Measuring commute patterns over time: Using administrative data to identify where employees live and work

We use administrative and survey data in the Integrated Data Infrastructure (IDI) to allocate workers to job locations (plants), which enables the production of over a decade of commute distance population statistics for New Zealand employees. We find that average commute distance (meshblock centroid-to-centroid) fell from 2005 to 2009 before rising again through to 2018 (the final analysis year), with most regions displaying this general temporal pattern. Census 2013 place of residence and work is used to test our methodology against the alternative of using pre-existing plant allocations from the Linked Employer-Employee Data (LEED) production system. For a consistent set of individuals, our estimate of the commute distance distribution closely matches the corresponding distribution in Census. In contrast, LEED-based estimates tend to significantly overestimate commute distances, including radically overestimating the likelihood of inter-island commuting. Our more plausible results are primarily due to re-engineering the job allocation process, as opposed to exploiting better administrative data, though we make marginal improvements to residential address identification through a new prioritisation method, allowing us to use a broader set of residential address sources than available in LEED.