[R] Still on poLCA

千早 ケンジ chihayakenji1893 at gmail.com
Fri Mar 5 17:19:11 CET 2010


Dear all,

I have just sent a message asking about poLCA but I thought of another question I wanted to ask. I get the G^2 statistic in my output and want to test for its significance. I get that the degrees of freedom for the test are (S-1-p) where S is the number of different patterns observed and p is the number of estimated parameters. Are these the "residual degrees of freedom" that I get in the output? 

========================================================= 
Fit for 2 latent classes: 
========================================================= 
number of observations: 1559 
number of estimated parameters: 25 
residual degrees of freedom: 1534 
maximum log-likelihood: -7419.601 
 
AIC(2): 14889.20
BIC(2): 15023.00
G^2(2): 1088.866 (Likelihood ratio/deviance statistic) 
X^2(2): 2284.071 (Chi-square goodness of fit) 
========================================================= 

If I use these degrees of freedom for the test I get really high probabilities for the model even with only 2 classes. Am I doing something wrong? If these are not the degrees of freedom for the test is there any way to calculate them (i.e.: finding the S  to substitute in the S-1-p formula)?

Kind regards

Guilherme Kenji


More information about the R-help mailing list