Retest reliability of recall measures of leisure-time physical activity in Australian adults

Background. Several studies have reported that short-term recall measures of physical activity participation have acceptable repeatability, but in most cases employed convenience samples and did not use optimal statistics. In this Australian study repeatability was assessed on participants recall of activity over two different time periods and over the same time period. Methods. Two 14-day recall measures of physical activity participation were administered in two studies to randomly selected population samples of adults in Adelaide, South Australia. Intraclass correlation coefficients (ICC), 80% and 95% limits of agreement and the kappa statistic were calculated. Results. For a continuous measure of energy expenditure the ICC was 0.86 using recall of the same 2-week period (N = 115), and was 0.58 for activity recalled over different 2-week time periods (N = 116). For categorized energy expenditure (Inactive, Low, Moderate and Vigorous categories), kappa was 0.76 for recall of the same period and was 0.36 for different recall periods. Similar results were observed for continuous and categorical forms of a measure of physical activity that recorded frequency of participation in vigorous and moderate activities and walking. The 80% limits of agreement were markedly smaller than 95% limits of agreement, but were still large. Conclusions. These data suggest that the variation in repeatability coefficients between recall of the same 2-week time period and activity recalled over different 2-week periods was due to actual variation in physical activity participation over different time periods, and not to poor recall or to poor measurement characteristics. The recall measures appear to have good repeatability for most respondents, but repeatability is poor for a substantial minority of the population.