Testing for group effect in a 2 x k heteroscedastic ANOVA model.

Area of Science:

• Statistics
• Biostatistics

Background:

• Traditional ANOVA models often assume homoscedasticity and Gaussian error distributions, which may not hold in real-world data.
• Relaxing these assumptions is crucial for robust statistical inference in complex experimental designs.

Purpose of the Study:

• To develop and evaluate a novel statistical test for the main effect of a two-level factor in a two-way ANOVA model.
• To address the limitations of traditional ANOVA by not requiring assumptions of homoscedasticity or Gaussian error.

Main Methods:

• Derivation of an approximate empirical likelihood ratio test (AELRT) for the group main effect in a 2 x k cell ANOVA.
• Utilizing simulation from the approximate empirical maximum likelihood estimate (AEMLE) under the null hypothesis to approximate test statistic distributions.
• Comparison of AELRT with the standard homoscedastic ANOVA F-test and a Box-type approximation for heteroscedastic data in terms of level and power.

Main Results:

• The AELRT procedure demonstrates appropriate type I error control when test statistic distributions are approximated via simulation from the constrained AEMLE.
• The test may exhibit conservative behavior regarding type I error control.
• Simulations indicate favorable performance compared to existing methods under heteroscedastic conditions.

Conclusions:

• The developed AELRT provides a viable alternative for testing main effects in ANOVA models with non-normal and heteroscedastic data.
• The methodology is applicable to real-world scenarios, such as analyzing factors influencing biological measurements like blood folate levels across different groups.
• Empirical likelihood offers a powerful framework for developing robust statistical tests beyond traditional parametric assumptions.