Comparative Quality of Aging and Psychological Predictors on Fall-Related Outcomes

To elucidate interpretations of regression data in health education, it is useful to invoke alternative models to understand effects of predictor variables. Partitioning explained variance within any dependent variable (univariate) or variate/function (multivariate) can make possible a better understanding of data. Unfortunately, this task is often difficult due to correlations among predictor variables. However, interpreting explained variance in the presence of correlated predictors should not be avoided due to difficulty. Correlations among predictor variables quite often honor an ecological reality time honored in health education circles, which purports there are multiple factors of influence on various health-related outcomes. A data analysis technique known as commonality analysis decomposes R2-type effect sizes into constituent, non-overlapping segments that visually depict unique and common explanatory powers of predictor variables in all possible combinations. The purpose of this presentation will be to concretely conduct and explain a commonality analysis and extend the analysis to the multivariate case, by way of a canonical commonality analysis. Data from a pilot study conducted to examine psychobiological and aging influences on unintended falls among older adults (n = 83) will be analyzed to demonstrate the utility of these non-traditional analytic methods. Data was collected on the following independent (age, trait anxiety, and falls efficacy) and dependent (data from the Berg Balance Test [BBT] and 1-repetition maximum leg press) variables. The commonality analysis revealed that noteworthy variance explained by falls efficacy (r2 = .245) and age (r2 = .306) was predominantly unique to both predictor variables. The canonical commonality analysis, however, revealed that age (rc2 = .355), as compared to falls efficacy, was the superior predictor when both BBT scores and 1-repetition maximum leg press were considered as outcomes simultaneously. Interestingly, the falls efficacy predictor possessed a larger effect size when both outcome variables were considered simultaneously (rc2 = .253) rather than when BBT scores were considered alone (r2 = .245). Parsing the composition of each predictor variable's effect size (through commonality analysis) illuminated what could not be ascertained by simply consulting the r2 or rc2 of the falls efficacy variable. A commonality analysis was necessary to understand the true predictive value of falls efficacy relative to the univariate versus multivariate context. These procedures are not widely used by health education researchers, perhaps because the interpretive value of such analyses is not widely known. These techniques can prove quite useful when multiple regression and canonical correlation analyses are conducted.