Non-normality of data PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Clio Berry posted on Tuesday, July 17, 2012 - 2:29 am
Another question:

I have variables which are non-normal so assume I should either transform them or use WLSMV estimation.

Many of the variables have univariate outliers (based on z scores >3, should I do something about the outliers by either removing cases, transforming the scores or transforming the variables as well as using WLSMV? Or should I just use WLSMV and leave the scores as they are?

Thank you,

Clio.
 Linda K. Muthen posted on Tuesday, July 17, 2012 - 11:06 am
When you say non-normal, do you mean continuous variables that are not normally distributed or categorical variables?
 Clio Berry posted on Wednesday, July 18, 2012 - 1:08 am
Apologies, I mean continuous variables that are not normally distritbuted, i.e. some are negatively skewed, some are positvely skewed and one is negatively skewed/approaching bimodal.
 Linda K. Muthen posted on Wednesday, July 18, 2012 - 6:13 am
You should treat the outliers in the regular way and use MLR.
 Clio Berry posted on Friday, July 20, 2012 - 8:10 am
Thank you.

And what if I have one categorical (ordered) variable and other continuous non-normal variables?

Should I then use WLSMV? Should I still transform univariate outliers if using WLSMV?

Thank you,

Clio.
 Linda K. Muthen posted on Friday, July 20, 2012 - 10:38 am
MLR is better for non-normal continuous variables. I would treat outliers the same with any estimator.
 Clio Berry posted on Friday, July 27, 2012 - 7:16 am
Thank you for your reply.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: