Setting up the Causal Model in AMOS

Setting up the causal model requires skills on how to navigate and use the AMOS interface. AMOS quickly performs the computations for the structural model and displays the results immediately. Your causal model can either be using the latent factors (from the initial CFA) or using the imputed factor scores (recommended if you have a common method bias). 

The section below gives the step-by-step procedure on dealing with these (A and B) two causal models as well as the (B) steps in determining significant effects and (C) some notes on how to make use of the modification indices to improve model fit.

A. Types of Causal Models

A.1 Using Imputed Factor Scores

Dataset to use: Imputed dataset (_C)

Applicable to: Both models (has method bias or none)

1. In AMOS, load the imputed dataset.
• Go to Select data -- File Name.
• Locate and click the dataset -- click Open -- then OK.
2. Click the List variables in the dataset button.
• Find the imputed variables.
• Drag to the center.
3. Build the model.
• Draw regression lines according to the hypothesized model.
• Use the Draw paths button.
• Put residuals (or error terms) to the dependent variable(s).
• Go to Plugins -- Name unobserved variables.
• Covary all the exogenous (independent) variables.
• Select all independent variables.
• Go to Plugins -- Draw 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 file 4 Causal Model Imputed.
6. Click Run -- then Up Arrow.
7. In the Outputs button, check for the Model fit.
• If there are model fit issues - Go to the Modification indices. See Part C.
• Check again the model fit values before proceeding to the effects.
• Make sure that at this point, your Chi-square and df will not be equal to zero.

Building the causal model using imputed factor scores

A.2 Using Latent Factors

A.2.1 If there is NO method bias

File to use: CFA initial

1. Remove all covariances only from the dependent variable(s).
2. Build the model.
• Move the dependent variable(s) objects to the right.
• Draw regression lines.
• Put residuals (or error terms) to the dependent variable(s).
• Go to Plugins -- Name unobserved variables.
3. Set the Output preferences.
• Go to Outputs.
• Check Standardized estimates, Squared multiple correlations, and Modification indices (set at 10).
4. Save as a new file 4 Causal Model Latent No Bias.
5. Click Run -- then Up Arrow.
6. In the Outputs button, check for the Model fit.
• If there are model fit issues - Go to the Modification indices. See Part C.
• Make sure that at this point, your Chi-square and df will not be equal to zero.
• Check again the model fit values before determining significant effects.

Building a causal model using latent factors - No method bias

A.2.2 If there is method bias

File to use: CFA CMB

1. Retain the CLF while removing the covariances from the dependent variable(s).
2. Build the model.
• Select all objects and move to the left.
• Move the dependent variable(s) objects to the right.
• Draw regression lines.
• Put residuals (or error terms) to the dependent variables(s).
• Go to Plugins -- Name unobserved variables.
3. Set the Output preferences.
• Go to Outputs.
• Check Standardized estimates, Squared multiple correlations, and Modification indices (set at 10).
4. Save as a new file 4 Causal Model Latent Bias.
5. Click Run -- then Up Arrow.
6. In the Outputs button, check for the Model fit.
• If there are model fit issues - Go to the Modification indices. See Part C.
• Make sure that at this point, your Chi-square and df will not be equal to zero.
• Check again the model fit values before determining significant effects.

Building a causal model using latent factors - With method bias

B. Determining significant effects

1. In the Outputs button, go to Estimates — Scalars — Regression weights
2. Report significant effects.
• Significance at 90% confidence level (P-value of at most 0.10)
• Significance at 95% confidence level (P-value of at most 0.05)
3. If you have zero df:
• You can delete regression lines of insignificant effects to free up df.
• Always check model fit if acceptable values are achieved.

Determining significant effects in AMOS 

C. Notes for Modification Indices In a Structural Model

1. You cannot covary errors in a structural model, only in the measurement model (CFA).
2. In a structural model, you can either do regression lines or remove variables.
3. You can regress a variable to another variable (even if not from the same construct).

In summary, setting up the causal model can be done by either using the imputed factor scores or the latent factors. For both settings, it also depends if your model has a common method bias or none. You need to retain the CLF as you do the causal modeling if you have a method bias. A Good model fit must also be assured first before determining significant effects. 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) 7. Setting Up Causal Model where I based these steps together with other videos on his YouTube Channel.