Teun Remmers

Investigating PA patterns in the transition from primary to secondary school | 155 Subsequently, we combined school-specific datasets into multiple PostgreSQL databases (http://www.postgresql.com) , where we performed additional queries to identify before school (i.e. 6 AM – start school time), during school (i.e. based on individual school's schedules), and after school (i.e. end school time – 11:59 PM) time-segments. Additionally, we also extracted data on weekend days. This resulted in eight time- segmented datasets; four datasets regarding primary and four datasets regarding secondary school. Spatial analyses Time-segmented datasets were integrated into ArcGIS version 10.4.1 (ESRI, Redlands, California), where we overlaid GIS-data from the municipality of 's-Hertogenbosch. First, we geo-referenced the parcels for each participant's residence and school, using ArcGIS geocode-functionality and the municipality's address-database (Figure 1). For each dataset, we identified the context ‘home’ by selecting data-points (records) that were within 10 meters of each respondent's residential-parcel. In addition, we identified the other contexts by identifying records within 10 meters from the school parcel, sports facilities, shopping centres or malls. For the identification of active and passive transport behaviours, we applied the above-described PALMS speed-thresholds. Finally, records that were not identified in the above described categories, were defined as records at other locations (e.g. at friend's homes or at parks). These context-definitions were based on the GPS-derived contexts of Klinker et al. (2014) and the Sensewear-derived contexts of De Baere et al. (2015) (16, 33). Finally, for each respondent, we computed the crow-fly distance between home and their primary- and secondary school parcel. Data reduction First, we selected children with longitudinal data in the transition from primary to secondary school (Figure 2). Subsequently, we omitted data from children that moved home between baseline and follow-up measurements, and with more than one home environment (often in the case of divorced parents). From these subsamples, data-points were excluded where the accelerometer was not worn (based on  20 consecutive minutes of zero counts), and data-points with extremely high accelerometer counts (i.e. > 9498 counts per minute) were omitted. Subsequently, we aggregated the data-points to the daily level (Figure 2). In accordance with previous studies investigating time- segmented PA (42-44), we ensured reliability of each time-segment by only selecting records from days with ≥ 50% of the potential wear time. For example, for children from schools that start at 8:00 AM, potential wear time in the before school time segment is two hours (as the definition of before school time starts at 6 AM). In this way, these children were required to have at least 60 minutes of wear time in this specific time segment. Finally, baseline and follow-up data were combined in four longitudinal datasets (i.e. before school, during school, after school and weekends; Figure 2).