8.2 Descriptive Statistics with PROC MEANS & PROC SUMMARY

8.2 Descriptive Statistics with PROC MEANS & PROC SUMMARY

Descriptive statistical analysis is a core data reporting task. In SAS, PROC MEANS and PROC SUMMARY are sister procedures designed to compute summary statistics for numeric variables across an entire dataset or within specific categorical subgroups. Both procedures share virtually identical syntax and analytical engines, but differ in their default output behavior.

---

1. PROC MEANS vs. PROC SUMMARY: The Key Difference

Understanding the fundamental distinction between PROC MEANS and PROC SUMMARY is a frequent SAS certification topic:

• PROC MEANS: Prints descriptive statistics to the active output destination (Listing, HTML, PDF) by default. To suppress printed output, you must explicitly add the NOPRINT option.
• PROC SUMMARY: Computes identical statistics but produces NO printed output by default. To print output from PROC SUMMARY, you must explicitly include the PRINT option on the procedure statement.

/* PROC MEANS prints output automatically */
proc means data=sashelp.shoes;
   var Sales;
run;

/* PROC SUMMARY computes statistics silently (typically paired with OUTPUT OUT=) */
proc summary data=sashelp.shoes;
   var Sales;
   output out=work.sales_summary mean=AvgSales;
run;

/* PROC SUMMARY with PRINT option behaves like PROC MEANS */
proc summary data=sashelp.shoes print;
   var Sales;
run;

---

2. Default Statistics vs. Explicit Keywords

When PROC MEANS is executed without specifying statistic keywords, SAS computes a standard set of five default statistics for every numeric variable in the VAR statement (or all numeric variables in the dataset if VAR is omitted):

1. N: Number of non-missing values.
2. MEAN: Arithmetic average.
3. STD: Standard deviation.
4. MIN: Minimum value.
5. MAX: Maximum value.

Specifying Explicit Statistics Keywords

If you list any statistic keyword on the PROC MEANS statement line, SAS overrides the five defaults and calculates ONLY the statistics you explicitly requested!

/* Requests ONLY N, NMISS, MEAN, and MEDIAN */
proc means data=sashelp.shoes n nmiss mean median maxdec=2;
   var Sales Returns;
run;

Frequently Tested Statistic Keywords:

---

3. Options on the Procedure Statement

• MAXDEC=n: Specifies the maximum number of decimal places (n from 0 to 8) to display in printed statistics.
• NONOBS: Suppresses the display of the total observation count column in printed output.
• NOPRINT: Suppresses all printed output (used primarily when creating output datasets via OUTPUT OUT=).
• DATA=dataset: Identifies the input dataset.

---

4. Subgroup Analysis: CLASS vs. BY Statements

Both CLASS and BY statements allow you to analyze data within categorical subgroups, but their underlying execution mechanics differ fundamentally.

/* CLASS Example - Unsorted Data OK */
proc means data=sashelp.shoes maxdec=2;
   class Region Product;
   var Sales;
run;

/* BY Example - Pre-sorting Mandatory */
proc sort data=sashelp.shoes out=work.shoes_sorted;
   by Region;
run;

proc means data=work.shoes_sorted maxdec=2;
   by Region;
   var Sales;
run;

---

5. Output Datasets: The OUTPUT OUT= Statement

The OUTPUT OUT= statement directs SAS to write computed statistics into a new SAS dataset.

proc means data=sashelp.shoes noprint;
   class Region;
   var Sales Returns;
   output out=work.shoes_summary
          mean=AvgSales AvgReturns
          sum=TotalSales TotalReturns;
run;

Syntax Rules for OUTPUT OUT=:

1. OUT=dataset-name: Names the destination output dataset.
2. statistic=new-variable-list: Maps statistics to output variable names. The order of variables in new-variable-list matches the order of variables in the VAR statement.
3. The AUTONAME Option: When added to the OUTPUT statement, SAS automatically creates output variable names by combining the original variable name with the statistic keyword name, eliminating the need to type manual variable lists.

proc means data=sashelp.shoes noprint;
   class Region;
   var Sales Returns;
   output out=work.shoes_autonamed
          mean= sum= / autoname;
run;
/* Creates variables: Sales_Mean, Returns_Mean, Sales_Sum, Returns_Sum */

Automatic Variables in Output Datasets: _TYPE_ and _FREQ_ when using CLASS

When a CLASS statement is used with OUTPUT OUT=, SAS creates two automatic variables in the output dataset:

• _FREQ_: Contains the number of observations included in that specific subgroup calculation.
• _TYPE_: A numeric variable indicating the combination of CLASS variables used to aggregate the statistics:
  • _TYPE_ is a binary code: with CLASS A B;, the leftmost CLASS variable owns the high-order bit.
  • _TYPE_ = 0: Grand total across all observations (no CLASS variable used for grouping).
  • _TYPE_ = 1: Statistics grouped by B only (the last CLASS variable listed).
  • _TYPE_ = 2: Statistics grouped by A only (the first CLASS variable listed).
  • _TYPE_ = 3: Statistics grouped by A and B together (every CLASS variable).

Restricting the Aggregation Levels with WAYS and NWAY

By default OUTPUT OUT= writes every _TYPE_ combination, which is 2^n rows-worth of levels for n CLASS variables. Two mechanisms narrow that down:

• WAYS n; — a statement that requests only the n-way combinations. ways 1; keeps single-variable summaries; ways 0 2; keeps the grand total plus every two-variable combination.
• NWAY — an option on the PROC MEANS statement that keeps only the highest-order combination (all CLASS variables crossed), equivalent to the maximum _TYPE_.

/* Only the one-way summaries: Region alone, then Product alone */
proc means data=sashelp.shoes noprint;
   class Region Product;
   ways 1;
   var Sales;
   output out=work.oneway_stats sum=TotalSales;
run;

/* Only the fully crossed Region * Product cells */
proc means data=sashelp.shoes nway noprint;
   class Region Product;
   var Sales;
   output out=work.crossed_stats sum=TotalSales;
run;

/* Example of filtering specific aggregation level in DATA step */
data work.region_totals_only;
   set work.shoes_summary;
   if _type_ = 1; /* Keeps only the level-1 CLASS summaries */
run;