Abstract:
|
By representing a data table and the additive relationships between the table cells as a network graph, Causey, Cox, and Ernst (1985) discuss how the solution to a number of sample design problems are equivalent to solving a problem that is referred to in the operations research literature as a transshipment problem. This paper briefly reviews the Causey et al. paper, discusses the usefulness of the network-graph representation, and illustrates the solution of associated transshipment problems for two examples, focusing on advantages and disadvantages of different software solutions. The first example is a randomization problem in which elements in multiple sub-populations are randomly assigned to equal-size treatment and control groups, subject to controlled-selection constraints with respect to a two-way stratification of each sub-population. The second example allocates stratum-level fielded sample sizes for a second wave of data collection such that in overlapping estimation domains the expected numbers of completed cases are equal to specified targets. The software solutions discussed include Excel Solver, OpenSolver, SAS's PROC NETFLOW, and SAS's PROC OPTMODEL.
|