Case Statement and Aggregation within a Group By classe in proc sql - sql

Question

I am having some trouble aggregating and using case within a group.
The objective is to check Indicator for each transaction key. If '1' indicator exists then we have to select the max(Change_Date). If all zeros then min(Change_Date). Along with that the Initial_key associated with that Change_date has to be populated as a Final_key.
The output looks like this

Answer 1

You can get the last two columns using aggregation. If I understand correctly:
select trxn_key,
coalesce(max(case when indicator = 1 then change_date end),
min(change_date)
) as final_date,
coalesce(max(case when indicator = 1 then initial_key end),
min(initial_key)
) as final_key
from t
group by trxn_key;
Then join this in:
proc sql;
select t.*, tt.final_date, tt.final_key
from t join
(select trxn_key,
coalesce(max(case when indicator = 1 then change_date end),
min(change_date)
) as final_date,
coalesce(max(case when indicator = 1 then initial_key end),
min(initial_key)
) as final_key
from t
group by trxn_key
) tt
on tt.trxn_key = t.trxn_key;

Answer 2

Could you try with below,
If i observe the test data provided by you
First we try to find the the max(indicator) within group by name and trxn_key.
Second based on the value above , we decide whether to take min(change_date) and min(initial_key) or max(change_date) and max(initial_key)
Because you don't need aggregated result we need to use analytic function which will not affect the final output rows.
SELECT t1.name
,t1.initial_key
,t1.change_date
,t1.indicator
,t1.trxn_key
,t1.trxn_date
,CASE
WHEN max_ind = 1
THEN
MAX(CASE WHEN indicator = 1 THEN change_date END) OVER (PARTITION BY NAME,trxn_key)
WHEN max_ind = 0
THEN
MIN(CASE WHEN indicator = 0 THEN change_date END) OVER (PARTITION BY NAME,trxn_key)
END final_date
,CASE
WHEN max_ind = 1
THEN
MAX(CASE WHEN indicator = 1 THEN initial_key END) OVER (PARTITION BY NAME,trxn_key)
WHEN max_ind = 0
THEN
MIN(CASE WHEN indicator = 0 THEN initial_key END) OVER (PARTITION BY NAME,trxn_key)
END final_key
FROM
(
SELECT NAME
,initial_key
,change_date
,indicator
,trxn_key
,trxn_date
,MAX(indicator) OVER (PARTITION BY NAME,trxn_key) max_ind
FROM table1
) t1
ORDER BY trxn_key,trxn_date,initial_key,change_date;

Answer 3

You can process groups with DOW loops (do until loops with SET & BY statement inside)
A DATA Step program with serial DOW loops (two in one step) can have the first loop process the group, measuring it in most anyway desired, and the second loop output records with values computed in the first loop.
Example:
data have;
input name $ initial_key change_date indicator trxn_key trxn_date;
attrib change_date trxn_date informat=date9. format=date9.;
datalines;
ABC 1 17feb20 0 1 16feb20
ABC 2 21feb20 0 1 16feb20
ABC 3 25feb20 0 1 16feb20
ABC 1 17feb20 1 2 20feb20
ABC 2 21feb20 0 2 20feb20
ABC 3 25feb20 0 2 20feb20
ABC 1 17feb20 1 3 22feb20
ABC 2 21feb20 1 3 22feb20
ABC 3 25feb20 0 3 22feb20
ABC 1 17feb20 1 4 26feb20
ABC 2 21feb20 1 4 26feb20
ABC 3 25feb20 1 4 26feb20
;
data want;
* first dow loop, compute min and max_ associated values;
do until (last.trxn_key);
set have;
by name trxn_key;
if missing(min_date) or change_date < min_date then do;
min_date = change_date;
min_key = initial_key;
end;
if missing(max_date) or change_date > max_date then
if indicator then do;
max_date = change_date;
max_key = initial_key;
max_flag = 1;
end;
end;
* compute final values per business rules;
if max_flag then do;
final_date = max_date;
final_key = max_key;
end;
else do;
final_date = min_date;
final_key = min_key;
end;
* second dow loop, output with final values;
do until (last.trxn_key);
set have;
by name trxn_key;
OUTPUT;
end;
format final_date min_date max_date date9.;
drop min_: max_:;
run;