Start and Stop Rules for Exploratory Path Analysis
Bill Shipley · Structural Equation Modeling A Multidisciplinary Journal · 2002
This article describes a method for choosing rejection probabilities for the tests of independence that are used in constraint-based algorithms of exploratory path analysis. The method consists of generating a Markov or semi-Markov model from the equivalence class represented by a partial ancestral graph (PAG) and then testing the d-separation implications of this model using either the test proposed in Shipley (2000a), for path models without correlated errors, or by applying a Bonferonni correction for path models with correlated errors.