Avoiding duplication when creating new variables

[Beanpot]

Hi,

I have a dataset that has multiple variables which can have the same values within different values. For example:

Obs	Var1	Var2	Var3
1	n1	n2	n3
2	n3	n4	.
3	n5	n1	.

In order to determine the frequency of each particular value I'm creating a new variable for each value using the following code:

data want;
set have;
if whichc('n3',of var_1-var_20) then n3=1;
else n3=0;
run;

The problem I'm encountering is that when I check the frequency of the values I create against the QC table I'm off by a small percent. For example, the QC table might say total observations with value n3 is 10,000 and when I do a proc freq on the new variable I get 10,020.

I'm wondering if the code might be double counting? If so I'm not sure how to fix this. Are there other reasons the counts could be off?

Thanks!

[Beanpot]

To clarify, if an observation had the value n3 in under multiple variables, would my code count that twice?

[PaigeMiller]

Your code using WHICHC would count it once per record, not once per variable.

A carefully examination of the results from your code would confirm this.

[Tom]

Your posted code is checking if ANY of the 20 variables is exactly equal to the string 'n3'.  It will produce the same result when all 20 of them have n3 as it will when only one of them does.

Do any of the values have leading spaces?  Or contain other invisible characters like TAB, LF, CR, FF, non-breaking space?  '090A0D0CA0'x.  For WHICHC to work they need to match exactly.   

[Quentin]

As an alternative approach, you could transpose your data into a vertical format, and the de-duplicate it to avoid double-counting, then run PROC FREQ.  Something like:

data have ;
  input id Var1 : $2. Var2 : $2. Var3 : $2. ;
  cards ;
1 n1 n2 n3
2 n3 n4 .
3 n5 n1 .
4 n1 n2 n2
;
run ;

proc transpose data=have out=vert ;
  var _character_ ;
  by id ;
run ;

proc sort nodupkey data=vert out=vert2 ;
  by id col1 ;
run ;

proc freq data=vert2 ;
  tables col1 ;
run ;