Regression and model-building in conservation biology, biogeography and ecology: The distinction between – and reconciliation of – ‘predictive’ and ‘explanatory’ models

Abstract

In many large-scale conservation or ecological problems where experiments are intractable or unethical, regression methods are used to attempt to gauge the impact of a set of nominally independent variables (X) upon a dependent variable (Y). Workers often want to assert that a given X has a major influence on Y, and so, by using this indirection to infer a probable causal relationship. There are two difficulties apart from the demonstrability issue itself: (1) multiple regression is plagued by collinear relationships in X; and (2) any regression is designed to produce a function that in some way minimizes the overall difference between the observed and ‘predicted’ Ys, which does not necessarily equate to determining probable influence in a multivariate setting. Problem (1) may be explored by comparing two avenues, one in which a single ‘best’ regression model is sought and the other where all possible regression models are considered contemporaneously. It is suggested that if the two approaches do not agree upon which of the independent variables are likely to be ‘significant’, then the deductions must be subject to doubt.