Methodology

Printed from https://destinito.com/research/caribbean-air-connectivity/methodology. The online page is the current version and carries any corrections.

What is counted, what is excluded, and where the judgement calls are. Every choice below changes a published number, which is why it is on this page rather than in the code.

Data period week beginning 19 October 2026
Method version v0.2
Islands 28
Airports 32
Licence CC BY-NC 4.0

What is measured

Scheduled departures from 32 airports serving 28 islands, for the reference weeks above. Each departure contributes to its origin island's departure count, to the distinct set of destinations that island reaches, and to the set of airlines serving it.

Destinations are counted per island, not per airport. An island that flies to both airports of St Lucia (UVF and SLU) has one destination there, not two; an airport outside the index counts as a destination in its own right. Departures with no identified carrier are not counted as an airline of their own.

Changes in v0.2

Method v0.2 differs from the unpublished v0.1 in six ways. Rankings from v0.1 are not comparable with v0.2 and were never published.

Destinations are counted per island, not per airport. An island that flies to both airports of St Lucia has one destination there.
North America is split into the United States and Canada in the market mix, by the destination airport's country.
Gateway metrics. For each gateway island the index reports the arrivals it receives from outside the Caribbean, how many are long-haul (at least 4,000 km), and how much of that can continue to another indexed island the same local day.
Eurostat directions. The two directions of an airport pair are different flights and are both kept; a partner airport in a French territory is classified by its own country, not by France.
Carrier gate. The name gate also fails on any departure whose carrier is not identified.
Island roles are defined on seats and islands served, not on departure counts. Evaluated in this order, first match wins: a shuttle serves exactly one destination island; a gateway has at least 9,500 weekly seats (estimated) on departures outside the Caribbean; a regional connector serves at least seven Caribbean islands, no single Caribbean partner takes half of its Caribbean departures, and it has at least 1,500 weekly seats (estimated) on departures to Caribbean destinations; everything else is a spoke. A nine-seat hop counts the same as a jet in a departure count, which is why departures were set aside. The thresholds were calibrated once on the 2027 data, and each edition's facts pack records the values used. The third role was called “hub” in an earlier draft and is renamed: the test measures connections within the Caribbean, not airline hub operations, and the index’s hub links metric (links to major airline hubs) is a different thing. The 1,500-seat floor was added so that an island touching seven islands on a handful of small-aircraft rotations is not labelled a connector.

Three exclusions

Flights between two airports on the same island

St Martin's SXM and SFG, St Lucia's UVF and SLU, the Dominican Republic's PUJ and SDQ, Jamaica's MBJ and KIN. Counting these would let an island gain a direct destination by flying to itself, and would rank a twenty-minute domestic hop above a genuine extra connection.

Departures whose destination cannot be named

The provider sometimes supplies a place name with no airport code. Where a name cannot be resolved to an airport, the departure is counted but excluded from route totals — it feeds the name gate instead. Dropping it silently would make an island look cleanly measured while part of its traffic was invisible.

Arrivals

Collected only at the gateway airports, and only to measure what they receive from outside the Caribbean and which onward connections it has. Mixing arrivals into departure counts would double the gateways and nobody else.

Multi-airport cities

Some carriers name a city rather than an airport — "Paris", "London", "Washington". These resolve to the IATA city code (PAR, LON, WAS) and count as one destination. Resolving "Paris" to Charles de Gaulle would be an invention: the schedule does not say which field. A city code is never used as the origin of a connection-time claim, for the same reason.

How seats are estimated

Seats are an estimate and are labelled as one. Each departure resolves through five tiers, in order, and the index records which tier produced every figure:

Aircraft model. The scheduled aircraft type matched to a seat count. An airline-specific rule beats a generic one — one operator's ATR 42 is not another's.
Route override. A hand-entered figure for one airline on one route, from that carrier's own timetable.
Airport constraint. Where the field physically cannot take anything larger. Saba's runway is 400 m and St Barthélemy's 646 m, so a flight touching either is Twin Otter class as a matter of physics. This outranks a fleet average: a physical limit is a stronger claim than a typical aircraft.
Airline default. A per-carrier figure, used only where an operator has confirmed it against the carrier's published fleet. An unconfirmed figure is never used.
Not estimated. No seats, and one tick against the seat gate.

There is no route average. An earlier version of this method gave unmatched flights the average size of the matched ones on their route. That is removed. The provider supplies an aircraft type on roughly two thirds of departures, so an average would have been applied to a third of every route — turning missing data into confident numbers. A cited dataset can survive a visible gap; it cannot survive an invisible invention.

Where a seat rule and an airport limit disagree, the estimate is kept and flagged, not quietly clamped. The conflict means one of the two inputs is wrong, and rounding it away would destroy the evidence of which.

The two gates

An island's figures publish only if it passes both:

Name gate. At most 1% of its departures may have an unresolvable destination, and every departure must have an identified carrier. This gates the COUNTS — departures, destinations, airlines.
Seat gate. At least 95% of its departures must carry an estimate from tiers 1–4. This gates the SEAT figures only.

The seat gate is all-or-nothing across the island set: if any island fails it, the seat components are excluded from every island's score. Mixing an island scored with seats against one scored without would make the ranking incomparable, which is worse than a ranking that leaves seats out and says so.

Editorial choices that move the headline

Puerto Rico, the US Virgin Islands, Guyana, Suriname and Belize count as Caribbean. Puerto Rico and the USVI on geography; Guyana, Suriname and Belize on CARICOM membership. This moves the "Caribbean versus international reach" split, and reasonable people draw the line elsewhere.

Nevis links to the St Kitts and Nevis country page, while keeping its own research page and its own measurement. The index treats the two islands separately because their air service is separate.

The reference week is a measurement, not a typical week. Schedules change; the week is stated on every page so a figure can be checked against the same week rather than against an average nobody can reconstruct.

The low week is a shoulder week, not the bottom of the season. The week beginning 19 October 2026 falls after the first winter seasonal services have begun but before the main winter schedule. Some carriers start or thicken Caribbean routes during October, so this week already contains winter service that a true trough week would not — and it is still missing the services that begin in November and December.

At Sint Maarten in this week, for instance, Delta and American both serve the island from their hubs but neither yet flies the New York–JFK route each adds later in the winter, and Air Transat serves Montréal but not Toronto. The low-season column should therefore be read as early shoulder season: higher than the quietest weeks of the year, lower than the winter peak, and not interchangeable with either. The peak week is collected separately and stated separately for the same reason.

Sources

Schedules — AeroDataBox, collected per airport per week.
Load factors — US BTS T-100 and Eurostat, monthly, by route where published. Reported separately from the seat estimates and never mixed into them.
Airport reference — OurAirports (public domain), used to resolve codes and countries. Never used to name a route on its own.
Regional timetables — carrier-published schedules, compiled by hand, used to check the feed rather than to replace it.

Known gaps in this edition

Stated because they are known, not because they are comfortable:

A small number of departures still carry a destination name the index cannot resolve. They are counted in the name gate and excluded from route totals.
Our own airport flight boards show more departures than the provider's schedule on at least one airport, concentrated in one carrier. Where that is confirmed over time it will raise the affected frequencies; until then the published figure is the provider's and this note is the caveat.
One airline default is a conservative floor rather than an exact figure, and is marked as such in the data notes. It understates that carrier's seats.

CC BY-NC 4.0. Back to Caribbean Air Connectivity Index 2027.