Neighborhood Effects: A LISREL Model for Clustered Samples

Abstract

This article presents a structural equation model for assessing the effects of group membership when observations are clustered by group. Within a LISREL specification, the net similarity in outcomes between a pair of observations in the same cluster is generated by an unobservable which captures the group effect. Analogies to the conventional ANCOVA model are noted, and an application to the assessment of “neighborhood effects” in reports of “neighborhood quality” from the 1976 Annual Housing Survey is presented. Estimates from the unobservable variable model are compared to those from a naive regression model and the model subjected to a specification test. The pitfalls of proceeding with the naive model are discussed, and the potential for future work exploiting the clustering in large-scale social surveys is emphasized.