NESUG 2010 Coders' Corner

Ordering PROC FREQ Around

Jonathan Kerman, Johns Hopkins University, Baltimore, MD

ABSTRACT
PROC FREQ is one of the most widely used SAS® procedures. The default output of PROC FREQ can compact-
ly summarize large data distributions in one-dimensional, two-dimensional or multi-dimensional frequency tables.
Yet the lesser-known ORDER= option gives the user added flexibility in controlling the order of the output. This
paper will demonstrate use of the ORDER= option and its four sub-options, showing how you can order PROC
FREQ to print the way you want.

INTRODUCTION
PROC FREQ is a trustworthy workhorse in the SAS toolbox. It may be old, but it still runs, simply and reliably.
Any user of PROC FREQ will be familiar with output looking like this:

Cumulative Cumulative
Year Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
1998 1317 10.64 1317 10.64
1999 1339 10.82 2656 21.45
2000 1403 11.33 4059 32.79
2001 1249 10.09 5308 42.88
2002 1320 10.66 6628 53.54
2003 1094 8.84 7722 62.37
2004 1328 10.73 9050 73.10
2005 968 7.82 10018 80.92
2006 1266 10.23 11284 91.15
2007 1096 8.85 12380 100.00

Figure 1. Number of food-borne illness outbreaks, by year

This example and the ones that follow use data adapted from the Centers for Disease Control (CDC) website of
food-borne illnesses (http://wwwn.cdc.gov/foodborneoutbreaks). This distribution shows the number of food-borne
illness outbreaks by year from 1998 to 2007.*
But suppose you want to change the order of your output, for example, by displaying the most recent year first.
PROC FREQ provides an option to give you additional control over the order of the output.
Before we discuss changing the sorting order, let‘s review how SAS orders things by default. The alphanumeric
sorting sequence, based on the ASCII system, sorts characters in the following order:
0 – 9, A – Z, a – z,
that is, numerals first, then capital letters, followed by lowercase letters. We‘ll see how this sorting order affects
output in some of the examples below. And we‘ll return to the exercise of printing the number of outbreaks from
latest year to earliest year at the end of the paper.

THE “ORDER=” OPTION

PROC FREQ has an option-- ―ORDER=‖. This option takes four possible values: DATA, FORMATTED, FREQ
and INTERNAL. Let‘s examine what each of these values does.

* The example dataset, named foodill, includes these fields: Year – year of outbreak; Month – month of outbreak;
Location– site where outbreak occurred (e.g.- home, school); Total_Ill – number falling ill

1
NESUG 2010 Coders' Corner

ORDER=INTERNAL
First, you should know that even without specifying it, PROC FREQ uses option ORDER=INTERNAL by default.
This means that the results are printed in the sorting order of the raw values—the internal representation of the
variables.
The output in Figure 1 above shows this. The years are displayed in numerical order, just as you would expect.
The program that generated this output could be written relying on the default behavior:
proc freq data=foodill;
table year;
run;

or by stating the option explicitly:

proc freq data=foodill order=internal;
table year;
run;

In the discussion below, I‘ll show the default output of several more variables. In these examples, I‘ll sometimes
specify the ORDER=INTERNAL option and sometimes leave it out to emphasize that either way produces output
sorted by the internal values.

ORDER=FREQ
The option ORDER=FREQ sorts the output by frequency of occurrence, from most common to least common.
This can be particularly useful if you have many values and you want to easily see which ones occur with the
greatest frequency. As an example, let‘s look at the distribution of settings where food-borne illnesses broke out.
Using the default ordering, here is the program to display the number of outbreaks by the place where they oc-
curred:
proc freq data=foodill;
table location;
run;

The output looks like this:

Cumulative Cumulative
Location Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
banquet 185 1.55 185 1.55
camp 104 0.87 289 2.43
church 226 1.90 515 4.32
day care 41 0.34 556 4.67
fair 74 0.62 630 5.29
grocery 14 0.12 644 5.40
home 2480 20.81 3124 26.22
hospital 74 0.62 3198 26.84
office 120 1.01 3318 27.85
other 1341 11.25 4659 39.10
picnic 125 1.05 4784 40.15
prison 127 1.07 4911 41.22
restaurant 5861 49.19 10772 90.41
school 477 4.00 11249 94.41
unknown 284 2.38 11533 96.79
wedding 33 0.28 11566 97.07
work (not caf) 331 2.78 11897 99.85
work cafeteria 18 0.15 11915 100.00

Frequency Missing = 465

Figure 2a. Number of food-borne illness outbreaks, by location

2
NESUG 2010 Coders' Corner

This output is fine, but it takes some work to look through and find the locations where the most or the fewest out-
breaks occurred. Instead, we can run the program with option ORDER=FREQ:
proc freq data=foodill order=freq;
table location;
run;

Now, the output will look like this:

Cumulative Cumulative
Location Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
restaurant 5861 49.19 5861 49.19
home 2480 20.81 8341 70.00
other 1341 11.25 9682 81.26
school 477 4.00 10159 85.26
work (not caf) 331 2.78 10490 88.04
unknown 284 2.38 10774 90.42
church 226 1.90 11000 92.32
banquet 185 1.55 11185 93.87
prison 127 1.07 11312 94.94
picnic 125 1.05 11437 95.99
office 120 1.01 11557 97.00
camp 104 0.87 11661 97.87
fair 74 0.62 11735 98.49
hospital 74 0.62 11809 99.11
day care 41 0.34 11850 99.45
wedding 33 0.28 11883 99.73
work cafeteria 18 0.15 11901 99.88
grocery 14 0.12 11915 100.00

Frequency Missing = 465

Figure 2b. Number of food-borne illness outbreaks, by location, ORDER=FREQ

This display shows us clearly where most outbreaks occur. Nearly half of all food-related incidents happened in
restaurants. We can also see that together, home and restaurant settings account for 70% of all reported out-
breaks. These statistics and similar groupings are readily apparent from the re-ordered output.

ORDER=FORMATTED
While the default option prints output in order of the internal values, ORDER=FORMATTED displays output in the
sorting order of the formats applied to the variables. Obviously, in order to demonstrate this option, we will need to
use a variable with SAS formats applied.*
Variable Total_Ill records the number of people falling ill at each outbreak. First, let‘s display the default output
without using any formats. Here is the program:

proc freq data=foodill order=internal;

table total_ill;
run;

* Using option ORDER=FORMATTED with an unformatted variable has no effect.

3
NESUG 2010 Coders' Corner

And a partial output would look like this:

Cumulative Cumulative
Total_Ill Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
2 2094 16.91 2094 16.91
3 1336 10.79 3430 27.71
4 1054 8.51 4484 36.22
5 670 5.41 5154 41.63
6 603 4.87 5757 46.50
7 462 3.73 6219 50.23
8 408 3.30 6627 53.53
9 349 2.82 6976 56.35
10 322 2.60 7298 58.95
.
.
.
1200 1 0.01 12379 99.99
1644 1 0.01 12380 100.00

Figure 3a. Number of food-borne illness outbreaks, by number falling ill (excerpt)

The complete output runs over 250 lines. Clearly, this variable would best be displayed by summarizing the raw
values, so we‘ll create a format to break the numbers down into categories.
Suppose the CDC had rated outbreaks by their severity. They might assign the following scheme:
Category 1 – 1000 or more ill
Category 2 – 100 to 999 ill
Category 3 – 25 to 99 ill
Category 4 – 1 to 24 ill

And we can create a format for these categories:

proc format;
value ctg 1000-high=’Category 1 (>= 1000)’
100-999 =’Category 2 (100-999)’
25-99 =’Category 3 (25-99)’
1-24 =’Category 4 (1-24)’;
run;

If we run PROC FREQ using our new format, the default output still uses the internal values to order the output.
The formats representing the smaller numbers print first, and the rows are ordered from ‗Category 4‘ down to
‗Category 1‘. Here is the program and the output:

proc freq data=foodill;

table total_ill;
format total_ill ctg.;
run;

Cumulative Cumulative
Total_Ill Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
Category 4 (1-24) 9877 79.78 9877 79.78
Category 3 (25-99) 2124 17.16 12001 96.94
Category 2 (100-999) 377 3.05 12378 99.98
Category 1 (>= 1000) 2 0.02 12380 100.00

Figure 3b. Number of food-borne illness outbreaks, by number falling ill, formatted

4
NESUG 2010 Coders' Corner

If we use the option ORDER=FORMATTED, we have the following program and output:
proc freq data=foodill order=formatted;
table total_ill;
format total_ill ctg.;
run;
Cumulative Cumulative
Total_Ill Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
Category 1 (>= 1000) 2 0.02 2 0.02
Category 2 (100-999) 377 3.05 379 3.06
Category 3 (25-99) 2124 17.16 2503 20.22
Category 4 (1-24) 9877 79.78 12380 100.00

Figure 3c. Number of food-borne illness outbreaks, by number falling ill, formatted, ORDER=FORMATTED

The rows are now printed in the alphanumeric order of the format‘s label. As a result, the categories are listed in a
more logical manner.

ORDER=DATA
The final sub-option prints output in the order that the values are encountered in the dataset. At first glance, this
might not seem so helpful.
Let‘s print the first ten lines of dataset foodill, noticing the arbitrary order of values for variable Total_Ill:
Total_
Obs Year Month Location Ill

1 1998 January 218

2 1998 January home 2
3 1998 January 8
4 1998 January home 3
5 1998 January home 2
6 1998 January other 26
7 1998 January 8
8 1998 January home 4
9 1998 January 2
10 1998 January day care 6

Figure 4. First ten lines of dataset foodill, selected variables displayed

Now, suppose we run PROC FREQ on variable Total_Ill, using option ORDER=DATA. Here is the program and
the first few lines of output:
proc freq data=foodill order=data;
table total_ill;
run;

Cumulative Cumulative
Total_Ill Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
218 1 0.01 1 0.01
2 2094 16.91 2095 16.92
8 408 3.30 2503 20.22
3 1336 10.79 3839 31.01
26 101 0.82 3940 31.83
4 1054 8.51 4994 40.34
6 603 4.87 5597 45.21

Figure 5. Number of food-borne illness outbreaks, by number falling ill, ORDER=DATA (excerpt)
5
NESUG 2010 Coders' Corner

The values are indeed tabulated and printed in the order they appear in the dataset. But this output doesn‘t seem
helpful at all. Why, you might wonder, would I ever want to use ORDER=DATA? Believe it or not, there are times
when this option can come in handy. Here‘s an example.
Let‘s display the months when food-borne outbreaks occurred. The program for this task is:
proc freq data=foodill;
table month;
run;

And the default output would look like this:

Cumulative Cumulative
Month Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
April 1115 9.01 1115 9.01
August 996 8.05 2111 17.05
December 1184 9.56 3295 26.62
February 929 7.50 4224 34.12
January 944 7.63 5168 41.74
July 1016 8.21 6184 49.95
June 1172 9.47 7356 59.42
March 1052 8.50 8408 67.92
May 1231 9.94 9639 77.86
November 1030 8.32 10669 86.18
October 905 7.31 11574 93.49
September 806 6.51 12380 100.00

Figure 6a. Number of food-borne illness outbreaks, by month

The months, which are stored as character strings in dataset foodill, print in alphabetical order—not the output we
were expecting. It would be much more helpful to have month print in calendar order.
Suppose we know that the records in our dataset are in chronological order (beginning with January) and that all
months have some outbreaks. (In other words, records for January 1998 come first, followed by those from Feb-
ruary 1998, March 1998, etc.) We can use these facts along with option ORDER=DATA to print the output in a
more intuitive order.
First, here‘s the program:
proc freq data=foodill order=data;
table month;
run;

And the new output is:

Cumulative Cumulative
Month Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
January 944 7.63 944 7.63
February 929 7.50 1873 15.13
March 1052 8.50 2925 23.63
April 1115 9.01 4040 32.63
May 1231 9.94 5271 42.58
June 1172 9.47 6443 52.04
July 1016 8.21 7459 60.25
August 996 8.05 8455 68.30
September 806 6.51 9261 74.81
October 905 7.31 10166 82.12
November 1030 8.32 11196 90.44
December 1184 9.56 12380 100.00

Figure 6b. Number of food-borne illness outbreaks, by month, ORDER=DATA

6
NESUG 2010 Coders' Corner

FORMATTED VARIABLES VS. ORDER=FORMATTED

It‘s important to keep in mind the distinction between using formatted variables and using the option OR-
DER=FORMATTED. Displaying formatted variables while using the default behavior of PROC FREQ keeps the
internal ordering; it just prints using the label given in the format.
To illustrate this, let‘s create a new format for the number of people becoming ill at an outbreak.
proc format;
value n_ill 1-9 =’Fewer than 10’
10-49 =’10-49’
50-99 =’50-99’
100-250 =’100-250’
251-high =’Greater than 250’;
run;

Here, we run PROC FREQ formatting variable Total_Ill and relying on the default ordering:
proc freq data=foodill;
table total_ill;
format total_ill n_ill.;
run;

The output produced is:

Cumulative Cumulative
Total_Ill Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
Fewer than 10 6976 56.35 6976 56.35
10-49 4372 35.32 11348 91.66
50-99 653 5.27 12001 96.94
100-250 304 2.46 12305 99.39
Greater than 250 75 0.61 12380 100.00

Figure 7a. Number of food-borne illness outbreaks, by number falling ill, formatted
This output is useful and straightforward. But suppose we had also specified option ORDER=FORMATTED.
Here is the program and the output:

proc freq data=foodill order=formatted;

table total_ill;
format total_ill n_ill.;
run;

The output produced is:

Cumulative Cumulative
Total_Ill Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
10-49 4372 35.32 4372 35.32
100-250 304 2.46 4676 37.77
50-99 653 5.27 5329 43.05
Fewer than 10 6976 56.35 12305 99.39
Greater than 250 75 0.61 12380 100.00

Figure 7b. Number of food-borne illness outbreaks, by number falling ill, formatted, ORDER=FORMATTED

This output is not very user-friendly. PROC FREQ does use the formats—so that we don‘t have 250-plus lines of
output—but the output rows are sorted by the label used for the format, in strict ASCII order. This is one example
of when not to use ORDER=FORMATTED.

7
NESUG 2010 Coders' Corner

As another example, let‘s create a format to represent the months as calendar quarters. Because the months are
character variables, we must create a character format (beginning with ‗$‘):
proc format;
value $qtr 'January','February','March' = 'Q1'
'April','May','June' = 'Q2'
'July','August','September' = 'Q3'
'October','November','December' = 'Q4';
run;

First, let‘s run PROC FREQ using the format together with the default ordering:
proc freq data=foodill order=internal;
table month;
format month $qtr.;
run;

Cumulative Cumulative
Month Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
Q2 3518 28.42 3518 28.42
Q3 2818 22.76 6336 51.18
Q4 3119 25.19 9455 76.37
Q1 2925 23.63 12380 100.00

Figure 8a. Number of food-borne illness outbreaks, by quarter, formatted

Why did PROC FREQ display the output rows in this order? Because, even though it‘s the labels that are being
printed, the underlying order is still the internal value of the variable. This means that behind the ―mask‖ of, ‗Q1‘,
‗Q2‘, ‗Q3‘ and ‗Q4‘, PROC FREQ still wants to order the rows ―April‘ (Q2), then ‗August‘ (Q3), then ‗December‘
(Q4), etc.
In order to get the output sorted by quarter, one way is to use ORDER=FORMATTED:
proc freq data=foodill order=formatted;
table month;
format month $qtr.;
run;

Cumulative Cumulative
Month Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
Q1 2925 23.63 2925 23.63
Q2 3518 28.42 6443 52.04
Q3 2818 22.76 9261 74.81
Q4 3119 25.19 12380 100.00

Figure 8b. Number of food-borne illness outbreaks, by quarter, formatted, ORDER=FORMATTED

A second way, again assuming that the records in our dataset are in chronological order beginning with ‗January‘,
is to use ORDER=DATA. This option will produce output identical to Figure 11 (because ‗January‖, represented
by ‗Q1‘, appears first in the dataset, and so on.)
In summary, then, formats use a different text to display the output—‗Q3‘ instead of ‗September‘; ‗Fewer than 10‘
instead of ‗6‘—but formats do nothing to change the order of the output. Option ORDER=FORMATTED, on the
other hand, will change the order of the output. You will probably only want to use this option when the formats
themselves have a meaningful alphanumeric order.

8
NESUG 2010 Coders' Corner

FORCING SAS TO PRINT IN REVERSE ORDER

After seeing all of these options in use, we can go back to the example mentioned at the beginning—displaying the
distribution of food-borne illness outbreaks by year, but in reverse order.
Here, we can use a bit of a trick. Knowing that ORDER=DATA will print in the order that each value is encoun-
tered in the dataset, we can first force a reverse sort:
proc sort data=foodill;
by descending year;
run;

Then use ORDER=DATA to get the output we want:

proc freq data=foodill order=data;
table year;
run;

And the re-ordered output is:

Cumulative Cumulative
Year Frequency Percent Frequency Percent
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
2007 1096 8.85 1096 8.85
2006 1266 10.23 2362 19.08
2005 968 7.82 3330 26.90
2004 1328 10.73 4658 37.63
2003 1094 8.84 5752 46.46
2002 1320 10.66 7072 57.12
2001 1249 10.09 8321 67.21
2000 1403 11.33 9724 78.55
1999 1339 10.82 11063 89.36
1998 1317 10.64 12380 100.00

Figure 9. Number of food-borne illness outbreaks, by year, most recent year first

CONCLUSIONS
With the ―ORDER=‖ option, PROC FREQ provides a method to modify the order of its output. The values that this
option can take-- DATA, FORMATTED, FREQ and, the default, INTERNAL—along with a clear understanding of
how each one functions provide the user with tools to help create clearer frequency tables.

ACKNOWLEDGMENTS
SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS
Institute Inc. in the USA and other countries. ® indicates USA registration.
Other brand and product names are registered trademarks or trademarks of their respective companies.

CONTACT INFORMATION
Your comments and questions are valued and encouraged. Contact the author at:
Author Name: Jonathan Kerman
Company: Johns Hopkins University, School of Public Health
Address: 615 North Wolfe Street, Room E7008
City, State ZIP: Baltimore, Maryland, 21205
Work Phone: (410) 955-4583
Fax: (410) 955-7587
Email: [email protected]