Bayesian optimal stratified sampling design
Abdul J. Sankoh · Communication in Statistics- Theory and Methods · 1992
An optimal allocation problem in a stratified finlte population is studied when the stratum variables are assumed to have independent Blackwell-MacQueen sequence priors. It is shown that if we have independent Blackwell-MacQueen sequence priors on the irh stratum parameter vector XI=(XII ,…, XINI), where Nl(i=l ,…, L) is the ith stratum population size, Ericson's (1969b) and, when the grand mean is the parameter o f interest, Khan's (1976) optimal allocation rc?sults for categorical finite populations can be generalized t o noncategorical finite populations. A comparative sensitivity analysis is also performed to determine the extent to which the allocated sample size in the single phase Bayesian and Neyman optimal allocation designs varies when the allocation parameters such as the stratum size Nj, standard deviation 5, and stratum sampling cost c, are varied over a wide range. The comparative analysis study shows a much higher sensitivity in the Bayesian design than in the Neyman design.