Auckland Urban Liveability Index

Background

This project started as my undergraduate dissertation at the University of Leeds, supervised by Roger Beecham. Auckland was a good test case: it’s ranked highly on global liveability indices despite a car-dependent, low-density urban form, and its own planning documents (the Auckland Plan 2050) call for a comprehensive, data-backed liveability measure that doesn’t exist yet. After the dissertation, I kept working on it with Roger and Luke Burns to get it to publication standard — extending the validation (adding the MGWR modelling), tightening the methodology writeup, and building the interactive webmap. It was published in Applied Spatial Analysis and Policy in 2025.

Structure (in short)

Research

Read into what urban liveability means in the literature and in Auckland’s own planning documents, then picked the components — social infrastructure, transportation, green space, safety, diversity — that would make up the index.

Data Collection

Collected 29 variables across those five components from OSM and official NZ data portals (Auckland Transport GTFS, NZ Police crime stats, 2018 Census, Education Counts), at Statistical Area 1 (SA1) level (n = 8,047, roughly 100–200 residents each).

Index Building

Transformed the raw variables into indicators — winsorising or log-transforming skewed distributions, standardising polarity, min-max scaling to 1–10 — then aggregated with the Mazziotta-Pareto Index (MPI), which penalises neighbourhoods with unbalanced scores across components rather than letting strong performance on one indicator offset weak performance on another.

Results

Visualised the index and tested it against variables it wasn’t built from (car ownership, population density, deprivation, education) using OLS, spatial lag, and multiscale geographically weighted regression (MGWR) models.

Discussion

Related the findings back to Auckland’s densification agenda and to what they mean for using AULI in other cities.

Methodology, in brief

The 29 indicators were mostly network-distance measures (walking distance from each neighbourhood to the nearest amenity - e.g. hospital, park, bus stop), computed with OSMnx over OpenStreetMap’s road network rather than straight-line distance. Distributions were cleaned (winsorised or log-transformed), reoriented so higher always means more liveable, then min-max scaled to a common 1–10 range before aggregation with the MPI - chosen over a simple mean specifically so a SA1 can’t get a high score by excelling on one component while neglecting another.

To sanity-check the index against variables deliberately left out of its construction, I fit OLS, spatial lag, and MGWR models relating AULI to commute mode, car ownership, ethnicity, education, deprivation, and (log) population density, at both SA1 and the coarser SA2 level.

Results, in brief

AULI shows a clear, strongly spatially autocorrelated pattern (global Moran’s I = 0.73): the highest scores cluster around the CBD and transit-oriented centres like Avondale, Papakura and Onehunga, and the lowest sit in low-density, peripheral, and industrial areas such as around the airport.

The regression checks broadly validated the index: liveability correlates positively with cycling to work and not owning a car, and negatively with driving to work — consistent with the densification and active-travel goals the index was built to measure. Controlling for spatial autocorrelation (via the spatial lag and MGWR models) weakened several relationships seen in the OLS fit, including a surprising one: deprivation, which looked positively associated with liveability under OLS, drops to near-zero or negative once space is properly accounted for. A further equity check — plotting population share by ethnicity against AULI percentile — found no strong disparity in access to liveable areas by ethnicity.

The full paper has the complete methodology, indicator list, and model diagnostics.