Fletcher, R.B. and Atkins, S., Carr, S., Massey University at Albany, New Zealand
A fundamental omission from the team dynamics research has been the application of multilevel models to understand the true relationships among individual and team level variables. Typically team data is analyzed by disaggregating all team level variables to the individual level. Analysis at the individual level ignores the team effects on each individual, which results in a violation of the assumption of independence of observations (de Leeuw, 1992). Alternatively one can analyze data at the higher level by aggregating the individual level variables to the team level. A concern with this approach is that prior to any analyses all within-team variation is thrown away, which means most the total variation is lost (de Leeuw, 1992). Aggregation of player level variables also produces exaggerated relationships between variables and therefore information is squandered, relationships are misrepresented and erroneous interpretations on the basis of aggregated data are made when extrapolating the results to the player level. A more serious issue of analyzing hierarchical data at the single level is the increased probability of making Type I errors. Thus, not taking into account team dependence among individuals is likely to lead to false conclusions regarding relationships among variables. Aggregation and disaggregation are both unacceptable techniques for analysing team data (de Leeuw, 1992). This paper will present a rationale for using multilevel models, which simultaneously analyze individual and team effects to produce more accurate relationships among variables, in the context of team based research using examples from current research (Wilkinson & Fletcher, in review, Fletcher & Dalgliesh, manuscript in preparation).