topic tests of equality after PROC MIXED or PROC GLIMMIX in Statistical Procedures

tests of equality after PROC MIXED or PROC GLIMMIX

srbotto — Mon, 20 Aug 2018 17:58:24 GMT

Hello. Is there a way to test for equality of coefficients in multilevel models (using SAS 9.3 or 9.4)? I have some multilevel (random effects) models that I am estimating with PROC MIXED and I would like to know if two coefficients are statistically different from one another.

If this was not a multlevel I would have something like the equation below and use the test option,

y=X1B1+X2B2+X3B3+e

proc reg data=temp1;

model zwfl=x1 x2 controllist;

t1: test x1=x2;

Is there a way to do this in a multilevel context? So after the following code?

proc mixed data=temp1;

class nbhid;

model zwfl=x1 x2 controllist/solution;

random intercept/sub=nbhid;

I would also like to do this after running a PROC GLIMMIX (similar situation with dichotomous outcome).

Thank you!

Betreff: tests of equality after PROC MIXED or PROC GLIMMIX

Daniel_Paul — Mon, 20 Aug 2018 18:38:58 GMT

Hey,

are the statements CONTRAST and ESTIMATE what you are looking for?

Bye,

Daniel

Betreff: tests of equality after PROC MIXED or PROC GLIMMIX

srbotto — Wed, 29 Aug 2018 12:39:57 GMT

I thought that might be it, but aren't those more for predicted probabilities and not for testing the difference between estimated coefficients/effects?

Betreff: tests of equality after PROC MIXED or PROC GLIMMIX

PaigeMiller — Wed, 29 Aug 2018 12:46:32 GMT

To matche the PROC REG example of testing x1=x2

You would use a contrast statement that looks at x1-x2 to see if the result is not significantly different than zero.

Betreff: tests of equality after PROC MIXED or PROC GLIMMIX

srbotto — Wed, 29 Aug 2018 16:30:41 GMT

Thank you! I guess I hadn't quite understood the contrast statement.