ABSTRACT
Sociologists are undoubtedly more familiar with that aspect of the problem of changing units labeled disaggregation. This chapter focuses on reformulating mathematical and statistical arguments in causal terms more familiar to sociologists and in trying to demonstrate that an explicit causal perspective elucidates some issues that tend to be hidden in the technical complexity of the original arguments. The author restricts himself to the logically prior problem of aggregation. He examines the implications of the close similarity of the problem of aggregation and conventional treatments of measurement error. The author discovers that the bivariate case is rather a special case in the sense that some of the most troublesome problems of aggregation do not arise. The chapter suggests that in most economic applications aggregation bias should not exceed 10% of the parameter values.
