How to conditionally back-cast based on a counter?

[thepushkarsingh]

I have a dataset(example below), with fields Scenarios, Date, Counter, Value and I need to create New Value. For each scenario, New Value takes value from latest counter(first counter from a continuous series) -

Scenarios	Date	Counter	Value	New Value
Scn1	Jan-22	0		189
Scn1	Feb-22	0		189
Scn1	Mar-22	1	189	189
Scn1	Apr-22	1		189
Scn1	May-22	1		189
Scn1	Jun-22	0		0
Scn2	Jan-22	0		167
Scn2	Feb-22	1	167	167
Scn2	Mar-22	1		167
Scn2	Apr-22	0		256
Scn2	May-22	1	256	256
Scn2	Jun-22	0		0
Scn3	Jan-22	1	23	23
Scn3	Feb-22	1		23
Scn3	Mar-22	1		23
Scn3	Apr-22	0		304
Scn3	May-22	1	304	304
Scn3	Jun-22	1		304

I am not sure how to achieve this conditional back-casting. May I request any help on this? 

[Tom · Accepted solution]

Clarify how you are defining the groups?

Looks like you want to increment the grouping when counter becomes zero.

First let's convert your listing into actual data. Lets use a SINGULAR name for that first variable to make the code clearer.

data have;
  input Scenario $ Date :monyy. Counter Value Expected;
  format date monyy7. ;
cards;
Scn1 Jan-22 0  . 189
Scn1 Feb-22 0  . 189
Scn1 Mar-22 1 189 189
Scn1 Apr-22 1  . 189
Scn1 May-22 1  . 189
Scn1 Jun-22 0  . 0
Scn2 Jan-22 0  . 167
Scn2 Feb-22 1 167 167
Scn2 Mar-22 1  . 167
Scn2 Apr-22 0  . 256
Scn2 May-22 1 256 256
Scn2 Jun-22 0  . 0
Scn3 Jan-22 1 23 23
Scn3 Feb-22 1  . 23
Scn3 Mar-22 1  . 23
Scn3 Apr-22 0  . 304
Scn3 May-22 1 304 304
Scn3 Jun-22 1  . 304
;

So the data is actually sorted by SCENARIO and DATE, but we can use the NOTSORTED keyword on the BY statement to have SAS calculated FIRST. and LAST. flags for COUNTER within SCENARIO.

data want;
  set have;
  by scenario counter notsorted;
  if first.scenario then group=1;
  else if first.counter and counter=0 then group+1;
run;

Now we can see that your expected new variable aligns with this new grouping variable.

Obs    Scenario       Date    Counter    Value    Expected    group

  1      Scn1      JAN2022       0          .        189        1
  2      Scn1      FEB2022       0          .        189        1
  3      Scn1      MAR2022       1        189        189        1
  4      Scn1      APR2022       1          .        189        1
  5      Scn1      MAY2022       1          .        189        1
  6      Scn1      JUN2022       0          .          0        2
  7      Scn2      JAN2022       0          .        167        1
  8      Scn2      FEB2022       1        167        167        1
  9      Scn2      MAR2022       1          .        167        1
 10      Scn2      APR2022       0          .        256        2
 11      Scn2      MAY2022       1        256        256        2
 12      Scn2      JUN2022       0          .          0        3
 13      Scn3      JAN2022       1         23         23        1
 14      Scn3      FEB2022       1          .         23        1
 15      Scn3      MAR2022       1          .         23        1
 16      Scn3      APR2022       0          .        304        2
 17      Scn3      MAY2022       1        304        304        2
 18      Scn3      JUN2022       1          .        304        2

So we could use a double DOW loop to find the first non-missing value per GROUP and attach it to each observation in the GROUP.

data want;
  do until(last.group);
    set want;
    by scenario group;
    new_value=coalesce(new_value,value);
  end;
  new_value=coalesce(new_value,0);
  do until(last.group);
    set want;
    by scenario group;
    output;
  end;
run;

Result

                                                                        new_
Obs    Scenario       Date    Counter    Value    Expected    group    value

  1      Scn1      JAN2022       0          .        189        1       189
  2      Scn1      FEB2022       0          .        189        1       189
  3      Scn1      MAR2022       1        189        189        1       189
  4      Scn1      APR2022       1          .        189        1       189
  5      Scn1      MAY2022       1          .        189        1       189
  6      Scn1      JUN2022       0          .          0        2         0
  7      Scn2      JAN2022       0          .        167        1       167
  8      Scn2      FEB2022       1        167        167        1       167
  9      Scn2      MAR2022       1          .        167        1       167
 10      Scn2      APR2022       0          .        256        2       256
 11      Scn2      MAY2022       1        256        256        2       256
 12      Scn2      JUN2022       0          .          0        3         0
 13      Scn3      JAN2022       1         23         23        1        23
 14      Scn3      FEB2022       1          .         23        1        23
 15      Scn3      MAR2022       1          .         23        1        23
 16      Scn3      APR2022       0          .        304        2       304
 17      Scn3      MAY2022       1        304        304        2       304
 18      Scn3      JUN2022       1          .        304        2       304

[thepushkarsingh]

Too many things to learn, in one single solution! Many thanks Tom.

[thepushkarsingh]

May I ask one follow up please? What if Value isn't missing in other cases, is there a neater way to pick only the relevant "Value"s in to New Value?

[Tom]

---

@thepushkarsingh wrote:
May I ask one follow up please? What if Value isn't missing in other cases, is there a neater way to pick only the relevant "Value"s in to New Value?

---

Sure. But you have to have a rule.

Is it the MIN value? The MAX value?

The value where "counter" variable first becomes 1?

All of those a easy.

[thepushkarsingh]

The value where counter becomes 1 at first in the series of 1s.

[Tom]

So use the same NOTSORTED trick.

data want;
  do until(last.group);
    set want;
    by scenario group counter notsorted;
    if first.counter and counter=1 then new_value=value;
  end;
  new_value=coalesce(new_value,0);
  do until(last.group);
    set want;
    by scenario group;
    output;
  end;
run;

You might be able to do it all in one step without adding the GROUP number.

[thepushkarsingh]

Thank you so much! You're a genius!