Multiobjective Optimal Allocation for Hard to Reach Populations
*Benjamin Phillips, Abt SRBI
Keywords: sample allocation, optimization, stratified sample
Efficiently allocating sample is a necessity for surveys of hard to reach populations is a necessity for reasons of cost and accuracy. However, matters are complicated when overall population and subdomain estimates within the population of interest are required, as these estimates objectives are frequently in conflict.
The example used in this presentation draws on current work on American Jews; a situation made more difficult by the absence of official statistics about the population. The survey objectives encompass both estimates of the population as a whole as well as of subdomains of interest (Orthodox and Russian Jews). Data from previous sample surveys as well as census-like information inform the development of an optimum sampling strategy of this difficult to reach population under cost constraints. Strategies for dealing with the imprecision of extant county-level survey data are discussed. The question of optimal allocation between cell phone and landline frames is also addressed. Nonlinear programming running under readily available software is used to solve the resulting optimization problem.