The hour-plus commute

For a city, yes: among places the Census Bureau reports, New York City’s 2024 mean of 40.6 minutes is the highest of any large city, and the surrounding metro has led the Bureau’s metro ranking for years. Claims that drivers in particular have the longest commute in the nation, which circulate in the congestion-pricing debate, rest on older private data and rank the city second for driving. The census figure is for all modes, and New York’s is long because of transit, not despite it: drivers here have a median trip of 30 minutes, which is ordinary; subway and bus riders have a median of 45, which is not.

Closely, and that is the check everything else rests on. The Bureau’s 2024 table reports 3,560,755 ± 22,605 commuters and 896,483 of them at 60 minutes or more, a share of 25.2%, with a mean trip of 40.6 minutes and 12.5% of workers at home. Rebuilding the same universe from the 2024 microdata gives 3,556,031 ± 26,625 commuters, 25.3% ± 0.5 over the hour, a mean of 40.7 minutes and 12.5% at home, all inside the margin of error . The five-year file, which most of the charts use, gives 3,245,001 commuters against a published 3,263,450, a difference of 0.6% that sits just outside the two margins; the microdata is a subsample of the full survey with its own weights, and a gap of that size on a count is usual. Its 25.1% ± 0.3 over the hour and 40.3-minute mean match the published 25.3% and 40.3. The full comparison, borough by borough, is in the companion files.

Compared with someone whose trip is under half an hour, they are far more likely to live outside Manhattan (92% against 72%), to ride the subway (53% against 17%) or the bus (18% against 6%), to be Black or Hispanic (56% against 41%), to be foreign-born (51% against 43%), and to have no degree (62% against 52%). They earn a little less, a median of about $49,000 against $53,000. They are twice as likely to leave home before 6 a.m. And at five commuting days a week, an hour each way amounts to ten hours in transit, against about six for the median commuter; the census records the usual one-way trip, not days worked, so that is a scenario rather than a measurement.

Subway riders are over the hour at 34% and bus riders at 38%; drivers at 18%; cyclists at 6%; walkers at 1%. Commuter rail, a small group of about 33,000 city residents, is over the hour for six in ten. The survey asks for the one means of transportation used for most of the distance, so a bus-to-subway trip is recorded once, as the subway if that was the longer leg; the two-fare trip is invisible as such, and “bus” means people whose main leg is a bus. Travel time is what the respondent writes down for the usual door-to-door trip, and it is heaped: far more people say 30, 45, 60 or 90 than 28, 47, 58 or 95. The 60-minute line is the Bureau’s own band edge, which is why it is used here; the dashboard slider lets you move it.

 

Measured

 

How to read this, and what it cannot show

One bar per usual mode, with the borough and earnings filters. Switch the measure to see how many people use each: the subway carries four in ten commuters, the car three in ten.

The census cannot say. What it measures is the share of bus riders reporting an hour or more, which rose from 27% in 1990 to 40% in 2024, and a rider’s trip can lengthen for reasons other than the vehicle’s speed: longer routes, more transfers, jobs that moved. The Comptroller’s separate operating data put the average bus at 8.17 miles an hour in 2024, down 0.6% over the previous ten years, so buses are very slow today but the speed series does not reach back to 1990.

A survey-weighted logistic regression of whether the one-way trip is 60 minutes or more, on borough of residence, usual mode, place of work, own-earnings band, race and ethnicity, occupation group, age band, sex, nativity and degree, fitted to the 125,000 commuter records in the five-year file with their person weights. What the chart shows is the average marginal effect of each category: the average change in the predicted chance of an hour-plus trip when every commuter is switched from the reference category to that one, everything else as observed. Reference groups are Manhattan, car, own borough, $60,000 to $100,000, white, office and administrative work, age 35 to 44, male, US-born and no degree. Margins come from refitting the model with each of the 80 replicate weights. The model explains a modest share of the variation (a pseudo-R² of 0.16), which is expected when the outcome depends on things the census does not record. It is an adjustment on observed variables, not a causal estimate: it says how much of a raw gap survives the controls, not why. Construction keeps a residual of about 6 points; every other occupation lands within a couple of points of office work.

The five-year file pools the 2020 to 2024 surveys, which is what makes cuts like Staten Island subway riders or Bronx commuters earning over $150,000 possible at all. That average contains the pandemic: 17% of its workers were at home against 12.5% in 2024 alone, which is why the working-from-home figure quoted above is the 2024 one. On the hour-plus share the two agree, 25.1% pooled against 25.3% ± 0.5 in 2024, and the borough, mode, earnings and race patterns recomputed on 2024 alone are the same within their margins: Black commuters 31%, white 18%; subway 33%, bus 40%, car 19%. The one-year figures are in the results file.

The questions are the same words. What differs is the survey: 1990 and 2000 are the decennial census long form, a 5% sample of housing units, and 2005 onward is the annual American Community Survey, about 1% a year with different weighting. The 1990 and 2000 points carry no margin because those files have no replicate weights; a 5% sample of the city’s workers is about 150,000 records and the sampling error on a citywide share is well under a point. The five boroughs are identified in every year (the microdata areas were redrawn in 2000, 2010 and 2020, always inside borough lines), so the city and borough totals line up; means-of-transportation codes were mapped to the current scheme. The 2020 survey used experimental weights after the pandemic wrecked its response rates, and the chart says so. One thing did change: before 2005 the survey asked about the week before Census Day in April, and since then it has asked every month of the year, which slightly favours the modern series with summer trips.

The hour-plus share peaked at 28% and the mean trip at 42 minutes in 2017 and 2018, and the subway’s own hour-plus share peaked at 38% in 2017. That is the period of the subway’s reliability crisis, the “summer of hell” and the governor’s state of emergency in June 2017. The census cannot attribute a cause, and part of the rise is composition, since the subway was carrying a record share of commuters. But the timing is what it is, and the Bronx, the borough most dependent on the longest lines, is where the share rose most, from 27% in 1990 to 37% in 2019.

Not on average. Working from home rose from 4.7% of workers in 2019 to 24% in 2021 and 12.5% in 2024, and the workers who stayed home were disproportionately the short-trip Manhattan office workers. Removing them from the commuting count should raise the hour-plus share among those left, and instead it fell from 27% to 25%. Two things could offset: fewer riders and, per the MTA’s on-time figures, a more reliable subway may have shortened the trips of the people still making them, while the departure of the short-trip workers lengthened the average of who remained. The census cannot separate the two, and no attempt is made to here.

The census asks people; the MTA counts turnstiles, fareboxes and tolls. The MTA series runs to 3 September 2026, twenty months past the last census year, and it is the only public data that says what happened after congestion pricing began in January 2025. Its weekday subway figure rose from 69% of 2019 in 2024 to 74% in 2025 and 75% in 2026 to date; the governor’s office reports nearly 1.3 billion subway trips in 2025, up 7% on 2024, and the citizens’ advisory committee puts that at 76.5% of 2019’s 1.7 billion. It cannot say how long anyone’s trip took, who was riding, or where they were going. So the 2026 section is about volumes: who is back and who is not.

Anything below the borough. The second leg of a two-seat trip. Why a trip takes an hour, whether that is distance, a slow bus, a transfer or a walk at either end. Reverse commuters into the city, who live in the suburbs and are not in a city survey at all. Whether the same people were over the hour in 2019 and 2024, which would need the same workers followed through time. And the microdata is a disclosure-protected sample of the survey, with a few top-coded values and some perturbation, so every relationship seen in it is an estimate of the city’s, not the city itself.

This is the methods note for everything above. It is deliberately more technical than the rest.

The published anchors. Table B08303 (workers 16 and over who did not work at home, by travel time), B08013 (aggregate travel time, whose ratio to B08303’s total is the mean) and B08301 (workers by means of transportation), for the city and the United States from the 2024 one-year survey and for the city and its five counties from the 2020–24 five-year survey. These are both the anchors and the reproduction targets.

Census microdata. The 2020–24 five-year public-use microdata sample for a 156-area study region, cut to residents of the five boroughs who were at work in the reference week and therefore have a means of transportation: 155,257 person records carrying 80 replicate weights and representing 3.92 million workers. Commuters are those who did not work from home and reported a one-way travel time — 125,312 records representing 3,245,001 workers, which is the universe of the published table. The usual means of transportation is a single mode, grouped as subway, bus, car (including trucks, vans and motorcycles), commuter rail, walked, bicycle and other. Place of work is coded at borough level in the 2020 vintage (Manhattan 04100, Bronx 04200, Brooklyn 04300, Queens 04400, Staten Island 04500) and grouped as Manhattan, own borough, another borough, or outside the city. Own earnings are in 2024 dollars in five fixed bands; race and ethnicity is Hispanic of any race, otherwise the census’s single-race groups; occupation is collapsed to the Bureau’s major groups; departure time before 6 a.m. is code 30 and below. The 2024 one-year file, with its own replicate weights, carries the reproduction and a sensitivity check.

The 1990–2024 series. The 1990 census 5% sample, the 2000 census 5% sample and the 2005–2024 one-year files. The five boroughs are identified in every one: by the 1990 area prefixes 050 to 054, verified against the 1990 census county populations to within 1%; by the 55 area codes 03701 to 04114 shared by the 2000 and 2010 vintages, verified against the 2000 equivalency file; and by the 2020 geography from 2022. Each year’s means-of-transportation codes are mapped to the 2019 scheme and the extraction script records every mapping. The 1990 and 2000 files have no replicate weights; the 2020 one-year file carries the Bureau’s experimental weights and is flagged. The two pre-2005 surveys refer to the week before Census Day in April, the later ones to a rolling reference week through the year.

Transit. MTA daily subway, bus, Long Island Rail Road, Metro-North and bridge-and-tunnel counts from 1 March 2020 to 3 September 2026, averaged by year and day type. The 2019 baseline for each mode and day type is recovered from the MTA’s frozen 2020–2025 file, which carries the MTA’s own share of the comparable pre-pandemic day for every date: the count divided by that share, averaged over 2024 weekdays and weekends. The series counts trips, not commuters, and has no sampling margin.

Estimators. For a statistic computed with the full person weight, the same statistic is recomputed with each of the 80 replicate weights and the variance is the Bureau’s successive-difference estimator, four-eightieths of the summed squared deviations of the 80 replicate estimates from the full-weight one. Margins are 90%, or 1.645 standard errors. Counts are weighted sums, shares and means weighted means, medians weighted medians without a margin. Every combination the chart controls can reach is precomputed with the same estimator and a margin on each bar; cells resting on fewer than 30 records are labelled too thin rather than drawn. The dashboard sums pre-aggregated cells and reads medians and threshold shares off a five-minute histogram, and carries no margins.

Does it reproduce? Every share and mean lands inside the margin, and the one-year commuter count lands inside the published margin. The five-year count differs by 18,449, or 0.6%, slightly outside the two margins: the microdata sample is a subsample of the full survey with its own weights, a difference of this size on a five-year count is usual, and it moves no share. The comparison is the reproduction table.

Geography and mode. The five-borough gradient, 9.8% in Manhattan to 31.5% in the Bronx, is almost entirely a transit gradient. Drivers are over the hour at 14.9 to 20.2% in every borough; subway riders at 10.9% in Manhattan, 34.5% in Brooklyn, 45.5% in Queens, 50.0% in the Bronx and 79.3% in Staten Island, the last on 330 records. By place of work, 55.0% of commuters work in their own borough and 13.9% of those are over the hour; 27.1% travel to Manhattan from another borough (40.4%); 10.3% work in another outer borough (39.6%); 7.6% work outside the city (32.6%). Bronx residents working in another borough are over the hour at 55.7%, Staten Island residents working in Manhattan at 70.4%.

Earnings. The citywide share is flat at 26.0 to 27.7% below $100,000, 22.8% between $100,000 and $150,000, and 15.4% above. Within Queens and the Bronx there is no gradient at all; within Staten Island it runs upward, 26.2% to 43.6%, because the better paid there travel to Manhattan; among subway riders it runs downward, 39.2% to 13.7%, because the highest earners board in or near Manhattan — 69.4% of commuters earning $150,000 or more work there. Earnings act on the hour through where a person lives, not through how fast they travel.

Race and ethnicity. Black commuters are over the hour at 31.4% ± 0.7%, Hispanic commuters 27.4% ± 0.5%, Asian commuters 26.4% ± 0.8% and white commuters 17.9% ± 0.4%. Holding the borough fixed narrows the Black–white gap without closing it (Brooklyn 32.3% against 19.3%, Queens 36.1% against 23.9%, the Bronx 31.6% against 27.9%); holding the mode fixed at the subway widens it, 45.1% against 20.2%. Black commuters take the bus at 17.0% against 6.9%, and 89.9% live outside Manhattan against 68.7%. This is a descriptive decomposition one variable at a time, not a model, and it does not rank the contributions of residence, mode and occupation.

The profile. Against commuters under 30 minutes, those over the hour are more often outer-borough residents (92.4% against 71.6%), subway riders (53.4% against 17.0%), bus riders (17.5% against 6.0%), Manhattan-bound (47.5% against 34.8%), Black or Hispanic (55.8% against 41.1%), foreign-born (51.2% against 43.1%), without a bachelor’s degree (62.4% against 51.7%) and out of the door before 6 a.m. (13.7% against 7.4%). Every one of those differences exceeds the sum of the two margins. Median own earnings are $49,240 against $52,793. Among occupation groups with at least 500 records the share runs from 37.4% in construction and 34.7% in cleaning and maintenance down to 15.7% in legal work and 19.2% in education and the arts.

The adjusted model. A weighted logistic regression of the probability of a 60-minute-or-longer trip on borough of residence, usual mode, place of work, earnings band, race and ethnicity, occupation group, age band, sex, nativity and degree: 50 parameters on 125,312 commuter records with person weights, fitted by iteratively reweighted least squares, pseudo-R² 0.16. What is reported is average marginal effects — for each category, the weighted mean across all commuters of the change in predicted probability when every commuter is set to that category rather than the reference (Manhattan, car, own borough, $60–100k, white, office and administrative, 35–44, male, US-born, no degree). Margins come from refitting with each of the 80 replicate weights, warm-started from the full-weight solution. Mode and geography survive: the subway is +19.0 ± 0.7 points and the bus +21.1 ± 1.1 against the car, where the raw gaps were +15.7 and +19.4, and the outer boroughs are +11.1 to +22.5 against Manhattan. Earnings largely do not: the top band’s raw −11.6 points becomes −5.8 ± 0.9 and every lower band sits within a point of the reference. Occupation effects shrink to within about two points of office work except construction, +6.0; a degree’s raw −7.9 becomes −2.2. The Black–white gap of +13.6 points becomes +6.0 ± 0.6, so about half is accounted for by residence, mode, workplace, earnings and occupation and about half is not. The residual is not attributed to any mechanism here, because the microdata record no station, transfer or service frequency. It is a descriptive adjustment on observed variables, not a causal estimate, and it cannot hold fixed what the census does not record. The code is article_commute_model.py.

The series. The hour-plus share was 22.3% in 1990 and 23.8% in 2000, held near 25% through 2014, rose to 28.4% ± 0.5% in 2017 and 27.1% in 2019, and stands at 25.3% ± 0.5% in 2024; the mean trip went from 35.7 to 40.7 minutes. The subway’s share of commuters rose from 36.7% to 46.7% in 2018 and fell back to 42.6%; the car’s fell from 33.0% to 29.0%. The subway’s own hour-plus share was 35.5% in 1990, 37.5% in 2017 and 32.5% in 2024. The bus’s rose from 27.4% to 39.5% — that is the share of riders reporting an hour or more, not a measure of vehicle speed; the Comptroller’s operating data put buses at 8.17 mph in 2024, down 0.6% over ten years. The car’s rose from 12.6% to 19.0%. Working from home went from 0.5% of workers to 12.5%, and 24.0% in 2021. The 2017 peak is not attributed to a cause; it coincides with the subway’s reliability crisis of that year and with the subway’s peak share of commuters.

2026 in the transit counts. Weekday subway ridership rose from 68.7% of the 2019 baseline in 2024 to 73.8% in 2025 and 74.6% in 2026 to date; buses 60.3%, 64.8%, 60.4%; bridge-and-tunnel crossings 101%, 102%, 100%; weekend subway ridership 80.3%, 87.7%, 88.5%. The Governor’s office reports nearly 1.3 billion subway trips in 2025, up 7%; the citizens’ advisory committee puts 2025 at 76.5% of 2019’s 1.698 billion and attributes much of the growth to congestion pricing, which began on 5 January 2025. The census cannot see 2025 or 2026, and these counts say how many trips were made, not how long they took or who made them.

Rerunning it. The analysis script computes every figure and number above, the charts and the dashboard in one run from a fixed seed; its docstring is the long-form method. The companion dataset is the census records themselves: the 155,000 city resident workers in the five-year file, with their survey weights and the derived variables, so a weighted mean of its hour_plus column among commuters reproduces the headline share.