Four studies out of a series of 6 previous studies have found that predictive models are more accurate at predicting actual match outcomes when the modelling assumptions are violated than when data are transformed to satisfy the assumptions. The current investigation produced two sets of two predictive models of the 2011 Rugby World Cup; one set of models used raw independent variables that violated the assumptions of the modelling techniques used and the other set used data that satisfied those assumptions. Each set of models contained a multiple linear regression based model where independent variables were included using the "Enter" method and a multiple linear regression based model where independent variables were included using the "Stepwise" method. The models created using data that satisfied the assumptions of linear regression were found to be more accurate at predicting actual match outcomes than when the data violated those assumptions. The assumptions strike back!
© Copyright 2012 International Journal of Computer Science in Sport. Sciendo. Alle Rechte vorbehalten.