ANALYSIS OF AGRICULTURAL FIELD TRIALS IN THE PRESENCE OF OUTLIERS AND FERTILITY JUMPS

We show how a Bayesian analysis of a fertility model incorporating many of the previously suggested models can account for uncertainty about which fertility model provides the best approximation in any given trial. We also show how uncertainty about anomalies such as outliers and fertility jumps can be accounted for. We argue that this is preferable to conditioning on an ''appropriate'' model, and show by examples how accounting for such possible anomalies can both influence support for a particular fertility model and reduce the dependence of treatment estimates on the choice of fertility model.