MLR = Automatic FIML or no? PreviousNext
Mplus Discussion > Structural Equation Modeling >
 Jake Lant posted on Friday, March 09, 2012 - 1:29 pm
Hello fellow Mplus-users and creators,

I am using MLR to run mediation models but it is the first time I do it with missing data. It runs it perfectly but it skips the observations with missing data. Is there any way around this?

I was under the impression MLR automatically estimated these but it doesn't seem to be doing that haha. I specified MISSING ARE variables (-999.00), etc.

Any help is appreciated.


 Linda K. Muthen posted on Friday, March 09, 2012 - 3:38 pm
I suspect that the cases that are being deleted have missing on the observed exogenous x variables. Missing data theory does not apply to these variables as the model is estimated conditioned on them.
 Jake Lant posted on Friday, March 09, 2012 - 6:22 pm
Thank you, Linda. And yes you're right (of course), that's precisely the error message that I got. So this means I will HAVE to exclude these variables?
 Linda K. Muthen posted on Friday, March 09, 2012 - 6:31 pm
You can bring them into the model by mentioning their variances in the MODEL command. They will then be treated as dependent variables and distributional assumptions will be made about them.
 Jake Lant posted on Friday, March 09, 2012 - 6:38 pm
Also, I think this is important.

These variables were somewhat "expected" to be missing because the participant didn't meet the criteria so we got them to check a box saying that it didn't apply just so that we have explanations for these missing variables.

Do you think there is a method that would probably fit this scenario better?

Thanks again.
 Jake Lant posted on Friday, March 09, 2012 - 8:14 pm
I did it (mentioned variances) and it worked, but I am not really liking the results because it really makes a somewhat big difference from excluding versus FIML, and I want to make sure there isn't a better way.

So I am thinking I should use the PATTERN IS command (missing by design) but I had set these as missing (i.e., -99, -100). But what kind of coding can I use to get it to accept it as a "pattern"?

I had done, for example,

MISSING ARE p1 (-999.00) p2 (-992.00) p3 (-100.99)

I tried:

PATTERN IS p1 (-999.00) p2 (-992.00) p3 (-100.99)

-looking for a shortcut, ;)- but of course it failed.

The example says,

PATTERN IS design (1= y1 y3 y5 2= y2 y3 y4 3= y1 y4 y5).

But I am not 100% as to what 1 = y1 y3 y5 would be here.

Any help to clarify this would be appreciated.

Thank you.
 Linda K. Muthen posted on Saturday, March 10, 2012 - 9:17 am
Missing data as you describe should not be treated as true missing data. You should not bring the variances of these variables into the MODEL command. Data like these should be analyzed in subsets.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message