Teun Remmers

128 | Chapter 8 Accelerometer measures Accelerometers provide reliable and accurate measures of youth's PA patterns (38, 39). In this study, accelerometers (GT3X, ActiGraph, Pensacola, Florida) were set to record data at 30Hz, and accumulated data into 10 second epochs. The manufacturer's software (ActiLife version 6.11.9) was used for initialization and initial screening of data-output. In order to manage the data-load of subsequent analyses, wear time criteria of => 600 minutes of wear time per day for at least two weekdays were applied (40). Weekend days were excluded. GPS measures The GPS logger used in this study (BT-Q1000XT, Qstarz International Co, Taipei, Taiwan) has demonstrated relatively good static spatial accuracy compared to other units (41), and acceptable dynamic accuracy (42). GPS and accelerometer devices were worn on the waist and taken off just before bedtime. We used the manufacturer's software (QTravel version 1.46) for initialization and downloading data-output. In order to optimize sample frequency while considering the limited data-storage capacity of the GPS when using a 7- day protocol, devices were set to record data at 10 second epochs. Furthermore, we configured the device to record date, time, longitude, latitude, elevation, speed, signal-to- noise ratio, number of satellites in reach, and to stop logging when storage capacity was full (18). Data analysis Data management in PALMS and data reduction Accelerometer and GPS data were processed using the Personal Activity and Location Measurement System (PALMS), which allows users control over most parameter settings in a web-based application (43, 44). Intensity of accelerometer activity was categorised into sedentary behaviour, light PA (LPA), moderate PA (MPA), and vigorous PA (VPA) according to Evenson's cut-points (45), which performed best in free-living activities of 5- 15 year-old children (46). We defined non-wear time as  20 consecutive minutes of zero accelerometer counts (47). PALMS processed GPS data by filtering invalid values according to extreme speed (i.e. threshold  130 kmph) and extreme changes in elevation (i.e. threshold  1000 meters). We applied the same PALMS algorithms (version 4) as Carlson et al. (2015) for trip and trip mode classification (e.g. pedestrian, bicycle) (48), with the exception of the speed-thresholds for bicycling. Namely, the present study applied the default of 10-25 kmph bicycling speed-thresholds, while Carlson et al. (2015) applied thresholds of 10-35 kmph because their sample consisted of commuting cyclists that were expected to accumulate higher cycling speeds. Invalid GPS points were imputed from the last known valid point, for up to 10 minutes. Finally, as our accelerometers had higher storage-capacities than the GPS loggers, and because in the present study we were only interested in their combined data, we ordered PALMS to match data based on start- and end-times of the GPS logger. The PALMS dataset resulted in 10-seconds GPS epochs (e.g. latitude, longitude, trip mode, speed) with timestamp-merged accelerometer data (e.g. activity counts, activity intensity classification).