Small Area Prediction for the Canadian Labour Force Survey

 joint work with Emily J. Berg (U.S. National Agricultural Statistics Service)

Small area prediction for a two-way table of counts, where the estimated counts are from a complex sample, is considered.
Modeling is difficult because distributions may differ by area, estimates of means may be correlated with estimates of variances, the parameter space is restricted, and mean models are nonlinear.
Possible procedures are considered and illustrated with data from the Canadian Labour Force survey.  The analysis demonstrates that large gains in mean squared error are possible.