Comparing Two Traffic Seasons: The Shifts That Distort Year-on-Year Variation

A bustling airport terminal at dusk, with passengers and staff amid planes being prepared for departure, highlighting the challenges of comparing different traffic seasons.

A traffic figure is rarely wrong on its own. What goes wrong is the sentence built around it: movements are up four per cent year on year. Two windows of identical length are almost never identical in content. The scheduling season did not begin on the same date, the month gained a weekend, Easter changed month, and the de-icing rig ran on eleven mornings instead of two. Before a variation means anything, the two periods have to be made comparable. Here is the catalogue of shifts that stop them from being so, and what each one does to the number.

A variation compares two definitions before it compares two realities

Every traffic number is the output of a counting rule applied to a window. Change the rule or change the window and the number moves without a single aircraft behaving differently. That is not a rhetorical caution: in practice, most of the surprising year-on-year variations that land on a management desk turn out to be composition effects, not demand effects.

The discipline is therefore reversed from the intuitive order. You do not start by explaining the variation. You start by proving the two periods were comparable, and only what survives that proof gets an explanation. Everything below is a way in which that proof fails.

Three families of shift do most of the damage: the scheduling calendar, which does not repeat on the same dates; the civil calendar, whose weekday and holiday composition changes every year; and the weather, which changes delivered capacity rather than demand. A fourth family — inconsistent definitions between two bulletins — is less common but far more embarrassing, because it survives every sanity check until someone external notices.

The scheduling season does not begin on the same date twice

Airline schedules are not organised by calendar year but by scheduling season, and the season boundaries are anchored to weekdays, not to dates. Under the worldwide slot guidelines, the Northern Summer season begins on the last Sunday of March and ends on the last Saturday of October; the Winter season fills the remainder. Because both ends are anchored to a weekday, a season always contains a whole number of weeks — but not the same number of weeks every year. Depending on where the last Sunday of March falls, a summer season can be one week longer or shorter than its predecessor, and the winter season absorbs the difference in the opposite direction.

A whole extra week of summer programme is not a rounding error. It is seven days of peak-season frequency added to one side of the comparison and removed from the other. Any season-to-season figure quoted without stating the number of weeks in each season is describing two different objects.

The same boundary problem reappears inside months. A March that contains the schedule change date behaves as a hybrid month, part winter programme and part summer programme, and the proportion of each depends on the year. Comparing March to March, or October to October, is comparing two different mixes of two different programmes.

Daylight saving adds a second, less visible layer. In the European Union the change occurs on the last Sunday of March and the last Sunday of October; in the United States and Canada it occurs on the second Sunday of March and the first Sunday of November. For several weeks each spring and autumn, the offset between the two continents is not what it is the rest of the year. Departure times expressed in local time therefore sit differently relative to night curfews, to slot windows and to connecting banks. A fixed local-time window compared across those weeks is not comparing the same operational conditions.

When comparing traffic numbers, the dates of the scheduling seasons matter because they can add or remove whole weeks of peak travel. If one year has an extra week of summer flights, the numbers won’t be directly comparable without adjusting for that difference.

In 2023, the Northern Summer season ran from March 26 to October 28, covering 31 weeks. In 2024, it ran from March 31 to October 26, covering 30 weeks. A 4% year-on-year increase in traffic could partly reflect that extra week of peak flights in 2023, not actual demand growth.

Calendar composition: weekdays, leap days and moving feasts

A thirty-day month contains four or five occurrences of any given weekday depending on which day it starts. Leisure traffic concentrates on Friday, Saturday and Sunday; business traffic on Monday and Thursday. A month that gains a full weekend relative to the previous year gains traffic that has nothing to do with the market, and a month that loses one loses it the same way.

A hand circles the date 8 on a calendar with a red pen.
Highlighting key dates reveals underlying traffic pattern distortions.

February brings the leap day, which adds a day to one side of every February and first-quarter comparison every fourth year. It is the easiest correction to make and the one most often forgotten in quarterly figures, where it hides inside a ninety-day aggregate.

Moving religious and cultural holidays are among the largest distortions in monthly series. Easter can fall anywhere between 22 March and 25 April: when it falls in March, March gains the outbound peak and April loses it, and the following year the transfer runs the other way. Ramadan and the Eid holidays advance by roughly eleven days each Gregorian year, moving a major traffic pattern across the Gulf, North Africa and South-East Asia through the entire calendar over three decades. The Lunar New Year moves between the second half of January and the third week of February, carrying with it what is routinely described as the world’s largest annual travel peak.

School holidays are staggered by design. Several European states split their national territory into zones with deliberately offset dates precisely so that the peak spreads; that means the position and the width of the peak both move from one year to the next. A public holiday that falls on a Saturday in one year and a Thursday in the next produces or removes a bridge weekend, which is a demand event created by nothing but the calendar.

Calendar and season shifts: what each one does, and how to neutralise it
Shift Effect on the figure Neutralisation
Season length differs by one week Whole week of peak programme on one side only Compare week counts first; normalise per operating week
Schedule change date inside the month Hybrid month, mix of two programmes Compare season to season, or split the month at the change date
Weekday composition (four or five Fridays) Leisure peaks over- or under-represented Align by ISO week, or compare same-weekday counts
Leap day One extra operating day in February and Q1 Compare daily averages, not totals
Easter changing month Traffic transferred between March and April Report March and April combined; state the transfer
Ramadan and Eid drift Peak moves across the calendar every year Align on the religious calendar, not the civil one
Lunar New Year Peak alternates between January and February Combine the two months for Asian markets
Daylight saving offset weeks Local departure times shift against curfews and banks Work in UTC for any time-of-day analysis

Schedule, slots and the base you are actually comparing against

At a coordinated airport, several different objects can legitimately be called « the schedule », and they do not move together. There is the slot portfolio allocated for the season; there is the schedule filed by the airlines; there is the schedule as it stood at some later snapshot, after cancellations and additions; and there is what actually flew. A comparison built on one of these against a comparison built on another is not a comparison at all.

Slot data carries its own history. Historic precedence at a coordinated airport is retained by operating the series to a baseline usage threshold — long set at eighty per cent of the allocated series in the European regime and reflected in the worldwide guidelines. During the pandemic that obligation was waived and then reintroduced progressively, which means any season compared against those years is compared against a schedule that was never required to operate in full. The comparison is not invalid, but it is not measuring the same thing, and it needs to be said in the same sentence as the percentage.

Commercial schedule databases are snapshots. A schedule extracted six months before the season contains flights that will later be cancelled; the same season extracted six weeks out contains far fewer of them. If the two years in your comparison were extracted at different lead times, the later extraction will always look weaker, and the difference is an artefact of the extraction date. Fix the lead time, record it, and publish it next to the figure.

Weather is not noise, it is a variation in delivered capacity

A published declared capacity is a fixed number. Delivered capacity is not. Low-visibility procedures increase separation on final and cut the arrival rate for as long as they last. A runway configuration forced by wind can remove a departure queue’s most efficient geometry. Winter operations impose de-icing throughput as a hard constraint on the departure rate, and the number of mornings on which that constraint binds varies enormously between two consecutive winters. Convective season re-routing does the same thing en route.

The consequence is that a mild winter and a severe winter are two different capacity years, not two different demand years. When a February shows a decline, the first question is not commercial. It is how many days of low-visibility procedures, how many de-icing mornings and how many regulation-hours the two Februaries contained.

Weather also hits the components asymmetrically. A cancellation removes a movement and its passengers from the totals. A delay removes nothing: late in the day it pushes traffic across the statistical day boundary, deflating one day and inflating the next, and it degrades punctuality without touching volume. A year-on-year punctuality comparison and a year-on-year volume comparison can therefore point in opposite directions for the same meteorological reason.

The trap. Two schedule extractions taken at different lead times will always show the later one weaker, because it already contains cancellations the earlier one could not know about. This artefact is indistinguishable from a genuine market decline unless you record the extraction date with every series. Recording it after the fact is not possible.

Weather affects how many flights can actually take off and land, not just how many people want to fly. A snowy winter might show fewer flights not because fewer people traveled, but because planes needed more time to be de-iced before takeoff.

Suppose that at a large airport, February 2023 had 12 days with low-visibility procedures, reducing arrival rates by 10%, and February 2024 only 3. If February 2024 still shows a 5% traffic decline, it cannot be blamed on the weather: 2024 actually had more usable capacity. The decline points to a real change in demand or in the schedule.

The same word, three different meanings between two bulletins

The vocabulary of traffic statistics is not standardised across airports, operators and States, and the differences are large enough to invert a sign. Before comparing two sources, or one source across a definition change, the counting rules must be read rather than assumed.

Counting conventions that differ between publishers
Term Where the definitions diverge Consequence if mixed
Aircraft movement One arrival and one departure count as two; some series exclude general aviation, State flights, training circuits or positioning sectors Two series of « movements » differing by a large fixed share
Passenger Transfer passengers counted once or twice; direct transit counted, counted once, or excluded Hub totals not comparable with point-to-point totals
Flight Scheduled, operated, or flown; cancelled flights included or removed; diversions attributed to origin or destination Cancellation-heavy periods look normal or catastrophic
Seats and capacity Filed seats, offered seats, or seat-kilometres; sub-charters and wet leases attributed to marketing or operating carrier Load factor computed on the wrong denominator
Statistical day Local midnight or 0000 UTC; some series use an operating day that runs past midnight Late-evening traffic shifts between days and months
Punctuality Off-block or on-block reference; threshold applied per flight or per delayed flight; cancellations counted as infinite delay or excluded Two « on-time performance » figures that cannot be reconciled

Base effects: the year you are comparing to had its own accident

Year-on-year variation measures the distance between two points, and one of those points is the past. A strike, an airspace closure, an airline failure, a runway rehabilitation, a terminal works phase or a one-off event that filled every hotel bed in the region all leave a mark on the base year. If the anomaly was in the earlier year, this year looks strong for no reason. If it was in the later year, the following year will look strong for no reason.

The remedy is not to remove the anomaly — that invites endless argument about what counts as abnormal. It is to publish two comparisons: against the previous year, and against a stated stable reference year. When the two disagree, the disagreement is itself the finding, and it points directly at which year contained the event.

Building a comparison that survives review

The choice of window is a methodological decision with consequences, not an administrative detail. Each option preserves something and breaks something else, and the right answer depends on what the figure is for.

Comparison windows: what each preserves and what it breaks
Window Preserves Breaks Best used for
Same calendar dates Public communication, regulatory reporting Weekday composition; moving holidays Statutory returns and long series
Aligned ISO weeks Weekday composition, weekend count Fixed-date holidays; some years contain a fifty-third week Operational and commercial analysis
Season to season The programme actually in force Unequal season lengths unless normalised Network and slot planning
Rolling twelve months All seasonality and moving holidays Responsiveness; a shock stays in the series for a year Trend detection and board reporting
Same-weekday matched days Day-of-week profile precisely Comparability with any published series Capacity and staffing studies

The right reflex. Before writing the percentage, write the two windows side by side and count what is inside each: number of weeks, number of Fridays and Saturdays, position of Easter and of school holidays, number of de-icing mornings, number of days under low-visibility procedures, and any known one-off event. If any line differs, the variation is not yet a finding — it is a list of corrections still to be made.

What to publish next to the number

A traffic figure that travels without its method will eventually be quoted against you. The minimum accompanying set is short and takes one line each: the exact definition used, the window and how it was chosen, the data source and its extraction date, the exclusions, the known one-off events in either year, and any adjustment applied with its direction and size.

This is not bureaucratic caution. It is what makes a variation usable by someone else — the capacity planner sizing next season’s stands, the ground handler building a roster, the regulator reading a performance plan. Practitioners comparing methodologies across airports and network operations will find the discussion continues on the FL330 professional forums, where the definitional differences between published bulletins come up more often than the traffic itself.

The final discipline is negative: when two periods cannot be made comparable, say so and stop. A stated impossibility is a result. A percentage computed anyway is a liability that will circulate long after the caveat has been forgotten.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

FL330, c’est toute l’aviation professionnelle — explorez nos autres espaces.