IMPACT OF SAMPLE SIZE, DIMENSIONALITY, AND NORMALITY ASSUMPTIONS ON SPHERICITY TESTS: A COMPARATIVE ANALYSIS OF BARTLETT’S AND MAUCHLY’S METHODS

Authors

  • RIDWANULLAHI, Abdulkareem Tope Author
  • USMAN Abubakar Author
  • SULEIMAN, Ganiyu Olayiwola Author

Keywords:

Sphericity, Mauchly's Test, Bartlett's Test, Type I error, Statistical power, Sample size, Dimensionality, Simulation study, Multivariate normal distribution, Repeated measures ANOVA

Abstract

This research investigates the performance of two widely used sphericity tests Mauchly's Test and Bartlett's Test under various simulated conditions. The study examines the impact of sample size, dimensionality, and distribution type on the effectiveness of these tests in detecting violations of sphericity, a key assumption in repeated measures ANOVA. Simulations were conducted using datasets with different sample sizes (30, 50, 100, 200, 500, and 1000) and varying numbers of variables (3, 5, 10, 15, 20, and 30), drawn from normal, exponential, and lognormal distributions. Both sphericity and non-sphericity conditions were tested, and each condition was replicated 10,000 times to ensure reliability. The primary performance metrics evaluated were Type I error rates, statistical power, and test stability. Type I error rates reflect the likelihood of incorrectly rejecting the null hypothesis when sphericity holds, while statistical power measures the ability to detect true violations of sphericity. Results show that both Mauchly's and Bartlett's Tests exhibit increasing statistical power and stability with larger sample sizes and higher dimensionality. Bartlett's Test consistently demonstrated lower Type I error rates particularly in smaller samples making it more conservative in rejecting the null hypothesis of sphericity. Conversely, Mauchly's Test showed slightly higher Type I error rates but offered comparable or superior statistical power in large sample sizes and high-dimensional settings. Overall, while both tests are effective, Bartlett’s Test may be preferable for small, normally distributed datasets due to its error control, whereas Mauchly’s Test proves advantageous in complex, high-dimensional data contexts for its greater sensitivity. Based on the evidence, stakeholders including applied researchers, statistical consultants, and policymakers are advised to prioritize Bartlett’s Test in small-sample research for its conservative reliability and adopt Mauchly’s Test in large-scale, high-dimensional applications. In all cases, appropriate corrections should be applied when the normality assumption is violated to ensure valid and robust statistical inferences.

Downloads

Published

2025-07-04

Similar Articles

31-40 of 88

You may also start an advanced similarity search for this article.