ST10
Using a SAS(R) IML Nonlinear Programming Procedure to Determine a Single Uniform Weighting Scheme for a Complex Survey Design
Contributed
Richard A. Moore, Jr.
U.S. Bureau of the Census
Abstract:
In probability-based surveys, each sampling unit is assigned a probability of being selected in to the sample. Each selected unit is assigned a weight equal to the inverse of the probability of selection. For estimation, all responses are inflated by the c
orresponding weight and summed. In a simple design, the same weight is used to weight all variables. In more complex designs, this weight can vary with the variable tabulated (3.00 for payroll, 3.50 for sales, ...). Although multiple weights produce more a
ccurate estimates, they make it very difficult for estimating other statistics (e.g., ratios or regressions), since traditional procedures require that the same weight used for all variables.
The 1997 Surveys Of Minority and Woman Owned Business Enterprise
s are tabulated using a multiple weighting scheme. Each record contains up to 9 different weights. Using the SAS IML procedure, PROC NLPNMS, a single weight is assigned to each unit. When units are tabulated using this single weight, nearly all published v
alues are preserved. This talk focuses on (1) the appropriate constraints and objective function, (2) diagnostic measures of efficiency, and (3) characterizing situations where the routine provides inaccurate solutions and offering some possible work aroun
d solutions.
Biography:
Richard Moore is a mathematical statistician with the US Bureau of the Census. He has a BS and MA from Wake Forest University and has 12 years of SAS programming experience.