topic Re: Retain in SAS Programming

Retain

Tommer — Fri, 26 Jun 2020 00:02:03 GMT

Hello, I am trying to generate same values for by group ie.

abc 1
abc 1
abc 1
def 2
def 2
xyz 3
xyz 3

I used the following code:

data dt;
input dat $;
datalines;

abc

xyz

abc

def

xyz

def
;
run;

proc sort data=dt;
by dat;
run;

data dt2;
set dt;
by dat;
retain ord;
if first.dat then ord+1;
run;

It does work well, but I was wondering since I used just first.dat, I was thinking the ord+1 will only apply to first value per the by group. However, it showed up for all values. Is this because of retain statement? I understand retain statement holds the value but how is it applying the ord+1 for values which are not first in the by group? Please help.

Re: Retain

Patrick — Fri, 26 Jun 2020 00:32:59 GMT

ord+1 gets only executed when the condition is true. That's why ord has the same value within your by group.

And yes, ord is not missing for the 2nd to nth value within a by group because it is retained. In your code you could even remove the Retain statement and the value would still be retained because SAS syntax ord+1 also implicitly leads to a retain. ...but: If the value wouldn't be retained then your counter wouldn't work at all as in the first case you would add +1 to a missing.

Also to be considered: Depending on the syntax adding +1 to a missing value returns either missing or 1.

Here the result from multiple syntax variations.

data dt2;
  set dt;
  by dat;
  retain ord1 ord2 ord3;
  retain ord2_B 1;
  if first.dat then 
    do;
      ord1+1;
      ord2=ord2+1;
      ord2_B=ord2_B+1;
      ord3=sum(ord3,1);
      ord4+1;
      ord5=ord5+1;
      ord6=sum(ord6,1);
    end;
run;

proc print data=dt2;
run;

Re: Retain

Tommer — Fri, 26 Jun 2020 00:36:10 GMT

Thanks! I applied the same logic of retain to do this where I want the same num values for each by group to be populated under ord. But now I don't see the num being carried over. What am I missing?

data dt;
infile datalines missover;
input dat $ num;
datalines;
abc 1
abc
xyz 2
abc
def 3
xyz
def
;
run;

proc sort data=dt;
by dat;
run;

data dt2;
set dt;
by dat;
retain num;
if first.dat then ord=num;
run;

Re: Retain

Astounding — Fri, 26 Jun 2020 01:01:00 GMT

You are missing the fact that you added NUM to your incoming data set. That means each time the SET statement reads in an observation, it replaces the previous value for NUM. If the new value is missing, it replaces a nonmissing value. You should not have NUM in your incoming data to apply this logic (which is otherwise perfectly good logic).

Re: Retain

mkeintz — Fri, 26 Jun 2020 05:12:23 GMT

As @Astounding noted, the fact that NUM in in the input dataset means that any RETAIN instruction you issued is superseded by the fact that NUM is read in for each iteration of the DATA set.

And in fact, you don't even need a retain statement for variables that are read via SET or MERGE - they are automatically retained, but it's not usually apparent.

You can use this attribute of SAS and produce the result you are looking for by using a conditional SET statement for the NUM variable, as in:

data dt;
  infile datalines missover;
  input dat $ num;
datalines;
abc 1
abc
xyz 2
abc
def 3
xyz
def
run;

proc sort data=dt;
  by dat;
run;


data dt2;
  set dt (drop=num);
  by dat;
  if first.dat then set dt (keep=num) point=_n_;
run;

This depends on two properties.

The proc sort retains the original record order within by groups. And since you issue a NUM value for the first record of each DAT value in the original unsorted data set, the sorted dataset will have the desired NUM at the beginning of each by group.
The unconditional SET statement doesn't read in NUM, but the subsequent conditional SET (if first.dat then set ...) does read in NUM. This value will be automatically retained until the next time the condition is satisfied.

It's not particularly recommended for you problem, but this conditional SET technique is very useful for "bringing history forward". You can read in, say, chronologically sorted monthly data set, a quarterly data set, and a yearly data set, such that the most recent yearly and quarterly values are brought forward to each monthly record.

Re: Retain

Tommer — Fri, 26 Jun 2020 15:19:46 GMT

@Patrick these are all great examples with really good scenarios to understand the concepts better!