How to apply the RESTRICT statement in proc FMM with k=4

[npandis]

Hi,

I am fitting a finite mixture model using the following code:

proc fmm data=mydataset;
class x1  ;
model y = x1  x2 / dist=poisson k=4;
restrict x2 1, x2 -1,x2 1,x2 -1;
run;

I  would like to restrict x2 to be the same across the 4 classes. I have seen examples for k=2 and in fact the code above makes the x2 coeffs equal in the first 2 components  but the coefs in the 3rd and 4rth components are not the same.

I have tried different combinations but they have failed.

It would be great if you could help me correctly add the restrict statement.

Thank you,

Nikos

SAS 9.4 

[Rick_SAS]

It looks like your problem is equivalent to modeling (y - x2). I suggest you create a new variable 

Y2 = Y - X2;

and then

MODEL Y2 = X1 / dist=Poisson k=4;

[npandis]

Dear Rick,

Thank you for the response which I tried to implement.

The model converged but provided no estimates for the predictor and only values for the intercepts of the 4 components.

The Pearson statistic is also strange 4.12E-15! I have about 400 observations.

For the above model I subtracted the continuous predictor X2 from the Y and X1 was binary.

Also, what if all my predictors are all categorical? 

Best,

Nikos

[Rick_SAS]

OK, thanks for the report. I haven't used the RESTRICT statement in FMM, so I thought I'd suggest a workaround. Hopefully, someone with more experience will provide an answer.

[npandis · Accepted solution]

I have found a solution by applying 3 restrict statements:

restrict X2 1 , X2  -1;
restrict X2 1, X2  0, X2 -1;
restrict X2  0, X2  0, X2 1, X2  -1;

Thank you,

Nikos