Statistical comparison of four effect sizes for single-subject designs

Area of Science:

• Behavioral Science
• Applied Psychology
• Autism Spectrum Disorder Research

Background:

• Debate exists on optimal methods for summarizing single-subject design treatment outcomes.
• Both nonregression and regression-based effect sizes are proposed, with advocates claiming superiority.
• Evaluating these methods is crucial for accurate interpretation of intervention efficacy.

Purpose of the Study:

• To compare the findings of different single-subject effect sizes.
• To assess the utility of regression-based effect sizes against nonregression methods.
• To determine if regression-based effect sizes improve understanding of treatment outcomes in autism interventions.

Main Methods:

• A review of 117 articles targeting problematic behavior reduction in 181 individuals with autism.
• Calculation of four effect sizes: mean baseline reduction (MBLR), percentage of nonoverlapping data (PND), percentage of zero data (PZD), and a regression-based d statistic.
• Analysis of statistical relationships between effect sizes using Pearson product-moment correlations.

Main Results:

• All effect sizes indicated behavioral treatment effectiveness.
• Only the PZD effect size detected moderating variables.
• Effect sizes showed differing statistical relationships.
• The regression-based d statistic did not improve understanding compared to nonregression effect sizes.

Conclusions:

• Nonregression effect sizes, particularly PZD, are valuable for identifying moderating variables in single-subject designs.
• Regression-based effect sizes may not offer superior insights into treatment outcomes for autism interventions.
• Further research should consider the comparative strengths of various effect size measures.