← SASnRReplicating the concept of usage of loops for dummy data creation

Introduction

A recurring need in early prototyping: build a skeleton dataset covering every combination of a few values — treatments x visits, groups x timepoints, subjects x parameters.
We want every pair to appear exactly once, so we can attach real data later without worrying about missing cells.
SAS builds this with nested DO loops and an OUTPUT statement, writing one row per iteration of the inner loop.
Tidyverse uses crossing() from tidyr, a one-liner that returns every combination of the supplied vectors.
Base R uses expand.grid(), the classic base-R way of producing a full-factorial data frame.
This lesson builds a 3 x 4 grid of treatments (1, 3, 5) and visits (0, 1, 2, 8) — 12 rows with every pair present once.

Replicating the concept of usage of loops for dummy data creation

SAS (Base SAS)+

data loop;
    do treatment=1,3,5;
        do avisitn=0,1,2,8;
            output;
        end;
    end;
run;

To build a dummy dataset with every combination of a set of values, we reach for a nested DO loop with an OUTPUT statement in between.
The outer loop iterates over treatments (1, 3, 5) and the inner loop over visit numbers (0, 1, 2, 8).
The output; statement inside the inner loop writes one observation for every combination — 3 x 4 = 12 rows.
Without an explicit output, SAS would only write a row at the end of each DATA step iteration; with it, we get a full cross-product.
This is the quickest way to prototype a skeleton dataset for testing — for example, a subject-by-visit grid before actual data is available.
After running, we check that loop has 12 rows with every treatment-avisitn pair appearing exactly once.

R (tidyverse)+

library(tidyverse)

 

loop <- crossing(treatment = c(1, 3, 5), avisitn = c(0, 1, 2, 8))

For the same cross-product in tidyverse, we call crossing() with one named vector per dimension.
crossing() returns a tibble of all unique combinations of the supplied vectors — the tidyverse equivalent of the SAS nested DO loops + output.
It automatically deduplicates the inputs and orders the output; no explicit loop, output, or row counter needed.
This is far more readable than a for-loop plus rbind() when we want a grid of values for prototyping or full-factorial testing.
A related function, expand_grid(), does the same without deduping — useful when the inputs intentionally contain repeats.
After running, we confirm loop has 12 rows (3 treatments x 4 visits) with every combination present exactly once.