RANDOM statements in PROC GLIMMIX for cluster, strata, and weight in a longitudinal mixed effects

[kai_cody]

Hi all,

I am using PROC GLIMMIX perform a longitudinal binary mixed effects model on complex survey data. I am unsure what I need to do with the RANDOM statements to account for everything. Completely new to mixed effects models, sorry!

First, I have cluster (PSU - 2 levels), strata (99 levels), and final weights for each ID for each of four timepoints (Wave 1, 2, 3, 4 -- "repeated measures"). Here is an example of my dataset structure for the first two ID participants:

ID        WAVE      PSU       STRATA     WEIGHT          STROKE 

1           1              2             94              34724.3           No

1           2              2             94              48284.2           No

1           3              2             94              52843.3           No

1           4              2             94              48848.4           Yes

2           1              1             55              553.6               No

2           2              1             55              364.1               No

2           3              1             55              152.4               No

2           4              1             55              868.0               No

How should my PROC GLIMMIX code look? Here is what I have below, but does this account correctly for everything? Should I account for weighting in a separate WEIGHT statement?

proc glimmix data=DATASET method=quad;
class ID WAVE PSU STRATA V1 V2;
model STROKE = V1 V2  WAVE / dist=bin link=logit solution;
random PSU(STRATA) / subject=ID weight=WEIGHT;
random WAVE / subject=ID weight=WEIGHT;
run;

Thank you very much!

[gcjfernandez]

Please refer the following SAS Global Forum paper; https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2020/4937-2020.pdf

[kai_cody · Accepted solution]

This is what I ended up with. I think it is correct, but now SAS is saying there is insufficient memory to run this, even after I adjusted MEMSIZE. But anyways, maybe this is helpful:

proc sort data=PATH_DATA;
by ID WAVE; run;

proc glimmix data=PATH_DATA;
class GENDER2 RACE2 HISPANIC2 AGE2 ALCOHOL CIG DIABETE EDUC1 EMPLOY1 DUAL VARPSU VARSTRAT INCOME MARITAL1 BMI INSURANCE1 REGION1;
model STROKE = GENDER2 RACE2 HISPANIC2 WAVE AGE2 ALCOHOL CIG DIABETE EDUC1 EMPLOY1 DUAL INCOME MARITAL1 BMI INSURANCE1 REGION1/ link=logit solution oddsratio;
random WAVE intercept / subject=ID;
random VARPSU(VARSTRAT);
weight WEIGHT;
run;