Statistical methods in performance analysis: an example from international soccer

Abstract

There are previous studies in performance analysis that have concluded significant effects by using a large number of events from a small number of matches. The statistical tests applied in such studies are inappropriate for non-independent data and significant results might not be achieved if performance indicators from whole performances were used. This paper illustrates the extent of the problem using an example where significant results are achieved using individual events from a sample of 9 matches. When the unit of analysis was a match rather than an event, no significant effects were found. Therefore, generalisations cannot be made from small numbers of matches where significance has been concluded based on inappropriate use of statistical procedures. It is recommended that performance analysis investigations use whole performances as units of analysis.