minimum for a given ID - sas

Question

I have my data sorted by ID and date. I have converted the date to a single numeric which has ordering (the year followed by the week in the year). i want to make a new variable that is a function of the minimum value in the finest partition. and example follows
ID Start listen
1 201134 201138
1 201204 201150
2 200905 200910
2 201005 201020
I want something like
ID Start listen weekSincestart
1 201134 201138 4
1 201204 201150 54
2 200905 200910 5
2 201005 201020 15
all im doing is taking (listen-min(start)) but i am assuming min() is taking the minimum start for a given ID. So, i am asking if there is a "by statement" for the min function

Answer 1

In my opinion there is no need to convert your start and listen values from dates using the method you have.
I converted your data back into dates using INTNX using the first day of the year in START and LISTEN variables and increment by week in the same variables. The dates might not be exactly what you have on your dataset, however it should result in something similar.
The following should do what you want, if I understand you correctly.
DATA WANT2;
SET HAVE;
BY ID START;
RETAIN _START;
FORMAT _START DATE9.;
IF FIRST.ID THEN _START = START;
WEEKSINCESTART = INTCK("WEEK",_START,LISTEN);
RUN;
In this instance your sample is sorted, however if you wish to conduct by statement processing to identify the first instance of a value in ID you will need to sort your dataset first. The retain statement will hold a value and by using the by statement we can specify when the value in the retained variable is altered. In this instance we want to alter the _START variablee when the first instance of an ID is encountered. I use the underscore prefix because it makes it easier to drop these variables en masse if necessary. This value will not be replaced until the next instance of an ID, which means it will be the value of subsequent observations for ID 1 and so on. The INTCK function measures the number of intervals, in this example the number of WEEKS, between period one and period two, in this instance between the first instance of START for each ID captured in _START and LISTEN for each observation.
The end result is:
ID START LISTEN _START WEEKSINCESTART
1 21AUG2011 18SEP2011 21AUG2011 4
1 29JAN2012 11DEC2011 21AUG2011 16
2 01FEB2009 08MAR2009 01FEB2009 5
2 31JAN2010 16MAY2010 01FEB2009 67
I hope this is useful.
Regards,
Scott

Answer 2

You could easily do it with proc sql:
proc sql;
create table RESULT as
select *, listen-min(start) as weekSincestart
from INPUT
group by id;
quit;
It will take the min of each id group to calculate min(start).
And since you select variables which are not in the group by, nor have an aggregation function, it will not aggregate multiple rows into one on the group by.

Answer 3

Your question is a bit confusing. If you want just listen minus start (what your 'result' is), then just do that. The min function does not cross rows; in SAS it's difficult to cross rows (or at least it's something you would have to do intentionally). Of course you do need to figure out how to deal with the year barrier; if I were you I'd leave the dates as actual dates and use INTCK to determine the difference in weeks.
If you actually do want the minimum for the entire ID, the data step solution (not quite as neat as the SQL solution but works roughly the same):
data want;
set have;
by id start;
retain _initial_start;
if first.id then _initial_start=start;
weeksincestart=listen-_initial_start; *or whatever you intended - this does not seem right;
drop _initial_start;
run;