Estimating effect size when there is clustering in one treatment group

Area of Science:

• Biostatistics
• Experimental Design
• Clinical Trials

Background:

• Experimental designs sometimes feature clustering within only the treatment group.
• This occurs in group tutoring, multi-therapist interventions, or clinic-administered treatments.
• Control groups typically receive no treatment, creating an imbalance.

Purpose of the Study:

• To address the consequences of ignoring clustering in single-group treatment designs.
• To provide methods for correcting biased effect size estimates and variances.
• To adjust significance tests for the effects of clustering.

Main Methods:

• Utilizing intraclass correlation (ICC) information.
• Developing correction factors for effect size and variance estimates.
• Proposing adjustments to significance testing procedures.

Main Results:

• Unaccounted clustering inflates actual significance levels (p-values).
• Biases lead to inflated effect sizes and underestimated variances.
• Intraclass correlation provides a basis for accurate statistical adjustments.

Conclusions:

• Ignoring treatment group clustering compromises study interpretation.
• Using intraclass correlation corrects biases in effect sizes and variances.
• Adjusted significance tests improve the validity of treatment effect evaluations.