Who still goes to the office

Closely, in every year, and that is the check everything else rests on. The Bureau’s 2019 table reports 4,032,673 ± 21,303 city workers and 187,030 ± 8,791 of them working from home, a share of 4.6%; rebuilding the same universe from the 2019 microdata gives 4,047,572 ± 25,555 workers and 188,639 ± 10,158 at home, 4.7% ± 0.3, inside the margin of error . For 2024 the table reports 4,070,607 ± 20,036 workers and 509,852 ± 12,449 at home; the microdata gives 4,065,043 ± 24,289 and 509,012 ± 13,017. The 2021, 2022 and 2023 tables and the five boroughs in 2019 and 2024 reproduce the same way, as does the subway’s share of workers in each year. The full comparison is in the companion files.

Because it asks a different question. The census asks how the person usually got to work last week and takes one answer; someone at home two days and on the subway three is a subway rider. The Current Population Survey asks whether any paid work happened at home, and the Bureau of Labor Statistics’ time-use survey asks where work happened on a particular day; both count the hybrid worker. So the three national figures for 2024 and 2025, 13% usually at home, 22% with any hours at home, 34.5% working at home on days worked, describe one arrangement three ways. For New York the first two are 12.5% and about 25%, and the monthly survey’s 10% whose whole week is at home is the nearest thing to the census’s one in eight.

Compared with a commuter, a worker who usually works from home in 2024 is far more likely to be in an office-based industry (49% against 27%), to hold a bachelor’s degree (69% against 48%), to earn $100,000 or more (38% against 25%), to be white (48% against 32%), to live in Manhattan (32% against 21%), to be self-employed (21% against 9%) and to be a woman (53% against 49%), and less likely to be Black or Hispanic (31% against 46%) or foreign-born (31% against 45%). Every one of those differences exceeds the sum of the two margins. They are equally likely to be under 35 and to work full time. In 2019 the remote worker was already whiter, more Manhattan and more often self-employed than the commuter, but earned the same and was no more likely to be paid $100,000; the pay and degree gaps are what 2020 added.

Not in these figures: the census counts residents by where they live, and the number of city residents at work rose slightly between 2019 and 2024, from 4.05 million to 4.07 million. Someone who left for Westchester and kept a Manhattan job is counted there, not here, which is why the Manhattan commuter count above adds the suburban files. Whether the same people who were at home in 2021 are still at home in 2024 cannot be seen either; the census does not follow individuals through time.

In 2019 the median worker at home, the median commuter and the median worker overall all earned about $66,000 in 2024 dollars, and 35% of remote workers were paid $100,000 or more against 33% of commuters, a difference inside the margin. The 2019 remote worker was a mix: 36% were self-employed, and the share at home was highest in arts and media, computing, personal care and sales. In 2024 the self-employed are still at home at the highest rate of any class of worker, 26%, but they are now a fifth of the remote workers rather than a third; the rest of the growth is salaried employees of private firms, whose share at home went from 3.5% to 12%. Government employees went from 1.8% to 5.7%.

Most of it. Bronx residents in the office-based industries are at home at 12% against 20 to 23% for residents of the other boroughs in the same industries, so a gap remains inside the industries, and a Bronx resident earning $100,000 to $150,000 is at home at 12% against 20% for a Manhattan resident in the same band. The census cannot say why. Occupation within an industry, the employer’s policy, the kind of home available for working in, and who moved where since 2020 are all candidates, and only the first of those is in the data.

Because it pools 2020 through 2024, and the thing being measured changed every year in that window: the pooled share at home is 17%, which is no year’s number. The one-year files are smaller, about 33,000 city workers each, so some cuts are thin and are labelled that way, but each one is a clean year. The 2020 file is drawn but flagged: the Bureau published it with experimental weights and no standard tables, and its 21% should be read as an order of magnitude.

As a level with a range, not as a monthly reading. Each month’s New York City sample is about 650 employed people, so a single month’s share has a margin of several points, and the survey re-interviews the same households in consecutive months, so adjacent readings are not independent. The margins drawn on the group chart come from the spread of the monthly estimates over the period and are approximate. Two more cautions. The Bureau updates its population controls every January, which shifts the weighted counts, so shares rather than counts are compared across years. And the sample shrank after the shutdown in the autumn of 2025, which is why the September to December 2025 figures are shown but not leaned on.

Jobs rather than residents, and days rather than usual arrangements. The Partnership’s survey asks Manhattan’s large office employers what share of their workers are in on an average weekday, which includes commuters from New Jersey and Connecticut and excludes every worker outside a Manhattan office; its 57% is an attendance rate, not a remote-work share. Kastle’s index counts badge swipes in the buildings it secures, relative to the same buildings before the pandemic, and most of the largest Manhattan portfolios are not its clients. The MTA counts trips. Each is a good measure of something the census cannot see, and none of them can say who the workers are.

Anything below the borough. How many days a week the census’s commuter actually goes in, since the census takes one usual answer. The second job. Whether the remote worker’s employer is in Manhattan, since a person at home has their home as place of work. Reverse commuters and suburban residents beyond the census study area. Whether the same people were at home in 2021 and 2024. And the microdata is a disclosure-protected sample of the survey, with a few top-coded values, 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 anchor. Census table B08301 (means of transportation to work, workers 16 and over; the final row is “worked from home”) for the city, its five counties and the United States, from the one-year survey: 2019 from the Bureau’s sequence-format summary file, 2021 to 2023 from the table-based summary files, 2024 from the Census Reporter mirror. No one-year table exists for 2020.

Census microdata. The one-year public-use microdata samples for 2015 to 2024 covering New York, New Jersey and Connecticut, cut to a metro study area of 156 microdata areas and then to residents of the five boroughs who were at work in the reference week and therefore have a means of transportation: about 33,000 person records a year, carrying 80 replicate weights and representing about four million workers. Working from home is the usual-means code 11; commuters are everyone else. Own earnings are wages plus self-employment income, adjusted by the Bureau’s within-year factor and then to 2024 dollars with the New York-metro consumer price index, and cut into five fixed bands. Industry is collapsed to ten sectors, with sectors 51 to 56 defined as office-based; occupation to the Bureau’s major groups, the 2015 to 2017 files carrying the 2010 codes whose major-group ranges coincide to within a few codes. Place of work is at borough level, Manhattan coded 3800 in the 2010 geography used through 2021 and 4100 in the 2020 geography from 2022. A worker physically commutes to Manhattan if the place of work is Manhattan and the usual means is not working from home; because a person at home reports home as their place of work, that condition binds only for Manhattan residents. The 2020 file carries the Bureau’s experimental weights and is flagged wherever it appears.

The monthly survey. Current Population Survey basic monthly public-use files from June 2024, when the telework items entered the basic questionnaire, to August 2026. The items ask everyone employed and at work whether they teleworked or worked at home for pay at any time last week and, if so, how many of the week’s hours across all jobs. Residents of the five counties are identified through the county code; the universe is employed-and-at-work, 16 and over, weighted by the final person weight. Monthly city samples run 570 to 720 people. No file exists for October 2025, when the survey was not conducted during the federal shutdown, and the sample is smaller from November 2025. “All hours at home” means the hours teleworked cover every hour actually worked that week at all jobs. The extraction reads the Bureau’s record layouts and the fixed-width files directly.

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 by day of the week. The 2019 baseline for each mode and each weekday is recovered from the MTA’s frozen 2020 to 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 by day of the week.

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 weighted means, medians weighted medians without a margin. The monthly survey’s public file has no replicate weights, so a pooled estimate from it is the mean of the monthly weighted estimates and its standard error is their standard deviation divided by the square root of the number of months; because consecutive months share three quarters of their sample this understates the true error, and the figures above are rounded accordingly. Every combination the chart controls can reach is precomputed on the microdata with the same estimators and a margin on each bar, and shipped as one file; cells resting on fewer than 30 survey records are labelled too thin rather than drawn. The dashboard sums pre-aggregated cells and carries no margins.

Does it reproduce? Every published count, share and borough share lands inside the margin in every year, as does the subway’s share of workers, which is what checks the mode coding. The year-by-year comparison is the reproduction table.

The series. The share at home ran 4.0% in 2015, 4.7% in 2019, 24.0% in 2021, 16.2% in 2022, 13.4% in 2023 and 12.5% ± 0.3% in 2024; the 2023 to 2024 difference sits inside the margin. Commuters fell by about 303,000 between 2019 and 2024 while the number of workers rose slightly. The subway’s share of workers fell 5.8 points, the car’s 1.0.

Who is at home. In 2019 the earnings bands were within 1.4 points of each other; in 2024 they run from 10.4% to 18.7%, and the median worker at home earns $74,213 against $53,598 for the median commuter — in 2019 both medians fell on the same heaped value, about $66,300. Inside the office-based industries the gradient largely vanishes (20.2%, 16.3%, 22.3%, 24.3% and 20.9% up the bands) and the borough gradient closes except for the Bronx, 12.2% against 20.4 to 22.6%. The 9.0-point difference between white and Black workers citywide is 7.8 points among degree holders and 8.7 points inside the office-based industries; the census cannot decompose it further than occupation and industry. The self-employed were the original remote workers, 18.7% in 2019 and 25.6% in 2024; private employees went from 3.5% to 12.0% and government employees from 1.8% to 5.7%.

Manhattan. The count of workers physically travelling to a Manhattan workplace fell by 305,409, or 12.0%, of which 231,081 are city residents and 74,328 suburban residents of the study area. Commuters from beyond the study area — eastern Suffolk, the upper Hudson Valley, Pennsylvania — are not counted, so the study area’s suburban count of 607,046 in 2024 is a floor for inbound commuting, not a total.

2026 in the monthly survey. The any-hours share is 24.4% in January to August 2025 and 24.9% in the same months of 2026; the all-hours share 8.5% and 9.9%; the some-hours share 15.9% and 15.0%. Every difference is inside the approximate margins. The gradient reproduces the census’s: 41.7% of degree holders against 6.1% of everyone else, 42.7% in management, business and professional occupations against 10.4%, Manhattan residents 42.0% against 20.0%. Government employees moved from 13.8% to 19.8%.

2026 in the transit counts. In 2026 to date the subway runs at 77 to 79% of 2019 on Tuesday to Thursday, 69% on Monday, 71% on Friday and 88 to 89% at weekends: about 4.19 million riders on a midweek day against a 2019 baseline of 5.40 million, and 3.85 million on a Friday against 5.44 million. The Long Island Rail Road is at 91% on Wednesdays and Thursdays and 77% on Mondays, and above 2019 at weekends; Metro-North 88% midweek and 74% on Mondays; bridge-and-tunnel crossings 95 to 105% every day. The midweek plateau has risen from 60% in 2022 through 70% in 2023, 71% in 2024 and 76% in 2025 to 78% in 2026.

One arrangement, three definitions. Nationally the same thing is 13% in the census, which asks for the usual means of getting to work; about 22% in the monthly survey, which asks about any hours last week; and 34.5% in the time-use survey, which asks whether any work was done at home on a diary day. None of the three is wrong, and none of them is comparable to the others.

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 66,000 city resident workers in the 2019 and 2024 files, with their survey weights and the derived variables, so a weighted mean of its worked_from_home column within a year reproduces the headline share.