Methodology

Every score uses the same inputs.

• City boundaries: GHSL Urban Centre Database R2024A from the Joint Research Centre. One polygon per city, matched by UCDB ID.
• Population: Kontur Population, an H3-indexed global dataset, intersected with the boundary.
• Public transport timetables: GTFS feeds from national or operator sources. Entur for Norway, Trafiklab for Sweden, HSL/Digitransit for Finland, opentransportdata.swiss via geOps for Switzerland, DELFI via public-transport.earth for Germany, transport.data.gouv.fr for France, OVapi for the Netherlands, BODS for the UK, and a long tail of per-city sources. Each feed is validated and sanitised before use.
• Street network: OpenStreetMap, used for walking, cycling and car routing.
• Destinations: OpenStreetMap points of interest, classified into five categories: employment, healthcare, education, retail, leisure.

Reference date is 2026-05-12, a Tuesday. All travel times are computed against the timetable in effect that day.

Socioeconomic indicators

Each city page shows a City Profile alongside the map: basic demographics, transit and mobility behaviour, economy, and environment. These indicators are context, not inputs to the Integration Index or to car-free liveability — the scores do not move when, say, unemployment rises.

The data comes from a single canonical source per country, so numbers stay comparable across the dataset:

• Continental Europe: Eurostat Urban Audit for population, median age, car ownership, modal split, GDP per capita, unemployment, and tertiary education. Eurostat nama_10r_3gdp for the NUTS3 GDP backfill.
• United Kingdom: ONS Census 2021 (population, median age, education TS067, car ownership TS045, modal split TS061), ONS Regional GVA(B) 2022 (GDP equivalent, resident-based, converted GBP→EUR at 1.17), NOMIS Claimant Count (unemployment), DfT Reported Road Casualties (road deaths). UK values are aggregated from Local Authority Districts to the city level using population weights.
• Air quality: WHO Ambient Air Pollution Database (Aug 2022 release) for PM2.5 annual mean, with EEA measurements used where they are fresher than the WHO release.
• Population density is derived: city population divided by the GHSL Urban Centre boundary area.

Indicators with coverage below two-thirds of cities are not displayed, even when they exist for some. Each value on the city page carries its source attribution and data year.

Each city is divided into small hexagonal cells (~460 metres edge to edge). Population comes from Kontur. Cells with at least 500 residents are included in the city total; sparser cells are still scored individually for the map, but don't move the city number.

Destinations are sampled to 2,500 per city using a stratified random method on a larger hex grid (~3.2 km). Category proportions are preserved, and the sample is deterministic (same seed every run). The grid stratification guarantees geographic coverage without favouring spatially dispersed cities, and without clustering where OpenStreetMap is mapped densely.

Destinations are weighted when they enter the reach calculation. Employment counts for 30%, healthcare for 20%, education for 20%, retail for 15%, leisure for 15%. Employment carries the most weight because reaching jobs is the central reason most people use transit.

Travel times come from r5py, a Python wrapper around Conveyal's R5 router. R5 combines GTFS timetables with the OpenStreetMap network and returns realistic door-to-door travel times.

For each origin–destination pair we compute median travel time across two two-hour windows on the reference day: AM peak (07:00–09:00) and midday (11:00–13:00). AM peak captures commuting load; midday captures off-peak frequency, which is a different signal in most networks.

We do not add a transfer penalty. R5's travel times already include the realistic waiting cost of a transfer. Adding more would double-count and punish networks designed around interchange.

Transit reach asks a simple question: of the destinations a person has nearby, how many can they actually get to by transit in a reasonable time?

For each cell we count destinations reachable within 60 minutes by transit. Closer destinations count for more than ones near the edge of the window: full credit up to 30 minutes, three-quarters credit out to 45, half credit out to 60. Category weights are applied on top.

That count is then divided by what's available within a 20 km straight-line radius, regardless of transit. The result is the share of locally available opportunities transit actually reaches, expressed 0–100. A cell needs at least five destinations within 20 km to be scored; sparser cells are dropped as noise.

Reach as a share penalises large cities, where 5,000 reachable jobs scores the same as 50 in a small town. Absolute reach corrects for that by counting the raw weighted destinations and putting them on a log scale: roughly 3,000 reachable destinations earns a score of 100, with everything below scaling logarithmically.

Reach and absolute reach measure quality of coverage and size of the opportunity set. They weigh equally in the quantitative score.

Competitiveness asks: when transit goes somewhere, is it slow? For each origin–destination pair we compare the actual transit time against a notional reference: how long the trip would take at a steady walking-and-waiting speed of 18 km/h along a straight line. If transit roughly matches the reference, the trip scores high. If transit takes twice as long, the score halves. Walkable trips under 1 km are excluded, as are trips longer than 20 km (not part of urban accessibility).

The city score is the population-weighted mean across cells, then adjusted for timetable consistency. We measure how much travel times jump around within the two-hour departure windows: if catching the next service or missing it makes a big difference, that's real luck-of-the-bus friction. Cities with Swiss-style clockwork timetables lose almost nothing here; cities with chaotic timetables can lose around 10–15%.

Equity is a tiebreaker, not a uniform-distribution metric. It rewards cities whose public transport is genuinely good across the whole urban area, and stays out of the way everywhere else.

The rule is simple. If a city's population-weighted mean transit reach is at or above 45, equity is computed as (1 − Gini) × 100 of per-cell transit reach. Uniformly excellent cities score high; cities that are excellent only in pockets score lower. If mean reach is below 45, equity falls back to a neutral 50 — neither a reward nor a penalty. The city still ranks on its other sub-scores; equity simply doesn't move the needle.

This shape stops cities from scoring high on equity by being uniformly mediocre, which earlier versions allowed. The Gini and the 10th-percentile floor are still published alongside each city as diagnostics, useful for planning even when the equity sub-score has reverted to the neutral baseline.

The four sub-scores combine into a 0–100 number using these weights:

Competitiveness carries the highest weight because a network that takes twice as long as a reasonable walking speed isn't doing its job. Reach and absolute reach split coverage breadth and depth. Equity carries the smallest weight and only fires as a bonus for cities whose public transport is genuinely good everywhere; for everyone else it sits at a neutral baseline. The Gini and 10th-percentile floor are still published per city as diagnostics.

City aggregation is a population-weighted mean across dense cells (≥500 residents).

Some integration qualities don't show up in a timetable. Whether one card works on every mode, whether a transfer means crossing a street or crossing a station, whether the evening service collapses into a separate fragmented network. These need human judgement against current evidence.

We score ten dimensions per city, 1–10 each, grouped into four areas:

• Physical interchange: signage and wayfinding, distance between modes at interchanges, physical experience (shelter, accessibility, crowding).
• Ticketing and fares: single ticketing or contactless platform, absence of price penalty for transferring, multimodal products and capping.
• Digital information: multimodal journey planning and real-time information across channels.
• System integration: timed connections (Taktfahrplan / pulse), evening and weekend integration, MaaS reach (bike-share, e-scooters, car-share inside the transit app or ticket).

Each city is also rated on parking difficulty (1–10), which feeds into the CDI rather than the qualitative score.

Scoring runs in two phases. First, we gather current evidence per area from public sources: operator websites, regulator publications, transport authority reports, and recent news covering each city. Each piece of evidence is captured with its source URL. Second, the evidence bundle is scored against the rubric for each dimension on a 1–10 scale, with a one-sentence justification per score and an overall confidence label (high, medium, or low) reflecting how strong the evidence for that city is.

Each area gets a 0–10 score (mean of its sub-dimensions), then is combined into a 0–100 qualitative number using these weights:

Ticketing carries the most weight because fare integration shows up consistently in ridership evidence. System integration matters because it captures whether the network was designed as a whole. Physical interchange is the visible face of integration. Digital information is real but less differentiating; Google Maps gives most cities a baseline.

The Integration Index is the top-level measure of how well a city's public transport works as a single connected system. It blends the quantitative and qualitative scores:

At 40%, the qualitative block has enough weight to register the integration qualities the timetable can't see. Each of its four sub-measures — ticketing, physical interchange, system integration, digital information — carries roughly 10% of the headline. A city with a strong timetable but a fragmented passenger experience won't rank as high as one where the system actually feels connected.

The Integration Index asks how well public transport works on its own. Car-free liveability asks the adjacent question: how much harder is daily life here without a car? It uses the Car Dependency Index framework — the technical name for the measure that compares what you can reach by transit against what you can reach by car.

For every neighbourhood we count the places people can reach within a reasonable time, first by public transport, then by car. Destinations close by count for more than ones further out, and the cutoff is gentle rather than a hard edge. We do this for the same set of destinations and from the same starting points, so the two counts are directly comparable.

Car times include extra minutes for the cost of parking. An easy-parking small city gets a small buffer; a tightly controlled centre like Paris gets a much larger one. The parking score comes from the qualitative research pass and ranges from a 10-minute buffer at the easy end up to nearly half an hour at the hard end.

If the car reaches more places, the neighbourhood leans car-dependent. If public transport keeps up or wins, it doesn't. We average across all neighbourhoods, weighted by population, and project the result onto a 0–100 scale. Higher means public transport holds its own against driving.

The underlying framework follows Campanelli et al. (2026); the exact formulas live in the code.

The Integration Index and Car-free liveability answer different questions and we keep them separate. The index measures how integrated public transport is. Car-free liveability measures how easy it is to live without a car. Both numbers matter, but folding them into a single headline mixes the quality of public transport with the quality of the road network, and penalises PT-strong cities for having good roads.

The list on the front page ranks by the Integration Index. Car-free liveability is reported on each city page so readers can answer either question without one obscuring the other.

Alongside the two numbers, each city carries a one-word label that synthesises Integration Index and car-free liveability into a quick orientation. The label is a reading aid, not part of the index itself.

• Strong PTIntegration Index at 40 or above, with public transport clearly outcompeting driving. Networks where transit is both broad and faster than the car for most everyday trips.
• BalancedIntegration Index at 25 or above, with transit and driving roughly even. A workable public transport system where neither mode dominates.
• Car leaningIntegration Index at 25 or above, but driving wins on most trips. Public transport exists and works, but most journeys are quicker by car.
• Car dependentIntegration Index under 25, and driving wins. Living here without a car is structurally hard.
• DevelopingSmall or thin networks that don't fit the categories above. Often early-stage systems where the data won't yet support a confident assessment.

• Reality, only the schedule. We measure what is planned. Delays, cancellations, and overcrowding are invisible.
• One reference day, two windows. 2026-05-12, AM peak and midday. Weekend, evening, and seasonal variation aren't in the score (some of it is reflected in the qualitative evening / weekend dimension).
• GTFS coverage varies. Some cities publish detailed and accurate feeds. Some don't. We validate every feed but the output is only as good as the input.
• POI coverage varies too. OpenStreetMap mapping is uneven across countries, especially for employment. The grid-stratified sampling helps with geographic clustering but can't fix gaps in the underlying data.
• Qualitative scores are evidence-based, not perfect. They're grounded in current public sources and calibrated against expert anchors, but they remain judgement calls against a rubric. The confidence label per city lets readers weigh scores accordingly.
• Car routing ignores congestion. We use free-flow speeds on OpenStreetMap. The parking buffer compensates roughly, but a city with 30 km/h average rush-hour speeds and a city with 50 km/h speeds look the same on the car side.