In designing a large multilevel state sample, there are a number of contending factors that shape the design of the survey. The School and Staffing Survey, to be conducted by the Bureau of the Census for the Center for Education Statistics, exemplifies many of these factors. The survey is designed to collect information form school districts, schools within sampled school districts, and teachers within sampled schools. At each level, analyses are planned for a number of characteristics. This paper presents a mathematical paradigm for allocating resources and determining a sample allocation which simultaneously provides a variety of estimates with low variances. Total costs are fixed for the survey, but variable costs are considered for the different units being contacted in the set of interviews.
-Cowan, Charles D. and Malec, Donald J.