Classification techniques in planning analysis

Abstract

PLANNING, like most social sciences is a relatively young discipline. Therefore quantitative techniques that serve taxonomic purposes are very useful for planners to refine, modify and reformulate their hypotheses and classification. This paper is an attempt to familiarize planning practitioners and researchers with a very powerful multivariate technique of analysis that has been used extensively in biometry and psychology [l], but only relatively rarely in the social sciences [2]. The exposition here is completely non-mathematical and is intended to familiarize the planner with the versatility and the applied aspects of the technique [3]. Multiple discriminant analysis is helpful mainly in answering questions on the validity and the dimensionality of any system of classification. The question that the technique principally answers is: “Is the a priori classification under consideration valid in terms of the variables that are being suggested as relevant to the classification ?” As such, the technique is applicable in a wide variety of situations that planners analyse. For example are the characteristics of rural poverty and urban poverty different from each other ? If so, what are the variables that define or express this difference, and what is the relative importance of these variables? Or to take another case : assume that different urban renewal treatments are based a priori on neighbourhoods being characterized by different groups of measurable socio-economic and physical characteristics. Multiple discriminant analysis can answer the question: Are the groups of neighbourhoods that receive different urban renewal treatments, e.g. total clearance, rehabilitation, conservation, really different from each other in terms of pre-hypothesized sets of characteristics? If they do differ, then the technique can tell us the degree to which different variables that have been introduced in the analysis explicate this difference. Discriminant analysis is similar to the better known multivariate technique factor analysis, in that it seeks to economize on the variables that validate the classification, picking out the more important ones and discarding those that are less significant. In this process discriminant analysis produces the “dimensions” of a classification, in much the same way as factor analysis establishes the major dimensions of variability in a set of data [4]. Discriminant analysis is similar to another familiar multivariate technique, multiple regression analysis to the extent that the validity of the analysis in both techniques is based on external criteria; multiple regression is only as good as the choice of the independent variables included in the analysis. Similarly, the validity of discriminant analysis depends on the variables chosen to explicate the classification [5]. However, while multiple regression predicts the degree of correlation and thence the degree of success (i.e. is this an