Introduction: Numerous studies have examined the validity of accelerometry-based activity monitors but few studies have systematically studied the reliability of different accelerometer units for assessing a standardized bout of physical activity. Improving understanding of error in these devices is an important research objective because they are increasingly being used in large surveillance studies and intervention trials that require the use of multiple units over time.
Methods: Four samples of college-aged participants were recruited to collect reliability data on four different accelerometer types (CSA/MTI, Biotrainer Pro, Tritrac-R3D, and Actical). The participants completed three trials of treadmill walking (3 mph) while wearing multiple units of a specific monitor type. For each trial, the participant completed a series of 5-min bouts of walking (one for each monitoring unit) with 1-min of standing rest between each bout. Generalizability (G) theory was used to quantify variance components associated with individual monitor units, trials, and subjects as well as interactions between these terms.
Results: The overall G coefficients range from 0.43 to 0.64 for the four monitor types. Corresponding intraclass correlation coefficients (ICC) ranged from 0.62 to 0.80. The CSA/MTI was found to have the least variability across monitor units and trials and the highest overall reliability. The Actical was found to have the poorest reliability.
Conclusion: The CSA/MTI appeared to have acceptable reliability for most research applications (G values above 0.60 and ICC values above 0.80), but values with the other devices indicate some possible concerns with reliability. Additional work is needed to better understand factors contributing to variability in accelerometry data and to determine appropriate calibration protocols to improve reliability of these measures for different research applications.