PROC GLIMMIX gave "0" random effect

[AJ8]

Hi there!

I was trying to run a 2-level model, but the random effects are showing up as zeros. Here are my codes:

proc glimmix data = ONE METHOD=RSPL NOCLPRINT;

CLASS Researcher_Identity Researcher_Factor1 Researcher_Factor2 Patient_Factor1 Pateint_Factor2;

MODEL Outcome = Researcher_Factor1 Researcher_Factor2  Patient_Factor1 Pateint_Factor2  / DIST=Binary LINK=logit;

RANDOM Intercept / subject= Researcher_Identity TYPE=VC G;

run;

THANK YOU SO MUCH!

[PaigeMiller]

It's not clear to me that you have actually stated a question or stated a problem.

Zero is a valid result, depending on your data.

Please elaborate. If possible, provide your data.

[AJ8]

Hi there!

Thank you for your response and sorry for not stating the issue.

Basically, I am trying to put people's identity at the second level and see if there is any random effect. I have a bunch of other variables mostly ordinal and dummy at the first level, such as age group, years of experience, gender etc. Data looks something like this:

ID   ID's_Gender   ID's_Age_Group   Outcome    Variable1   Variable2

1          F                           1                    Yes            1                3

1          F                           1                     No             2                3

1          F                           1                     No             1                2

2         M                            2                    Yes            2                2

2         M                            2                    Yes            1                3

So here are my quetsions:

1) My random effects turned out to be all zero. How do I tell if the zero is a valid result?

2) If it is, can I conclude that there is no random effect and change the model to a logistic regression and simply place people's identity as a vairable? If I do this then I am not able to control for any variables related to the ID. 

Because the question of interest is really to see how a perosn did comparing to another.

THANK YOU SO MUCH!

[PaigeMiller]

I'm still not sure I can help ... your variable names in PROC GLIMMIX don't match the sample data you provided, so I have no idea how they relate to the model statement.

Nevertheless, it still is possible for 0 to be the result. This happens either because you have mis-specified the model (something I won't comment on, as I have no idea what the experiment is or what you want to do with this data), or because the variability is truly zero.

Again, to answer your questions, we'd need a discussion of the problem, a discussion of the experiment, and why this model fits the experiment, and we probably need the entire data set, which you probably can't provide.

[AJ8]

Hi there,

using the data I have provided, the code would look something like this:

proc glimmix data = ONE METHOD=RSPL NOCLPRINT;

CLASS ID ID's_Gender Variable1 Variable2

MODEL Outcome = ID's_Gender ID's_AgeGroup Variable1 Variable2  / DIST=Binary LINK=logit;

RANDOM Intercept / subject= ID TYPE=VC G;

run;

[PaigeMiller]

There's a lot more information I asked for, and it hasn't been provided.

In other words, without a description of the experiment, and the full data (which you may not be able to provide), there's no way I can say if the 0 value is correct. You might want to find a statistician where you work/where you go to school and see if that statistician can help.

And I am curious to understand how your variable names in SAS can have an apostrophe in the variable name ...

[Rick_SAS]

Assuming that you haven't accidentally specified the NOINT option, a possible explanation is that you have a linear dependency (multicollinearity) in your explanatory variables.  Are there any NOTEs in the SAS log about matrices not being positive definite?  Please copy/paste the complete SAS log so we can see the exact messages.

[AJ8]

Hi there!

Thank you so much for your response! I realized that I actually included a wrong variable by accident and I am all good now.

THANK YOU!