Apparent and real bias in numerical transfer functions in palaeolimnology

We present an evaluation of the procedure by which model prediction bias is examined in palaeolimnological transfer function inference models. We argue that most of the prediction biases commonly reported in the literature are, in fact, fallacious, and are the artificial consequence of the inappropriate manner in which residuals are traditionally examined. We show that the extent of the specious model bias is entirely predictable from first principles and is essentially determined by the strength of the predictive model. We suggest that the analysis of residuals should always be examined as a function of the model's predictions and we discuss the implications of the old and new approaches.