Checking Variables for Normality in SAS

Overview

Many statistics require that the variable you are analyzing have a normal distribution about the mean. Without going into the details this requires that the values for a variable across all cases take on certain properties. When analyzing data it is always wise to verify that your variable does in fact have a normal distribution about the mean. If not, then you may have to use a different set of statistical procedures when you do more complex analysis. Within SAS you may use the proc univariate to look for normality.

Procedure Univariate

Proc univariate will generate a page with assorted statistics for each variable specified. The statistics generated by proc univariate fall into three categories, moments which provide statistics such as mean, variance, etc.; quantiles which show the distribution of the values; and extremes which highlight the highest and lowest observed values.

To test for normality you must add the keyword normal to the procedure declaration of proc univariate. Proc univariate will display two statistics at the end of the moments section. These are W:Normal and Pr. The values it will show range from 0.0 <= value <=1.00 . Values closer to 1 indicate a higher degree of normality. SAS uses the shapiro-wilk test when the sample size is below 2000 and the Kolmogorov test above 2000.

How low a value you can accept is dependent on the size of your sample and the accuracy you require in your analysis. For more details on interpreting this consult a statistics manual.

The basic format of this procedure is:

proc univariate data=dataset normal;
var variable-list;
run;

If you do not specify a list of variables then all possible variables will be summarized.

Example 6. Using Proc univariate with normal

Back to Sas Index Page
Author - Jack Suess
UMBC University Computing Services
Created - 1/15/96