Identification of Influential Cases in Structural Equation Models Using the Jackknife Method

Abstract

Relatively little attention has been given to detecting influential cases (ICs) when estimating structural equation models (SEMs). Most techniques examine individual cases using covariance-based techniques such as the Mahalanobis distance, which examine the distributional characteristics of the cases but ignore the model. Cases identified using such model-free techniques are usually referred to as out-liers. In SEM, however, the model is of central importance. The characteristics of the model (number of latent variables, etc.) have an effect on which cases are influential. The authors propose applying the well-known jackknife procedure to detect model-based ICs, which may be influential with respect to overall fit, particular model parameters, or both. The procedure is illustrated by two studies—one using simulated data, the other empirical data.