This may be overkill for your situation, but when I used to work with US Medicare data, which has separate files for each year within a given domain (e.g., inpatient, outpatient, etc.), I built a macro that generated SAS views on the fly given the start and end year of the analysis in question. What it generally did was align the variables according to the least granular data available for that time range. So, in your example, if for example lang_cantonese and lang_mandarin replaced lang_chinese in 2021, and your current analysis uses data from 2021 to 2024, then the view would keep the more granular variables only. However, if that same analysis went back to 2020, then the view would translate the granular variables into courser ones:
** NOTE: this is a *very* rough, untested sketch and the assumption is
that this data step is inside a macro ;
%macro mkview(startyear=, endyear=);
proc format;
value fvers
2016-2020="J"
2021-2024="K"
;
run;
data _null_;
call symputx("version", put(&startyear, fvers.));
run;
data vw.want view=vw.want;
set
%do yyyy=&startyear %to &endyear;
have&yyyy (in=year&yyyy)
%end;
;
array yr {&startyear:&endyear} year&startyear-year&endyear;
length year 3;
do yyyy=lbound(yr) to hbound(yr);
if yr[yyyy] then year=yyyy;
end;
%if &version=J %then %do;
* do version J stuff, e.g., ;
lang_chinese=max(lang_chinese, lang_mandarin, lang_cantonese);
* other version J stuff... ;
%end;
%else %if &version=K %then %do;
* version K stuff ;
%end;
run;
%mend;
Again, that's just a sketch - obviously some situations will be more complicated, but that's the general premise. These views would then be saved in a permanent directory (if needed) that's specific to a given analysis and used as if they were regular datasets.
