Homogeneity of variance assumption
Assumes the variance of dependent variable is equal in each population.
In the case of unequal sample sizes and heterogeneity of variances, standard error will be:
- Overestimated when largest variance occurs in largest group
- Conservative F-Test (ANOVA) and true Type I error is smaller than nominal significance level
- Underestimated when smallest variance occurs in largest group
- Liberal F-Test (ANOVA), larger than desired Type I error
In the above case, choose a non-parametric statistical test.
Test of Homeogeneity of Variances:
- Levene Test
- Null hypothesis: Assumption of homogeneity of variance is met
- If reject null hypothesis options:
- Data transformation
- Adjust significance level
- Choose Welch Method instead of conventional F-test (ANOVA)
- Choose non-parametric test (Kruskal-Wallis — no assumptions)