ABSTRACT
Many data analyses in the social and behavioral sciences focus on unobserved variables, more formally called latent variables. The prototypical latent variable is an attitude, typically assessed via a self-report attitude survey that may contain sets of items. The responses to each set for each respondent are summed or averaged to estimate the underlying latent variable. The proposed measurement structure is assessed with factor analysis, first exploratory factor analysis, extraction and rotation of the factors, and then ideally confirmatory factor analysis. The application of factor analysis to items from an attitude survey is item analysis, which involves an assessment of not just the factorial structure that underlies the items but also the reliability of the emergent scales. The evaluation of a proposed confirmatory factor analysis is of the variables’ covariance structure implied by the underlying model. The measurement models analyzed here are multiple-indicator models that assign each variable to a single factor, representing an underlying attitude in this context. An application to more general path analysis software is also presented.
