Paper 248-31

Programming with the KEEP, RENAME, and DROP Data Set Options
Stephen Philp, Pelican Programming, Los Angeles, CA

ABSTRACT
One of the more frustrating things for a new user learning SAS can be the multitude of ways of accomplishing the
same thing, each with its own subtleties. The topic of dropping, keeping and renaming variables in data sets is no
exception. Using a DATA step, there are two ways of manipulating variables for keeping, dropping, or renaming:
DATA step statements and data set options. First we will review the basic workings of the DATA step, which will help
you understand how each approach differs, which will then help you to create more efficient code and avoid some
potentially costly mistakes.
INTRODUCTION
As you likely know, a SAS data set is a table. It has rows (observations) and columns (variables). Besides adding
and deleting rows, the modifications you can make to the structure of a table are adding columns, deleting columns
and renaming columns. To modify columns in a SAS data set you use DROP/KEEP and RENAME. There are two
ways to apply a column modification to a data set: using DATA step statements and data set options. In order to
understand the differences in the behavior between the statements and the options we will first review the basic
workings of the DATA step, then apply that knowledge to create clearer, more efficient code using the data set
options.

DATA STEP PROCESSING
SAS stores its data in tables called data sets. These tables have observations (rows) and variables (columns). Data
sets can be thought of as having two logical parts: a data part and a descriptor part. The descriptor part holds a
description of the data set including variable attributes. The data part holds the actual data. When working with
data sets in DATA steps, it is helpful to peek under the hood and understand how DATA step processing works.

First of all, a DATA step has two phases: the compile phase and the execution phase. When you submit a DATA
step for execution SAS first checks the syntax, compiles the statements and then sets up the Program Data Vector
(PDV). The PDV is an area of memory where SAS stores variable values and attributes. It is the PDV that stores an
observation as it is being processed in the DATA step.

Name _N_ _ERROR_ Var1 Var2 Var3 Var4 Var5
Type N N N $ $ $ $
Length 8 8 8 20 5 12 1
Retain yes yes no no no no No
Format best12. best12. best12. $w. $w. $w. $w.
Value 1 0 64 Trader ggyy moment X
Figure 1. Example Program Data Vector
1
Tutorials SUGI 31


When a DATA step is preparing to read from a data set, it uses the information in the data sets descriptor section to
find out what variables to include in the PDV.

After the compile phase the DATA step executes. During execution the DATA step loops by first reading values into
the PDV, executing statements that may change the values in the PDV and then eventually writing the values in the
PDV out as an observation into the new data set.




























DATA
Counts loops and
resets PDV.
SET
Reads values into
PDV.
EXECUTABLE
STATEMENTS
Manipulates values
in the PDV.
RUN
(OUTPUT;RETURN;)
Pushes PDV to new
data set as an
observation.
Output
Data
Input
Data

Figure 2. Simplified DATA Step Loop.

Corresponding to the two phases of the DATA step, there are two types of DATA step statements: compile time
statements and execution time statements. As their names suggest compile time statements do their work at compile
time and execution statements do their work during the execution of the DATA step. Some examples of execution
time statements are:
Assignment statements (variable =value;).
If-then/else statements.
Do loops.
Generally any statement which relies upon the values of variables stored in the PDV.

Some examples of compile time statements are:
Retain statement.
Array declarations.
Drop statement.
Keep statement.
Rename statement.
2
Tutorials SUGI 31


SYNTAX
Remember, a SAS statement is syntactically different from an option. SAS statements are defined as beginning with
a keyword and ending with a semicolon. Data set options appear in parentheses next to a data set. Beyond
dropping, keeping and renaming variables there are a number of data set options available to the programmer.
Multiple options are separated with spaces.

(option-1=value1 <option-n=value-n)

Some Examples:
Data test(keep=answer1 answer2 key);
Proc summary data =students(keep=grade passFail) nway;
Set travelers(rename=(luggage =bags) );

Notice the RENAME=option has an additional set of parentheses where you specify old variable name =new
variable name.
WHAT THEY DO
Now that we have examined the syntax, lets review what these do.
DROP: specifies variables to be dropped (filtered out) from a table.
KEEP: specifies which variables will be kept in a table.
RENAME: renames a variable. OldVariable =NewVariable.

All three of these actions affect the structure of a table when it is being input or output. The KEEP/DROP/RENAME
compile time statements always affect the output table. This can cause some confusion in that the order of these
statements generally does not affect the DATA steps behavior. Consider the following two DATA steps:

dat a one;
dr op wor ds;
set gr ammar s;
i f subst r ( var 1, 1, 4) = THI S t hen do;
. . .
r un;

dat a t wo;
set gr ammar s;
i f subst r ( var 1, 1, 4) = THI S t hen do;
. . .
dr op wor ds;
r un;

Both DATA steps have exactly the same behavior even though the DROP statement appears in different places.
3
Tutorials SUGI 31






























DATA
Counts loops and
resets PDV.
SET
Reads values into
PDV.
EXECUTABLE
STATEMENTS
Manipulates values
in the PDV.
RUN
(OUTPUT;RETURN;)
Pushes PDV to new
data set as an
observation.
drop;
Output
Data
Input
Data
Figure 3. Drop Statement Affects Output Data.


Remember, the KEEP, DROP and RENAME statements always affects the data being output.

Unlike their equivalent data step statements, data set options apply only to the data set with which they appear. That
way they can affect the data set being read or written. This is one of the more compelling reasons for using the
equivalent data set options.

Whether you use an option or the equivalent statement, there is one additional timing rule that always stays the same:

KEEP/DROP always happens before the RENAME.
4
Tutorials SUGI 31


USING THE OPTIONS
The KEEP=, DROP=and RENAME=options are more flexible than their respective statements in that they can be
applied to the data set being written, the data set being read or both. This gives the programmer much more control
over how the DATA step processes data and can lead to more efficient code. Given the DATA step:

dat a hal f Year ;
set QTRone
QTRt wo;
r un;

The data set being written is halfYear and there are two data sets being read: QTRone QTRtwo. This is illustrated in
figure 4.































DATA
Counts loops and
resets PDV.
SET
Reads values into
PDV.
EXECUTABLE
STATEMENTS
Manipulates values
in the PDV.
drop=
QTRtwo
QTRone
halfYear
RUN
(OUTPUT;RETURN;)
Pushes PDV to new
data set as an
observation.
Figure 4. Reading two data sets.

5
Tutorials SUGI 31

When a data set option is applied to a data set being read from, it applies its action before the data is read into the
DATA step. Such as:

dat a hal f Year ;
set QTRone( dr op = st ar t Dat e )
QTRt wo;
r un;

In this example the variable startDate is being dropped (filtered out) from QTRone before being read into the data
step. In fact, it keeps the DATA step from knowing the variable even exists in the data set. This keeps it from being
created in the PDV.































DATA
Counts loops and
resets PDV.
SET
Reads values into
PDV.
EXECUTABLE
STATEMENTS
Manipulates values
in the PDV.
drop=
QTRtwo
QTRone
halfYear
RUN
(OUTPUT;RETURN;)
Pushes PDV to new
data set as an
observation.

Figure 5. Drop= option affecting one input data set.
6
Tutorials SUGI 31


Using data set options you can also specify that a variable should be dropped when the observation is being written to
the new data set. In this case you would put the option next to the data set on the data statement.

dat a hal f Year ( dr op = st ar t Dat e ) ;
set QTRone
QTRt wo;
r un;

This makes the variable available to the DATA step but prohibits it from being written to the data set halfYear.
































DATA
Counts loops and
resets PDV.
SET
Reads values into
PDV.
EXECUTABLE
STATEMENTS
Manipulates values
in the PDV.
RUN
(OUTPUT;RETURN;)
Pushes PDV to new
data set as an
observation.
drop=
halfYear
QTRtwo
QTRone
Figure 6. Drop= option affecting output data set.
7
Tutorials SUGI 31


The same can be applied to KEEP=and RENAME=as well. Consider the following DATA step:

dat a hal f Year ( r ename = ( st ar t Dat e = begi nDat e ) ) ;
set QTRone
QTRt wo;
r un;

In this case, the variable startDate is being renamed when the observation is being written to halfYear. There will no
longer be a variable named startDate in halfYear. But startDate is still available during the processing of the DATA
step.

Again, notice the different syntax of the RENAME=option. The variables being renamed are placed within ( and ).
This allows you to rename multiple variables. Such as:

( r ename= ( var 1=newVar 1 var 2=newVar 2 var 3=newVar 3 ) )

You can combine options on the same data set. Consider:

dat a hal f Year ( dr op= DI SCOUNT pr i ce
r ename= ( st ar t Dat e = begi nDat e ) ) ;
r et ai n DI SCOUNT = . 30;
set QTRone
QTRt wo;
newPr i ce = pr i ce ( pr i ce * DI SCOUNT ) ;
st ar t Dat e = i nt nx( MONTH , st ar t Dat e, 0) ; * set t o begi nni ng of mont h;
r un;
8
Tutorials SUGI 31


A big drawback to using the statements instead of the options is that the statements apply equally to all data sets
being created. Consider a data step that is creating two data sets. One data set has only unique observations and
the other data set has all the duplicates. For the unique data set we only want to keep the key variables, but for the
duplicates data set we want to keep all the variables. For this case we would use a keep=option on the unique data
set. Since the option only applies to the data set it is next to, the other data set is not affected.

dat a uni que( keep=key)
dupl i cat es;
set myDat a;
by key;
i f f i r st . key and l ast . key t hen out put uni que;
el se out put dupl i cat es;
r un;
































DATA
Counts loops and
resets PDV.
SET
Reads values into
PDV.
EXECUTABLE
STATEMENTS
Manipulates values
in the PDV.
RUN
(OUTPUT;RETURN;)
Pushes PDV to new
data set as an
observation.
duplicates
keep=
unique
myData
Figure 7. Keep= option affecting one output data set.
9
Tutorials SUGI 31


USING DROP/KEEP/RENAME IN A MERGE
These data set options can be especially useful in a data step merge. Consider the following two data sets.
FirstNames LastNames
ID NAME ID NAME
1 Robert 1 De Niro
2 Tom 2 Hanks
3 Reese 3 Witherspoon
4 Nicole 4 Kidman

Here we have two data sets firstNames and lastNames that each has a variable called name. What happens if we try
to merge them to get both names onto one data set? The DATA step will overlay any variables that exist on multiple
data sets. If we tried to merge them by ID, which name would be in the resulting data set?

dat a names;
mer ge Fi r st Names Last Names;
by i d;
r un;

We would end up with one name variable in the data set which would have the last names in it. It is
important to note that by default DATA steps will overwrite values on overlapping variables in a merge without even a
warning. You can change this by specifying msgLevel=i in your SAS system options to force a warning in your log
whenever a merge overwrites variable values.

Names
ID NAME
1 De Niro
2 Hanks
3 Witherspoon
4 Kidman

Whenever variables overlap in a merge like this, the value from the last data set to contribute an observation to the
merge is the one you get; but in this case, not what we want. We want both the first name and the last name. So we
need to rename the variables before they get overlaid. To do this, we need to use the RENAME=option on the data
sets in the merge statement.

dat a names;
mer ge f i r st Names( r ename= ( name=f i r st Name) )
l ast Names( r ename= ( name=l ast Name) ) ;
by i d;
r un;

Now we get a resulting data set with both names.
Names
ID FIRSTNAME LASTNAME
1 Robert De Niro
2 Tom Hanks
3 Reese Witherspoon
4 Nicole Kidman

10
Tutorials SUGI 31

The same applies to performing a join in the SQL procedure. However, in SQL you will get a warning in your log if
you have overlapping variables. Also, the logic of which value gets kept is opposite of the merge: the value comes
from the first named data set, not the last.

pr oc sql ;
cr eat e t abl e names as
sel ect * f r omf i r st Names as L,
Last Names as F
wher e F. i d = L. i d;
qui t ;

Names
ID NAME
1 Robert
2 Tom
3 Reese
4 Nicole

pr oc sql ;
cr eat e t abl e names as
sel ect * f r omf i r st Names( r ename=( name=f i r st Name ) as L,
Last Names( r ename=( name=l ast Name ) as F
wher e F. i d = L. i d;
qui t ;

USING OPTIONS IN PROCEDURES
Another reason to use the data set options is that they can be used with procedures as well as DATA steps. This
adds a lot of flexibility and can often result in reduced steps and much more efficient code. Lets say you have a big
data set with 100 variables and many thousands of observations. You want to create a temporary data set out of it
with only 5 variables. You also want your temporary data set sorted. One (inefficient) way to do this would be:

dat a wor k. myTempDat a;
set saved. bi gDat aSet ;
keep var i abl e1
var i abl e2
var i abl e3
var i abl e4
var i abl e5
;
r un;

pr oc sor t dat a = myTempDat a;
by var i abl e1 var i abl e2;
r un;

The first improvement you could make would be to keep the variables you want before the data is read into the data
step using your KEEP=option in the set statement. Remember, the KEEP statement allows all the variables to be
processed in the DATA step and then restricts which variables are written to the new data set. The KEEP=option on
the set statement allows only the variables you specify into the DATA step and is more efficient.

dat a wor k. myTempDat a;
set saved. bi gDat aSet ( keep= var i abl e1
var i abl e2
var i abl e3
var i abl e4
var i abl e5
) ;
r un;
11
Tutorials SUGI 31


But we still have an extra step that we dont need. We could apply the KEEP=option on the data set in PROC SORT
and then use the procedures OUT=option to create our temporary data set.

pr oc sor t dat a = saved. bi gDat aSet ( keep= var i abl e1
var i abl e2
var i abl e3
var i abl e4
var i abl e5
)
out =wor k. myTempDat a;
by var i abl e1 var i abl e2;
r un;

Now we are only reading the data set once and we are restricting the procedure to process only the variables we
want.

MULTIPLE OPTIONS IN ONE PROCEDURE
Many times we use the SUMMARY procedure to do a count of observations. We can accomplish this by combining
the DROP=and RENAME=data set options.

pr oc summar y dat a= st udent s NWAY;
cl ass gender ;
out put out = count s( dr op=_t ype_ r ename=( _f r eq_=count ) ) ;
r un;

In this case, we are using the RENAME=option to take advantage of the automatic _freq_ variable that PROC
SUMMARY creates. We are also dropping the _type_ variable since it is not needed.

WARNINGS
If you use the KEEP, DROP and RENAME statements you should know that SAS will issue a WARNING and not an
ERROR in the log if the variable does not exist. This can lead to some potentially costly errors if that variable is used
somewhere else in the code and the warning is missed. The data set options on the other hand, will issue an ERROR
if a variable is specified that does not exist. In fact, there are two SAS system options that control how SAS responds
in this situation.

Input Data Sets-- DKRICOND =Error | Warning | NoWarning
Output Data Sets-- DKROCOND = Error | Warning | NoWarning

CONCLUSION
Understanding how the DROP=, KEEP=and RENAME=data set options work helps the programmer to understand
better how the DATA step is working. They can also lead to clearer and more efficient code. Data set options offer
the programmer much more flexibility in specifying how SAS processes data than their statement counterparts.
POINTS TO REMEMBER
The statements always affect how the data is written to the output data step. The options can be applied to
either the input or the output data set.
The statements affect all output data sets equally if more that one is being created. The data set options
can be applied individually to each data set.
Specify a data set option in parentheses after a data set name.
When renaming a variable always remember old =new.
Keep/Drop always happens before the rename.
Restricting the number of variables going into a DATA step can be more efficient.
REFERENCES
Heaton, E., (2003) SAS System Options Are Your Friends, in SUGI28 Conference Proceedings, Cary, NC: SAS
Institute.
12
Tutorials SUGI 31

ACKNOWLEDGMENTS
Thank you Orla Hayden and Michael Shreve for your generous time proofreading.

CONTACT INFORMATION
Your comments and questions are valued and encouraged. Contact the author at:
Stephen Philp
Pelican Programming
Los Angeles, CA
[email protected]
www.pelicanprogramming.com
http://datasteps.blogspot.com

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 trademarks of their respective companies.

13
Tutorials SUGI 31