How to add subset data step by step

[lichee]

Hi all,

I wanted to examine data with subset cohort included step by step. For example, there are five cohorts, A-E, and I wanted to run regression with the first cohort, and then I would run regression with an additional cohort each time. How can I conduct this using MACRO or iteration efficiently?

proc reg data=mydata;

     model dependent=independents;

     where cohort in ('A');

run;

proc reg data=mydata;

     model dependent=independents;

     where cohort in ('A','B');

run;

proc reg data=mydata;

     model dependent=independents;

     where cohort in ('A','B','C');

run;

proc reg data=mydata;

     model dependent=independents;

     where cohort in ('A','B','C','D');

run;

proc reg data=mydata;

     model dependent=independents;

     where cohort in ('A','B','C','D','E');

run;

I would think I can probably start with creating the list like below:

proc sql noprint;
select distinct quote(cohort) into :st_list separated by ','
from mydata;
quit;

[Astounding]

As part of your plan, note that the IN operator does not require commas. You could shoot for (for example):

where cohort in ('A' 'B' 'C')

That will simplify subsequent programming.

Also note, SQL automatically creates &sqlobs which would be 5 in this case (number of items extracted). So your program could continue by writing a macro along these lines:

%macro regloop;
%local n;
%do n=1 %to &sqlobs;

Then have the macro generate PROC REG, using the first &n items from &stlist.

Definitely needs some details worked out, but that should give you a way to approach the problem.

[PaigeMiller]

I assume that the code you showed works. What would a macro do that this code does not do? You mention "iteration", what would be iterating?

I also question the statistical methodology here about adding cohort after cohort into the data in sequential fashion like you are doing. I would imagine a better way to see the impact of cohort is to add it into the model so that you can have different intercepts and/or different slopes for the different cohorts.

[lichee]

Thank you both! I'm testing out my code, but cannot make it work. Just not good at iteration, especially when it's in macro.

[PaigeMiller]

---

@lichee wrote:
I'm testing out my code, but cannot make it work. Just not good at iteration, especially when it's in macro.

---

Its still not clear to me what "iteration" you mean, where in this problem is there "iteration"? Please describe in detail.

[lichee]

I'd like to see how to add cohorts one by one given the list of cohorts. Thanks!

[lichee]

Thank you Paige! I know it's statistically not correct way to add cohort after cohort into regression analysis. But I'm just giving a quick and simple example.

[Astounding · Accepted solution]

OK, here's a bit more of what a macro could look like.  All of this would follow the code you originally posted.  It assumes you remove the comma and use a blank as the "separated by" character.

%macro regloop;
   %local n subset;
   %do n=1 to &sqlobs;
      %let subset = &subset %scan(&stlist, &n);
      proc reg data=mydata;
        model dependent = independents;
       where cohort in (&subset);
      run;
   %end;
%mend region;

%regloop

This will generate 5 PROC REGs, just like you have.  But the cohort values (and the actual number of PROC REGs) will depend on the data set you are processing.