Discussion by J.A. Hartigan

Dr. Murtagh paints a grim picture of the typical astronomical database, containing data of various types and complexity from various sources with varying reliability. And I do not think much comfort is available in standard statistical theory for handling such data; most multivariate analysis presupposes data to be in a rectangular data matrix and begins by assuming that points are sampled independently from some population. So, yes, usually we begin by getting data into shape by crude procrustean means, lopping off bits of data in one place, and interpolating bits of data elsewhere.