Ground truth · the record re-derived from primary data
The Hour Nobody Spends
There is a claim, repeated for fifty years, that every human society spends about one hour a day travelling, and that this is an anthropological invariant. This page tests it against 252,808 American time diaries, 32 English survey years and 22 European countries. The level turns out to be right, almost eerily so. Nearly everything stronger than the level turns out to be wrong, and the way it is wrong is the interesting part.
The claim
The modern statement of it belongs to Cesare Marchetti, writing in 1994 and drawing on Yacov Zahavi's fieldwork for the US Department of Transportation. Marchetti thought this was a fact about the species rather than about any country:
And then the sentence that this page exists to test. Read it twice, because it makes two separate claims and only the first one is famous:
The first sentence is about a mean. The second is about a shape: it says the distribution is narrow, that most people are near the hour. That second sentence is checkable, and it is the one that fails.
Before testing it, one thing about where the claim comes from, because it turned up while reading the paper and it is not usually mentioned. Marchetti collected no data of his own here, and his article cites eight works in total. The one-hour finding rests on two of them: Zahavi's The "UMOT" Project (Report DOT-RSPA-DPB 2-79-3, 1979) and The UMOT-Urban Interactions (DOT-RSPA-DPB-10/7, 1981). Both are unrefereed contractor reports for the US Department of Transportation, and neither appears to be on the open web in any form. The most-cited invariant in transport behaviour rests on two documents a reader cannot obtain. That is not an accusation of error. It is a reason to go and measure.
What is already known, before anything below
The most important thing to say early, because it would be the easiest thing to imply falsely. That American travel time has been falling since 2003 is not a finding of this page. It is published, on this same survey and over this same window, by Morris, Speroni and Taylor, and the larger shift it belongs to has been mapped by others too. This page recomputes their trend from the primary file rather than quoting it, because that is what this venue does, but the trend is theirs.
What is left, and what the rest of this page is for, is narrower and is stated as such: an exact decomposition of the American fall into how often anyone goes anywhere against how long they go for, and the observation that the first of those is doing about seventy per cent of the work while the second held near Marchetti's hour until 2020. If that too turns out to be published somewhere we did not find, the finding is not ours and the page should be corrected.
First, the level, which is right
Give the claim its due before breaking it. Three independent instruments, three countries or continents, three different survey designs, and all three land near an hour.
Three separate surveys, none of them talking to the others, none of them designed to test this claim. The English figure needs one caution before it is remarkable. That table stacks two series: rolling multi-year averages up to 2000-2002, then single years from 2002 on, and several of the rolling rows overlap each other, so a mean over all · published rows would count the late 1990s five times over. Take a non-overlapping cover of the same rows instead and the mean daily travel time of England, across half a century, is · hours. Marchetti said "around one hour". He was not wrong about the level.
Now the shape
Here are real American days, drawn at random from the file. Each bar is one person's total travel time on one diary day, against the 1,440 minutes the day contains. The green bar is the average that every summary of this literature reports.
Instrument · draw a day
Twelve actual American diary days, sampled with probability proportional to their survey weight, which is what makes them a fair picture of the country. Press again for twelve more. The bars are totals, not timelines: the survey records how many minutes of the day went on travel, not when.
loading 252,808 diaries…
You can press that button for a long time without drawing an hour. The reason is in the distribution, and the distribution is not narrow. Recomputed from every diary in the file:
·
This is not a new objection in kind. The standing review of the field said so twenty years ago, and said it about the whole literature:
Mokhtarian and Chen also caught the field in the act, in a passage worth having in full, because it is what a narrow range looks like when you are hoping for one:
And they note that the man whose fieldwork the whole edifice rests on did not himself claim a constant. Zahavi and Talvitie, in 1980, expressed the
The bench
Everything above and below is recomputed in your browser from the same 2.4 MB of packed diaries, one row per respondent. Move the controls and the numbers are re-derived from the microdata in front of you, not looked up in a table.
Instrument · the travel-time bench
Pick a year and a slice of the population. The mean, the share who went nowhere, and the mean among those who went somewhere are recomputed from the selected diaries, weighted by the survey's own final weights.
·
The hour moved
If the budget were a behavioural constant, the yearly means would wander around a flat line. They do not. Two surveys, two countries, two instruments that share nothing, and both are falling over the same window.
| series | window | slope | standard error | t or z |
|---|
American daily travel time fell by · minutes per year from 2003 to 2019, before the pandemic touched it, at · standard errors from zero. English travel time fell by · minutes per day per year over the comparable window, at · standard errors from zero. The two surveys were not built to agree and they agree.
How it fell, which is the part worth having
A daily mean can fall two ways. Either people make shorter journeys, or fewer of them make any journey at all. Those are different worlds and the arithmetic separates them exactly. Write the mean as
mean = (1 − share who went nowhere) × (mean among those who went somewhere)
and any change in the mean splits, with no remainder, into a part from the first factor and a part from the second.
That split is not a new idea and it is worth saying whose it is, because it turns out to sit exactly on the crack in the original evidence. Zahavi did not measure travel time per person. He measured it per traveller, and Mokhtarian and Chen record why. The reason for using travellers as the unit
And Goodwin had already put a finger on what the choice hides, in 1981, in a sentence that reads now like a prediction:
So the second factor above is Zahavi's number, the one that held still. The first factor is the one he set aside because it would not. Here is what each has done since 2003.
Instrument · give a year back an earlier habit
Take 2019 and hand it one of 2003's two numbers, leaving the other alone. Which one has to be restored for the fall to disappear?
·
·
The obvious objection is that the country simply got older, and older people travel less, so this is demography wearing a behavioural costume. It is not, or at least not mostly. Splitting every year into 28 cells of age by sex by employment and holding the cell shares fixed at 2003's, the pre-pandemic slope goes from · to · minutes per year. Composition accounts for about of the fall. The rest is people of the same age, sex and employment status going out less often than people like them used to.
Who stopped going out
| group | mean 2003 | mean 2019 | mean 2024 | went nowhere 2003 | 2019 | 2024 |
|---|
·
What kind of travel stopped
The survey codes every trip by the purpose of the activity it served, so the loss can be itemised. This is where the finding stops being a statistic.
·
2020, and what did not come back
2020 has to be handled separately and the survey itself insists on it. Collection was suspended from 18 March to 9 May 2020, so no annual estimate for that year exists at all: the ordinary final weight is set to −1 on every 2020 record and cannot be used, and what the Bureau supplies instead is a partial-year figure covering the ten months that were collected. The Bureau supplies a special weight instead, and supplies it for 2019 as well, so that exactly one comparison is legitimate: 2019 against 2020, under the pandemic weighting. Here is that comparison, and then the years since.
·
What survives
The constant fails as a constant. Its most useful corollary survives, and it is worth stating plainly because it is the part that actually predicts things. If time is roughly stickier than distance, then when transport gets faster people do not bank the saving, they spend it on going further. The English record measures that directly, because it reports distance and time separately for fifty-two years.
·
And the stickiness is real, if modest. Over the clean English window, the coefficient of variation of travel time is · against · for distance. Time is about · times steadier than distance. That is the honest residue of the conjecture: not a constant, but the stiffer of the two.
What a constant is a fact about
One last number, because it is the one that reframes the rest. Take every American diary day in the file and ask how much of the variation in one person's travel time is explained by knowing things about them. Knowing the year explains · of it. Knowing their age band, sex and employment status explains ·. Knowing all four together explains ·.
So the invariant is real in the only sense it was ever measured: as a property of a mean over a population over a year. It is not a property of a person, or of a day, or even of a country, and the aggregate stability that made it look like an anthropological law is the same aggregation that hides everything underneath. What is underneath, in the United States between 2003 and 2019, is a country that went out less: not shorter trips, fewer of them, and the ones that went first were the trips to other people.
The check
Every number on this page is recomputed from primary records, twice: once offline by verify-the-hour-nobody-spends.mjs and once in your browser from the same packed diaries. The verifier reads this page's served HTML and asserts that what it prints matches what came out of the data.
- recomputing…
What this page cannot tell you
A diary day is not a person
Each respondent reports one randomly assigned day. A person with zero travel on their diary day is not a person who never travels. So the "went nowhere" share is a share of days, not of people, and the spread of the daily distribution overstates how different people are from one another over a month. This is the single most important limitation here, and it does not touch the trend: the diary design is the same in every year, so a change in the share of zero-travel days is a change in how often anyone goes anywhere.
The response rate fell by a quarter over the same window
·
That is the Bureau naming this exposure and saying it was not measured. The published rates are also the optimistic ones: processing moves poor-quality diaries into non-response afterwards, · The weights are post-stratified to population controls, which fixes the demographic part, and the finding survives standardising on age, sex and employment directly. Neither of those can rule out a residual bias in mobility, which is not a variable anyone weights to. This is the soft spot, and it is where a critic should go first.
And rising immobility has been called a survey artifact before
This is the sharpest specific objection available, so it belongs here rather than in a footnote. McGuckin, reporting the 2017 National Household Travel Survey, found the same rise in people reporting no travel on their travel day, and warned in the same breath that zero-trip reporting differs significantly by collection mode: that survey moved from telephone to web during the period, and the shift alone moves the number. If the same thing had happened here, the participation finding would be an artifact of the instrument. The reason to think it has not is that the American Time Use Survey has run the same computer-assisted telephone time diary throughout, so it is not exposed to that particular break. That is an argument, not a proof, and the objection stands on the record.
A travel purpose is a destination, and one destination per trip
·
So the purpose split is a picture of where trips were going, not of what kind of trip they were, and a chained journey attributes its legs to the stops between. Two more coding rules matter and cut the same way: driving as a job is coded as work, not as travel, and a run or a recreational cycle is coded as recreation, not as travel. The totals are unaffected by all of this. The purpose lines are conditioned by it.
One number we could not reconcile
·
Self-report, and one country at depth
These are diaries, not trackers. People report their own days, and the coding lexicon changed once in a way that matters: the first digit of the travel codes moved from 17 to 18 in 2005. The multi-year file used here is already harmonised by the Bureau onto the later scheme and the purpose groups map one to one across the break, so totals and purpose groups are comparable end to end; individual six-digit sub-codes are not, and none is used. The deeper limit is geographic: the English and European figures corroborate the level and the direction, but the decomposition, the standardisation, the purpose split and the group cuts are all United States only, and nothing here says they hold elsewhere.