Estudio de simulación sobre la potencia y sensibilidad de dieciséis pruebas de normalidad en distintos escenarios de no normalidad
In data analysis, validating the normality assumption is crucial for determining the suitability of applying parametric methods. The objective of this research was to compare the power and sensitivity of sixteen normality tests, classified according to various aspects. The methodology involved simul...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Instituto Tecnológico Metropolitano
2025-03-01
|
| Series: | TecnoLógicas |
| Subjects: | |
| Online Access: | https://revistas.itm.edu.co/index.php/tecnologicas/article/view/3293 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In data analysis, validating the normality assumption is crucial for determining the suitability of applying parametric methods. The objective of this research was to compare the power and sensitivity of sixteen normality tests, classified according to various aspects. The methodology involved simulating data using the Fleishman contamination system. This approach allowed us to evaluate the tests under non-normality conditions across ten distributions with varying degrees of deviation from normality. The results obtained showed that tests based on correlation and regression, such as Shapiro-Wilk and Shapiro-Francia, outperform the others in power, especially for large samples and substantial deviations from normality. For moderate deviations, the D’Agostino-Pearson and skewness tests performed well, while for low deviations, the Robust Jarque- Bera and Jarque-Bera tests were the most effective. Additionally, some tests exhibited high power across multiple distribution types, such as Snedecor-Cochran and Chen-Ye, which performed well for both symmetric platykurtic and asymmetric leptokurtic distributions. These findings offer valuable insights for selecting appropriate normality tests based on sample characteristics, which improves the reliability of statistical inference. Finally, it is concluded that this research demonstrates scenarios in which the most commonly used statistical tests are not always the most effective |
|---|---|
| ISSN: | 2256-5337 |