Large variances PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Jan Newman posted on Monday, March 04, 2013 - 10:30 pm
I am running a logistic regression with 19 covariates and 1 outcome. The covariates are based on measures with different scales. I am using MLR with a monte carlo integration. My input file would not run, and I was advised by customer support (who was able to get it to run) to rescale my continuous variables by dividing by a constant.

Do I rescale all variables using the same constant? Is the input just DEFINE x1 = x1/2 or something like that? Is this a better option than standardizing the variables?

I apologize for all of these questions. I am new to mplus and new to programming-based stats programs. Thank you in advance.
 Linda K. Muthen posted on Monday, March 04, 2013 - 10:54 pm
You should not standardize the variables. Divide by a constant that results in the variances being between one and ten. You can use different constants to meet this goal. You should use the same constant if the variables are repeated measures of the same variable.
 Jan Newman posted on Tuesday, March 05, 2013 - 1:36 am
I have tried different constants, all smaller than my SDs.

I have not been able to get the input file to run on my 64-bit computer. I have let run for 35 minutes and rebooted. I am not sure what's going on there.

I have sent to customer support and they can run. Support is still getting the Fisher Information Matrix error message as well - probably due to the large variances. I have missing data for 3 of my covariates/predictors at arouond 23-25% hence the reason for the integration.

Any other thoughts/ideas?
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: