Moderation Analysis in AMOS

on

Interaction effects are the joint effects of two predictor variables in addition to the individual main effects. This step-by-step procedure of testing for interaction (or moderation) makes use of the imputed factor scores which is the recommended process as it is easier to perform.

This will walk you through (A) Standardizing and computing variables, (B) Setting up the path interaction, (C) Refining the model, and (D) Plotting and interpreting the path interaction.

A. Standardizing and Computing Variables

Dataset to use: Imputed dataset (_C)

  1. In the imputed SPSS dataset, got to Analyze — Descriptive statistics — Descriptives.
  2. Throw in the imputed variables -- check Save standardized values -- click OK.
  3. Ignore the output -- go to SPSS variable view if new variables were created.
  4. To compute variables, go to Transform -- Compute variable.
    1. Fill in the target variable box interacting variables.
      • Example: Know_x_Attitude.
      • The moderating variable here is Know interacting Attitude.
    2. Multiply the Z values of the target variables in the “Numeric expression” box.
      • As for the above example, it would be: ZKnow * ZAttitude.
      • Click OK.
    3. Ignore the output -- go to SPSS variable view if new variables were created.
      • Repeat the computation with other interaction terms.
      • Always refer to the model for the interaction required.
  5. Save the dataset.
Standardizing variables in SPSS

Computing interaction variables in SPSS


B. Setting up the Path Interaction


File to use: Causal Model Imputed

Dataset to use: Imputed dataset (_C) 

  1. In the Causal model imputed, re-load the dataset.
  2. Go to 'List of variables' button.
    • Find the interaction variables created.
    • Drag to the causal model.
  3. Integrate the interaction variables to the model.
    • Arrange the interaction variables accordingly.
    • Draw regression lines from interaction to the dependent variable(s).
    • Add covariances to the added independent variables.
      • If it finds confusing, you may remove the existing covariances.
      • Select all independent variable objects.
      • Go to Plugins -- select draw covariances.
        • This is to make sure no duplicate covariances.
  4. Set the Output preferences.
    • Go to Outputs.
    • Check Standardized estimates, Squared multiple correlations, and Modification indices (set at 10).
  5. Save as a new model 6 Path Interaction.
  6. Click Run -- then Up Arrow.
Setting the path interaction in AMOS

C. Refining the model


Purpose: To free up degrees of freedom, most especially if you have zero df

  1. In the Output, go to Estimates -- Regression weights
    • Look for insignificant interaction effects.
      • Not significant at 90% confidence level (P-value of at least 0.10)
      • Not significant at 95% confidence level (P-value of at least 0.05)
    • Start with the highest P-value.
  2. Go back to the model, remove insignificant interaction effects.
    • Do this one at a time.
    • Repeat this process if there are still insignificant interactions.
    • Make sure most of the remaining interactions are significant.
  3. Check for the Chi-square and df values.
  4. Check for model fit and make sure good values.
  5. In the model, look at the R square of the dependent variable(s).
    • You may find it the one above the box, or go to...
    • Estimates -- Scalars -- Squared multiple correlations.
  6. Significant interactions are now ready for plotting.
  7. Quick question: Can you do interaction with the combined latent causal model and imputed factor scores? Yes, but it would be messier. I suggest not to do it. Stick with the imputed model. Make sure you have a good dataset from the EFA.
Refining the path interaction model in AMOS


D. Plotting the interactions


Excel file: Stats Tools Package

  1. In the excel file, go to the 2-way interactions tab.
    • Enter the regression information:
      • Name of independent, moderator, and dependent variables investigated.
      • Unstandardized regression coefficients (under Estimates column).
  2. Look at the graph on the right side and note the interpretation of results.
  3. Report on your paper if your hypothesis is supported or not.
  4. Do this plotting with one interaction variables at a time.
    • Refer to this video on how to plot and interpret the results.
Plotting significant interactions in Excel Stats Tools Package

In summary, testing the interaction effects using imputed factor scores required to standardize first the variables in order to compute for the interaction terms. These interaction terms will be added to the causal model to see if they have significant effects or none. In some cases, you will be run out of degrees of freedom (i.e. df = 0), and so it is justifiable to remove insignificant interaction effects to free up degrees of freedom. Interpreting interaction effects is made easy using the Stats Tools Package. Hope this helps.

Big thanks to Dr. James Gaskin for helping me learn on this topic. You may check this YouTube video SEM Series (2016) 9. Interactions where I based these steps together with other videos on his YouTube Channel.

King Mongkut's Institute of Technology Ladkrabang

Comments

Post a Comment