In this study, we have taken GPS to represent the best-available method to objectively record actual routes travelled by commuters.
In reality, GPS data is subject to positional error and this has been shown to vary depending on the device, its use, and according to setting,
such as the amount of sky or number of satellites visible to the device (Kerr et al., 2011). A recent study by Schipperijn et al. (2014) tested
the accuracy of the QStarz BT-Q1000X for recording routes taken by different transport modes in a range of settings, finding an overall
median error of just 2.9 m, ranging from 0.7 m in open areas to 5.2 m in heavily built-up urban areas, and from 3.9 m for walking trips to
0.5 m for car journeys. In this analysis, we undertook a comprehensive data cleaning exercise and used buffers to negate such positional
inaccuracies, but it is likely that actual routes delineations are subject to a small error. It is also important to remember that while GPS data
could help to inform or validate route choice models, its use may be inappropriate in some situations such as scenario planning, as
observed behavioural patterns may be determined by factors other than those under study (Chaix et al. 2013).