Querying dataset for unique column values

[serge68]

I have an input dataset with two columns(system and class). I want to create a dataset from that containing only those systems that only have a class of x. The same system can have multiple entries. If one entry has a class of X, and another entry a different class, then I don't want the system returned. There are 36 potential values of class(a-z and 0-9).

My input looks like this:

  data have;                  
    input system $ class $ ;  
    datalines;                
 fred  a                      
 fred  x                      
 wilma a                      
 wilma b                      
 wilma b                      
 betty x                      
 betty x                      
 ;                            
 run;                         

From this I would only want Betty returned. As Fred has a class of both A & X, while Wilma doesn't have X, I don't want either of them. Betty is the only one that only has a class of X.

I'm assuming PROC SQL would be the way to go here. Though I'm struggling to get a query that can return what I'm after.

[PeterClemmensen]

Try this

  data have;                  
    input system $ class $ ;  
    datalines;                
 fred  a                      
 fred  x                      
 wilma a                      
 wilma b                      
 wilma b                      
 betty x                      
 betty x                      
 ;                            
 run;                   

proc sql;
   create table want as
   select * from have
   group by system
   having sum(class='x')=count(system);
quit; 

[Patrick]

Assuming you want to select rows for any case where there is only one class per system and not only when it's 'x'

proc sql;
  create table want as
    select * 
    from have
    group by system
    having count(distinct class)=1
  ;
quit;

[novinosrin · Accepted solution]

HI @serge68  A way to approach is to check

1. Distinct values of class within BY GROUP system

2. We need the distinct value as 1 and that value should be 'x'

So in SQL syntax

data have;                  
    input system $ class $ ;  
    datalines;                
 fred  a                      
 fred  x                      
 wilma a                      
 wilma b                      
 wilma b                      
 betty x                      
 betty x                      
 ;                            
 run;      

proc sql;
create table want as
select *
from have
group by system
having count(distinct class)=1 and class='x';
quit;

[serge68]

Perfect! Thanks.

[Kurt_Bremser]

Data step solution:

data want;
ind_x = 0;
do until (last.system);
  set have;
  by system notsorted;
  if ind_x ne . and class = 'x' then ind_x = 1;
  if class ne 'x' then ind_x = .;
end;
do until (last.system);
  set have;
  by system notsorted;
  if ind_x then output;
end;
drop ind_x;
run;

[RichardDeVen]

A DATA step can process pre-grouped data and keep one row for each system having only class='x' cases

DATA want(keep=system);
  do until (last.system);
    set have;
    by system notsorted;
    if not not_x then not_x = class ne 'x';
  end;
  if not not_x;
run;