Abstract:
|
McQuitty in the 50s observed that if classes of people, PSUs in sampling terms, were formed where PSUs within strata were identical with regard to stratification variables, the number of such variables that could be used would typically be much smaller than the number available with relevant information, since one would rapidly have more potential strata than PSUs. He proposed Similarity Analysis. This was programmed for the ILLIAC. Starting with a matrix of similarity scores the two most similar PSUs or PSU group are combined at each iteration to form a developing stratum. At the 1969 ISI meetings, Beale presented an alternative procedure. Starting with random strata, each PSU of each stratum is tried in each stratum and left in that stratum that gives the largest between sum of squares, iterating until the gain is small. Census is using such a method. We have programmed a variant of each method for large, fast PCs, and have planned a series of studies on the utility of each and combinations of the two. We are using data for 220 variables from the 1994 County and City Data Book. Initial results, where sets of strata have been formed and estimates made therewith, will be reported.
|