Build an origin-destination matrix and identify unobserved demand
7 min read
Lesson objectives
- Construct an auditable directional journey matrix and distinguish observed demand from demand constrained by current service.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
An origin-destination matrix turns journey records into a picture of where people start and where they intend to go, but it does not by itself explain their mode choices or the absence of travel by others. A row represents an origin zone and a column a destination zone; a cell holds complete person journeys within a declared period. Direction belongs in the key. A journey from neighbourhood A to employment centre B is distinct from B to A because peak times, constraints and purposes can differ. Combining the two cells and designing symmetric service may hide that distinction. Select zones that balance decision detail, privacy and sample size. A very large zone hides access differences within it; a very small zone gives unstable cells and may expose individuals. Record the geographic boundary, period, sampling frame, weights, purpose and transfer-counting rule. A phone feed or ticket file does not automatically observe all residents or journeys. The matrix identifies demand patterns worth testing; it is not a rule that the largest cell must receive funding.
Start with a clean journey table containing a journey identifier, origin, destination, departure time, mode, sampling weight and completion status. Keep one record per complete journey even if it contains three boarding legs. If the announced denominator is journeys wholly inside the study area, exclude a row beginning outside it; if the aim is journeys originating in the area, state that different inclusion rule. Preserve an unknown origin rather than silently assigning it to the centre. Aggregate weights by origin-destination pair, period and mode, while retaining the unweighted observation count in each cell. A large weighted estimate built from a few raw observations signals uncertainty. An origin row total is journeys leaving that zone, a destination column total is journeys entering it, and the grand total must equal the weighted eligible-journey total. Reconcile transit cells with boardings, expecting extra boarding legs where transfers occur rather than forcing totals to match.
Separate three concepts before using the matrix to design an intervention. Observed demand consists of journeys that happened and were captured by the source within its boundary. Potential demand consists of journeys that might become feasible if service, cost or land use changed. Induced demand consists of additional journeys that may arise after capacity or speed improves. Unmade potential journeys cannot be obtained by a simple subtraction from fare transactions: a person who did not travel is absent from that file. Use travel interviews, job and service locations, walk and wait times, and surveys that include nonusers to build a testable hypothesis. Ask whether a long commute reflects free preference or lack of closer opportunities, and whether few trips by older residents indicate nearby services or a barrier to travel. Store each hypothesis, source and limit separately. Do not rename a latent-demand estimate as observed fact, and do not expand a handful of responses into a city percentage without a representative design.
Read the matrix with peak direction, service resource and alternatives in view. A high-volume pair may already have frequent service, while a smaller pair's missing link to a hospital can be important despite its count. Disaggregate by purpose and time; then compare door-to-door public-transport, car and walking times, rather than in-vehicle speed alone. Calculate transit share inside each cell using known-mode journeys in that cell. A citywide denominator does not belong under a cell numerator. Inspect the reverse direction separately. A difference may reflect work or education schedules, or a sampling defect; it is not automatic permission to erase half the network. Require a minimum raw observation count before declaring a difference for a district or group, and display uncertainty for sparse cells. If zone definitions change in a later year, rebuild both periods to a common set of zones before claiming a trend.
Turn cells into intervention hypotheses, not preapproved projects. If A-to-B journeys cluster in the morning and B-to-A journeys return in the evening, test a direct connection or coordinated transfer. Calculate vehicle hours and usable capacity in both directions, not just the morning journey count. If a short origin-destination pair has a high car share, inspect continuous walk routes, crossings, heat exposure and available services before promising mode shift. If a service zone shows few observed journeys, ask about opportunities people cannot reach rather than dismissing it as low priority. Predeclare a measurement for each proposal: complete journeys, access time, traveller cost and outcomes for affected groups. Note alternative explanations such as a new school or a relocated employer. A matrix then remains a diagnostic tool connected to later evaluation instead of letting one large cell dictate the budget.
Lesson summary
A directional matrix separates observed journeys from hypotheses about unmade travel and checks denominators, weights and transfers before service design.