From Isolation to Inclusion

Introducing the Multidimensional Transport Poverty Index tool

Rosa Félix
TML - Transportes Metropolitanos de Lisboa

CERIS, Instituto Superior Técnico
University of Lisbon

Mauricio Orozco-Fontalvo
Gonçalo F. Matos
Miguel Alvelos
Camila Garcia
Filipe Moura

November 2nd 2026 — Equity in Transport Workshop, Barcelona

Why Transport Poverty?

Transport Poverty


Transportation is essential — its absence or inadequacy creates:

  • 🚫 Limited access to work, education, health, leisure, services and opportunities
  • 🏘️ Social exclusion & territorial inequality
  • 💸 Disproportionate financial burden on low-income households
  • ☠️ Unequal exposure to road risk

Transport Poverty (Lucas et al. 2016) has direct consequences in social exclusion and territorial cohesion (Pritchard et al. 2014; Allen and Farber 2019; Mejia-Dorantes and Murauskaite-Bull 2022)


EU Regulation 2023/955 (Social Climate Fund) provides the first legally binding definition of Transport Poverty at European scale (European Parliament and the Council of the European Union 2023)

A Multidimensional Problem

Based on Lucas et al. (2016), Mejia-Dorantes and Murauskaite-Bull (2022) and the European Commission (2024):


🗺️ Accessibility
Ability to reach work, health, education, and other essential services within reasonable time

🚌 Mobility
Quality of transport supply: frequency, coverage, travel times, active travel infrastructure

💶 Affordability
Households’ capacity to afford transport (+ housing) relative to income

🦺 Exposure to Externalities
Exposure to road externalities: safety, crash severity, air pollution, noise

What is Missing?

  • ❌ No composite index integrating all 4 dimensions, across multiple transport modes
  • ❌ Existing indexes rely on costly primary data (surveys)
  • ❌ No operational tool translating research into policy support


👉 We wanted to create an index
Multi-dimentional · Multi-modal · Scalable · Replicable

Modes considered per dimension

Methodology

Case Study: Lisbon Metropolitan Area

  • Polycentric structure (strong employment / opportunities core in Lisbon)
  • Strong urban/peripheral PT supply contrast
  • High car modal share in periphery

Source: Wikipedia
  • 18 municipalities · 124 parishes · ~2.8M inhabitants, 3,000 km²
  • Analysis unit: freguesia (parish) — lowest admin level with full data coverage
  • Also computed at municipality and H3 grid level, resolution 8 (~3,686 hexagons)

res 8 grid example

Open Data

Source Used for
INE Census 2021 / IMOB 2017 Income, housing, commuting OD pairs, modal share
OpenStreetMap Road network, cycling/pedestrian infra, POIs
GTFS (9 operators) PT frequency, travel times, coverage
ANSR (road safety) Accident locations and severity (2019-2023)
Carris Metropolitana Health centres, schools
PMMUS / TML Shared mobility docks
GlobalBuildingAtlas Buildings height (population density)


All computations done in R using odjitter, r5r, accessibility, sf, tidytransit, FactoMineR (Carlino and Lovelace 2022; Pereira et al. 2021, 2026)

Computing Accessibility

  • Travel time matrix for all modes × all H3 res 8 grid cells
  • 📍 Points of interest: healthcare, schools, groceries, green spaces, recreation, jobs

Walking time to the closest groceries

Cumulative accessibility by walking to health centres

Cumulative accessibility by bicycle to health centres

Cumulative accessibility by public transport to health centres

Cumulative accessibility by car to health centres

Normalization & Aggregation

All indicators (>1700) normalized to [0, 100] using min-max scaling, depending if is a cost or a benefit:

\[X_{norm} = \frac{X - X_{min}}{X_{max} - X_{min}} \times 100\ \text{ or }\ X_{norm} = \frac{X_{max} - X}{X_{max} - X_{min}} \times 100\]

Intra-dimension: PCA (PC1 score) aggregates multiple indicators (OECD et al. 2008)


Inter-dimensional aggregation

Variant Method Best for
🔵 Contrasting Entropy-weighted mean Territorial contrast
🔴 Critical Geometric mean Worst-dimension penalization
🟡 Balanced Arithmetic mean Compensatory / communication
⚙️ Custom Analytic Hierarchy Process (AHP) weighting Decision-maker preferences

The IMPT Dashboard

Dashboard: 🎮 Live Demo

Available at:
🔗 ushift.pt/apps/impt


Features:

  • 🗺️ Interactive maps at parish, municipality & grid scales
  • 📊 All dimensions × transport modes × indicators
  • ⚙️ Political IMPT: user-defined AHP weights
  • 📥 Downloadable results (>1700 indicators)

Results

🗺️ Accessibility



  • High accessibility in the center
  • Deficiencies concentrated in rural/low-density areas

🚌 Mobility


  • Low scores in most AML except the center
  • Periphery is severely underserved
  • Highlights the centripetal structure of the LMA

Mobility justice?

💶 Affordability


  • High scores overall
  • The Navegante monthly pass greatly mitigates costs (inexpensive)
  • Worst cases in low-income, car-dependent areas
  • Paradox: Housing more expensive where Transportation is more affordable and vice-versa

🦺 Safety



  • Urban core scores poorly (high volume of crashes): inverse spatial pattern
  • Worst rates in coastal EN6 corridor (Oeiras/Cascais), and Lisbon centre, with very high incomes

IMPT Index: Three Aggregation Methods

Contrasting IMPT

Critical IMPT

Custom Weights IMPT

All modes · Parish level · Higher score = higher transport poverty


The Critical IMPT (geometric mean) is best for identifying specific dimensional deficiencies

🧩 A single score, or many indicators?

“multiple indicators are better than any single metric or composite” (Lowans et al. 2021)


Risk of one single number
A composite can hide which dimension is failing.

Value of one single number
It ranks places, targets funds and communicates quickly.


What we did: publish the composite index together with every dimension and indicator, so users can further investigate.

Validation

Conclusions

The IMPT allows

  • To diagnose the territory and quantify transport poverty
  • To support decision-makers in prioritizing interventions and investments
  • To compute transport poverty disregarding the indicators available
  • To compute transport poverty at different scales (region, city, parish, grid)

Contributions

  • First multi-dimensional, multi-modal composite index of Transport Poverty for Portugal
  • 4 dimensions (Accessibility, Mobility, Affordability, Safety) × 4 transport modes × 3 aggregation methods
  • Entirely based on open, replicable data
  • Policy-ready: AHP weighting + interactive dashboard
  • Validated with stakeholders
  • Replicable to other metropolitan areas

Limitations

  • Affordability: too few indicators for PCA
  • Digital literacy dimension dropped (data gaps)
  • Environmental externalities not included (air pollution)

Future Work

  • 🌍 Transferability: application to other metropolitan areas
  • 🧠 Include perceived transport poverty as a dimension
  • 💨 Include air pollution and noise exposure as environmental externalities
  • 📡 Longitudinal monitoring: IMPT updates as data becomes available (eg.APIs) to assess interventions’ impacts

Your turn!

  • Explore the IMPT dashboard
  • Start the AHP survey
  • Apply your own weights to the indicators!

Where do people actually live?

Census block data are coarse

Resident population redistributed using buildings

Considering the average over a parish can mislead when people live in a small part of it. Distributing the population across buildings (weighted by its volumes) gives better proxies.

What is PCA?

From Mauricio Orozco lectures’ slides

A dimension is described by several indicators, and many of them move together (e.g. jobs reachable in 30 and in 45 minutes). Adding them up counts the same information twice.

Principal Component Analysis (PCA) finds the direction along which places differ most. The first component (PC1) is that direction. Each place gets one score: its position along it.

PCA is the approach recommended by the OECD/JRC Handbook on composite indicators (OECD et al. 2008).

PC1 should capture a large share of the variation, and every indicator should push it in the same direction.

Rotate the axis
Variation kept by this axis:
Each dot is a district, and the values are accessibility and mobility scores (standardised). The dots stay fixed: the axis rotates, and each district gets a score where it lands on the axis.