ST11 Using SAS(R) to Estimate Variance by the Jackknife Method     Contributed

Thomas Mule
U.S. Bureau of the Census
Abstract: Calculating variances for estimates from survey sample data usually can not be done by using traditional SASŪ statistical procedures like MEANSŪ or REGŪ . These traditional procedures assume that the data is drawn from a simple random sample. Most survey s amples involve clustering, stratification and unequal probabilities of selection. There are two methods which can be used to account for these when estimating variance: Taylor Expansion or Repeated Replication. SASŪ introduced the SURVEYMEANSŪ and SURVEYRE GŪ procedures starting in version 8 which uses the Taylor Expansion methodology. For analysts who want to use repeated replication methods like the Jackknife, there is currently no SASŪ procedure to do this. I present a SASŪ macro which uses the Jackknife methodology to estimate the variance for totals and ratios. Reference: Wolter, Kirk M., Introduction to Variance Estimation, Springer-Verlag, New York, 1985.

Biography:
Thomas Mule is a mathematical statistician at the Census Bureau. He works on sample design and variance estimation for surveys to evaluate Census 2000. He's used SAS to analyze survey data for the past 6 years.