Art's suggestion is the right one IMO for your specific problem.

As a general critique, most of the time when a "utility" macro contains a full datastep, it's the wrong answer. Utility macros should generally be less than one datastep; ie, called inside a datastep. Otherwise you're just using the macro to do data step looping, which is much more cheaply handled with data step functions and statements.

For example, you could do

data want; set test; n=_n_; if missing(x) then do; do until (not missing(x)); n=n-1; set test(keep=x) point=n; end; end; run;

which is nearly identical to your macro in its approach, but uses data step looping instead of the macro, and will be substantially more efficient.

A utility macro to accomplish this would be something like this:

%macro rollback(var); %do i = 1 %to 8; &var=ifn(&var,&var,0,lag&i(&var)); %end; %mend rollback;

data want; set test; %rollback(x); run;

Note how it is called from inside a data step. I don't recommend this particular solution as there are better ways [in particular, this executes many statements needlessly and has to be hardcoded with the maximum possible number of lag values] but it illustrates my point.

-Joe

On Wed, Feb 3, 2010 at 10:47 AM, Robert Feyerharm <robertf@health.ok.gov>wrote:

> Has anyone else struggled with the lag function in SAS? The lag function > often doesn't work in an intuitive manner in conditional statements, or if > missing values are involved. > > Anyways, I wrote a short macro code to handle lag calculations which I'd > like to share. It works fairly well, although I may be reinventing the > wheel. A colleague was working with a large dataset containing missing > values, and wanted to impute the last known value into the missing values > for a particular field. For example, given the following test dataset, the > macro fills in the missing values of x with the last known value of x > (1,2, or 3): > > data test; > input x y; > datalines; > 1 1 > . 2 > . 3 > . 4 > . 5 > 2 6 > . 7 > . 8 > 3 9 > . 10 > . 11 > . 12 > . 13 > . 14 > ; > run; > > /* Let num = no. of records in target dataset. */ > %let num=14; > > %macro lagvar(num); > %do i = 1 %to &num; > data test; > set test; > z=lag(x); > if x=. then x=z; > run; > %end; > > data test; > set test; > drop z; > run; > %mend; > > %lagvar(&num) > > I'm curious if there is a shorter program that can handle the same > operations without resorting to a macro? > > Robert Feyerharm > Oklahoma State Department of Health >