@ARTI1 wrote: Hello I have a dataset with ~ 130K records. I want to pick a sample size of 11000 depending upon the city flag I have. City A will have 40% of records, City B will have 40% records and City C will have 20% records. (4400, 4400, 2200). I can use the obs method, but I want to randomly select these records. Is there a way in survey select to accomplish this, or is there any other method? Thanks in advance! This sounds to me like a stratified sample issue. So something like; Proc sort data=have; by city; run; proc surveyselect data=have out=selected samprate=(40 40 20); strata city; run; would be a generic approach. There may be issues with the order of samprate values. They need to match the sorted order of the stratification variable. So if your "city" values are character the alphabetical value of the name needs to match the samprate. The first sorted name will be selected at 40%, the second at 40% and the third at 20%. OR you could specify SAMPSIZE = (4400 4400 2200) which may be preferable as the rates sometimes will be off by a couple if you need an exact number.   The sort is needed to use the strata approach. Of course use your data set and actual variable names. ... View more

Another way to approach this is to maintain a dataset that has a single value...the date it last ran. You then pass that value into a start_date variable that can be used in your program. At the very end, you overwrite the dataset with today's date, that becomes the start date, the next time you run it. This way, if you happen to have a day it can't run during the week, you pick up that data the next day it does run.  libname mydata "directory_path"; data _null_; set mydata.last_run_date; call symputx('start_date',"'"||put(date_last_run,date9.)||"'d"); run; data want; set have; where date_var >= &start_date. and date_var < date(); run; data mydata.last_run_date; date_last_run=date(); run; ... View more