Estimation procedures for repeated surveys

Abstract

The aim of this thesis is to examine the methods of estimation that have been developed for use in sample surveys repeated at regular intervals, with particular emphasis on the time series methods that have recently been proposed. The practice of surveying a population at regular intervals, to provide timely estimates of characteristics changing over time and the effects of changing conditions, have been of interest throughout the relatively short history of the sample survey method. As long ago as 1942, Jessen showed that by retaining the same sample units from one survey to the next, an improved population estimate could be derived, the gain in efficiency being due to the correlation evident between responses obtained from the same individual unit on different occasions. This idea led to the wider use of overlapping samples in repeated surveys and the development of more general estimates to make efficient use of the overlapping design. While the correlation between responses given by individual units in the population was accepted and used, the population characteristic being estimated was assumed to be a fixed value. Any correlation between them over time is then ignored. This rather anamolous situation was pointed out by Blight and Scott (1973) and estimation procedures have since been developed which incorporate the stochastic properties of the population characteristic into the estimates. Show more