A Comparison of Methods for Estimating Confidence Intervals for
Omega-Squared Effect Size
W. Holmes Finch and Brian F. French
The paper use simulation study to compare the performance of three methods - parametric, percentile bootstrap, and bootstrap bias-corrected and accelerated confidence interval for calculating confidence intervals for w2, which is the preferred effect size estimate for ANOVA.
Manipulated factors:
population effect size (0, 0.01, 0.059, 0.138 )
dependent variable: standard and seven non-normal conditions
group variance homogeneity: 0, 0.5, 1.0, 1.5
number of groups:3,4,5
independent variables: 1, 2 with no interaction, 2 with interaction
sample size: 5,10, 20, 50, and 100
The results show that sample size and magnitude of the effect influence coverage rates and interval width. The percentile bootstrap method produced the widest intervals and had higher coverage rates in the smaller
effect size conditions compared with the parametric and BCA approaches
future study:
1 The sample size can be unequal among these groups
2 the independent variable can be 2 variable with different effect size.