Dunnett's many-to-one test and least square means

Area of Science:

• Biostatistics
• Statistical Methods
• Clinical Trial Design

Background:

• Dunnett's many-to-one test is widely used in dose-finding studies for comparing multiple treatment groups to a control.
• Exact Type I error calculation is possible for raw means under normality and independence assumptions.
• Least square means (LSMs) in ANCOVA and two-way ANOVA models, particularly with unbalanced data, violate the independence assumption.

Purpose of the Study:

• To investigate the dependence between least square means in ANCOVA and two-way ANOVA models without interaction for unbalanced data.
• To develop novel procedures for accurately calculating the joint distribution of the test statistic in Dunnett's test under these conditions.

Main Methods:

• Analysis of least square means dependence in unbalanced ANCOVA and two-way ANOVA models.
• Derivation of new statistical procedures for joint distribution calculation.
• Simulation studies or theoretical derivations to validate the proposed methods.

Main Results:

• Quantification of the dependence structure between least square means in the specified models.
• Development of accurate methods to compute the joint distribution of Dunnett's test statistic, accounting for LSM dependence.
• Demonstration of the impact of this dependence on Type I error rates.

Conclusions:

• The independence assumption for least square means is often violated in complex experimental designs with unbalanced data.
• The proposed methods provide a more accurate approach to controlling Type I error rates when using Dunnett's test with LSMs in ANCOVA and ANOVA.
• This research enhances the reliability of statistical inference in dose-finding and other comparative studies.