The assessment of children's physical activity: a comparison of methods

The authors assessed the convergent validity and the sources of error in an electronic single-plane accelerometer (i.e., Caltrac Personal Activity Computer (Hemokinetics, Inc., Vienna, VA). The device was validated against observed all-day physical activity levels of children in their natural environment. Thirty pre-school children were observed in non-structured activity for periods of approximately 9 h while wearing the accelerometer. The results revealed moderately high but variable Spearman rank-order correlations between hourly readings of the accelerometer and the observational system (range of correlations = 0.62 to 0.95). An all-day accelerometer reading significantly correlated with the observational instrument (rho = 0.54). A step-wise regression analysis revealed that the best behavioral predictor of the all-day accelerometer reading was the observed behavior of walking, explaining 32% of the total variance. Older vs younger children (i.e., greater than 32.5 months), females vs males, and overweight (i.e., 75 percentile or greater) vs normal weight children tended to show higher correlations between direct observation and accelerometer readings. Implications of these findings and the utilization of the accelerometer in epidemiologic research are discussed.