Penalized power approach to compare the power of the tests when Type I error probabilities are different

Çavuş, Mustafa; Yazıcı, Berna; Sezer, Ahmet

doi:10.1080/03610918.2019.1588310

Cited by 10 publications

(10 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Maruo et al (2017) also show that this method controls the type I error of the statistical test for the model median difference, and it has moderate or high performance for power compared with the existing methods (ordinary MMRM, MMRM for the log-transformed outcome, etc.) from their simulation studies (the simulation program is provided in https://github.com/kzkzmr/Maruo_2017_StatMed_Simulation with penalized power results proposed by Cavus et al (2019)). Thus, bcmmrm function analysis results with high power and high interpretability for longitudinal randomized clinical trials with skewed outcomes.…”

Section: Summary and Discussionmentioning

confidence: 99%

bcmixed: A Package for Median Inference on Longitudinal Data with the Box–Cox Transformation

Maruo¹,

Ishii²,

Yamaguchi³

et al. 2021

The R Journal

View full text Add to dashboard Cite

This article illustrates the use of the bcmixed package and focuses on the two main functions: bcmarg and bcmmrm. The bcmarg function provides inference results for a marginal model of a mixed effect model using the Box-Cox transformation. The bcmmrm function provides model median inferences based on the mixed effect models for repeated measures analysis using the Box-Cox transformation for longitudinal randomized clinical trials. Using the bcmmrm function, analysis results with high power and high interpretability for treatment effects can be obtained for longitudinal randomized clinical trials with skewed outcomes. Further, the bcmixed package provides summarizing and visualization tools, which would be helpful to write clinical trial reports.The remainder of this manuscript is organized as follows: In section Methods, we describe the methods proposed by . Then, we illustrate the usage of the bcmixed package with the example data in section Illustrations. Finally, our contributions are summarized in section Summary and discussion. MethodsWe briefly introduce the method proposed by in . The detailed expressions can be found in .

show abstract

Section: Summary and Discussionmentioning

confidence: 99%

bcmixed: A Package for Median Inference on Longitudinal Data with the Box–Cox Transformation

Maruo¹,

Ishii²,

Yamaguchi³

et al. 2021

The R Journal

View full text Add to dashboard Cite

show abstract

“…However, Cavus et al (2021) suggested that Lloyd's adjusted power does not account for how close the test size is to the nominal size 𝛼 * . Instead, Cavus et al (2021) proposed to penalize the test power by the degree of deviation of the test size from 𝛼 * . The proposed penalized power is estimated by…”

Section: Power Adjusted or Penalized For Sizementioning

confidence: 99%

“…However, Cavus et al. (2021) suggested that Lloyd's adjusted power does not account for how close the test size is to the nominal size

\alpha ^*

. Instead, Cavus et al.…”

Section: Power Adjusted or Penalized For Sizementioning

confidence: 99%

“…When the size of the tests being compared is either different from one another or from the nominal size

\alpha ^*

, comparing the power of the tests alone without accounting for the differences in size may be misleading. A well‐established but underutilized approach to compare tests with regard to power and size simultaneously is to estimate power adjusted or penalized for size (Cavus et al., 2021; Lloyd, 2005a). The general principle is to adjust or penalize according to the deviation of test size from

\alpha ^*

.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Relative likelihood ratios for neutral comparisons of statistical tests in simulation studies

Huang

Trinquart

2023

Biometrical J

View full text Add to dashboard Cite

When comparing the performance of two or more competing tests, simulation studies commonly focus on statistical power. However, if the size of the tests being compared are either different from one another or from the nominal size, comparing tests based on power alone may be misleading. By analogy with diagnostic accuracy studies, we introduce relative positive and negative likelihood ratios to factor in both power and size in the comparison of multiple tests. We derive sample size formulas for a comparative simulation study. As an example, we compared the performance of six statistical tests for small‐study effects in meta‐analyses of randomized controlled trials: Begg's rank correlation, Egger's regression, Schwarzer's method for sparse data, the trim‐and‐fill method, the arcsine‐Thompson test, and Lin and Chu's combined test. We illustrate that comparing power alone, or power adjusted or penalized for size, can be misleading, and how the proposed likelihood ratio approach enables accurate comparison of the trade‐off between power and size between competing tests.

show abstract

“…However, any comparison of the powers is invalid when Type I error probabilities are different. Cavus et al (2021) proposed the penalized power approach in (7) to compare the power of the tests when Type I error probabilities are different. where β is Type II error rate, α i is Type I error of the test and α 0 is the nominal level.…”

Section: Monte-carlo Simulation Studymentioning

confidence: 99%

A revised generalized F-test for testing the equality of group means under non-normality caused by skewness

Çavuş¹,

Yazıcı²,

Sezer³

2021

Communications Faculty of Science University of Ankara Series A1Mathematics and Statistics

Self Cite

View full text Add to dashboard Cite

The non-normality may occur in the data due to several reasons such as the presence of the outlier or skewness. It leads to lose the power and fail control Type I error probability of the tests which are used to test the equality of the group means under heteroscedasticity. To overcome this problem, a revised generalized F-test (RGF) is proposed to test the equality of group means under heteroscedasticity in which non-normality is caused by skewness in this study. An extensive Monte-Carlo simulation study is conducted to investigate and compare the performance of the proposed test with non-parametric alternatives under several values of skewness, and different number of groups. The proposed RGF is the best choice in the high level of skewness for k = 3, 4, 5. The Kruskal-Wallis test shows better performance than the others in small and moderate sample sizes for k = 6, and 7. It is shown that the proposed RGF test is superior than the non-parametric alternatives in the most of the conditions.

show abstract

Penalized power approach to compare the power of the tests when Type I error probabilities are different

Cited by 10 publications

References 34 publications

bcmixed: A Package for Median Inference on Longitudinal Data with the Box–Cox Transformation

bcmixed: A Package for Median Inference on Longitudinal Data with the Box–Cox Transformation

Relative likelihood ratios for neutral comparisons of statistical tests in simulation studies

A revised generalized F-test for testing the equality of group means under non-normality caused by skewness

Contact Info

Product

Resources

About