The 2006 reverse record check sample allocation

Sample allocation can be optimized with respect to various goals. When there is more than one goal, a compromise allocation must be chosen. In the past, the Reverse Record Check achieved that compromise by having a certain fraction of the sample optimally allocated for each goal (for example, two thirds of the sample is allocated to produce good-quality provincial estimates, and one third to produce a good-quality national estimate). This paper suggests a method that involves selecting the maximum of two or more optimal allocations. By analyzing the impact that the precision of population estimates has on the federal government's equalization payments to the provinces, we can set four goals for the Reverse Record Check's provincial sample allocation. The Reverse Record Check's subprovincial sample allocation requires the smoothing of stratum-level parameters. This paper shows how calibration can be used to achieve this smoothing. The calibration problem and its solution do not assume that the calibration constraints have a solution. This avoids convergence problems inherent in related methods such as the raking ratio.