The Application of Time Series Methods to the Analysis of Repeated Surveys

Abstract

Summary An estimate y, of a population value 0t from a survey carried out at time t can be expressed as y, = 8, + e,, where e, is the error of estimation. By representing 0, as a time series, possibly non-stationary, the problem of estimating 0 +. becomes one of signal extraction in the presence of noise. This paper extends the earlier theoretical results of Scott and Smith (1974) to surveys of complex design and applies the results to examples of both non-overlapping and overlapping surveys. The empirical results suggest that worthwhile gains can be obtained by using time series methods. This paper is particularly directed to the requirement of paragraph 3(a) in the statement of Editorial Policy.