A Distribution-Free Approach for Inference about Structural Equations Models

• Activity Number: 641
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Social Statistics Section
• Author(s): Douglas David Gunzler*+
• Companies: University of Rochester

Abstract

We will discuss a new class of functional response models (FRM) to address limitations of current distribution-based structural equations models (SEM) within the context of mediation analysis. Unlike existing SEM that rely on parametric distribution assumptions for inference, the FRM-based approach does not require any distribution assumption and thus provides robust inference for a wider class of data distributions. Since SEM is often applied to longitudinal study data to infer causal relationships, severely biased estimates arise when missing data follows the missing at random (MAR) mechanism and the distribution assumptions are not met. The distribution-free nature of the proposed approach is especially important for practical applications. We discuss inference for this new approach and illustrate its application and robustness using real as well as simulated study data.