Adding new factors to predictive models: the effect on the risk of extrapolation

Abstract

Two aspects of the addition of new environmental factors to predictive microbiology models are discussed. The concept of risk of extrapolation is introduced to characterize the probability that a prediction is outside the model interpolation region. For an empirical model, the interpolation region is defined by the data which are used to estimate the parameters of the model. It is shown that the risk of extrapolation can become unexpectedly high during extending a model to describe the effect of newer factors, if the extension is supported by insufficient data. A convenient method of extending predictive models is also presented to accommodate observations on the effect of additional environmental factors. New factors can be added to a basic common model, in such a way that the basic model will be a special case of the extended models. Conditions under which this approach is useful are discussed. Published data are used to illustrate both points.