An overview of applied robust methods.

Area of Science:

• Social and Behavioral Sciences
• Statistical Modeling

Background:

• Social science data often exhibit non-normal distributions and heavy tails.
• Conventional statistical methods relying on sample means and covariances can lead to inaccurate conclusions with such data.

Purpose of the Study:

• To provide an overview of robust statistical procedures for social science research.
• To highlight robust methods effective for commonly used models, including latent variable and moderated mediation analyses.

Main Methods:

• Overview of robust statistical procedures.
• Application examples using latent variable models and moderated mediation analysis with categorical covariates and product terms.
• Demonstration of R and SAS programs for practical implementation.

Main Results:

• Robust methods offer substantial performance improvements over conventional techniques for non-normally distributed data.
• These methods lead to more efficient parameter estimates and reliable model evaluation.
• Improved model/data diagnostics and more trustworthy conclusions, especially for replication studies.

Conclusions:

• Robust statistical methods are essential for accurate analysis of non-normally distributed social science data.
• Adoption of robust procedures enhances the reliability and trustworthiness of research findings.
• Software programs are available to facilitate the routine application of these recommended methods.