Approaches to the Aggregation Problem

Abstract

The author describes and analyzes different sorts of problems from aggregation bias, a type of composition errors, that can result when shifting from group-level data to individual-level effects. He develops three approaches, grouping, causal modeling, and specification error approach. The analysis shows that all approaches are satisfactory for simple cases, and the latter two are preferable (although not entirely satisfactory) for cases where ordinary methodological difficulties appear.