Finite Mixture Regression for Segment...
Message/Author
 Pierre S. posted on Wednesday, December 30, 2015 - 1:52 am
Hi Drs Muthen,

I'd like to clarify whether I have the correct syntax or not.

Background: I have a model with several IVs (observed), say X1, X2, X3, and 1 latent DV (say: Y by y1, y2, y3 - I assume that the measurement model is equal for all respondents). I want to find out whether (a) there are different segments/groups of respondents that differ with regards to their b1, b2 and b3 path coefficients and/or (b) there are different segments/groups of respondents that differ with regards to their b1, b2, b3 AND x1, x2, and x3 (intercepts?).

As far as I understand, I need a finite mixture model, try different numbers of classes (“segments”) and chose the best one based on various criteria (AIC, BIC etc.)

I created this syntax:

Classes = c (2);
Analysis:
type = mixture;
starts = 500 10; !increase once syntax correct

MODEL:

%overall%
Y by y1 y2 y3;
Y on x1 x2 x3;
%c#2%
Y on x1 x2 x3;

Questions
Q1: Is this syntax correct to identify different segments based on the b1, b2 and b3 path coefficients?
Q2: To try whether more groups/segments are ‘better’, I change the number of classes (e.g, to three) and add %c#3% & Y on x1 x2 x3; below – correct?
Q3: How should my syntax look like if I want to create my segments based on b1, b2, b3 AND x1, x2, x3?

Thx!
 Bengt O. Muthen posted on Thursday, December 31, 2015 - 5:42 pm
Q1. Segments will be based on those 3 slopes but also the y intercept being different across segments.

Q2. Right.

Q3. If you want the means of the x's to contribute to the segmentation you should bring the x's into the model by mentioning their means or variances.