Pascal,
You have 10 observed indep. variables, 3 (1 model + 1 weights for each
class) {W123}
-> You can have for each segment the probability of observing the Dep.
Variables 5 levels (either by importing results from Latent class or by
computing the likelihood in SPSS) {P123}
-> You have the weights of each class..Its the probability taht an
observation belong to the class
-> You can perform the overall prob for each Dep. Variables 5 level
({W123}*{P123} a 1x3 by 3x5 -> 1x5 prob
-> Either you specify the expected level with the highest prob, or compute
the weighted mean (if it make sense)
Hope I've been clear
Naji
-----Message d'origine-----
De : [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
la part de Pascal van Hattum
Envoy� : jeudi 29 janvier 2004 15:46
� : [EMAIL PROTECTED]
Objet : [edstat] Latent Class Regression using Latent Gold
Hello,
Who can help me to rebuild the classificationmodel of Latent Gold in
SPSS. The following is the problem:
I used Latent Gold to make a latent class regression model with 10
nominal independent variables and 1 nominal dependent variable (with 5
levels). In Latent Gold I specified 3 latent classes. So my model will
include 3 nominal regression models (for each latent class one model).
Now I want to rebuild the model in SPSS to predict the levels of so
called 'hold out cases' (using the model and the independent variables
I want to predict the level of the dependent variable and the latent
class) Using the latent gold outputdata I can make a 5x3 matrix with
conditional probabilities P(Y=y|X=x) for y=1,...,5 (number of levels
dependent variable) and x=1,...,3 (number of latent classes). If I
know the latent class I can obtain the predicted level of the
dependent variable and vice versa if I know the predicted level of the
dependent variable I can obtain the latent class). So I need to know
one of the predicted variables to obtain the other. But how can I get
one of these variables with the latent gold output and the 5x3 matrix
with conditional probs?
I hope that someone can help me.
Thanks in advance,
Pascal
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
