|
Efficiency of Pearson, Spearman and Kendall’s Correlation Coefficients When Data is Non-normal Distributed |
|---|---|
| รหัสดีโอไอ | |
| Creator | Chaninan Pruekpramool |
| Title | Efficiency of Pearson, Spearman and Kendall’s Correlation Coefficients When Data is Non-normal Distributed |
| Contributor | Nattika Jaroentaku, Siwachoat Srisuttiyakorn |
| Publisher | Centre for Education Innovation, Print and Online Media |
| Publication Year | 2563 |
| Journal Title | An Online Journal of Education, Chulalongkorn University |
| Journal Vol. | 15 |
| Journal No. | 2 |
| Page no. | OJED1502040 (16 pages) |
| Keyword | non-normal distribution, efficiency, Pearson’s correlation coefficient, Spearman’s correlation coefficient, Kendall’s correlation coefficient |
| URL Website | https://so01.tci-thaijo.org/index.php/OJED |
| Website title | An Online Journal of Education, Chulalongkorn University |
| ISSN | ISSN 1905-4491 (Online) |
| Abstract | This research aimed to compare the efficiency in estimation and hypothesis testing of correlation analysis among Pearson, Spearman and Kendall’s correlation coefficients when the data were non-normal distributed. The Monte Carlo data simulation was used in this study to generate the data from a log-normal probability distribution. The total numbers of simulations were 288 situations repeating 1,000 rounds in each situation. The efficiency in estimation was considered from Relative Bias (RB) and Monte Carlo Standard Errors (MCSE). The efficiency in hypothesis testing was considered from the probability of a type I error and power of testing. The results revealed that (1) Pearson’s correlation coefficients had tended to present the estimating values with the lowest relative bias while Kendall’s correlation coefficients had tended to present the estimating values with the highest relative bias. (2) Pearson’s and Spearman’s correlation coefficients had tended to present the values of Monte Carlo standard errors close to each other when there was no relationship between two sets of the data and Kendall’s correlation coefficients had tended to have the lowest values of the Monte Carlo standard errors. (3) The significance testing of these three correlation coefficients had tended to be able to control the type I error under .05 significant level, and (4) Pearson’s correlation coefficients had tended to present the highest power of testing. |
