Wednesday, 4 March 2026

Asymptotic and Bootstrap Implementation of the D'Agostino-Berlanger-D'Agostino K2 Normality Test in R | Chapter 3 | Mathematics and Computer Science: Research Updates Vol. 9

 

The K² test could be one of the best tests for assessing normality, yet its use is limited, likely because it is not commonly included in standard statistical software, despite being implementable in R through the moments package. Moreover, its asymptotic approximation has been questioned for small samples, and no bootstrap version currently exists, even though it is feasible in R. This simulation study aimed to: (1) verify the linear independence and nonlinear relationship between √b₁ and b₂; (2) develop an R script for the K² test in both asymptotic and bootstrap versions; (3) assess the fit of the bootstrap distribution of the K² statistic to a chi-square distribution with two degrees of freedom; (4) compare the power of both implementations against non-normal distributions; and (5) contrast the bootstrap version of K² with the Shapiro–Wilk test in small samples. A Monte Carlo simulation with 10,000 replications was conducted, using 16 non-normal distributions as alternative hypotheses and sample sizes (n) ranging from 20 to 2,000 in increments of 20. Linear independence and a parabolic relationship between √b₁ and b₂ were confirmed, and the R script was verified to be functional. The script is available for download as a Word document from a GitHub repository. The bootstrap distribution of K² converged to a chi-square distribution for n ≥ 120. The asymptotic version of K² and the Shapiro–Wilk test showed greater power than the bootstrap version, except for mesokurtic asymmetric distributions. Bootstrap implementation is recommended in these cases for n < 120, while the asymptotic version is generally more powerful and appropriate for n ≥ 20. The developed R script is highly useful for assessing the normality assumption required by many parametric tests, such as t-tests and F-tests for comparing means and variances, as well as for characterising the distribution of a sample of quantitative data; therefore, its use is recommended for these purposes.

 

Author(s) Details

José Moral de la Rubia
School of Psychology, UANL, Mexico.

 

Please see the book here :- https://doi.org/10.9734/bpi/mcsru/v9/6969

 

No comments:

Post a Comment