In this chapter, I move on from the role played by statistics in informing methods of data collection in sociology as a population science – that is, through sample survey research – to the role statistics plays in informing data analysis – that is, through what has become known as ‘multivariate analysis’. In fact, close links exist in the social sciences between sample survey research and multivariate data analysis; they have to a large extent evolved together.

In order to use surveys to capture the degree of variability in human social life or, in other words, population heterogeneity, the nature of this heterogeneity, or of such part of it as is of research interest, must be specified: that is to say, variables must be envisaged. This entails, to revert to the discussion of Chapter 5, the formation of appropriate concepts and then the development of classifications or scales through which these concepts can be made operational as variables with an adequate degree of reliability and validity. In turn, data from survey research expressed in variable form become the material to which methods of multivariate analysis can be applied in order to bring out – to make visible – the perhaps quite complex relations existing among variables. And it is through the statistical modelling of social data in this way that, in the manner suggested by Stigler, the objects of study of sociology are formed: the population regularities emergent from individual variability that constitute appropriate sociological explananda.

The term ‘variable sociology’ has often been used pejoratively in attacks on quantitative sociology, whether from ‘anti-positivist’ or other positions. It is, however, important to note that two quite different lines of criticism arise, although they are not infrequently confounded.

One objection (see e.g. Abbott, 1992; Esser, 1996; Sørensen, 1998) is concerned with the way in which, in sociological analysis conducted in terms of relations among variables, the action and interaction of individuals that underlie these relations can be, and often are, lost from sight.