One possibility is to use Pearl regular expressions. E.g.: data test; Address = '44 Bird Street'; PrxID=PrxParse('/\d+\s+(.*)/'); if PrxMatch(PrxID,Address) then StreetName=PrxPosN(PrxID,1,Address); run; This function just finds one or more digits ("\d+"), followed by one or more blanks ("\s+"), and puts the rest of the string into a capture buffer ("(.*)"), the contents of which is returned by the PrxPosN function. You will probably find that more sophistication is needed (which is why I suggest that you get into PRX functions), but this is a start.   ... View more

Rezza is of course correct.   Remember, the data step is an implied loop through every record in the input datasets.  If while in a DATA step loop it executes a SET, MERGE, UPDATE, MODIFY, or INPUT statement, it reads the next record into the PDV. If there is another records after the one read, it remembers that.  It then progress through the rest of the data step code.  If it finds an OUTPUT statement, it writes out the current record and remembers it has been written.  When it hits a "run" statement (or another DATA or PROC statement if you forget the RUN), and it hasn't already been output, it outputs the record.  If there are still more records from the last input dataset, it goes back to the top of the code and starts over.  You can have more than one OUTPUT and it will output the record every time it sees one. ... View more

90% plus of requests when I get to process similar data the user would like the "race" responses in some specific order. The easiest way I have found to do that consistently is to assign a numeric value to the "final race" variable and an associated format that displays the text in the desired order. So I would start with creating the Format first such as: Proc format; value raceorder 1 = "Asian" 2 = "Black" 3 = "Native American" 4 = "White" 5 = "Multiple Race" 6 = "Unknown" ; run; And then use the Sum/ whichn information to assign the above numeric values. My example includes Unknown category because I get too much garbage and always have some in most of my data. That would be the sum of races le 0. (less than in case your system has MISSING values) or Whichn result of 0.   Example of creating data with "race" variables not in the possibly desired report order and making the final race variable match a report order as defined in that format above. data have; input ID race1 race2 race3 race4 ; label Race1 = 'White' Race2 = 'Black' Race3 = 'Asian' Race4 = 'Native American' ; datalines; 1 1 0 1 0 2 0 1 0 0 3 0 0 0 1 4 1 0 0 0 5 0 0 1 0 6 0 0 0 0 ; data want; set have; array r race: ; if sum(of r(*))=2 then final_race=5; else select (whichn(1, of r(*))); when (1) final_race =4; when (2) final_race =2; when (3) final_race =1; when (4) final_race =3; otherwise final_race=6; end; run; proc freq data=want; tables final_race; format final_race raceorder.; run; I use an explicit value for the "Unknown" or missing or what ever your system may want because if you use an actual missing value then the sort order place the "Unknown" first.   Style choices abound in variations of this. The format approach can become extremely powerful if you get to work with one of the data sets that has subgroups of Asian (one I have seen has something like 24) or Hispanic split up into geographic sub groups. Then Multilabel formats could provide the way to do subgroup/main "race" categories.     ... View more