|
Abstract:
|
In analyses of complex sample surveys, using balanced repeated replication (BRR) is a commonly used method for variance estimation. To create BRR weights, two primary sample units (PSU) are required per stratum. Creating pseudo-stratum and/or pseudo-PSU is needed when there are not exactly two PSUs per stratum. In NHANES, it is not straightforward to create BRR weights for single year samples as variance strata are formed for two-year data releases with one PSU per year. The purpose of this paper is to compare alternative methods for creating BRR weights using a one-year NHANES sample. Three approaches are considered: 1) randomly assigning two PSU to form a pseudo-stratum, 2) using the upper sampling stage as pseudo-stratum, and 3) treating the sampling PSUs as pseudo-strata and the lower sampling stage as pseudo-PSUs. Simulation results of relative bias and stabilities of the variances are estimated for the proposed methods. Simulation samples are generated from 16 years of NHANES samples. The initial results suggest the instability of single year data is more important than choice of BRR method.
|
