|
Abstract:
|
Statistical matching is a method for combining survey databases on distinct topics that is widely used in microsimulation modeling and media research. Several techniques have been used to calculate sampling variances for statistically matched results, including multiple imputation (Rassler 2002), generalized variance functions (Reilly 2000) and resampling methods such as random groups and the jackknife (Ingram et al. 2000; Reilly 2003), but little work has been published comparing these methods. A variant of multiple imputation proposed by Robins and Wang (2000) is newly applied to statistical matching, and the resulting variance estimates are compared with those from traditional multiple imputation and other methods. A New Zealand media research service provides an illustrative application of these techniques.
|
