A MULTILEVEL ANALYSIS OF BMI, GENDER, PHYSICAL ACTIVITY, SCHOOL SES AND QUALITY OF FACILITIES IN HEALTH-RELATED FITNESS

Abstract

The identification of predictors of interindividual differences in children's health-related fitness is of relevance not only for physical education teachers, but also for the promotion of health-related programs within community settings. When dealing with survey data about health-related fitness, the hierarchical structure of the information is usually ignored in data analysis. PURPOSE:(1) To analyze the heterogeneity in the 1-mile run/walk performance, a health-related fitness item concerning cardio-respiratory endurance;(2) show the relevance of hierarchical linear models to deal with nested data. METHODS subjects (n = 1255) of both gender, aged 10 to 18 years, from two main cities-northeast region of Portugal were tested on 1-mile run/walk, after parents consent. Total physical activity (TPA) was assessed with the Baecke et al. questionnaire. They belonged to 33 different schools. Schools were classified according to their SES as well as to the quality of their facilities. HLM 5.0 software was used following a data analytic strategy that included:(1) a random effects anova;(2) a level-1 random-intercept and random-slopes model;(3) a level-2 intercept and slopes-as-outcomes model. RESULTS Not only is data highly reliable across schools (R = 0.90), but also showed that 22.5% of total variance in the 1-mile run/walk is due to differences in performance between schools (overall mean performance is 9.97 ± 0.33 minutes). The level-1 random model showed significant (p < 0.05) effects for age (−0.15 minutes), gender (−1.30 minutes), TPA (−0.08 minutes), and BMI (0.09 minutes). The level-2 model showed only a significant impact of SES (a unit change in SES was associated with a decline of 0.47 minutes in performance across schools). CONCLUSION At individual level, gender, age and BMI are the most important predictors of heterogeneity in the 1-mile run/walk. School SES is relevant in explaining differences in performance. Hierarchical linear models are very well suited to analyze data having a nested structure.