ_reviewer = same grid as investigator
    Grid used to average Bayesian OCs; mismatch with investigator prior drives the sensitivity claims.
  • Efficacy/futility thresholds e_i, f_i = 0.05 unadjusted; scenario-specific when controlled
    Unadjusted e_i=0.05 and various f_i; later Pocock or simulation-optimized thresholds for controlled designs.
assumptions (5)
  • domain assumption Observations X_i are i.i.d. N(μ,1) with known variance (w.l.o.g. by rescaling).
    Stated in §2.3; enables closed-form conjugate posteriors and asymptotic transfer to other estimators via CLT (§6.6).
  • domain assumption Investigator prior is normal N(0, σ_investigator²), conjugate to the likelihood.
    §2.3 equations (1)–(2); yields analytic posterior mean/variance and exact frequentist equivalence as σ→∞.
  • domain assumption Trial decisions are binary efficacy/futility conclusions from comparing posterior Pr(μ>0|data) or one-sided p-values to fixed thresholds at pre-specified looks.
    §2.1–2.3; the paper explicitly focuses on decision-related metrics rather than estimation bias/MSE.
  • ad hoc to paper Bayesian operating characteristics are defined by averaging over a reviewer prior π_reviewer that may differ from the analysis prior.
    §2.2 introduces the dual-prior device; standard Bayesian decision theory uses one prior, but the dual-prior framing is the paper’s evaluation construct for stakeholder disagreement.
  • standard math Standard group-sequential multiplicity methods (e.g., Pocock boundaries) control frequentist Type I error under the stated correlation structure of sequential z-statistics.
    Used in §5.1 and supplement §1.2; classical result of Pocock and multivariate normal boundary calculation.
invented entities (1)
  • Positive true discovery odds (pTDO)
    purpose: Summarize the informative value of an efficacy conclusion as Pr(μ>0|V)/Pr(μ≤0|V), equal to the Bayes factor of V for μ>0 when the prior is symmetric about zero.
    Defined in §2.2 following Storey’s pFDR/pTDR; not a physical entity but a named operating characteristic central to the Bayesian claims. Independent evidence is definitional/mathematical rather than empirical outside the simulations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Impact of interim analyses on Bayesian and frequentist operating characteristics of Bayesian clinical trials." pith.science (2026). https://pith.science/paper/NSDQ4CR7

@misc{pith2026260704976,
  author       = {Pith},
  title        = {Pith review of: Impact of interim analyses on Bayesian and frequentist operating characteristics of Bayesian clinical trials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NSDQ4CR7}},
  note         = {Machine review of arXiv:2607.04976}
}
read the original abstract

While Bayesian methods are increasingly used in clinical research, confusion persists as to whether Bayesian designs are affected by repeated interim analyses, and how such effects should be evaluated. We aimed to clarify this question by evaluating both frequentist and Bayesian operating characteristics of group-sequential trials. We conducted simulation studies with normally distributed outcomes, examining designs with repeated analyses, with and without futility stopping rules and multiplicity adjustments. We show that repeated interim analyses alter both frequentist and Bayesian operating characteristics in group-sequential trials, regardless of the inferential framework adopted. Without proper adjustment for multiplicity, the Type I error rate increases with the number of analyses. Bayesian operating characteristics such as the risk of erroneous conclusions and the informative value of an efficacy conclusion are meaningful alternatives to classical frequentist metrics, but they are sensitive to prior divergence between stakeholders, and this sensitivity grows with the number of analyses. Even under full prior agreement, the informative value of an efficacy conclusion is reduced. While it is possible to control frequentist operating characteristics of Bayesian trials with appropriate methods, it is not possible to guarantee such control for Bayesian operating characteristics because they are prior-dependent and different stakeholders may adopt different priors. Contrary to claims that Bayesian inference is immune to multiplicity, our results show that Bayesian clinical trials are no less affected by repeated analyses than frequentist ones, regardless of the framework used to evaluate them.

Figures

Figures reproduced from arXiv: 2607.04976 by the authors.

Figure 1
Figure 1. Type I error rate and power of frequentist and Bayesian trial designs with efficacy stopping rules (frequentist: one-sided 𝒑𝒗𝒂𝒍𝒖𝒆 < 𝟎. 𝟎𝟓, Bayesian: 𝑷𝒓(𝝁 > 𝟎|𝑿𝟏,…,𝒌𝒊 ) > 𝟎. 𝟗𝟓,𝒊 ∈ {𝟏, … , 𝒎}), no futility stopping rules, and no correction for multiplicity of analyses. 𝝈𝒊𝒏𝒗𝒆𝒔𝒕𝒊𝒈𝒂𝒕𝒐𝒓 is the standard deviation of the prior 𝓝(𝟎,𝝈𝒊𝒏𝒗𝒆𝒔𝒕𝒊𝒈𝒂𝒕𝒐𝒓 𝟐 ) used for the Bayesian analyses. Panel A: Type I error rate (y-axis) as a fu… view at source ↗
Figure 2
Figure 2. Type I error rate and power of frequentist and Bayesian trial designs with efficacy stopping rules (frequentist: one-sided 𝒑𝒗𝒂𝒍𝒖𝒆 < 𝟎. 𝟎𝟓, Bayesian: 𝑷𝒓(𝝁 > 𝟎|𝑿𝟏,…,𝒌𝒊 ) > 𝟎. 𝟗𝟓,𝒊 ∈ {𝟏, … , 𝒎}), and added futility stopping rules (frequentist: one-sided 𝒑𝒗𝒂𝒍𝒖𝒆 > 𝟎. 𝟗𝟎, Bayesian: 𝑷𝒓(𝝁 > 𝟎|𝑿𝟏,…,𝒌𝒊 ) < 𝟎. 𝟏𝟎,𝒊 ∈ {𝟏, … , 𝒎}), with no correction for multiplicity of analyses. 𝝈𝒊𝒏𝒗𝒆𝒔𝒕𝒊𝒈𝒂𝒕𝒐𝒓 is the standard deviation of the pr… view at source ↗
Figure 3
Figure 3. Probability of concluding efficacy (binary event 𝑽) conditional to 𝝁 > 𝟎 (panels A, B, C) and 𝝁 ≤ 𝟎 (panels D, E, F), according the number of analyses, in Bayesian trials with efficacy stopping rules 𝑷𝒓(𝝁 > 𝟎|𝑿𝟏,…,𝒌𝒊 ) > 𝟎. 𝟗𝟓,𝒊 ∈ {𝟏, … , 𝒎}, no futility stopping rules, and no correction for multiplicity of analyses. 𝝈𝒊𝒏𝒗𝒆𝒔𝒕𝒊𝒈𝒂𝒕𝒐𝒓 is the standard deviation of the prior 𝛑𝒊𝒏𝒗𝒆𝒔𝒕𝒊𝒈𝒂𝒕𝒐𝒓 = 𝓝(𝟎,𝝈𝒊𝒏𝒗𝒆𝒔𝒕𝒊𝒈𝒂𝒕𝒐𝒓 𝟐 ) used for … view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 3 canonical work pages

  1. [1]

    𝟎𝟓 and 𝑷𝒓𝛍∼𝑵(𝟎,𝟎.𝟐𝟓𝟐)( 𝑽 ∣∣ 𝛍 > 𝟎 ) = 𝟎. 𝟖𝟎

  2. [2]

    non-informative

    Discussion In this work, we show that repeated interim analyses of accumulating data in clinical trials, without adjustment for multiple testing, fundamentally alter their operating characteristics, regardless of whether a frequentist or Bayesian framework is adopted for the design of the interim analyses, the interpretation of the operating characteristi...

  3. [3]

    Conclusion Consistently with the current literature, we confirmed that group-sequential Bayesian trials are not immune to Type I error rate inflation. Moreover, we showed that Bayesian operating characteristics such as the risk of erroneous conclusions and the informative value of an efficacy conclusion degrade as the number of interim analyses increases ...

  4. [4]

    The testing of statistical hypotheses in relation to probabilities a priori

    Neyman J, Pearson ES. The testing of statistical hypotheses in relation to probabilities a priori. Math Proc Camb Phil Soc 1933; 29: 492–510

    1933

  5. [5]

    Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

    Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B (Methodological) 1995; 57: 289–300

    1995

  6. [6]

    A Sharper Bonferroni Procedure for Multiple Tests of Significance

    Hochberg Y. A Sharper Bonferroni Procedure for Multiple Tests of Significance. Biometrika 1988; 75: 800–802

    1988

  7. [7]

    Multiple Comparison Procedures

    Hochberg Y, Tamhane AC. Multiple Comparison Procedures. 1st edn. Wiley. Epub ahead of print 21 September 1987. DOI: 10.1002/9780470316672

    work page doi:10.1002/9780470316672 1987

  8. [8]

    A Simple Sequentially Rejective Multiple Test Procedure

    Holm S. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics 1979; 6: 65–70

    1979

Show all 37 references
  1. [9]

    The problem of multiple comparisons

    Tukey JW. The problem of multiple comparisons. Princeton University

  2. [10]

    Multiple testing in clinical trials

    Bauer P. Multiple testing in clinical trials. Statist Med 1991; 10: 871–890

    1991

  3. [11]

    Repeated Significance Tests on Accumulating Data

    Armitage P, McPherson CK, Rowe BC. Repeated Significance Tests on Accumulating Data. Journal of the Royal Statistical Society Series A (General) 1969; 132: 235

    1969

  4. [12]

    A case for bayesianism in clinical trials

    Berry DA. A case for bayesianism in clinical trials. Statistics in Medicine 1993; 12: 1377–1393

    1993

  5. [13]

    Bayesian clinical trials

    Berry DA. Bayesian clinical trials. Nat Rev Drug Discov 2006; 5: 27–36

    2006

  6. [14]

    Bayesian adaptive trials offer advantages in comparative effectiveness trials: an example in status epilepticus

    Connor JT, Elm JJ, Broglio KR. Bayesian adaptive trials offer advantages in comparative effectiveness trials: an example in status epilepticus. Journal of Clinical Epidemiology 2013; 66: S130–S137

    2013

  7. [15]

    A practical guide to adopting Bayesian analyses in clinical research

    Gunn-Sandell LB, Bedrick EJ, Hutchins JL, et al. A practical guide to adopting Bayesian analyses in clinical research. Journal of Clinical and Translational Science 2024; 8: e3

    2024

  8. [16]

    Characteristics, Progression, and Output of Randomized Platform Trials: A Systematic Review

    Griessbach A, Schönenberger CM, Taji Heravi A, et al. Characteristics, Progression, and Output of Randomized Platform Trials: A Systematic Review. JAMA Netw Open 2024; 7: e243109

    2024

  9. [17]

    Characteristics, design, and statistical methods in platform trials: a systematic review

    Massonnaud CR, Schönenberger CM, Chiaborelli M, et al. Characteristics, design, and statistical methods in platform trials: a systematic review. Journal of Clinical Epidemiology 2025; 184: 111827

    2025

  10. [18]

    Application of Bayesian approaches in drug development: starting a virtuous cycle

    Ruberg SJ, Beckers F, Hemmings R, et al. Application of Bayesian approaches in drug development: starting a virtuous cycle. Nat Rev Drug Discov 2023; 22: 235–250

    2023

  11. [19]

    Bayesian analysis in confirmatory clinical trials: A narrative review and discussion of current practice

    Turner RM, Tweed CD, Duong T, et al. Bayesian analysis in confirmatory clinical trials: A narrative review and discussion of current practice. Clinical Trials 2026; 17407745261437669

    2026

  12. [20]

    Interim Analysis in Clinical Trials: The Role of the Likelihood Principle

    Berry DA. Interim Analysis in Clinical Trials: The Role of the Likelihood Principle. The American Statistician 1987; 41: 117–122

    1987

  13. [21]

    A Bayesian group sequential design for a multiple arm randomized clinical trial

    Rosner GL, Berry DA. A Bayesian group sequential design for a multiple arm randomized clinical trial. Statistics in Medicine 1995; 14: 381–394

    1995

  14. [22]

    Do we need to adjust for interim analyses in a Bayesian adaptive trial design? BMC Med Res Methodol 2020; 20: 150

    Ryan EG, Brock K, Gates S, et al. Do we need to adjust for interim analyses in a Bayesian adaptive trial design? BMC Med Res Methodol 2020; 20: 150

    2020

  15. [23]

    Comparison of Bayesian with group sequential methods for monitoring clinical trials

    Freedman LS, Spiegelhalter DJ. Comparison of Bayesian with group sequential methods for monitoring clinical trials. Controlled Clinical Trials 1989; 10: 357–367

    1989

  16. [24]

    Problems of multiplicity in clinical trials

    Simon R. Problems of multiplicity in clinical trials. Journal of Statistical Planning and Inference 1994; 42: 209–221

    1994

  17. [25]

    Comparison of Bayesian and frequentist group-sequential clinical trial designs

    Stallard N, Todd S, Ryan EG, et al. Comparison of Bayesian and frequentist group-sequential clinical trial designs. BMC Med Res Methodol 2020; 20: 4

    2020

  18. [26]

    Bayesian Analytical Methods in Cardiovascular Clinical Trials: Why, When, and How

    Heuts S, Kawczynski MJ, Sayed A, et al. Bayesian Analytical Methods in Cardiovascular Clinical Trials: Why, When, and How. Canadian Journal of Cardiology 2025; 41: 30–44

    2025

  19. [27]

    FDA. Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products, https://www.fda.gov/regulatory-information/search-fda-guidance-documents/use-bayesian- methodology-clinical-trials-drug-and-biological-products (accessed 14 June 2026)

    2026

  20. [28]

    The Positive False Discovery Rate: A Bayesian Interpretation and the q-Value

    Storey JD. The Positive False Discovery Rate: A Bayesian Interpretation and the q-Value. The Annals of Statistics 2003; 31: 2013–2035

    2003

  21. [29]

    R: A Language and Environment for Statistical Computing

    R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing, https://www.R-project.org/ (2023)

    2023

  22. [30]

    Group sequential methods in the design and analysis of clinical trials

    Pocock SJ. Group sequential methods in the design and analysis of clinical trials. Biometrika 1977; 64: 191–199

    1977

  23. [31]

    The choice of sequential boundaries based on the concept of power spending

    Bauer P. The choice of sequential boundaries based on the concept of power spending. Controlled Clinical Trials 1991; 12: 637

    1991

  24. [32]

    A Multiple Testing Procedure for Clinical Trials

    O’Brien PC, Fleming TR. A Multiple Testing Procedure for Clinical Trials. Biometrics 1979; 35: 549–556

    1979

  25. [33]

    Interim analysis: The alpha spending function approach

    Demets DL, Lan KKG. Interim analysis: The alpha spending function approach. Statist Med 1994; 13: 1341–1352

    1994

  26. [34]

    Group Sequential Designs: A Tutorial

    Lakens D, Pahlke F, Wassmer G. Group Sequential Designs: A Tutorial. Preprint, PsyArXiv. Epub ahead of print 27 January 2021. DOI: 10.31234/osf.io/x4azm

    2021 doi

  27. [35]

    Control of Type I Error Rates in Bayesian Sequential Designs

    Shi H, Yin G. Control of Type I Error Rates in Bayesian Sequential Designs. Bayesian Anal; 14. Epub ahead of print 1 June 2019. DOI: 10.1214/18-BA1109

    2019 doi

  28. [36]

    A Bayesian sequential design using alpha spending function to control type I error

    Zhu H, Yu Q. A Bayesian sequential design using alpha spending function to control type I error. Stat Methods Med Res 2017; 26: 2184–2196

    2017

  29. [37]

    analysis.idx

    Amrhein V, Greenland S, McShane B. Scientists rise up against statistical significance. Nature 2019; 567: 305–307. Supplementary material for the article Impact of interim analyses on Bayesian and frequentist operating characteristics of Bayesian clinical trials Clément R. Mas...

    2019

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.