Purpose: Relative and absolute decisions in physical activity research are commonly based on a single assessment period that may not represent habitual activity behavior. This study used the generalizability theory to 1) quantify multiple sources of measurement error and 2) estimate the number of days and seasons needed to characterize long-term levels of activity.
Methods: Between 2005 and 2006, youth participating in an intervention program were asked to wear a pedometer across seven consecutive days during three separate months (September, January, and May). Total variance in activity was partitioned and quantified according to differences among participants, inconsistency across days, relative differences among seasons, and the interactions among variables. Two coefficients (generalizability and phi) were calculated from multiple decision studies using a random and mixed design to determine the study protocol needed to achieve a reliability of 0.80 for relative and absolute decisions, respectively. Data were analyzed in 2009.
Results: Complete data were available for 42 boys and 38 girls. Residual variance accounted for the largest source of measurement error (55.64%), whereas smaller amounts were attributed to the participant (18.74%), season (6.59%), and day (2.67%) terms. Using a random design, both coefficients failed to reach an acceptable level of reliability using a single season. In contrast, using a mixed design, an acceptable level of reliability could be reached using 7-8 d from a single, fixed season.
Conclusions: This study demonstrates the potential for using the generalizability theory to make decisions regarding the rank order of activity among individuals (relative decision) and compliance rates for physical activity recommendations (absolute decision).