Structural Covariance Models in Criminology - A Comparison of LISREL and PLS - Final Report

• NCJ Number: 87559
• Author(s): B McGarvey; P M Bentler; E H Freeman; W F Gabrielli
• Date Published: 1980
• Length: 46 pages
• Grant Number(s): 79-NI-AX-0086
• Sponsoring Agency: US Dept of Justice
• Corporate Author: University of California, Los Angeles
• Sale Source: National Institute of Justice/; NCJRS Photocopy Services
• Publication Type: Report (Study/Research)
• Language: English
• Country: United States of America

Annotation

This report describes the importance of structural covariance models in criminology, as well as the characteristics of two algorithmic competitors: the full-information, maximum-likelihood LISREL IV model and the partial least squares (PLS) model.

Abstract

Using a substantive model hypothesizing certain structural relationships among the constructs of criminality, present socioeconomic status (SES), educational attainment, social strain, family status, and parental SES, the paper examines the computational characteristics of the two algorithms. The assessment includes measures of goodness of fit, as well as an examination of residual correlations and covariances. While LISREL IV seems to provide both more options in use, as well as a generally better fit to the data, PLS provides a comparable, highly correlated set of parameter estimates in considerably less computational time. Implications of these results are discussed. (Author abstract modified)