Factors Influencing the Accuracy of Predictions of the 2014 FIFA World Cup

The purpose of this paper was to compare the accuracy of different simulation models of the 2014 FIFA World Cup. There were 12 (2 x 3 x 2) models altogether (2 data sets of previous matches, 3 sets of variables and models where the data either satisfied the assumptions of linear regression or not). One set of previous data consisted of 440 matches from all international tournaments played since the 2006 FIFA World Cup. The second data set was a subset of 96 of these 440 matches that were from inter-continental tournaments. There were three predictor variables used; difference in FIFA World ranking points (PD) between the 2 teams contesting a match, difference in distance travelled to tournaments (DD) and difference in recovery days from previous matches (RD). The goal difference (GD) between the two teams contesting a match was modelled using (a) PD only, (b) PD and DD and (c) PD, DD and RD. Six models were produced without any changes in the data meaning they violated the modelling assumptions. The other 6 models were constructed using data that had been transformed and outliers being removed in order to satisfy the modelling assumptions. The standard deviation of residual values was used to add random variation about expected results within the simulator. The models that satisfied the assumptions of linear regression were not as accurate at predicting the outcomes of the 2014 World Cup matches as models where the assumptions were violated. Models based on 2 or 3 variables were more accurate than models based on PD alone. Finally, models based on the complete set of previous tournament data were more accurate than those based on the subset of data from inter-continental tournaments.