Merging Data Seven Different Ways

David Franklin, Independent Consultant, New Hampshire, USA

Abstract Hash Tables

Merging data is one of the fundamental functions carried out when manipulating data 5 Since SAS version 9.1, the use of hash tables to merge data has been
to bring it into a form for either storage or analysis. The use of the MERGE statement available.
inside a datastep is the most common way this task is done within the SAS language DATA alldata0;
but there are others. This paper looks at the seven possible methods, including the IF _n_=0 THEN SET patdata;
use of the MERGE statement, for a one to one, or one to many merge, introducing the IF _n_=1 THEN DO;
SAS code needed to combine the data. No one method is better than another but DECLARE HASH _h1 (dataset: "PATDATA");
some pointers will be given on choosing a method for your data. rc=_h1.definekey("SUBJECT");
rc=_h1.definedata("TRT_CODE");
Introduction rc=_h1.definedone();
call missing(SUBJECT,TRT_CODE);
Merging variables from one dataset into another is one of the basic data manipulation END;
tasks that a SAS SET adverse;
programmer has to do. The most common way to merge on data is using the MERGE rc=_h1.find();
statement in the DATA step but there are six other ways that can help. First though, IF rc^=0 THEN trt_code=" ";
some data: DROP rc;;
RUN;
Dataset: PATDATA
SUBJECT TRT_CODE
Loading Unique Dataset into an Array
124263
124264
A
A
6 Unique dataset is put into an array first, then array is used to attach
124265 B TRT_CODE to ADVERSE when reading dataset ADVERSE in.
124266 B DATA _null_;
Dataset: ADVERSE SET sashelp.vtable;
SUBJECT EVENT WHERE libname='WORK';
124263 HEADACHE WHERE ALSO memname in('PATDATA','ADVERSE');
124266 FEVER CALL SYMPUT('X'||memname,put(nobs,8.));
124266 NAUSEA DATA alldata0;
124267 FRACTURE LENGTH trt_code $1;
ARRAY f{&xpatdata.,2} $6 _TEMPORARY_;
This data will be used throughout the poster for each method described. DO i=1 TO &xpatdata.;
SET patdata (RENAME=(trt_code=trt_code_dict));
f{i,1}=PUT(subject,6.); f{i,2}=trt_code_dict;
MERGE Statement
1 Most common code used to merge data – maximum control but data must be
END;
DO i=1 TO &xadverse.;
sorted or indexed before DATA statement. SET adverse;
DATA alldata0; trt_code='';
MERGE adverse (in=a) patdata (in=b); DO j=1 TO &xpatdata.;
BY subject; IF subject=INPUT(f(j,1),best.) THEN DO;
IF a; trt_code=f{j,2}; OUTPUT;
RUN; END;
IF ^MISSING(trt_code) THEN LEAVE;
SQL END;
2 Another common way, but can be resource hungry if large datasets are IF MISSING(trt_code) THEN OUTPUT;
END;
involved.
DROP i j trt_code_dict;
PROC SQL;
RUN;
CREATE TABLE alldata0 AS
SELECT a.*, b.trt_code
FROM adverse a LEFT JOIN patdata b MODIFY Statement
ON a.subject=b.subject; 7 Using PATDATA with the unique records, to update ADVERSE using a MODIFY
QUIT; statement, then putting the data out to a new dataset ADVERSE2 – the
RUN; MODIFY statement will not add TRTCD to the ADVERSE dataset, but the new
dataset ADVERSE2 contains it:
KEY= option DATA adverse adverse2;
3 A method that uses two SET statements, reading data from ADVERSE and then DO p=1 TO totobs;
looping through PATDATA to find a match. _iorc_=0;
SET patdata point=p nobs=totobs;
DATA alldata0;
DO WHILE(_iorc_=%sysrc(_sok));
SET adverse;
MODIFY adverse KEY=subject;
SET patdata KEY=subject /UNIQUE;
SELECT (_iorc_);
DO;
WHEN (%sysrc(_sok)) DO; /*Match Found*/
IF _IORC_ THEN DO;
SET patdata POINT=p; OUTPUT adverse2; END;
_ERROR_=0; trt_code='';
WHEN (%sysrc(_dsenom)) _error_=0; /*No Match*/
END;
OTHERWISE DO; /*A major problem somewhere*/
END;
PUT 'error: _iorc_ = ' _iorc_ /
RUN;
'program halted.'; _error_ = 0; STOP; END;
END;
FORMAT Procedure END;
4 Creates a format from PATDATA and then adds TRT_CODE to ADVERSE using END;
the format. STOP;
DATA fmt; RUN;
RETAIN fmtname 'TRT_FMT' type 'C';
SET patdata; Conclusion
RENAME subject=start trt_code=label; There are a number of methods which can be used to merge data, beyond the
PROC FORMAT CNTLIN=fmt; MERGE statement within a DATA step. No one method is better than another, and the
DATA alldata0; methods shown here are by no means exhaustive. It is only though trying these
SET adverse; different methods at your site that you will see resource efficiencies between the
ATTRIB trt_code LENGTH=$1 LABEL='Treatment Code'; methods.
trt_code=PUT(subject,$trt_fmt.);
RUN; Contact Information
Your comments and questions are valued and encouraged.
David Franklin
16 Roberts Road, Litchfield, NH 03052
Tel/Fax: 603-262-9160 Email: 100316.3451@compuserve.com
Web: http://ourworld.compuserve.com/homepages/dfranklinuk