Binomial Logistic Regression
Message/Author
 anny fenton posted on Wednesday, July 02, 2014 - 11:08 am
Hi there,

I am attempting to run a model with a rate as an outcome. I believe that specifying a binomial distribution with a logit link would be best, but I'm unsure how to specify this distribution in MPlus. Thank you, Anny
 Tihomir Asparouhov posted on Wednesday, July 02, 2014 - 6:17 pm
You would have to spread out the data and use a two level model. if you have a rate of 2/5 record that as cluster 1
binary cluster
1 1
1 1
0 1
0 1
0 1
etc. if the next observation is 1/3 record this as cluster 2
1 2
0 2
0 2
Then you simply analyze the data as binary two-level data and you can add predictors etc.
 anny fenton posted on Wednesday, July 09, 2014 - 8:47 am
Thank you so much for your response. I had two follow up questions:
(1) How would I interpret my coefficients? My outcome is a person's estimate of the foreign population size so how might I interpret say a one year change in education?
(2) Do you have a suggestion for what estimator I should use? The outcome is a mediator for several outcomes in my SEM which is already clustered at a regional level and so effectively a two level model.

Thank you for any advice in advance!
 Tihomir Asparouhov posted on Wednesday, July 09, 2014 - 11:41 am
(1) To obtain model estimates for specific question you might have to use the factor scores for the between level part of the variable.

(2) You can use MLR and add the extra level of clustering like this using

type=complex twolevel;
and
cluster=region record;
 anny fenton posted on Friday, July 11, 2014 - 8:31 am
Thank you again Tihomir. I'm still struggling to understand what is on the between level versus the within level, however.

Say my model is x -> y -> z where there is also a relationship between x and z (x -> z). y and z are at the person level and x is at the region level. y is the rate so it also has the binary cluster.

As I understand it, I would want to use the following code, but this code has failed to work. If you have any suggestions, it would be much appreciated. Thank you.

CLUSTER is y.clust x.clust;
BETWEEN = x z;

ANALYSIS:
TYPE = COMPLEX TWOLEVEL;

Model:
%WITHIN%
y;
%BETWEEN%
y ON x;
z ON y x;
 Linda K. Muthen posted on Friday, July 11, 2014 - 10:07 am
Please send the output and your license number to support@statmodel.com.