Evaluation of the ADAM Post-Sampling Stratification Weighting Procedure

Michael Yang, National Opinion Research Center
Dean R. Gerstein, NORC at the University of Chicago
Bruce Taylor, National Institute of Justice

ABSTRACT
The Arrestee Drug Abuse Monitoring (ADAM) program uses post-sampling stratification weighting as a means of eliminating or reducing bias. Unlike most probability-based sample designs, ADAM does not start with a sampling frame and known selection probabilities. To develop compensatory weights in the form of , ADAM estimates the product of selection probabilities and response rates from the achieved sample. The current procedure defines post-sampling strata and compute as the ratio of the number of completed interviews to the population size of each stratum. This procedure might have been effective in reducing bias; but it has certainly produced survey weights with substantial variance, leading to large design effects. The effectiveness of the ADAM post-sampling stratification weighting procedure depends on the validity of two key assumptions: (1) are different in different strata; and (2) the means of major survey variables are different in different strata. This paper empirically evaluates these assumptions using ADAM data collected in the last seven quarters. It estimates the bias for major survey variables and discusses the tradeoffs between bias and variance. This evaluation will have significant implications for the improvement of the ADAM weighting procedure and the precision of ADAM estimates.

(Return to Program Resources)

Updated 05/20/2006