Abstract
Selecting the number of observations from each stratum is a primary decision in stratified random sampling design. Typically, allocation schemes aim to minimize or bound the variance associated with estimating some overall population parameter, subject to a limitation on sampling resources. This paper examines the impact of further constraints on allocations for stratified sampling; it was motivated by an application requiring all stratum means to be estimated with equal precision. Simple procedures for analyzing trade-offs between equal-precision allocations and those optimizing total cost or precision of overall population estimates are presented. The effect of equal-precision allocation is illustrated within the context of an anthropological study of eight strata defined by villages. In this example, the equal-precision allocation greatly improves the precision of estimating the stratum-specific means over the customary proportional allocation, with only mild degradation in the precision used to estimate the population mean.
Acknowledgements
The authors would like to thank Dr S. Crane for posing such a simple question that led to interesting discussions.