An Assessment of Some Technical Problems in the Comparative Study of Residential Segregation

Abstract

It is common ground among urban ecologists that a segregation index used in comparative analysis should not have a 'built-in' relationship with any of the variables that are studied as correlates of segregation (Taeuber and Taeuber, 1965). The 'index war' (Peach, 1975), which raged between 1947 and 1955, was settled in favour of the index of dissimilarity (D) largely on account of the fact that it was perceived to measure segregation independently of the effect of population composition (Duncan and Duncan, 1955a). But the basis of the widespread use of D has recently been challenged by Cortese, Falk and Cohen (1976) and Winship (1977) who have demonstrated that D is sensitive to the size of the areal unit and proportion black2 if segregation is measured from the baseline of a random, rather than an even, distribution of groups. This critique has most substance when small areal units are employed: the choice of baseline makes little practical difference when the average population of areal units exceeds 1000 people. Now Johnston (1981) has used a recent paper by the author (Morgan, 1980) to identify a new problem in using D, and the closely related segregation index (S.I.; Duncan and Duncan, 1955b), in comparative studies of segregation. This is potentially an important critique since if it has substance it undermines the basis of nearly all comparative analyses of segregation at all scales (for example, Taeuber and Taeuber, 1965; Roof, Van Valey and Spain, 1976). The object of this note, therefore, is to assess its validity. My paper (Morgan, 1980) demonstrated on the basis of a correlation and regression analysis of 50 Standard Metropolitan Statistical Areas (S.M.S.A.) that, inter alia, the logarithm of the population size and the manufacturing ratio influenced the degree of occupational segregation in metropolitan areas in the United States. Johnston suggests that these are not substantive findings, but are artifacts of the procedures that were used. He employs hypothetical examples to demonstrate that, ceteris paribus, when occupational groups are distributed over census tracts of constant size so that each group is as isolated as possible from all other groups, segregation increases with the population size of the S.M.S.A. and decreases with its manufacturing ratio. These findings have a common origin. Both population size, through its impact on the number of tracts, and the manufacturing ratio, through its impact on the size of occupational groups, influence the proportion of the population that is able to live in tracts occupied exclusively by members of the same occupational group. The second criticism is much the weaker. The manufacturing ratio is not an amalgam of three of the occupational groups, as Johnston assumes in order to demonstrate the existence of a negative linear relationship; it is the percentage of the labour force engaged in manufacturing and is drawn from the industry tables in the United States census. In any case, Johnston's