|
Abstract:
|
The Annual Survey of Local Government Finances is conducted by the U.S. Census Bureau to collect data on state and local government financial activity. Every five years a full census of all local governments is taken, and two years subsequent to that the Economic Statistical Methods Division designs and selects a new sample using the last census to create a frame. In the 2019 sample selection cycle ESMD introduced major changes to the sample design of ALFIN; new certainty criteria were added, optimal allocation was used for the first time, and a two-phase probability proportional-to-size design was replaced by a single phase stratified simple random sample. We show that while both designs meet initial requirements the new design gives superior performance along multiple dimensions. In particular the new design improves precision for key variables over the long term life of the sample, allows for unbiased estimation of sampling variances, and also facilitates incorporation of alternative estimators that are robust to influential units. Both sample designs are evaluated through a Monte Carlo simulation experiment using data from the 2012 and 2017 Census of Governments-Finance.
|
