SEM Moderation Model in Smart PLS
SEM Moderation Model in Smart PLS
Theoretical model developed by a study is tested using Structural Equation Modelling (SEM). SEM is capable of testing complicated relationships among variables. Moderation analysis tests the influence of a third variable on the relationship between independent variables and dependent variables.
Here, job characteristic is considered as exogenous (independent) variable, job satisfaction is considered as endogenous (dependent) variable and position is considered as moderator variable. Smart PLS software is used to test whether position moderates the relationship between the exogenous and endogenous variables.
To create a new project in smart PLS, click Select File > Create new project or click on New Project icon in the top.
A dialog box would appear as shown in the following image. Name the project and click on Ok.
To import the data, double click on the option ‘Double-click to import data’. Then, browse the folder where the data file is located.
Here, SEM_Model_Moderation_Data.csv file is being imported.
Then, click Ok on the ‘Import Datafile’ dialog box
Click the highlighted icon to create the path model.
Then select the variables from the indicators by pressing “shift” button.
Drop the variables in the path model screen
Then, rename the latent variable through the ‘Rename’ option.
Here, the latent variable is renamed as Job Characteristics. Then, repeat the same procedure to create the other latent variables.
Here, two latent variables, namely, Job Characteristics and Job Satisfaction, are created.
Use ‘Connect’ icon to connect the two latent variables
The latent variable turns into blue colour when the two variables are connected.
Click on ‘Calculate’ icon and select ‘PLS Algorithm’ option to get the measurement model.
Partial Least Squares Algorithm dialog box would open. Select ‘Factor’ option in weighting scheme and mention the maximum iterations. Here, 500 maximum iterations are considered. Now, click on ‘Start Calculation’ button.
The results for the PLS Algorithm would be obtained.
All the results of the PLS algorithm can be checked by clicking these options.
Here, AVE value for Job Satisfaction is less than 0.5. Hence, some items are deleted from the Job Satisfaction latent variable.
Here, outer loading value for JS_11 and JS_3 is less than 0.6. Hence, these two items are deleted.
After deleting JS_11 and JS_3, 0.533 is obtained as AVE value for Job Satisfaction.
The result can be saved in excel sheet by clicking on ‘Excel’ option.
Add moderate variable in the structural model. Here, Position is considered as a moderate variable. Click on ‘Add Moderating Effect’ option in dependent variable (Job Satisfaction) to add a moderating effect.
Then, Moderating effect dialog box would open.
Here, select Position as the moderating effect and Job Characteristics as the independent variable. Click on Continue.
Then, moderating effect variable would be obtained.
The variable could be renamed through the ‘Rename’ option. Here, variable is named as “JC × Position”.
Click on ‘Calculate’ icon and select ‘Bootstrapping’ to get the structural model.
Then, Bootstrapping dialog box would open. Select Path option in weighting scheme and mention the maximum iterations. Here, 500 maximum iterations are considered. Now, click on ‘Start Calculation’ button
The result for Bootstrapping would be obtained.
Click on ‘Calculate’ icon and select ‘Blindfolding’ option to get predictive relevance.
Then, Blindfolding dialog box would open. Select Path option in weighting scheme and mention the maximum iterations. Here, 500 maximum iterations are considered. Now, click on ‘Start Calculation’ button.
The result for blindfolding would be obtained.
Output of SEM model in Smart PLS
Assessment of measurement model
In the Construct Reliability and Validity table, Cronbach’s Alpha, rho_A and Composite Reliability values for all study variables are greater than the cut-off value of 0.7. Therefore, it could be said that reliability is established. Average Variance Extracted (AVE) values for the two variables are higher than the cut-off value of 0.5, thereby satisfying the model’s convergent validity.
Table for Fornell-Larcker Criterion would be obtained under discriminant validity. Here, all the values are higher than the highest correlation of the specific variable with the other variable.
In the HTMT table, values are less than the cut-off value of 0.85. This assures the optimal difference established between the constructs and the validity of factors.
Evaluation of model fitness
R Square and R Square Adjusted values are obtained for the dependent variable (Job satisfaction).
Here, R Square value of 0.175 indicates that 17.5% of the variation is explained by job characteristics.
Then, f Square value is obtained. Here, f square value is equal to 0.212, which indicates that Job characteristics has a medium impact on Job satisfaction.
Here, all VIF values are less than 5, implying the nonappearance of collinearity among the indicators.
The Construct Cross validated Redundancy table is obtained when the blindfolding algorithm is run. The value for Q2 is obtained from this table. Predictive relevance of the model is estimated using the Q2. The model’s significantly predictive Q2 value is greater than 0. Here Q2 = 0.102 and thus it is evident that the model had a good predictive relevance.
Testing of the model
The table for Path Coefficients is obtained when the Bootstrapping model is run.
It can be inferred from this table that Job Characteristics had a significant influence on Job Satisfaction (t=7.200, p=0.000). Position had a significant influence on Job Satisfaction (t=3.397, p=0.001). However, the moderating effect JC × Position did not influence Job Satisfaction. Thus, it can be concluded that relationship between Job Characteristic and Job Satisfaction is not moderated by Position.