Andrea <alamont082@GMAIL.COM> wrote
>I am trying to run a LGM with an ordinal outcome.
I'm guessing you mean a logistic regression model? Using PROC LOGISTIC? Or what?
> Its not running,
What do you get in the log?
>and I realized that the problem was empty cells in the bivariate
>correlations. Indeed, when I run proc corr, correlations come up
>blank (.) for these two variables. One variable has n=171 and the
>other has n=71.
The overall N is irrelevant if the problem is what you said it was. If you have
cells with 0 cases, then logistic models will have problems. You can't divide by 0.
But you should show the log.
> Can this be the problem? This is only happening with
>the last time point.
Well, it certainly COULD be. But the log will tell us.
And if you have multiple time points, how are you dealing with that?
Correlations run nicely for all other time
>points for the same variables (although N's may differ due to planned
>missingness). How can I fix this problem?
First, tell us what you are trying to do.
What's your DV? What are your IVs? What is your question of interest? What hypotheses have
you got? Or is this purely exploratory. You allude to multiple time points ... how many have you got?
What's the design? Is it repeated measures, or are these different subjects?
Then maybe we will be able to help
Peter L. Flom, PhD
Website: http://www DOT statisticalanalysisconsulting DOT com/
||1/8/2010 10:24:32 PM