XLSTAT: Consumer satisfaction analysis with PLS-PM in Excel

 

Background

The PLS-PM approach is a statistical method for modeling complex relationships between latent variables and manifest variables, belonging to the family of Structural Equation Models (SEM). It emphasizes a predictive dimension to estimate relationships and test causal hypotheses, while allowing theoretical knowledge to be incorporated through variants such as the hierarchical or confirmatory approaches.

 

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

• Select the New project option. When you create a new project, its default name starts with PLSPMBook.

A raw XLSTAT-PLSPM project contains two sheets that cannot be removed:

D1: This sheet is empty. You will need to add all the input data that you want to use into that worksheet.

PLSPMGraph: This sheet is blank and is used to design the model. When you select this sheet, the "Path modeling" menu is displayed on the upper left part of the page.

 

• Select the marketing display mode in the XLSTAT-PLSPM options.

• Copy the data from a standard Excel sheet to sheet D1.

• Build the model:

  • Go to the PLSPMGraph sheet. The command bar is at top left:

• To create several latent variables in a row, double click on the circle button so that it stays pressed while you add variables:

• Add arrows linking latent variables to each other. To add an arrow, simply click on the latent variable from which it starts and select the destination latent variable while holding down the Ctr+L button. Once both variables have been selected, use either the toolbar button shown in the following figure or the keyboard shortcut Ctrl+L.

• Define the manifest variables associated with each latent variable. Click on the latent variable and use the MV button on the bar:

In the General tab:

• Select the quantitative data columns linked to the latent variable.

• Choose the reflective mode.

For example, for the latent variable Expectations, the dialog box will look like this:

The obtained model has the following form:

• Click on the icon to select the calculation options.

In General tab:

• Select the Standardized, weights on standardized MV option for the Treatment of the manifest variables and the Values of the first eigenvector option for the Initial weights.

In Output tab:

• Activate the Model option.

• Activate the Latent variables scores and select Standardized option.

 

Interpreting the results – part 1

The first part of the results gathers information on the data and on the created model (descriptive statistics related to the manifest variables and model specification).

First, it is important to check blocks unidimensionality. In our case (reflective variables), blocs must be unidimensional. The composite reliability table provides this information:

 The Cronbach alpha is below the 0.7 threshold for the Expectation and Loyalty variables. However, Dillon and Goldstein’s rho is above the 0.7 threshold for all variables. Finally, the first eigenvalue is higher than the second in many cases. Those results let us consider that blocks are unidimensional, although it could be interesting to investigate supplementary dimensions in Expectation and Loyalty. The Complaint latent variable does not appear in the table as it is associated with one single manifest variable.

The next table contains information on the goodness of fit index.

 The absolute GoF = 0.471. It is very close to its bootstrap estimation. This value is not easy to interpret. It is mostly used to compare different groups of individuals or different models.

The two following tables display outer weights, and the correlations associated with the measurement model.

 

Interpreting the results – part 2

Once the measurement model is studied, the structural model can be analyzed. Each latent variable is associated with several outputs. We will comment on the outputs of the Satisfaction variable:

We may consider that the latent variable is well explained (R²=0.680). We see that the Perceived quality has the most important effect on Satisfaction followed by the Perceived value. The Expectation impact is insignificant. The last table summarizes the preceding results. We see that the Perceived quality has a 60% contribution to the R².

The following chart illustrates the tables we just mentioned:

Next, outputs linked to the marketing analysis are displayed starting with the IPMA (Importance Performance Matrix Analysis). The IPMA allows you to visualize the importance and performance of latent variables on a target variable. Here is the chart for the Satisfaction variable:

We see that Perceived quality is the most important and performant latent variable. It will be difficult to increase this variable’s performance since it is already quite high. It is followed by the Expectation and Image variables that share the same importance. Image seems interesting because it has a lower performance, and it could be improved more easily than other variables.