# Getting All Duplicates within a SAS Data Set

by on ‎01-29-2015 12:52 PM - edited on ‎10-13-2016 07:59 AM by (653 Views)
• ### SAS Programming

#### The following blog post was originally published on my blog, The Chemical Statistician.

Introduction

A common task in data manipulation is to obtain all observations that appear multiple times in a data set - in other words, to obtain the duplicates.  It turns out that there is no procedure or function that will directly provide the duplicates of a data set in SAS*.

*Update: As Fareeza Khurshed kindly commented, the NOUNIQUEKEY option in PROC SORT is available in SAS 9.3+ to directly obtain duplicates.

The Wrong Way to Obtain Duplicates in SAS

You may think that PROC SORT can accomplish this task with the nodupkey and the dupout options.  However, the output data set from such a procedure does not have the first of each set of duplicates.  Here is an example.

data test;

input var1 \$;

cards;

111

222

333

444

444

444

555

555

666

666

666

777

888

888

999

999

999;

run;

The data set test has 5 sets of duplicates.  A simple PROC FREQ shows the frequency counts of each possible value of var1.

proc freq

data = test;

tables var1 / nocum nocol norow nopercent;

run;

 111 1 1 1 3 2 3 1 2 3

Here is the WRONG way to get the duplicates from test.

proc sort

data = test

nodupkey

dupout = dups;

by var1;

run;

proc print

data = dups;

run;

 1 444 444 555 666 666 888 999 999

The PROC FREQ of the data set dups will show a different frequency count than that of test as shown previously.

 444 2 1 2 1 2

When you use nodupkey and dupout in PROC SORT, you will remove the first of each set of duplicates in the output data set.  That's why each duplicated value has a frequency in dups that is one less than its frequency in test.

The Right Way to Obtain Duplicates in SAS

To obtain ALL duplicates of a data set, you can take advantage of first.variable and last.variableHere is the code to do it with the above example data set of test; you will get both the single observations and the duplicate observations.

(Update: As Robert Matthews commented, you can alter this code to just obtain the duplicates - this would be more efficient if you have many single observations and few duplicate observations, and you don't care about the singles.)

data single dup;

set test;

by var1;

if first.var1 and last.var1

then output single;

else output dup;

run;

Here is the logic behind the code.  We are producing 2 data sets: single for the observations that appear only once, and dup for the duplicate observations.

If first.var1 and last.var1 both equal to "1" for the same observation, then there is no duplicate, and this observation belongs to single.

Otherwise, it's a duplicate, and this observation belongs to dup.

Here is what you get as the output data set.

 1 444 444 444 555 555 666 666 666 888 888 999 999 999

Notice that it has the correct number of duplicates for each value.

by
on ‎10-13-2016 02:01 AM

Here's an alternative if the original data is not sorted by the variable whose duplicates are wanted, but big enough to make using proc sort expensive (in terms of computer resources).  This single DATA step outputs all duplicates without pre-sorting the data set, but produces the duplicates in sorted order.

The fundamental logic is to (1) record-by-record, build a hash table V tracking all values of VAR as they are encountered, noting the record number of the first encounter. (2) compare each incoming value of VAR record to the hash table.   If not found add it to V.  But if it is found then add the current record to hash object DUPLICATES, and (if neccessary) use SET ... POINT= to go back and get the first duplicate and put it also in DUPLICATES.

``````data _null_;
set have end=eod;
if _n_=1 then do;
declare hash v ();
v.definekey('var');
v.definedata('var','_point');
v.definedone();
declare hash duplicates (dataset:'have (obs=0)',ordered:'a',multidata:'y');
duplicates.definekey('var');
duplicates.definedata(all:'y');
duplicates.definedone();
end;

else do;                 /* Not new var? Then we are processing a duplicate       */
if _point^=. then do;  /* Then go back to add the first dupe if not already done*/
set have point=_point;
v.replace(key:var,data:var,data:.);
end;
end;
if eod then rc=duplicates.output(dataset:'duplicates2');
run;``````

Note by declaring hash object DUPLICATES as "multidata:'Y'", it can hold multiple items ("rows") per key value.  Making it "ordered:'A'" means that it is stored in sorted order.

Regards,
Mark

Contributors