1. Update the WAIS-III Model
While examining our four-factor model of the WAIS-III, we found that the perceptual factor should be collapsed from two-factors into one. Since the model is now stable, we can look for other potential areas to improve the model.
2. Three-Factor WAIS-III
In the previous exercises, you made a visualization of the current three-factor model of the WAIS after finding a Heywood case on the latent variables. Now that we have a stable model, we can explore the loadings and fit indices.
3. Factor Loadings
Using the summary() function, we find that the loadings for the WAIS subscales are mostly strong indicators for their factors. However, letter number sequencing and digit symbol coding only weakly load onto their factors with completely standardized loadings below 0.300. We could remove these subscales as poor indicators, but since this is a standardized test, we will leave them in the model and explore other areas of improvement.
4. Variances
It is always helpful to check the variances for any potential issues, even though we did not get a warning about a negative variance. Variances will be based on the scale of the data, but very large variances or standard errors should be explored, as they can indicate potential problems with the model. Digit symbol coding was a poor fitting variable on the loadings, and here the variance estimate is quite high. However, if we use the var() function, we can see that this estimate is pretty close to the actual variance and is likely just a variable with a lot of variance, rather than a model error.
5. Fit Indices
Overall, the model fit is poor, as the goodness of fit measures with CFI and TLI are in the 70s, while the badness of fit measures, the RMSEA is high, and SRMR is OK. We can use modification indices to see if we can improve model fit with the addition of a new parameter estimate.
6. Modification Indices
The modification indices give us a good place to start to improve the model. It is important to think about the implications of what you are adding to the model. Similarity and information are two subscales on the verbal comprehension factor, so it is logical that they might have correlated error terms.
7. Let's practice!
Let's try adding some new parameters to our three factor model to see if it improves model fit.