1. Removing loadings to improve fit
The last lesson was about adding loadings to improve fit. You can also remove loadings to improve fit by getting rid of weak item/factor relationships.
2. What does it mean to remove a loading?
For this lesson, you'll learn how to remove a loading by deleting the loading of item O4 to the Openness factor. If you inspect the CFA results, you'll notice that this was the weakest loading, with a value of 0-point-287.
3. What does it mean to remove a loading?
Here's what deleting loadings from the model looks like conceptually. In this example, you are effectively not only deleting the loading, you are also excluding the O4 item from the model since there are no other loadings associated with that item. By deleting this loading, you are effectively revising the measure by removing that item.
4. Removing a loading in the syntax
The process for changing up the syntax and rerunning the CFA is just like when you were adding an item. First, you'll remove item O4 from the equations used to create the original syntax. This actually tells the package not to estimate the relationship between item O4 and the Openness factor. Then, as before, use the cfa() function to convert those equations to sem-compatible syntax.
5. Running the revised CFA
You'll notice when you run the CFA with the revised syntax; you get a new warning. Since you've effectively removed item O4 from your model, the sem() function doesn't like that it's still in your input dataset. Don't worry - it's just a warning, not an error. It's difficult to turn this off, so just be sure to review the warning messages and make sure nothing unexpected pops up.
6. Comparing the original and revised models
You may recall that when you added loadings, the first type of comparison was to run a likelihood ratio test. Since you've discarded an item, though, the models aren't run on the same dataset, and the likelihood ratio test can't be run.
Instead, let's first look at the fit indices. As you'll see, the model with the deleted item and loading has a higher CFI and is therefore preferred.
7. Comparing the original and revised models
You can also still consult the RMSEAs. The difference here isn't as noticeable, but you can see that the RMSEA is a teeny bit lower for the revised model.
Again, for more information about calculating and interpreting fit indices, check out this website, which covers the indices in this lesson as well as several others.
8. Let's practice!
Okay! This is conceptually very different from adding loadings, but the code required is pretty similar. Let's go try this out.