•  
  •  
 

Abstract

Research in MIS often focuses on the relationships among latent variables of interest that cannot be directly measured. Because of potential error in measurement and associated confounding, indirect measurement of latent constructs requires formal assessments of reliability and validity. Without these measures, resultant paths in causal implications may be inaccurate, biased, and unstable. However, even with favorable metrics of validity and reliability, it is still possible for estimated models to be confounded. In many cases, such confounding occurs when a measurement item reflects more than one latent construct, that is, when there is a lack of unidimensionality. This problem can lead to false assumptions regarding the strength of paths between latent constructs and patterns of causality within a nomological network. While assessing unidimensionality is a critically important aspect of validity, it is not always formally tested in MIS research. This tutorial introduces the concept of unidimensionality from a LISREL Confirmatory Factor Analysis (CFA) perspective. Assuming that the underlying data distribution assumptions and model used are correct, the tutorial provides a step-by-step example of how to assess unidimensionality with LISREL. The tutorial also shows how a CFA can detect problematic multidimensional items and the problems that can occur if undetected.

Share

COinS