Stepwise multiple regression requires that model selection (i.e. deciding which regression variables should be included in the final MAM) is conducted through parameter inference (i.e. testing whether parameters are significantly different from zero) (Chatfield 1995), which can lead to biases in parameters, over-fitting and incorrect significance tests. To see this, consider a simple example, using a single parameter. Consider the linear model that models an observation yi as a function of parameters α and β, predictor value xi and some error ɛ: