ABERRANT DATA IN FIELD EXPERIMENTS

Abstract

A field experiment will sometimes give data so unexpected that their correctness is doubted. It would be wrong to discard them because they go against the experimenter's preconceptions but sometimes enquiry will reveal why they are aberrant. Sometimes no reason can be found but they still raise doubts. The problem is approached using a published body of data.

The calculation of residuals and total residuals is commended as a means of enquiry. Exclusion of data is to be avoided if possible. It can have serious consequences and methods are given for assessing them. It is important to seek alternative approaches and one possibility is the use of covariance adjustments on fertility trends shown by study of the residuals. It is shown also that extreme values sometimes arise from unsuitable data and can be eliminated by a suitable transformation, but it must be meaningful and not just a statistical expedient to avoid difficulties.