SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

© Webmaster

Contributions: Abstract

Fitting nonstandard structural equation models: Modeling multiple-sample data having unequal numbers of indicators per sample

Keith Widaman
University of California at Davis

Most approaches to multiple-group structural equation modeling require that the same number of manifest variables are included in each sample. In a current study of European American and Mexican American families, the Mexican American sample had additional variables (e.g., maternal and paternal acculturation) that were not included in the European American sample and would be inappropriate for such a sample.

However, we were interested in making valid cross-sample comparisons of factor loadings and regression weights in our analyses. This presentation will consider the general problem of unequal numbers of variables per sample, particularly in the presence of missing data (and therefore fitting of models directly to raw data).

I will then (a) offer one reasonable approach to handling the problem - through the addition of stochastic random variables to the sample with fewer variables, (b) discuss conditions that must hold so that overall test statistics and the parameter-by-parameter statistics are unaffected by the addition of the random variables, and (c) illustrate the approach using the analysis of empirical data utilizing the approach. General implications of this approach for ensuring that optimal multiple-sample comparisons are conducted will be stressed.