topic Moving average based on condition in SAS Programming

Moving average based on condition

hhchenfx — Fri, 11 Mar 2022 03:03:05 GMT

Hi Everyone,

For each id, I want to calculate moving average using values not equal to 0.

For the data below with moving average of 3 lag 1, the average for record 4 should take (3+9)/2; record 5 will have 9/1.

The proc expand with WHERE v2>0 is not working that way.

Can you please help?

Thanks,

HHC

data have;
input date id v2;
datalines;
1 1 3
2 1 0
3 1 9
4 1 0
5 1 5
1 2 0
2 2 0
3 2 9
4 2 0
5 2 5
;run;

*NOT right;
proc expand data=have out=want;
where v2>0;
convert v2 = avg/transformout=(MOVave  3 lag 1);
run;

Re: Moving average based on condition

mkeintz — Fri, 11 Mar 2022 03:43:46 GMT

Shouldn't it be

convert v2 = avg/transformout=(MOVave 1 lag 3);

Re: Moving average based on condition

hhchenfx — Fri, 11 Mar 2022 04:13:34 GMT

convert v2 = avg/transformout=(MOVave 1 lag 3);

will take average of 3 prior cell regardless of cell value, so it will not work.

Thanks.

HHC

Re: Moving average based on condition

hhchenfx — Fri, 11 Mar 2022 04:22:34 GMT

Hi,

I do the vertical/Excel kind of way.

It works.

If you have a shorter method, please let me know.

Thanks,

HHC

data have;
input date id value;
datalines;
1 1 3
2 1 0
3 1 9
4 1 0
5 1 5
1 2 1
2 2 0
3 2 9
4 2 0
5 2 5
;run;

proc sort data=have; by id descending date;run;

data want; set have;
drop id1 date1 v1;
SUM=0;
Count=0;
do i=_N_ +1 to _N_+3;

	set have (rename = (date=date1 id=id1 value=v1)) point=i;
		if id1=id and v1^=0 then do;
			sum=sum+v1;
			count=count+1;
		end;
		average=sum/count;
end;
output;
run;

Re: Moving average based on condition

ballardw — Fri, 11 Mar 2022 11:41:53 GMT

Easiest is very likely 1) add a new variable that is missing when the variable you want to average is missing

2) expand that new variable

3) (may want to remove the added variable)

Re: Moving average based on condition

mkeintz — Fri, 11 Mar 2022 13:42:05 GMT

You want the average of the previous three observations, excluding thoses with V2=0. How about a DATA step:

data have;
input date id v2;
datalines;
1 1 3
2 1 0
3 1 9
4 1 0
5 1 5
1 2 0
2 2 0
3 2 9
4 2 0
5 2 5
;run;

data want;
  array prev_three {3} _temporary_;
  set have;
  by id;
  if first.id then call missing(of prev_three{*});
  avg=mean(of prev_three{*});
  prev_three{mod(_n_,3)+1}=ifn(v2>0,v2,.);
run;

The important thing here is that AVG is calculated prior to replacing the 3rd previous value of V2 in the PREV_THREE array with the current value.

Re: Moving average based on condition

hhchenfx — Fri, 11 Mar 2022 18:13:43 GMT

Your code is amazing.

It work much much faster than my code!!!

Thank you,

HHC