SmartPLS - Measure Validity

We need to evaluate the measurement model before we can know how reliable and valid the model is. As I explained in the earlier post HERE, reliability can be measured by assessing the composite reliability and Cronbach's Alpha whereas validity can be measured through convergent and discriminant validity.

Have you download SmartPLS software? If "YES", you can just proceed to test the validity. If "NO", you can download the software and install it into your computer. In this post, I'll show you step by step on how to calculate validity. 

First, we need to examine the cross-loadings since they are the dominant approaches to measure discriminant validity.

Step 1: Asses indicator loadings by clicking "Default Result" --> "PLS" --> "Calculation Results" --> "Outer Loadings"

The table below shows a complete table of the outer loadings

From the table, we can see that all reflective indicators have loadings of 0.636 and higher. SmartPLS provide outer loadings and outer weights for all construct in the PLS path model, regardless of whether they are measured reflectively or formatively.

Step 2: Assess the Fornell-Larcker criterion, click "Report" --> "PLS" --> "Quality Criteria" --> "Latent Variable Criterion

This is the last step in analysing the cross-loadings. For reflective measurement model, we use Fornell-Larcker criterion. Fornell-Larcker criterion is the second and more conservative approach in assessing discriminant validity. It compares the square root of the AVE values with the latent variable correlations. The square root of each construct's AVE should be greater than its highest correlation with any other construct. The logic of this method is based on the idea that a construct shares more variance with its associated indicators than with any other construct.

The table below shows the Fornell-Larcker criterion before we calculate the square root of AVE

From the table above, we include the values of the square root of AVE

Finally, the table of Fornell-Larcker criterion is complete

These are the needed step in measuring the validity through SmartPLS. So, how can we explain those number? What are those number supposed to mean? Actually, you can refer these papers, HERE and HERE to understand how to analyse our collected data.

Till next time. Be safe and take care! 

Let's fight Covid-19 together!

Reliability and Validity Analysis

Previous scholars have shown that Partial Least Squares (PLS) is a robust technique that has been frequently used in the literature. In PLS, to analyse a model, we need to follow the two-step procedure which are the measurement model and structural model (Hair et al. 1998).

1. Measurement Model
In this step, we measure convergent validity and discriminant validity. 

What are the purpose of measuring convergent validity? The reason for analysing convergent validity is to measure the close relations exist among the items of the same construct. We need to measure composite reliability (CR) and average variance extracted (AVE).

In PLS, the factor loading should be greater than 0.7 (Hair, 2013). We need to remove items loading with less than 0.7 to increase CR or AVE in the first order of the component of independent variables (IVs).

Fornell and Larcker (1981) stated that the CR values should be more than 0.7 and AVE should be greater than 0.5 in order for the result to be accepted. Convergent validity is established if all the values of CR are greater than 0.7 and all the values AVE in a study are greater than 0.5. To know more about how to calculate CR and AVE, you can click HERE (Scopus-indexed journal).

Discriminant validity explains the degree of irrelevance between constructs. To measure discriminant validity, there are two steps which are confirmatory factor analysis and AVE analysis. All items should have high loading on their corresponding constructs. Then, we calculate the square root of the AVE that exceed the inter-correlation of the construct in the proposed model. In order to support discriminant validity, each construct's AVE square root should be greater than its correlations with other constructs. 

To see more about the explanation in analysing convergent and discriminant validity you can either click HERE or HERE. 

2. Structural Model
Structural equation modelling tested the hypothesised paths of the structural model. All coefficient estimates were significant (p<0.05) in accordance with the hypothesised directions. In testing the proposed hypotheses, the standardised path coefficient is expected to be at least 0.2 and preferably greater than 0.3 (Chin and Newsted, 1999).

The reliability of the coefficients in the study is obtained through a bootstrapping procedure (500 resamples). All t values need to be higher than the theoretical t value of 2.57 for a 5% probability of error. Meanwhile, the p values of 0.000 indicate that all path relationship are significant at a 5% probability of error.

Predictive accuracy of a model can be measured by analysing the coefficient of determination. A rule thumb on the acceptable coefficient of determination is 0.75, 0.50 and 0.25, explaining substantial, moderate or weak level of predictive accuracy, respectively (Hair et al., 2014).

For further explanation on how to calculate the structural equation model analysis and how to explain the predictive accuracy of the model, you can click HERE (Scopus-indexed journal).

Love,
Dr SAA
2