An Empirical Examination of the Classical Assumptions concerning Errors in Data

Abstract

A method is presented to examine the classical assumptions concerning errors in data; namely, that the errors are independently and normally distributed with zero mean and constant variance. The only requirement is that the investigator have two sets of observations on a single set of events. Applied to a set of international trade data, the results indicate that, at least for the data under consideration, the classical assumptions cannot be accepted. Alternative error distributions are hypothesized and tested with a new sample.