|
|
 |
| Mixture Modeling with control variables |
 
|
| Message/Author |
|
|
|
|
Greetings;
I am new to Mplus and have a question about mixture modeling using continuous indicators. I am trying to conduct a latent profile analysis using eight continuous indicators. I have an additional categorical variable, indicating different groups of observations, that significantly alters the eight continuous indicators. Therefore, I would like to control for its effect in the LPA. However, I am not sure how I can include the categorical variable as a control variable in LPA. I see two approaches based on what I found in Mplus User's Guide:
(1) Inclusion of the categorical variable as a "covariate" in the mixture model (similar to EXAMPLE 7.12 in Mplus User's Guide); Or
(2) Doing a multilevel mixture modeling, with the categorical variable specified as the "CLUSTER" (similar to EXAMPLE 10.6 in Mplus User's Guide).
I was wondering if any of these approaches is appropriate and whether there is an easier way to do this.
Thank you |
|
|
| Jon Heron posted on Friday, May 13, 2016 - 12:26 am
|
|
|
Hi Hamed
two comments
[1] It's not very clear what type of variable your grouping is. Are we talking about something like gender, or something like classroom which represents a nested structure to your sample?
[2] I'm not sure what you mean by control. Do you want to avoid the situation in which the classes you extract merely reflect your categorical variable? Does this categorical variable have only a few different categories? might you expect totally different profiles within each of these categories? what sort of sample size do you have?
hmm, perhaps more than two comments there
best, Jon |
|
|
|
|
Answer to Hamed Saremi:
If you have only a few categories on the categorical variable then I would use it as a covariate. If you have at least say 25, you could use it as a cluster variable. |
|
|
|
|
Hello Jon and Bengt;
Thank you for your responses. I am sorry for lack of clarity in my first post.
The data have been collected from social media users and include their perceptions and attitudes toward the social media. I would like to use these variables in LPA to create profiles of users. However, participants have been using different types of social media platforms (Facebook, LinkedIn, etc.), which is my categorical variable with 10 categories. This is the effect I would like to control (as Jon said, I do not want my profiles of users be just a representative of the social media platform they are using). Users are technically nested within different social media platforms.
The sample size is 470.
So, based on Bengt suggestion, I presume I should go with covariate (similar to EXAMPLE 7.12 in Mplus User's Guide) to control the effect of social media platform variable. Correct?
Best,
Hamed |
|
|
|
|
| Right. |
|
|
| Back to top |
|
|
