The influence of aggregation on the ordering of distributions

Researchers often compare distributions using some criterion or other of inequality, poverty or welfare. It is standard practice to base such comparative analysis on aggregated data. But will the results obtained be dependent on the degree of aggregation of the data? This paper argues, on the basis of a simulation study, that they will be, since the probability of obtaining a ranking can increase rapidly with the degree of aggregation. Aggregation exaggerates the differences between similar distributions and overlooks crossings at the lower tail. A change in research strategy is warranted: aggregation should be avoided where possible.