Using principal component analysis to develop performance indicators in professional rugby league

Previous research on performance indicators in rugby league has suggested that dimension reduction techniques should be utilised when analysing sporting data sets with a large number of variables. Forty-five rugby league team performance indicators, from all 27 rounds of the 2012, 2013 and 2014 European Super League seasons, collected by Opta, were reduced to 10 orthogonal principal components with standardised team scores produced for each component. Forced-entry logistic (match outcome) and linear (point’s difference) regression models were used alongside exhaustive chi-square automatic interaction detection decision trees to determine how well each principle component predicted success. The 10 principal components explained 81.8% of the variance in point’s difference and classified match outcome correctly ~90% of the time. Results suggested that if a team increased “amount of possession” and “making quick ground” component scores, they were more likely to win (β = 15.6, OR = 10.1 and β = 7.8, OR = 13.3) respectively. Decision trees revealed that “making quick ground” was an important predictor of match outcome followed by “quick play” and “amount of possession”. The use of PCA provided a useful guide on how teams can increase their chances of success by improving performances on a collection of variables, instead of analysing variables in isolation.