|
|
 |
| Factors based on zero-inflated and no... |
 
|
| Message/Author |
|
|
| Alex R posted on Monday, June 11, 2018 - 3:57 pm
|
|
|
Hello Mplus team,
I wonder if you can advise on the appropriate analyses for the following scenario:
- My variables represent personality traits (Likert-based) and task-based behaviors
- Behaviors are assessed as latent variables using (3) indicators from three analogous tasks
- A few behaviors will be assessed with zero-inflated count variables, which reflect how often an extreme behavior was observed (most people have 0). The rest are based on continuous and more or less normally distributed variables.
Analysis 1 involves a CFA model with latent correlations among 8 behavioral factors and residual correlations among all the task-specific variables to account for method covariance (similar to this graphic imgur.com/YOZs9Jw).
Analysis 2 involves regressing each behavior (factor) on a number of personality variables, and this will be done across a series of models, one per behavior. (There is also analysis 3 to test for mediation, but I'll set that aside).
The challenge in both cases is that I have to work with factors based on zero-inflated count variables.
[continued] |
|
|
| Alex R posted on Monday, June 11, 2018 - 3:59 pm
|
|
|
[continued]
I've looked over example 3.8 in the manual and the recommendations here (http://www.statmodel.com/discussion/messages/9/11280.html?1398186317) and I imagine that I will probably want to declare those inflated variables as COUNT = x1-x3 (nbi), then specify two factors with count indicators (f1 by x1-x3) and inflation indicators (f1i by x1#1 x2#1 x3#1).
If appropriate, this seems straightforward to do in the regression case, but complicated in the CFA case where I need to specify residual correlations, as I believe these require additional factors of "integration" (if this is even an option with 0-inflated counts).
Apologies for the extended msg and thanks for your thoughts. |
|
|
|
|
| This should work fine. There are many alternative models to explore here - we don't want to give advice on general analysis strategies. |
|
|
| Back to top |
|
|
