Door-to-door time against in-vehicle time across European corridors

Research question

Every public timetable, and almost every journey planner built on one, reports the time a service spends running. That is not the quantity a traveller spends. The traveller also reaches the station, waits for whatever connection the timetable happens to offer, and leaves the station at the far end.

The gap has been named and measured before. Krygsman, Dijst and Arentze define an interconnectivity ratio, the ratio of access, egress and transfer time to in-vehicle time, and find it large enough to dominate short multimodal trips. Valuation research finds the same asymmetry from the demand side, with waiting and transfer time carried at a multiple of in-vehicle time in both revealed and stated preference.

Two questions follow that are not settled, and that matter to anyone building a ranking.

1. Across a corridor network rather than a single city, how much of door-to-door time is actually spent in a vehicle?
2. Is the difference large enough, and unevenly enough distributed across modes, to change which mode a traveller would be shown as fastest? On which corridors?

The hypothesis was that reversals would concentrate on short corridors, where a fixed access penalty is large relative to the journey, and that they would depend heavily on the access and egress assumption, which is the parameter most studies argue over. The first expectation is supported. The second is not, and that turns out to be the more useful finding.

Related work

Door-to-door measurement is well established in urban accessibility research, where multimodal travel time is computed from schedules and street networks and compared across modes. That literature is mostly intra-urban and mostly concerned with accessibility surfaces rather than with which option a planner ranks first.

The intercity literature asks the ranking question directly, but almost exclusively for the rail against air comparison, where the airport access penalty is large and well studied. The recurring result is a break-even distance beyond which air wins, and the recurring sensitivity is to the airport access assumption.

What is comparatively unexamined is the ground-mode case: rail against coach against ferry, on corridors where the access penalties are similar and so ought to cancel. If access and egress were the only source of divergence, those corridors would show no reversals at all. They do, and the reason is the other term. The full reference list, with a DOI on every entry, is in the PDF.

Data

The analysis reads a timetable snapshot released with this paper. It holds 545 corridors of filed rail, coach, ferry and air itineraries collected from operator feeds, each itinerary carrying real departure and arrival stamps per leg, the calling points, the operator and the number of changes. It was collected on 2026-09-16 for service on 2026-09-28 and 2026-09-30.

Restricting to corridors whose origin and destination are both European leaves 493 corridors, 936 itineraries and 2765 legs, over 166 cities in 25 countries. City coordinates are WGS84. Corridor distance is the great-circle distance between city centroids.

The snapshot also carries a per-leg detour factor. It takes only the four values 1.00, 1.05, 1.10, 1.15 and is assigned by mode rather than by corridor, so it carries no corridor-specific information and route distance is a constant multiple of great-circle distance within each mode. Great-circle distance is used directly.

Two integrity checks run inside the analysis and are asserted by the test suite: no leg has a negative duration, and the elapsed time computed from the first departure and the last arrival equals the snapshot's own stated journey time on every itinerary in scope. Both pass at zero exceptions (0 and 0 respectively).

Method

For each itinerary three quantities are computed.

In-vehicle time

The sum over legs of arrival minus departure. This is the quantity a timetable reports.

Interchange time

Elapsed time, last arrival minus first departure, minus in-vehicle time. This is filed data rather than an assumption: it is the gap the operators’ own published times leave between one service and the next.

Access and egress

An allowance applied once at each end, by the mode of the first and last leg. The central case uses the allowances the Orlero ranking engine applies in production: 22 minutes at a station for rail and coach, 45 at a port, 95 at an airport. These are allowances, not measurements, and the robustness section shows how little the result depends on them.

Door-to-door time is the sum of the three.

Each itinerary is assigned a headline mode, the mode accounting for the largest share of its in-vehicle minutes. A Berlin to London journey that opens with eleven minutes of S-Bahn is a rail journey; classifying it by its first leg would record a fact about leg order rather than about travel.

A corridor is comparable when it carries filed itineraries under two or more headline modes. On a comparable corridor each mode is represented by its own fastest itinerary, which is what a search result shows, and the minimising itinerary is allowed to differ between the two clocks. On several corridors the fastest train in the vehicle is not the fastest train out of the door, and forcing one itinerary to stand for both would suppress part of the effect being measured.

A reversal is a comparable corridor whose first-ranked mode differs between the in-vehicle ordering and the door-to-door ordering.

How much of a journey is spent in a vehicle

In-vehicle time accounts for a median of 77% of door-to-door time across the 936 itineraries, mean 70.4%, interquartile range 57 to 87.3%. The median itinerary carries 107 minutes that a timetable does not report: 47 minutes of filed interchange and 67 minutes of access and egress. The median interconnectivity ratio is 0.299.

The distribution is wide and it differs by mode. Ferry itineraries spend the least of their time in a vehicle, because a port allowance is charged at both ends of what is often a short crossing. Coach itineraries spend the most, because the journeys are long enough for a fixed allowance to disappear into them, even though coach carries by far the longest filed interchanges.

The air row rests on 7 itineraries, and no claim in this paper depends on it.

Grouping by the number of changes shows the two components trading places. Direct itineraries carry no interchange by construction, yet have a lower in-vehicle share than itineraries with two changes, because many of them are short crossings where the fixed allowance dominates. Filed interchange time then rises steeply with the number of changes. 29.5% of the itineraries in the sample are direct.

Whether the ranking changes

Of the 493 corridors, 79 carry filed itineraries under two or more headline modes. On 13 of those, 16.5%, the mode ranked fastest changes when the clock changes. All 13 order changes are changes of the leading mode rather than reshuffles below it, which follows from most comparable corridors carrying exactly two modes.

These are not marginal cases. Several corridors where one mode leads by ten hours in the vehicle lose by seventeen hours out of the door, because the land alternative to a sailing is a long detour with long filed waits. The scatter of every comparable corridor against both clocks is Figure 2 of the PDF.

Ten of the thirteen are under 400 km, which supports the hypothesis that reversals concentrate on short corridors. Eight are Mediterranean island or peninsula corridors, and two are the same city pair measured in both directions.

The same question asked of itineraries rather than modes gives a consistent answer. Of the 443 corridors carrying more than one filed itinerary, the single fastest itinerary changes on 76, 17.2%.

Which term does the work

This is the result that was not expected. Setting the access and egress allowance to zero for every mode leaves filed interchange time as the only difference between the two clocks. Under that setting there are 14 reversals, and all 13 of the reversals found in the central case are among them. The allowances create 0 reversals and cancel 1.

In other words, the effect is carried by a quantity that is published in the timetable and needs no assumption at all. The access parameter, which is where the intercity literature concentrates its uncertainty, is close to irrelevant to the ranking question on these corridors, because the corridors where two ground modes compete charge both of them the same allowance and it cancels.

Robustness

The allowance was swept: the ground allowance from 10 to 45 minutes, and the terminal premium, the extra charged at a port over a station, from 0 to 60 minutes, with the airport premium held at three times the port premium. Seventy-two combinations. The reversal rate never leaves the range 16.5 to 20.3%, and it does not respond to the ground allowance at all, because a change applied equally at both ends of two competing ground options cancels exactly.

A second check addresses the long tail. Aegean and Corsican corridors connect through sailings a day or more apart, and an itinerary that asks a traveller to sleep somewhere is arguably not waiting in the sense the interconnectivity literature means. Dropping the 50 itineraries containing any single interchange longer than twelve hours leaves 76 comparable corridors and 11 reversals, 14.5%. The headline result is not an artefact of overnight layovers.

Limitations

Two adjacent service dates

The snapshot covers 2026-09-28 and 2026-09-30. Filed interchange time is a property of the timetable on the day, and two days in late September are not every day. Frequency effects, seasonal sailings and weekend timetables are all outside what this measures. A repeated snapshot is the obvious next step and the released code runs unchanged on one.

493 corridors, not the European network

These are the corridors for which a feed answered, which skews toward corridors somebody asked about. Not a random sample of European city pairs.

Access and egress are allowances

The allowances are the platform’s operating constants, not measurements. The conclusion survives the whole plausible range; the minute counts in Table 1 move with it.

Detour factors are derived and coarse

The released network assigns one of four detour values by mode. It is not a corridor-level measurement and is not used as one here.

No fares

This paper measures time, not price, so it says which mode is fastest rather than which a traveller would choose.

Air is not measured

Seven itineraries, so the rail against air comparison is not attempted.

Headline-mode classification

On a short strait crossing, an itinerary with a ferry leg can be classified by the land mode carrying more of its minutes. Reggio Calabria to Messina is such a case. The rule is applied identically to both clocks, so it cannot by itself manufacture a reversal.

Conclusion

Across 493 European corridors, in-vehicle time captures a median of 77% of what a traveller spends, and the shortfall is distributed unevenly enough across modes to reverse the fastest-mode ranking on 16.5% of the corridors where two modes compete. The reversals concentrate below 400 km, as the accessibility literature would predict.

The finding worth acting on is the decomposition. The reversals are produced by filed interchange time, which every timetable already publishes, and not by the access and egress allowance, which nobody can measure without a survey. A journey planner that wants to stop misranking modes does not need to resolve the access-time question first. It needs to stop reporting the sum of the leg durations and start reporting the time between the first departure and the last arrival, which it is already holding.

How to reproduce this

Every number on this page and in the PDF is generated. Nothing is transcribed.

1. Clone the repository and take the three data files named above at the version stated below.
2. Run node analysis/analyse.mjs from research/papers/door-to-door-europe. It writes results.json, the table bodies under tables/ and the figure data under figures/.
3. Run make in the same directory. It reruns the analysis, typesets the paper with tectonic and copies the PDF to the path this page serves.
4. Run make verify to confirm that a fresh run reproduces the committed outputs byte for byte.

Reproducing requires Node 20 or later and tectonic, and no network access beyond what tectonic needs to fetch its own TeX packages on first run.