The Role of Sample Size to Attain Statistically Comparable Groups - A Required Data Preprocessing Step to Estimate...

Area of Science:

• Observational studies
• Causal inference
• Biostatistics

Background:

• Propensity score methods are crucial for bias reduction in observational data.
• Existing guidelines for propensity score use primarily address large sample sizes.
• Limited guidance exists for scenarios with small treated groups and numerous covariates.

Purpose of the Study:

• To investigate the efficacy of propensity score methods for bias removal in small treated groups.
• To assess the impact of numerous covariates on bias reduction with small treated samples.

Main Methods:

• Simulation studies were conducted to evaluate propensity score performance.
• Factors examined include control-to-treated group size ratios, covariate numbers, and initial covariate imbalances.
• Empirical evaluation using real-world data validated simulation findings.

Main Results:

• Selection bias can be effectively removed even with small treated samples.
• The required control-to-treated group size ratio increases significantly with more covariates and smaller treated groups.
• A study with 8 treated units and 10 covariates needs a 10:1 control ratio, while 500 treated units need only 2:1.

Conclusions:

• Propensity score methods are viable for bias reduction in small treated group studies.
• Specific conditions, particularly larger control group sizes, are necessary for effective bias removal.
• Findings offer practical guidance and highlight areas for future research in causal inference.