Active adults recall their physical activity differently to less active adults: test–retest reliability and validity of a physical activity survey

Abstract

This paper determined the test-retest reliability and criterion validity of a modified version of the Active Australia Survey (AAS) and whether these properties varied across participants' activity levels. Participants (n = 63) responded to repeat administrations of the AAS and wore an accelerometer for 7 days. Analyses used Spearman's rho (rs,) or weighted kappa (k) and Bland-Altman methods. Variation in mean difference and 95% limits of agreement (LOA) across average levels of activity were tested by linear regression. Reliability correlations (rs; 95% confidence intervals (CI)) for minutes per week ranged from 0.40 (0.16, 0.59) to 0.80 (0.68, 0.87). For days per week, the agreement (k; 95% CI) between administrations ranged from 0.43 (0.34, 0.73) to 0.83 (0.61, 0.93). There was a small mean difference between administrations (−8.46 moderate-vigorous minutes per week); 95% LOA widened as participants' average activity levels increased. Validity correlations (rs; 95% CI) for minutes per week ranged from 0.50 (0.28, 0.66) to 0.61 (0.43, 0.75). For days per week, the agreement (k; 95% CI) ranged from 0.35 (0.10, 0.50) to 0.61 (0.29, 0.87). The mean difference between the AAS and accelerometer and 95% LOA both varied with participants' activity levels. The reliability and validity of the modified AAS were better than those of previously published versions, but varied according to participants' activity levels. In this study, participants who engaged in more activity had more measurement error than less active participants. This proportionality will have important implications for cross-sectional and intervention studies. This phenomenon needs to be examined for other self-reported physical activity measures.