← SASnRCross-Tabulation: Two-Way Frequency Table

Introduction

A two-way frequency table counts how many records fall into each combination of two categorical variables — treatment by sex, AE severity by system organ class, and so on.
Clinical programmers run cross-tabulations daily to verify that subject counts balance across treatment arms, to spot data-entry errors, and to feed downstream TFL programs.
SAS handles this with PROC FREQ and the TABLES statement; R offers count() plus pivot_wider() in tidyverse and table() in base R.
We build a small ADSL-like dataset with 12 subjects split across three treatment groups and two sexes, then cross-tabulate them side by side.
After running, we confirm that every cell count adds up to the total number of subjects — no rows lost, no rows duplicated.

Cross-Tabulation: Two-Way Frequency Table

SAS (Base SAS)+

*==============================================================================;
* Cross-Tabulation: Two-Way Frequency Table
*==============================================================================;

*------------------------------------------------------------------------------;
* Step 1 - Create a small ADSL-like dataset
*------------------------------------------------------------------------------;
data adsl;
    infile datalines dlm='|' dsd missover;
    input USUBJID : $10. SEX : $1. TRT01P : $12. AGEGR1 : $10.;
    datalines;
S-001|M|Placebo|<65
S-002|F|Drug A|<65
S-003|M|Drug B|>=65
S-004|F|Placebo|<65
S-005|M|Drug A|>=65
S-006|F|Drug B|<65
S-007|M|Placebo|>=65
S-008|F|Drug A|<65
S-009|M|Drug B|>=65
S-010|F|Placebo|<65
S-011|M|Drug A|>=65
S-012|F|Drug B|<65
;
run;

*------------------------------------------------------------------------------;
* Step 2 - Two-way frequency table (default format)
*------------------------------------------------------------------------------;
proc freq data=adsl;
    tables SEX * TRT01P / nocum nopercent;
run;

*------------------------------------------------------------------------------;
* Step 3 - Cross-list format (one row per combination)
*------------------------------------------------------------------------------;
proc freq data=adsl;
    tables SEX * TRT01P / crosslist nocum nopercent;
run;

*------------------------------------------------------------------------------;
* Step 4 - Save the cross-tabulation to a dataset
*------------------------------------------------------------------------------;
proc freq data=adsl noprint;
    tables SEX * TRT01P / out=freq_sex_trt;
run;

proc print data=freq_sex_trt;
run;

We start by building a 12-row ADSL dataset with USUBJID, SEX, TRT01P and AGEGR1 so we have realistic clinical variable names.
To produce a two-way frequency table, we write TABLES SEX * TRT01P inside PROC FREQ — the asterisk crosses the two variables.
NOCUM suppresses the cumulative frequency column, and NOPERCENT removes the percentage column, giving us a clean count-only table.
The default output is a matrix with SEX on the rows and TRT01P across the columns — exactly the layout we see in clinical tables.
Adding the CROSSLIST option switches to a listing format with one row per SEX-TRT01P combination, which is easier to scan when there are many categories.
To save the counts for downstream processing, we add OUT=freq_sex_trt and NOPRINT — the output dataset contains SEX, TRT01P, COUNT and PERCENT columns.
After running, we open freq_sex_trt and verify that the COUNT column sums to 12 — our original number of subjects.

R (tidyverse)+

#==============================================================================
# Cross-Tabulation: Two-Way Frequency Table
#==============================================================================

library(tidyverse)

#------------------------------------------------------------------------------
# Step 1 - Create a small ADSL-like dataset
#------------------------------------------------------------------------------
adsl <- tribble(
  ~USUBJID, ~SEX, ~TRT01P,   ~AGEGR1,
  "S-001",  "M",  "Placebo", "<65",
  "S-002",  "F",  "Drug A",  "<65",
  "S-003",  "M",  "Drug B",  ">=65",
  "S-004",  "F",  "Placebo", "<65",
  "S-005",  "M",  "Drug A",  ">=65",
  "S-006",  "F",  "Drug B",  "<65",
  "S-007",  "M",  "Placebo", ">=65",
  "S-008",  "F",  "Drug A",  "<65",
  "S-009",  "M",  "Drug B",  ">=65",
  "S-010",  "F",  "Placebo", "<65",
  "S-011",  "M",  "Drug A",  ">=65",
  "S-012",  "F",  "Drug B",  "<65"
)

#------------------------------------------------------------------------------
# Step 2 - Two-way count in long format
#------------------------------------------------------------------------------
freq_long <- adsl %>%
  count(SEX, TRT01P)

print(freq_long)

#------------------------------------------------------------------------------
# Step 3 - Pivot to a wide cross-tabulation
#------------------------------------------------------------------------------
freq_wide <- freq_long %>%
  pivot_wider(names_from = TRT01P, values_from = n, values_fill = 0)

print(freq_wide)

#------------------------------------------------------------------------------
# Step 4 - Quick alternative with janitor::tabyl
#------------------------------------------------------------------------------
# install.packages("janitor")  # run once if not installed
library(janitor)

adsl %>%
  tabyl(SEX, TRT01P)

We build the same 12-row ADSL tibble using tribble() so the data matches the SAS version exactly.
To get the raw counts, we call count(adsl, SEX, TRT01P) — this returns a long tibble with one row per SEX-TRT01P combination and a column n for the count.
The long format is handy for further joins and calculations, but it does not look like a traditional cross-tabulation grid.
To reshape into the familiar matrix layout, we pipe into pivot_wider(names_from = TRT01P, values_from = n, values_fill = 0) — treatment groups become columns.
values_fill = 0 fills any missing combinations with zero so no cell is NA.
For an even quicker one-liner, janitor::tabyl(SEX, TRT01P) produces a cross-tabulation with counts and optional percentages, plus an adorn_totals() method for row and column totals.
After running, we confirm that all cell values in freq_wide sum to 12 — our total subject count.

Happy learning!

Search lessons

Introduction

Cross-Tabulation: Two-Way Frequency Table