What is the most efficient way to fill in blank cells w/ NA?

[Tom]

You should have handled the missing values before you transposed it.

[Hello_there]

My actual data came from PROC means and when I transposed it by visit, with id being treatment, that’s when the blanks appeared bc not every treatment had a visit

[Tom]

---

@Hello_there wrote:
My actual data came from PROC means and when I transposed it by visit, with id being treatment, that’s when the blanks appeared bc not every treatment had a visit

---

Show the code.  I have never seen PROC MEANS make some thing that looks like '3 (4)' as output.  I generally only creates numeric values.

[Hello_there]

That was an extra step I took to get it into the format I needed in an intermediary data step. This is my personal computer so I don’t have access to the actual output.

But when you use PROC means to get summary statistics by visit for each treatment, it only summarizes the actual data in the raw data set. When you transpose it, there will be missing values for treatments that didn’t have a visit.
For example, subjects in treatment one can have data for visit 001, but if treatment 2 doesn’t have data for visit 001, then it will be blank.

My example wasn’t really accurate in the OP bc one of the treatments has to have data for a visit to populate in the data step. If there is no data for all treatments, there are no visits.

[Tom]

So do the step to add the visits before the proc means.  Then the step that is making it pretty for the transpose can deal with the missing values.

If you don't have a dataset with the set of valid id*visit combination you can easily make one.

Example:

data have;
  input id visit value ;
cards ;
1 1 100
1 2 112
2 1 200
3 2 300
;
proc sql ;
create table for_analysis as 
select * 
from have 
natural full join 
  ( select * from (select distinct id from have),(select distinct visit from have) )
;
quit;

[Hello_there]

This works also!

[Hello_there]

The array would be more efficient though right? Because if you did the extra step, you would still have to address the missing values at the end. But maybe this way is less lines of code.