A number of geographical studies in the past fifteen years have used local or regional unemployment series in the analysis of different aspects of the spatial economy. The common factor linking this work together has been the attempt to use unemployment series in the monitoring, analysis, and forecasting of the behavior of local economies, especially in the analysis of the impact of national economic fluctuations on the structure and performance of regional systems. Most of these studies have used quantitative methods to summarize the relationship between unemployment series, the commonest methods being regression analysis, spectral analysis, and Box-Jenkins methods.This paper is concerned mainly with regression analysis; in particular, since unemployment series are highly and positively autocorrelated, the use of standard regression methods is often inappropriate, and procedures that incorporate dependence in the error term need to be considered. As an empirical example, local unemployment responses in the Severnside area of Great Britain, for the period 1961-78, are estimated by different methods. The central argument of this paper is that the quantitative analysis of local unemployment series may provide important empirical evidence concerning the performance of local economies, but that when quantitative methods are used the statistical assumptions of these methods must be considered. It will be shown that the use of statistically inappropriate approaches may produce conclusions that are quite misleading.
. Regional Response ModelsThe two most important contributions to the statistical analysis of regional and local unemployment are those by Brechling (1967) andThirlwall (1966), and the models developed in those papers have become known as regional response (or Brechling) models. Calibration of the theoretical model proposed by Brechling has usually used multiple regression analysis, or ordinary least squares (OLS), on the following functional form: *Thanks are due to a number of colleagues at the University of Bristol, in particular Leslie Hepple and Ed Thomas, and to two referees of a previous version of this paper.Richard Dunn is research associate, Department of Geography, University of Bristol.