PROC COMPARE in SAS: Compare Two Datasets

PROC COMPARE is the SAS procedure that compares two data sets, or two sets of variables inside one data set, and reports every difference it finds. The short version: name the two data sets with BASE= and COMPARE=, add an ID statement so observations are matched by key, and read the summary reports and the SYSINFO return code. This guide covers the sequence from first run to an automated check, and what the output means. For the wider analysis context, see the guide to ANOVA in SAS and the comparison of SAS and SPSS for research analysis.

What PROC COMPARE is used for in research

PROC COMPARE answers one question: are these two data sets the same, and if not, where do they differ? In research practice that covers checking that a re-run of a cleaning program reproduces the earlier data set, confirming that two versions of an analysis file differ only where you expect, and validating a derived data set against an independently programmed one. If your data sets follow a standard such as CDISC clinical data standards, the same comparison applies to the standardized files.

According to the SAS 9.4 documentation, PROC COMPARE sends its results to the SAS log, the SYSINFO macro variable, procedure output, and output data sets. That means one run can serve a person reading a report and a program deciding whether to continue.

Install or access (official source only)

PROC COMPARE runs inside a SAS session, so it is available wherever you already run SAS code through an installation or environment licensed to you or your institution. The reference page for the procedure is part of SAS’s own documentation on support.sas.com (listed under Sources below). Check which SAS release your site runs, because option behavior is documented per release.

Step-by-step: compare two data sets

1. Name the data sets. BASE= is the base data set and COMPARE= is the comparison data set. SAS documents the basic call as proc compare base=SAS-data-set compare=SAS-data-set; run;.
2. Run the minimal step. The documentation’s own example is proc compare base=proclib.one compare=proclib.two; run;.
3. Match observations by key with ID. Values of the ID variables label each observation in the output. Duplicate ID values produce notes or warnings, so use a key that is unique within each data set.
4. Restrict or pair variables. The VAR statement limits which variables are compared. The WITH statement names the variables in the comparison data set to pair against the VAR variables, and VAR is required alongside WITH. If you omit COMPARE=, the procedure compares variables within the one data set and you must use WITH (documented in the SAS 9.1.3-era overview).
5. Write differences to the log or a data set. WARNING, PRINTALL and ERROR write a description of the differences to the log. OUT= creates an output data set, by default one observation for each pair of matching observations; OUTSTATS= writes summary statistics for each numeric variable pair.
6. Control the report. NOVALUES suppresses the value comparison, NOSUMMARY suppresses the summary reports, and TRANSPOSE prints the report by observation instead of by variable.

proc compare base=work.analysis_v1
             compare=work.analysis_v2;
   id subject_id;
run;

Output and how to read it

The default summary reports

By default PROC COMPARE prints four summary reports: a data set summary, a variables summary, an observation summary, and a values comparison summary (SAS 9.1.3-era overview). Read them in that order. The first two tell you whether the structure matches (which variables exist in only one data set, and whether types, lengths, labels and formats differ). The last two tell you whether the same observations are present and whether the values agree. The BRIEFSUMMARY option suppresses the first four reports.

The OUT= data set

An OUT= data set carries the ID and BY variables and the compared variables, and each row has a _TYPE_ value: BASE, COMP, DIF or PERCENT (SAS 9.1.3-era overview). Options such as OUTBASE, OUTCOMP, OUTDIF and OUTPERCENT add the corresponding observation types. Filtering on _TYPE_='DIF' gives a compact list of the values that differ.

Which numeric variables get statistics

STATS prints statistics only for the numeric variables that are unequal; ALLSTATS and PRINTALL print them for all numeric variables.

The SYSINFO return code

PROC COMPARE stores a return code in the automatic macro variable SYSINFO. The value is the sum of the codes for each condition that is true, so one number can encode several differences. The documented codes are:

Condition	Code	Meaning
DSLABEL	1	Data set labels differ
DSTYPE	2	Data set types differ
INFORMAT	4	Variable has different informat
FORMAT	8	Variable has different format
LENGTH	16	Variable has different length
LABEL	32	Variable has different label
BASEOBS	64	Base data set has observation not in comparison
COMPOBS	128	Comparison data set has observation not in base
BASEBY	256	Base data set has BY group not in comparison
COMPBY	512	Comparison data set has BY group not in base
BASEVAR	1024	Base data set has variable not in comparison
COMPVAR	2048	Comparison data set has variable not in base
VALUE	4096	A value comparison was unequal
TYPE	8192	Conflicting variable types
BYVAR	16384	BY variables do not match
ERROR	32768	Fatal error: comparison not done

For an automated check, SAS shows testing the value right after the step, for example %if &sysinfo >= 64 %then %do; handle error; %end;. Because the codes are sums of powers of two, a result of 4096 means only values differ, while 4104 means values differ and a format differs (4096 + 8). A result of 0 means no differences were found in the conditions listed above.

Troubleshooting

• SYSINFO reads zero or an unexpected value later in the program. PROC COMPARE must run before you check SYSINFO, and every SAS step resets it. Copy it into your own macro variable (%let rc=&sysinfo;) immediately after the step.
• Notes or warnings about duplicate ID values. The ID variables label observations, and duplicates in them are flagged. Choose a key that is unique in both data sets.
• Numeric values flagged that look identical. In the SAS 9.1.3-era overview the default CRITERION= is 0.00001 and the default METHOD= is EXACT (RELATIVE when CRITERION= is used). Check your release’s reference page before relying on these defaults in a validation plan.
• Missing values. By default a missing value is equal only to a missing value of the same kind. NOMISSBASE, NOMISSCOMP and NOMISSING change this.
• Variables or observations that exist in only one data set. LISTVAR and LISTOBS list them.

Free alternatives

If you do not have SAS, the same check can be done in free tools, with different matching rules. The two below are documented by their own projects.

Tool	Function	What the official documentation says
Python (pandas)	DataFrame.compare	Compares two DataFrames element-wise and returns a DataFrame highlighting the differences. It can only compare identically-labeled DataFrames (same shape, identical row and column labels), and matching NaNs do not appear as a difference.
R (base)	all.equal	Tests near equality of two R objects, with a default numeric tolerance of sqrt(.Machine$double.eps), close to 1.5e-8. It returns TRUE or a character description of the differences, so wrap it in isTRUE() inside if conditions.

FAQ

What does PROC COMPARE do in SAS?

It compares a base data set with a comparison data set and reports differences in structure, observations and values, writing results to the log, the SYSINFO macro variable, printed output and optional output data sets.

How do I compare two datasets in SAS with PROC COMPARE?

Use proc compare base=dataset1 compare=dataset2; run;, adding an ID statement for the key variable. The output is described above.

How do I compare only some variables?

List them in a VAR statement, and add WITH to pair them with differently named variables in the comparison data set.

How do I know programmatically whether two data sets matched?

Check &sysinfo immediately after the step; the value is the sum of the condition codes in the table above.

Sources (official documentation)

• SAS Institute, SAS 9.4 COMPARE procedure, Results (accessed 2026-10-29): support.sas.com, COMPARE Procedure, Results
• SAS Institute, COMPARE procedure overview (SAS 9.1.3-era documentation, accessed 2026-10-29): support.sas.com, COMPARE Procedure overview
• pandas documentation, pandas.DataFrame.compare (accessed 2026-10-29)
• R Core documentation, all.equal (accessed 2026-10-29)

Related reading: SPSS statistical analysis software and JASP statistical software for other ways to work with analysis data.