Tek Yönlü Bağımsız Grup Tasarımı için Varyans Homojenliği Testleri

Abstract

In this thesis, a comparison of Type I error rates and power values for 20 different variance homogeneity test methods was conducted using a Monte Carlo simulation approach. This was done to assess the homogeneity of the variance measures used in comparing distributions. The study computed Type I error rates and power values for normal, negatively skewed, and positively skewed distributions under both equal and unequal sample sizes through simulation. Additionally, functions for the 20 included tests were developed, resulting in the creation of the "vartest" package in the R programming language. This package, which focuses exclusively on variance homogeneity tests, is made available as an open-source tool for users. Given the absence of a library solely dedicated to the application of variance homogeneity tests in the literature, this package aims to facilitate user access. The variance homogeneity of a real dataset was analyzed using the tests included in the vartest package. Based on the findings of the study, it is recommended to use Bartlett, Modified Z Variance, Levene 2, and Z Variance tests for normally distributed data and Capon and Klotz tests for non-normally distributed data.