Final class count and proportions wit...
Message/Author
 John Woo posted on Wednesday, October 20, 2010 - 4:36 pm
Hi,
I have two latent categorical variables, cd and ch.
cd and ch are indicated by their respective outcome-variables with conditional independence.

cd is 4 classes.
ch is 5 classes.

controlling for covariates, i am regressing cd on ch.

In the output, I get:

FINAL CLASS COUNTS AND PROPORTIONS FOR THE LATENT CLASSES
BASED ON ESTIMATED POSTERIOR PROBABILITIES

Latent Class
Pattern

1 1 0.00000 0.00000
1 2 4.37333 0.00554
1 3 0.00000 0.00000
1 4 20.38640 0.02581
2 1 40.18266 0.05086
2 2 117.64014 0.14891
... (cut)

The estimated size of group "1 1" and "1 3" are zero.
Because of this (I think) I am also getting the following regression output:

CD#1 ON
CH#1 -37.236 0.000 999.000 999.000
CH#2 -2.009 0.703 -2.858 0.004
..

CD#3 ON
CH#1 -19.090 0.000 999.000 999.000
..

How should I interpret the coefficients above that are associated with p-values of 999s?
 Linda K. Muthen posted on Thursday, October 21, 2010 - 2:30 pm
The coefficients with p-values of 999 have been fixed so have no standard error or test of significance. You would interpret them as those in CH class 1 have a zero probability of being in CD class one for example.
 John Woo posted on Thursday, October 21, 2010 - 3:21 pm
Thank you, Linda.

I have one more follow-up question.

If I am manually computing for multiple-imputed results (i.e., averaging the parameter coefficients), what value should I use for the coefficient value of those that have p-values of 999s.

For example, for the dataset above, should I use "-37.236" for cd#1 on ch#1 coefficient?
 Linda K. Muthen posted on Friday, October 22, 2010 - 7:39 am
I would average over the fixed parameters the same way as for the free parameters.
Post:
This is a private posting area. Only registered users and moderators may post messages here.