Causal Inference Using Nonnormality

Abstract

Path analysis, often applied to observational data to study causal structures, describes causal relationship between observed variables. The path analysis is of confirmatory nature and can make statistical tests for assumed causal relations based on comparison of the implied covariance matrix with a sample covariance one. Estimated path coefficients are useful in evaluating magnitude of causality. The traditional path analysis has potential difficulties: there exist equivalent models and the postulated model is assumed to represent “true” causal relations. One cannot determine the causal direction between two variables if only two variables are observed because they are equivalent to each other. The path coefficient estimates are biased if unobserved confounding variables exist, and one cannot detect their existence statistically in many cases. In this paper we develop a new model for causal inference using nonnormality of observed variables, and provide a partial solution to the above-mentioned problems of causal analysis via traditional path analysis using nonnormal data.