Background

Confirmatory factor analysis (CFA) is a statistical method used to check whether data confirms a predefined theoretical model by assessing the relationships between observed and latent variables. In the context of PLS-PM, CFA validates the quality of measurement blocks by confirming that the selected indicators accurately reflect the latent constructs they are intended to represent.

 

Dataset for Example

250 customers of a mobile phone operator were surveyed to answer a large number of questions to model their loyalty to the operator. The PLS-PM model is based on the European Customer Satisfaction Index (ECSI).

 

Setting Up

• Open XLSTAT.

• Click on the Marketing/PLS Path Modeling menu and select the New Project feature.

• Define the latent variables and associated manifest variables of your model using the command bar available on the graph sheet. See the tutorial Create and run a basic PLS-PM project for a more detailed explanation of how to build a PLS-PM model.

• Click on the button with the double arrow to create two-way links between all latent variables and set up your model for Confirmatory Factor Analysis.

• The result is the following model:

• Click on this button to start analysis.

• In the General tab, select the PLSPM method.

• Click on the OK button to launch the computations.

• The results are displayed on a new Excel sheet.

 

Interpreting the Results

The goal of Confirmatory Factor Analysis is to verify whether the measurement model—that is, the relationships between the manifest variables and the latent variables—accurately describes the latent constructs to which these variables are linked.
To interpret the results, you should therefore focus on the composite reliability table and the loadings table to assess the reliability of the variables.

In our case, Cronbach’s alpha is below its threshold for the constructs Expectations and Loyalty. However, Dillon–Goldstein’s rho remains above 0.7. Finally, the first eigenvalue is much larger than the second in many cases. These results lead us to consider that the blocks are unidimensional and confirm the validity of the model.

Most of the correlations are close to 0.7, which indicates that the manifest variables contribute satisfactorily to the explanation of the latent variable. However, the correlation of the variable CUSL2 with the Loyalty construct is low. This may suggest that the item was poorly designed or does not truly reflect the concept of Loyalty. In this case, it would be advisable to remove this indicator to retain only variables that sufficiently explain the latent construct.