Safa Aouinti Thanks a lot for your help.

Dear Jibin;

You may also find this documeny useful:

https://www.cdc.gov/nchs/tutorials/NHANES/NHANESAnalyses/AgeStandardization/Task1b_SAS92.htm

Sincerely,

@Edris Haghir, Thanks a lot for your suggestion!

why not using something like this based on a surveymeans (adapted to NHANES)? Here also, you can code as 1/0 or 100/0 your categories to have the prevalence and keep the original metric for continuous (to have age adjusted estimates for continuous)

proc surveymeans data=/*dataset name*/;

STRATA sdmvstra;

CLUSTER sdmvpsu;

CLASS age_c; /*your age class variable (10 years age class)*/

WEIGHT WTINT2YR;

var X; /*the variable you want to have (continuous or 1/0 or 1/100 for prevalence)*/

run;

I add a little piece. I feel awkward with the regression procedure because you should have something like...

proc surveyreg data=f1;

STRATA sdmvstra;

CLUSTER sdmvpsu;

CLASS age_c;

WEIGHT WTINT2YR;

MODEL edu=age_c/noint solution;

estimate "age standardized prevalence of edu" age_c 0.0696 0.0747 0.1329 0.2408 0.2855 0.1965;

run;

here the numbers 0.0696 0.0747 0.1329 0.2408 0.2855 0.1965 are the coefficients of my estimate statement (here I have my relative frequencies) ok if I use this reg I obtain 38.3% but using the observed frequencies is not very rigorous

if I use age classes frequencies estimated using a survey freq then I have other coefficients and my procedure became

proc surveyreg data=f1;

STRATA sdmvstra;

CLUSTER sdmvpsu;

CLASS age_c;

WEIGHT WTINT2YR;

MODEL edu=age_c/noint solution;

estimate "age standardized prevalence" age_c 0.077829 0.106935 0.208546 0.247302 0.233024 0.126364;

run;

Now I obtain 36.3 %

If I run that simple proc surveymeans (the first procedure I told) then I get the same 36.3%

basically using the surveymean you have the same result... But less work...

Does it make sense for you?