Count games not athletes? The problem of pseudoreplication when analysing small-sided team games

Small-sided games (SSG) are common across a number of team sports and are used for physical conditioning and technical and tactical skill acquisition purposes. As athletes within and between teams interact with each other due to the task structure, individual athletes' performances are inherently interdependent. Accordingly, measurements across athletes are correlated with each other and do not represent independent pieces of information. When data dependencies are not correctly analyzed, this can result in what is called pseudoreplication and lead to erroneous, overly liberal statistical inferences. To assess the magnitude of possible errors, a Monte Carlo simulation was performed comparing analyses using a repeated-measures ANOVA, a Friedman test, and a linear mixed-effects model (LMM). Only the LMM maintained the nominal Type-I error rate whereas increasing the dependency between data points led to large increases in Type-I error rates for the other two methods. A re-analysis of a published small-sided games study indicated high within-game correlations between athletes (ICC = 0.77). To assess the scope of pseudoreplication in SSG research, 102 studies from four leading sport science journals were surveyed. Over 66% of the studies showed indicators of possible pseudoreplication, whereas only 6% reported an appropriate analysis. Small-sided team games researchers accordingly need to be aware of pseudoreplication and how best to avoid this issue through appropriate statistical analysis methods discussed in the present work.