The UNIVARIATE Procedure

Example 3: Computing Robust Estimators

Procedure features:

PROC UNIVARIATE statement options:

	ROBUSTSCALE
	TRIMMED=
	WINSORIZED=

Data set:

STATEPOP

This example

computes robust estimates of location
computes two trimmed means
computes a Winsorized mean.

Program

options nodate pageno=1 linesize=80 pagesize=72; proc univariate data=statepop robustscale trimmed=6 .25 winsorized=.1;

var citypop_90;

title 'United States 1990 Census of Population and Housing'; run;

Output

Because each value of population is unique, the mode is missing.
Both the trimmed and Winsorized means are smaller than the arithmetic mean. This may be due to the positive skewness of the data. PROC UNIVARIATE trims 6 observations or 11.76 percent of the data from the tails. When you request to trim 25 percent of the data, PROC UNIVARIATE trims 13 observations or 25.49 percent of the data from the tails. This is because the number of observations trimmed is the smallest integer greater than or equal to 12.75 (.25×51). Likewise, when you compute a Winsorized mean for 10 percent of the data (.1×51=5.1), PROC UNIVARIATE uses 6 observations or 11.76 percent of the data from the tails.

Chapter Contents
Previous
Next
Top of Page