Response adaptive designs with asymptotic optimality

Yanqing Yi, Xuan Li

Research output: Contribution to journalArticle

Abstract

This article discusses the asymptotic optimality of statistical inference for response-adaptive designs, which has ethical advantages over traditional methods for clinical trials. The upper bound of statistical power of asymptotically level α tests is derived and the Wald statistic is shown to be asymptotically optimal in terms of achieving the upper bound of the asymptotic power. The rates of coverage error probability of the confidence interval are proven to depend on the convergence rate of the allocation proportions for non-normally distributed responses. When the response density functions are normal density functions, it is proven that the coverage error probability and type I error rate are of the order n−1. The Canadian Journal of Statistics 46: 458–469; 2018

Original languageEnglish (US)
Pages (from-to)458-469
Number of pages12
JournalCanadian Journal of Statistics
Volume46
Issue number3
DOIs
StatePublished - Sep 1 2018

Fingerprint

Asymptotic Optimality
Adaptive Design
Coverage Probability
Error Probability
Density Function
Wald Statistic
Upper bound
Asymptotic Power
Normal Function
Statistical Power
Type I Error Rate
Response Function
Asymptotically Optimal
Statistical Inference
Clinical Trials
Confidence interval
Rate of Convergence
Proportion
Statistics
Asymptotic optimality

Keywords

  • Confidence interval
  • coverage error probability
  • most powerful test
  • order of type I error rate
  • response adaptive designs

Cite this

Response adaptive designs with asymptotic optimality. / Yi, Yanqing; Li, Xuan.

In: Canadian Journal of Statistics, Vol. 46, No. 3, 01.09.2018, p. 458-469.

Research output: Contribution to journalArticle

@article{2f3092369eb949099f859718036b84c0,
title = "Response adaptive designs with asymptotic optimality",
abstract = "This article discusses the asymptotic optimality of statistical inference for response-adaptive designs, which has ethical advantages over traditional methods for clinical trials. The upper bound of statistical power of asymptotically level α tests is derived and the Wald statistic is shown to be asymptotically optimal in terms of achieving the upper bound of the asymptotic power. The rates of coverage error probability of the confidence interval are proven to depend on the convergence rate of the allocation proportions for non-normally distributed responses. When the response density functions are normal density functions, it is proven that the coverage error probability and type I error rate are of the order n−1. The Canadian Journal of Statistics 46: 458–469; 2018",
keywords = "Confidence interval, coverage error probability, most powerful test, order of type I error rate, response adaptive designs",
author = "Yanqing Yi and Xuan Li",
year = "2018",
month = "9",
day = "1",
doi = "10.1002/cjs.11460",
language = "English (US)",
volume = "46",
pages = "458--469",
journal = "Canadian Journal of Statistics",
issn = "0319-5724",
publisher = "Statistical Society of Canada",
number = "3",

}

TY - JOUR

T1 - Response adaptive designs with asymptotic optimality

AU - Yi, Yanqing

AU - Li, Xuan

PY - 2018/9/1

Y1 - 2018/9/1

N2 - This article discusses the asymptotic optimality of statistical inference for response-adaptive designs, which has ethical advantages over traditional methods for clinical trials. The upper bound of statistical power of asymptotically level α tests is derived and the Wald statistic is shown to be asymptotically optimal in terms of achieving the upper bound of the asymptotic power. The rates of coverage error probability of the confidence interval are proven to depend on the convergence rate of the allocation proportions for non-normally distributed responses. When the response density functions are normal density functions, it is proven that the coverage error probability and type I error rate are of the order n−1. The Canadian Journal of Statistics 46: 458–469; 2018

AB - This article discusses the asymptotic optimality of statistical inference for response-adaptive designs, which has ethical advantages over traditional methods for clinical trials. The upper bound of statistical power of asymptotically level α tests is derived and the Wald statistic is shown to be asymptotically optimal in terms of achieving the upper bound of the asymptotic power. The rates of coverage error probability of the confidence interval are proven to depend on the convergence rate of the allocation proportions for non-normally distributed responses. When the response density functions are normal density functions, it is proven that the coverage error probability and type I error rate are of the order n−1. The Canadian Journal of Statistics 46: 458–469; 2018

KW - Confidence interval

KW - coverage error probability

KW - most powerful test

KW - order of type I error rate

KW - response adaptive designs

UR - http://www.scopus.com/inward/record.url?scp=85052674733&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85052674733&partnerID=8YFLogxK

U2 - 10.1002/cjs.11460

DO - 10.1002/cjs.11460

M3 - Article

AN - SCOPUS:85052674733

VL - 46

SP - 458

EP - 469

JO - Canadian Journal of Statistics

JF - Canadian Journal of Statistics

SN - 0319-5724

IS - 3

ER -