What the analysis found
This page is the short version of the whole site. Each finding below states its result in the heading, carries the same status label used on the charts themselves (what the labels mean), and links to the field note that backs it up. Where a figure is not pinned down, the finding says why. Every figure on this page is read from the published chart data.
Air travel grows roughly in step with GDP in the US, faster in Australia and far faster in Brazil
descriptive Evidence strength varies by row; each row carries its own published status label.
Brazil's long-run income elasticity is 2.84 and Australia's 1.70, while Singapore (0.87) and the United States (0.94) sit close to one, so in those two markets flying grows roughly in step with the economy. Singapore and the United States are a statistical near-tie. The rows are not equally firm: the United States and Brazil are confirmed, Singapore's long-run link is confirmed to exist without its form being pinned down, and Australia is provisional, because its main test passes only with the pandemic quarters in the sample. Every ranked market fails at least one re-check of the long-run link on pre-COVID data alone. Part of Brazil's figure may come from new airlines entering the market: a rough check on the airlines flying before that entry gives a much lower figure. What counts as a passenger also differs by market, from airport movements in Singapore to boardings per flight leg in the United States, so rows that sit close together are not ranked against each other.
China is shown beside them but not ranked: it is tested on a shorter pre-COVID window and the result is inconclusive. Canada is left off because its passenger and passenger-kilometre series give elasticities far apart, so no single figure can be ranked. Japan, Mexico, South Korea, India and the United Kingdom are excluded too, on grounds that include a nearly flat GDP series, structural breaks in short samples and a gap in the UK's source data.
In the United States, air travel tracks income closely
confirmed The estimate moves with specification choices the data cannot reject.
A one percent rise in US GDP goes with roughly 0.94 percent more passengers over the long run. The model has income in it and no ticket prices, and real fares fell over the same years, so the figure also counts the effect of cheaper fares where they came with rising incomes. The bootstrap interval runs from 0.884 to 0.995, but it measures sampling noise under one specification. Choices the data cannot reject, such as whether consumer prices enter the long-run relationship, whether a trend is allowed and how the pandemic quarters are handled, move the estimate further: across the alternative specifications recorded with the headline it runs from about 0.90 to 1.31. Flying grows roughly in proportion to income without the exact size being settled.
Distance still deters flying on US routes
confirmed
The structural gravity estimate of distance decay is -0.49, with a standard error of 0.047, clustered by airport at both ends. It rests on 130,167 US domestic route observations, and origin and destination fixed effects absorb the economy at each end of a route. The other models on the chart are references with known weaknesses, shown for comparison and not as independent confirmation of the headline.
Per mile, long flights are cheap on US routes
descriptive A pattern in the data, not a causal estimate.
Across 83,815 US route-quarter observations, the distance coefficient is 0.23: doubling the miles far less than doubles the fare, so a transcontinental trip costs less per mile than a short flight. The regression reports what each route feature is associated with, not what it causes.
After COVID, the seasonal model predicts US passenger demand best among the traditional models compared, and the error-correction model worst
descriptive Accuracy is conditional on feeding models the realised same-quarter macro inputs, not a real-time forecast test.
On the post-COVID expanding-window comparison, the seasonal SARIMAX model has the lowest root-mean-square error of the traditional models compared, 0.081, and the error-correction model, the site's former headline specification, the highest, 0.243. Every model is given the actual economic data for the quarter it predicts, so the scores measure how well each fills in a quarter once the economy is known, not how well it forecasts ahead. Before COVID the leading models' errors are close together. The ranking covers the models on this chart only: pretrained foundation models scored on the same windows beat all of them, and no ARDL model, the kind that now gives the headline elasticity, has a post-COVID score here.
The United States projection follows income and population alone, and leaves airline expansion out
projection Income-only central projection from income per person and population; supply-side expansion, network formation and low-cost entry are not modelled.
Letting income and population alone drive the projection, United States traffic reaches roughly 1,338 million passengers a year by 2044. By design it excludes new routes, cheaper fares and airline expansion, so it is an income-only projection, not a full forecast.
The model has run low on past data. Re-run from earlier years with actual income, on the curve fitted to all years, it projected less traffic than was flown in most windows of 5 years or more ending by 2019.
The security-count nowcast fits closely mainly because both counts cover the same passengers
descriptive The compared series are near-identical tallies by construction, so a tight fit is expected rather than evidence of skill.
Airport security screenings and official passenger counts move together almost perfectly, with a statistical fit of 0.9995, close to the maximum possible. That is because both count the same passengers, so the official count is close to a fixed multiple of the screenings. The close fit is two counts agreeing, not forecasting skill. It is still useful for the quarters the official counts have not yet covered. Tested by leaving out each quarter in turn, it fills the gap to within 1.17 percent on average.