DOMAIN vs WHERE Statement

[kmardinian]

I am working with a dataset where I only want ages 18-45, so I created a flag for it. But when I seem to run a surveylogistic with it, I get this WARNING with the DOMAIN statment;

1 OPTIONS NONOTES NOSTIMER NOSOURCE NOSYNTAXCHECK;

61

62 proc surveylogistic data=assign ;

63 repweight rakedw1--rakedw80 / jkcoefs=1;

64 weight rakedw0;

65 class fruitveg (ref="ALWAYS") age (ref='Ages 18 to 26') ethn (ref="White") BMI (ref='UNDERWEIGHT/NORMAL WEIGHT') sugar

65 ! (ref='0 to 100 sugar drinks per month') exercise (ref="Less than 30 min walking/week") educ (ref="Less than High School")

65 ! income (ref="0 tp 50,000") / param=ref ;

66 model diabetesoutcome(ref= "NO") = fruitveg age ethn BMI sugar exercise educ income;

67 domain flag;

68 format diabetesoutcome diabetesoutcomex. fruitveg fruitvegx. age agex. gender genderx. ethn ethnx. BMI BMIx. sugar

68 ! sugarx. exercise exercisex. smoke smokex. educ educx. income incomex. geo geox.;

69 run;

NOTE: PROC SURVEYLOGISTIC is modeling the probability that diabetesoutcome='TYPE 2'.

WARNING: Convergence was not attained in domain flag=1.

WARNING: No analysis for domain flag=1.

NOTE: Convergence criterion (GCONV=1E-8) satisfied.

NOTE: PROC SURVEYLOGISTIC is modeling the probability that diabetesoutcome='TYPE 2'.

NOTE: Convergence criterion (GCONV=1E-8) satisfied.

NOTE: PROCEDURE SURVEYLOGISTIC used (Total process time):

real time 3.03 seconds

cpu time 3.00 seconds

 But when I switch the DOMAIN statement to a WHERE statement I do not get a WARNING, and I am trying to understand why that is and whether it will affect my results. Thank you!

[PeterClemmensen]

The DOMAIN statement requests analysis for each level/domain in your domain variable. In your case you have two domains: 1(Age 18-45) and 0(outside age 18-45).

The where statement simply excludes observations where the condition in the where statement is not satisfied.

Provide some sample of your data if you want a more usable answer 🙂

Also consult the PROC SURVEYLOGISTIC Documentation

[SAS_Rob · Accepted solution]

Unless the sample sizes of the age groups are fixed, that is part of your study design, then you should not use a WHERE statement.  My suggestion to avoid the issue is to set all the response variables for the group that you are NOT interested in to missing and then use the NOMCAR option on the SURVEYLOGISTIC statement.  This will give you a valid subgroup analysis (since the NOMCAR option treats missing values as a subgroup) and fit only the model you are interested in.