topic non-monotone missing data proc gee in Statistical Procedures

non-monotone missing data proc gee

Sanscrit_2prov — Tue, 27 Jul 2021 16:38:25 GMT

I am trying to use the Proc GEE missmodel statement and keep getting an error about my data being non-monotone. But when I look at the pattern table SAS creates it looks monotone to me. (see log below)

Distribution Poisson
Link Function Log
Dependent Variable hosp_outcome

Missing Data Patterns

----------------Occasion-----------------
Group 1 2 3 4 5 6 7 8 9 Freq Percent

1 X X X X X X X X X 2805 74.74
2 X X X X X X X X . 153 4.08
3 X X X X X X X . . 125 3.33
4 X X X X X X . . . 123 3.28
5 X X X X X . . . . 119 3.17
6 X X X X . . . . . 128 3.41
7 X X X . . . . . . 116 3.09
8 X X . . . . . . . 106 2.82
9 X . . . . . . . . 78 2.08

These are the statements I used:

/*Generate Point Estimates*/
ods output GEEEmpPEst/*parameterestimates*/=point_JointParms1;
proc gee data=combine2 desc plots=histogram;
class spid ctime;
missmodel asthma ;
model hosp_outcome=
newtime
asthma
/dist=poi;
repeated subject=spid/within=ctime corr=ar(1);
output out=hosp_parms_whites xbeta=xbetaout predicted=predout;
title 'Hospitalization Robust SE Whites Poisson Weighted ';
run;

Has anyone seen this before and how to fix?

Re: non-monotone missing data proc gee

ballardw — Tue, 27 Jul 2021 16:51:53 GMT

Please show the LOG with the code and all the messages for Proc GEE. Copy from the Log and paste into a text box opened on the forum with the </>. The text box preserves formatting of the text which helps if any SAS diagnostic characters are included. The main message windows will reformat text and may lose meaning depending what is removed.

Re: non-monotone missing data proc gee

SteveDenham — Tue, 27 Jul 2021 17:52:29 GMT

I am not sure at all if this is the issue, but the counts are non-monotonic by group 1 to 9, so it could be that for some group or occasion combination, there is something unusual. In particular, I would examine Group 6, which has a count greater than Groups 3, 4 and 5. The documentation says "If the pattern of missing data is intermittent (not dropout), the GEE procedure terminates and does not perform an analysis." I would suspect that there is at least one intermittent missing value for asthma in there.

SteveDenham.

Re: non-monotone missing data proc gee

MichaelL_SAS — Tue, 27 Jul 2021 18:20:48 GMT

It might be the case that PROC GEE is actually returning the incorrect ERROR message.

It is mentioned in the "Missing Data" section of the PROC GEE documentation, but the WGEE methods also require equal size clusters/subjects. That is, for each level of the SUBJECT= variable there must be the same total number of observations (observed + missing). It might be the case that you have a cluster with less than 9 total observations and that equal cluster condition is violated but PROC GEE is returning an incorrect and unhelpful error message. I would recommend contacting SAS Technical Support about this issue.