Answers are grounded in this paper and cite their sources.
In 2022, a normality test based on the parametric seven-number summary was proposed. Its test statistic is the sum of squared standardized quantiles and was initially approximated by a chi-square distribution with seven degrees of freedom, without accounting for correlation among quantiles. Objective: To improve the test by incorporating these correlations. Two alternatives were proposed: (1) estimating the sampling distribution of the Q-statistic via bootstrap (QB), and (2) using a quadratic form with a correlation matrix of quantiles under normality (QT). Methods: The three variants (Q, QB, QT) were compared with the Shapiro-Wilk W-test in terms of accuracy (hit ratio) and statistical power. A total of 372 random samples were generated across 31 sample sizes from twelve continuous distributions. Correct classifications were compared using Cochran’s Q test, and power was assessed via repeated-measures ANOVA. Results: QB was significantly the most accurate and showed the highest average power compared to Q and QT. Its accuracy was equivalent to that of the Shapiro–Wilk W-test, although the latter outperformed all three Q variants in average power. Conclusions: QB is a suitable inferential extension of the seven-number summary for testing normality.
No takes yet. Share an insight, caveat, or question.
José Moral de la Rubia (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: