体育・スポーツ科学における適切な多変量解析手法の選択とは？

抄録

In physical education and sports science, multivariate analysis is needed when outcome variables are more than two at a time. Multivariate analysis is a generic term; several analyses (e.g. multivariate regression, factor analysis) are encompassed with multivariate analysis. Each analysis of multivariate analysis is based on different each theory. Because researchers can select an analysis from multivariate analysis group, research conclusions may change by types of multivariate analysis although when dataset is identical. In this article, we discussed the influence that the differences in types of multivariate analysis affected research findings. We analyzed the relationship between 8 physical fitness variables and physical activity and gender by multivariate analysis of covariance (MANCOVA), exploratory factor analysis (EFA), and structural equation modeling (SEM). There were small differences on results that were concerned with significance of estimated parameters by each analysis; however, effect size of each parameter was equal in MANCOVA, EFA, and SEM. When statistical assumptions (e.g. multivariate normality, statistical independence) are valid, selection of an analysis may be not a serious problem. On the other hand, when statistical assumptions are invalid, researchers should report details of dataset and the rational reason that selected the analysis because research conclusions may easily change by types of multivariate analysis.