Corrections and changes

1. 2026-10-26 · v1.0 · Published — Published.

Read the article →

1. 2026-10-23 · v1.0 · Published — Published.

Read the article →

1. 2026-10-21 · v1.0 · Published — Published.

Read the article →

1. 2026-10-19 · v1.0 · Published — Published.

Read the article →

1. 2026-10-16 · v1.0 · Published — Published.

Read the article →

1. 2026-10-14 · v1.2 · Update — The story now carries the history the counts sit in, from the City's own documents (owner, 14 October): the Vision Zero plan of 2014 (over 250 killed a year, pedestrians 56% of the dead, the ask for a 25 mph limit, cycling 'quadrupled' over the decade before), the 25 mph default limit of 7 November 2014, the State's legalisation of e-bikes and e-scooters on 1 April 2020 and the Council's that June, DCWP's count of more than 60,000 app delivery workers (2022), Sammy's Law and DOT's first 20 mph streets (October 2024), the 15 mph e-bike limit in force from 24 October 2025, the Council's hearing of 30 September 2026 with its 23 bills, DOT's estimate of daily cycling trips (over 620,000, up 64% from 2013 to 2023) and its bike network (1,550 lane miles, 555 protected lanes), and DCWP's 2022 study of app delivery (orders up by more than 50% with the pandemic). Each is a registered source with a saved copy and a checked quote; none is used in a comparison. A new section tests four things said about e-bikes against the counts, and a new question checks the readings in the recent coverage: the hearing's 13 e-bike riders killed in the first half of 2026 against 10 (no clear change under the model, 34.9% fewer to 159.7% more), percentage rises from before the records named e-bikes, the illegal-device shares, the Class III premise, and the Comptroller's 4.5% of pedestrian injuries (2020-2023) as the Council quotes it. Explanations are attributed (DOT's 31% since 2014 and its reasons; the Council's own account of the growth in riding) or given as a likely reading with the limit said once. The two-wheel turn is now a 'However' with the vehicles named; the 'since 2023' reading sits with chart 2; the close returns to the plan. No death count moved.
2. 2026-10-14 · v1.1 · Update — Rewritten as one story, with the same sources, figures and charts (owner, 14 October): the fall in road deaths belongs to people on foot and in cars, people on two wheels never joined it, and the new deaths are on e-bikes, e-scooters and mopeds. No figure moved. The headline now states the finding; the dek, the In-brief box and the scope box are cut to the house limits; the roadmap sentence is gone; each caveat is said once, where its number is, and the accounting of the rise since 2020 by vehicle (36, 7, 7 and 28 a year), the comparisons ending before the 2022 change (71 and 243 a year), DOT's September counts (17 e-bike and 11 e-scooter riders) and the sources' counts for 2024 moved to the questions. Chart 2 gains a line for all deaths on two wheels, the series the fall since 2023 and the 2013 to 2019 level are said of, and both chart titles now state the story's claims. Page fix: the methods' model range for this year's total was left at the 4 October week (21% fewer to 16% more); it now prints from the results (19.3% fewer to 17.6% more).
3. 2026-10-14 · v1.0 · Published — Published.

Read the article →

1. 2026-10-15 · v1.3 · Update — Revised after the owner's outside review of 15 October, with the same sources and charts; no count moved. Added: the 1998 benchmark's reading per resident beside the count (392 felony assaults per 100,000 residents in 1998 against 355 in 2024, on the State's own denominators), so a count passed is not read as a per-person rise; the 2019 base of the count without strangulation (17,280) beside its 45%; a note under the places chart that the complaint files run about 200 above the yearly table for 2019; the under-18 shootings (120 in 2025 against 64 in 2019) in the story, as the part of the shooting fall that did not arrive. Checked and declined: the review's recalculation of the strangulation-excluded rise (44.23%) took 2020's row (17,395) for 2019's (17,280); 25,089 against 17,280 is 45.2%, and within the complaint files alone 43.5%, as the methods say. Rewritten: a shooting defined where the word first appears (a person struck by a bullet, not shots fired); felony assault defined by the law's elements (injury with a weapon, not a weapon alone); the opening carries the two numbers the story turns on (688 shootings, 44% more felony assaults); the street-stop counts move from the murder section to the belief they test; the strangulation test is called one test of the classification, not a fixed definition; a home named as a place, not a relationship; the close states what the records settle and what they leave open instead of restating the opening, under a heading of its own. After the editor's pass on this version: a shooting is an incident in which a bullet struck at least one person, not a count of people shot; the under-18 comparison names its baseline (2019) and its fall from the 2022 peak, which could still be chance; felony assault is defined by its principal kinds, without claiming every kind needs an injury; the per-resident comparison says it ends in 2024 and carries the estimate-vintage caveat beside it; the line without strangulation is called an estimate joining two sources, above every year after 2000 and below 2000; the Commissioner's "revolving door" moves beside the bail-law belief it bears on; the close no longer says the misdemeanour count rules out reclassification or that the department publishes no case records; the homepage paragraph rewritten to match.
2. 2026-10-14 · v1.2 · Update — Rewritten as one story, with the same sources, figures and charts (owner, 14 October): New York can set a shooting record and still record far more felony assaults than before the pandemic; the two counts have parted since 2020, and a record for gun violence is not a verdict on all violence. The headline states the finding; the dek, the In-brief box and the scope box are cut to the house limits; the hook and this year's counts are said once; the roadmap sentence is gone; the three-counts chart moves from the questions into the story as its close, and the last section no longer restates the In-brief; one-sentence paragraphs are merged. The story now carries the history the counts sit in, from the City's and the State's own documents, each a registered source with a saved copy and a checked quote: the strangulation law of 2010 (a class D felony, in force ninety days after signing), the bail law in force from 1 January 2020 (Part JJJ of the 2019 budget bill), the NYCLU's tally of the NYPD's stops (685,724 in 2011, 13,459 in 2019), the Blueprint to End Gun Violence of January 2022 (more than 6,000 guns removed in 2021), the NYPD's year-end release of 6 January 2026 (felony assault up 0.4% in its preliminary count, 'largely driven by assaults on public sector employees and domestic violence', public employees up roughly 25%, domestic violence 41%) and the Police Commissioner's briefings of January 2025 and 2026. None of their figures is set against a count from the files. A new section tests four things said about the violence on the files alone: shootings from January to May 2020 ran 18% above 2019, a difference that could still be chance, and from June to December at more than double (1,209 against 504); shootings fell 49% from 2011 to 2019 while the stops fell 98%; the complaint files' two public-employee codes rose 5% in 2025 (2,960 to 3,114), a rise that could still be chance, where the release says roughly 25%; and all felony-assault complaints ran 1% above 2024, which could still be chance. A new question checks the department's own readings against the records. Explanations are attributed (the NYPD's and the Commissioner's) and the counts are said not to test them; where a department figure is set beside the files (its roughly 25% against the files' 5%), the page says the two are not like for like. After an editor's outside pass on the first draft of this version: the thesis is the divergence of the recorded counts, not a claim about how the city has measured its safety; 2025 is the lowest complete year for shootings, not the safest year by every measure (2017 had fewer murders); no claim that laws and policing were constant; each belief's verdict is what the records can say (timing, not cause; not testable as stated); the Bronx's comparison year is named. The Bronx, the under-18 shootings and the gun's share of murders move to the questions. Every earlier qualifier stays: this year's 11% fall in shootings and the under-18 fall since 2022 could still be chance. No count moved.
3. 2026-10-14 · v1.1 · Correction — Correction, after an outside review. The felony-assault chart's second line, without strangulation, took the complaint files' strangulations away from the complaint files' own total, which runs a few hundred a year off the NYPD's yearly count drawn beside it, so in 2009 it sat above the total. It now takes them away from the yearly count, so the gap between the lines is the strangulations. Moved: 2025 without strangulation 25,089 (was 25,092), 2019 17,280 (was 17,480), and the rise since 2019 without strangulation 45% (was 44%). Every named change is now also rechecked with the extra swing its series shows, as a sensitivity; this year's 11% fall in shootings and the fall in shootings of people under eighteen since 2022 do not clear no change under it, so both are now called gaps that could still be chance. After the outside review of this version: the methods say the line without strangulation joins two sources (44% within the complaint files alone), that fatal shootings stand in for murders in the recheck, and that from 2013 the jails' complaints are counted apart; the first definition of felony assault names its other kinds; the code no longer computes a population change across Census estimate vintages or keeps the dropped 'killed in other ways' and 'bound' figures; the 2022 transit-worker law is quoted from its saved text. The year to date stays the weekly report of 4 October, as published: this is a correction, not a refresh. No other figure moved.
4. 2026-10-12 · v1.0 · Published — Published.

Read the article →

1. 2026-10-15 · v1.1 · Update — Told as one story, to the writing rules of 15 October: the spine is that last winter's record was not just the cold, with the survey and the inspectors pointing the same way and public housing the part the complaints leave out. The opening carries the new inspection rule as its hook; four explanations on offer (the cold, a handful of buildings, easier complaining, the 2017 overnight standard) are tested on the files in a section of their own, with the repeat-building chart; the inspectors' violations chart moves from the questions into the story; the close says what the records settle and what the 2026 survey will add. Every date, figure and quotation from a document is now read from its saved copy: the City's Rental Ripoff report and the Mayor's release of 16 July 2026, Gothamist's report of 21 September, Local Law 86 of 2017, HPD's Heat Sensors Program page, the Comptroller's 2025 update, NYCHA's releases, HPD's heat page and Int 0252. The record on the winter before (22%) and the violations against the pre-pandemic winters are rechecked against the swing of their series. No count, chart or source figure moved.
2. 2026-10-09 · v1.0 · Published — Published.

Read the article →

1. 2026-10-08 · v1.0.1 · Correction — Correction: the text under Figure 7 said that a few incidents without a borough were counted in the whole-city figure only. In fiscal 2024 to 2026 every incident in its averages has a borough, so the five boroughs together are the whole city, as the chart's own note already said. No number changed.
2. 2026-10-07 · v1.0 · Published — Published.

Read the article →

1. 2026-10-05 · v1.0 · Published — Published.

Read the article →

1. 2026-10-01 · v1.0 · Published — Published.

Read the article →

1. 2026-10-04 · v1.4 · Correction — Corrections, from a review of the published articles on the rebuilt code. The 1.3 correction had not reached the summary of the 1990 results at the end: working from home was 2.5% of workers at work in 1990, not 0.5%, and working in Manhattan 47.3% of those travelling to a workplace, not 47.5%, as the story and charts already said; the 1990 median on the national research index is $54,037, not $54,038. The borough medians in the questions are $85,000 in Manhattan, $61,000 on Staten Island, $55,000 in Brooklyn and $51,000 in Queens, not $87,000, $62,000, $56,000 and $52,000, which no output matched; 76% of employed residents work for a private employer, not 77%; information has the most graduates, with education and finance close behind, not information and finance; in 1990 health care was not much more than a third of its present size, not a third; in the monthly survey wholesale and retail trade is 9% of employed residents, not 10%, the shares are compared with 2025 rather than the same months of 2025, and the sample is about 570 to 720 people a month, not 650 to 750. The payroll count was unchanged on 4 October 2026: July is still the record and August preliminary, revised when September's figures are published, and the release of 30 September is no longer described as coming. The earnings chart's note now says, as the methods do, that 1990 earnings are deflated with the 1989 index. The note on rerunning the figures names the steps that make the template charts and the dashboard.
2. 2026-10-03 · v1.3 · Correction — Correction: in 1990 about one worker in forty usually worked from home, not one in two hundred, and in 2000 about one in thirty-five rather than one in two hundred. Our extract of the 1990 and 2000 census samples had swapped the codes for working at home and for travelling by other means, so it counted home workers as commuters with no travel time; the extract was fixed on 2 October 2026. The works-from-home line of every group in 1990 and 2000 rises with it, and the shares working in Manhattan and in the five boroughs in those years move by less than a point (47% in 1990 either way). Also corrected: one occupation code, other financial specialists, now counts with management, business and finance, which is 18.9% of employed residents in 2024 (18.8% before) and 17.4% in 2019 (17.3%).
3. 2026-09-28 · v1.2.1 · Correction — Corrections of wording: Manhattan's share of residents who travel to a workplace is almost, not exactly, what it was in 1990 (47% then, 46% in 2024, a difference beyond the margin); 1990 manufacturing is about 9% on the other reasonable mapping of the 1990 codes (8.8%), not 'no less than 9% on any other'; Figure 1's caption said professional services grew 'by half as much again', when they added about half as many jobs as health care. No figure changed.
4. 2026-09-28 · v1.2 · Update — Charts: the question on which sectors recovered from the pandemic now has a table of every sector's payroll month by month since January 2019, with its August 2026 count and change on August 2019, as Figure 14; the monthly survey's chart is now Figure 15. No figure changed.
5. 2026-09-27 · v1.1 · Update — Charts: Figure 9, what residents do in the monthly household survey, is now one row per group with a dot for each period, so the three periods are visible at once and the chart fits one screen; the bar chart moved into the questions at the end. No figure changed.
6. 2026-09-27 · v1.0 · Published — Published.

Read the article →

1. 2026-10-04 · v1.6 · Correction — Corrections, from a review of the published articles on the rebuilt code. Rent burden left out renters whose rent is 101% or more of their income, which the census codes as a computed ratio like any other: counted, 50% of renting households with a child under six pay 30% of income or more in rent, not 45% (the figure had also been per child rather than per household), and in the full results 50.3%, 55.6% and 51.1% for households with a child under six, with older children only and with none, not 43.6%, 48.5% and 44.3%; Figure 2 and the dashboard move with them, and such households are still rent-burdened about as often as households without children. The number of young children has been flat since 2022 in the Bureau's estimates and was about the same in the 2023 and 2024 surveys, so the count is no longer said to keep falling; births are. Children aged one to five leave at a higher rate than any group we measured but adults aged 25 to 44, where it said young adults, and the rate was higher in 2019 and 2024 than in 2015, where it said every year, which the chart does not show. Fathers are employed more than eight times in ten at every age of the youngest child, not nine in ten. The infant and toddler places and vouchers could serve about 13% of under-threes counting unfilled seats, and served about 12%. The first 2-K seats were offered for the fall of 2026; most of them, not all, are full-day. Smaller corrections: day care workers' margins of about 4,000 and 3,000, not 4,000 each; the record counts in the method note; the household-income sensitivity of up to about two points; at the Comptroller's family-care average the 7% share sits above both rates here, not between them; one 2021 reproduction row is inside the combined margin only; the kindergarten cohort is the smallest of the five years on the register. The enrolment chart's value axis now has a title. The note on rerunning the figures names the steps that make the template charts and the dashboard.
2. 2026-09-28 · v1.5 · Correction — Corrections: Staten Island, not Manhattan, has the fewest children under six (Manhattan has the lowest share of its residents); the parent at home was 36% of children under six in 2015, not 37%; the enrolment of three-year-olds between $25,000 and $50,000 is 50%, not 51%, and the clause that tied it to parents at home, which mixed two measures, is removed; in the older estimate series the Bronx's under-fives fell as fast as Manhattan's, so Manhattan fell fastest only in the newer one; the share of children with every parent in the labor force rose from under two thirds, 64%, not from six in ten; half of all new mothers worked full time, not half of the employed ones; one reproduction row, in 2021, is inside the combined margin but outside the published one alone, and we now say so. Staten Island's public enrolment of three-year-olds rests on too few records to report. Charts: 2020 is drawn without a margin, because the Bureau warns that the experimental replicate weights for that year understate the uncertainty. The analysis script's method note names the births source used since 1.1.
3. 2026-09-27 · v1.4 · Update — Charts: Figure 10, the Census Bureau's under-five estimates, is now two lines, one per estimate vintage, that both have a July 2020 point and are never joined, so the whole series fits one screen; the bar chart moved into the questions at the end. No figure changed.
4. 2026-09-27 · v1.3 · Update — Narrative, after an editorial review of how the series makes its point; no figure changed. The opening now says what the records add up to before laying out the evidence: fewer small children every year; families in which every parent is working or looking for work, nearly as often in a baby's first year; care before age three that fits few family budgets by the conventional yardstick, with the least public help of any age; and small children leaving at one of the highest rates of any group. The section headings state what each section finds, the In-brief box leads with that point, and the conclusion ends on it. Two new charts: mothers of young children out of the labor force by the income of the rest of the household, 2015 against 2024, drawn as a dumbbell so the fall at the top shows at a glance, with the household's whole income and the fathers as other views; and three- and four-year-olds' school enrolment drawn as a slope from 2015 to 2024. The bar chart of enrolment by year and borough moved into the questions at the end.
5. 2026-09-26 · v1.2 · Correction — Correction: where the one-to-five-year-olds who left in 2024 went was quoted beyond what the survey supports. Florida's share, given as about one in twenty (946 children), rests on six survey records, and the counts for Pennsylvania, the rest of the South and the Midwest on seven to fourteen, all under the thirty-record floor the charts use; the split of the suburban half between Long Island, the Lower Hudson Valley and northern New Jersey was given without margins. We now say that half went to the suburbs, on 78 of the 149 records, and that the rest are spread too thinly to rank; the full results keep each estimate with its record count. The chart of where the youngest leavers go moved from the FAQ on the cost of care to the FAQ on which children leave, where it belongs, and its caption no longer names Florida, which the chart itself leaves undrawn. Second correction: household and family incomes and earnings, which the census files already hold in 2024 dollars, were adjusted for inflation a second time, overstating 2015 figures by about 28% and 2024 figures by 1.5% (leavers' incomes were right). Corrected: a household with a child under six earned about $102,000 at the median in 2024, not $103,000, and one with no children $78,000, not $79,000; by borough $254,000 in Manhattan (not $258,000), $124,000 on Staten Island (not $126,000), $112,000 in Queens (not $113,000), $99,000 in Brooklyn (not $101,000) and $57,000 in the Bronx (not $58,000); in 2015 such a household earned $74,000 in 2024 dollars, not $96,000. Among mothers whose household earns more than $300,000 without her, 45% were out of the labor force in 2015, not 43% (on 143 records with a margin of about eight points, not 217 and six and a half), against 21% in 2024; the whole 2015 series up the bands is 28%, 45%, 37%, 27%, 33%, 41% and 45%, not 26%, 41%, 43%, 31%, 29%, 37% and 43%, and the 2024 series 21%, 38%, 34%, 23%, 23%, 22% and 21%, not 21%, 37%, 35%, 22%, 24%, 22% and 21%; by household income the second and third bands are 37% and 36%, not 36% and 37%. Infant centre care fits within 10% of family income for 20% of families, not 21%, and within 20% for 44%, not 45%; family day care for 17%, 27%, 52% and 65%, not 18%, 27%, 53% and 66%; the 7% test passes for 3% of Staten Island's families, not 4%; the median family income is $107,000, not $108,000. Preschool teachers in day care centres report $23 an hour, not $24; nannies $32,000 a year, not $33,000, in households with a median income of $71,000, not $72,000. The households of the small children who stayed have a median income of $92,000, not $94,000. The 7% result (13%), the nanny and day care medians of $17 and $14 an hour, and the halving of the at-home share among affluent mothers stand.
6. 2026-09-25 · v1.1 · Correction — Brought up to date to September 2026, with one correction. Correction: the births series was described as births to city residents, but the City Summary's Table PC1 counts every birth registered in the city, 12,701 of the 98,389 in 2023 to mothers who live elsewhere. It is replaced by the State Health Department's count of births to mothers living in the five boroughs, wherever the birth took place: 116,040 in 2015 and 90,869 in 2023, the same fall of a fifth. Added for 2025 and 2026: the City's provisional birth counts to March 2026 (fewer in 2025 than in 2024, and the fewest of any first quarter on the provisional record in January to March 2026), and the monthly Current Population Survey to August 2026, which finds about seven in ten mothers of a child under six in the labor force, no detectable change from 2024 or 2025. The summary and the In-brief box open with these. No census figure changed: the Census Bureau has postponed its 2025 one-year release.
7. 2026-09-25 · v1.0 · Published — Published.

Read the article →

1. 2026-10-05 · v1.7 · Update — Brought up to date to 3 October 2026. The shelter census now runs to 3 October rather than 24 September: September is complete, still about 84,000 people, and the first days of October are in the chart, whose title and subtitle name the census's last day where the title said September. No other figure moved.
2. 2026-10-04 · v1.6 · Correction — Corrections, from a review of the published articles on the rebuilt code. The census codes for Bermuda and Canada had been counted with the Caribbean, Mexico and Central America: by residence a year ago that group was 19% of arrivals from abroad in 2015 and 20% in 2024, not 22% and 23%, with Canada, Oceania and elsewhere at 6% and 4% and Europe ahead of the group in 2015, so Figures 6 and 12 now say Asia leads in each of the three years shown rather than Asia and the Caribbean in every year. A suburban leaver who works from home has the suburb as a workplace, so the share still working in the city is now stated among those who do not work from home: 55% in 2024, about the same as the 57% of 2019, where it was 47%, down from 55%, among all the employed. The Comptroller's 1% for filers above $500,000 is the share who left the state, not a net rate. Population-estimate comparisons now stay inside one series: net international migration rose from about 20,000 in the year to July 2021 to 220,000 in the year to July 2024 and fell back to 66,000, its lowest since 2021, where the text had set those figures against the 2010s of an earlier series. Also corrected: the shelter series starts in March 2021, not January; Florida takes one leaver in thirteen, not fourteen; Figure 2 covers the three years it shows, not every year; the monthly survey's margins are under one and a half points, not one to one and a half. Two figures that rested on no saved source were removed, a 2017 revision of 184,000 and the number of movers to Puerto Rico. The note on rerunning the figures names the steps that make the template charts and the dashboard.
3. 2026-09-28 · v1.5 · Correction — Corrections: South America and Africa went from about a sixth of arrivals from abroad, not a seventh (the 1.4 correction had reached the captions but not the text); in 2015 arrivals from abroad slightly outnumbered those from other states, so Figure 2's 'other states first in every year' is withdrawn; the 23% of leavers who usually worked from home is a share of employed leavers, not of all leavers; Florida is the third destination by state, not the seventh; the 2024 net domestic loss is within the survey's margin of 2015's rather than almost exactly its size, and it is departures, not both flows, that were back where they started; the largest net loss and the highest leave rate are for racial and ethnic groups, not for any group, and within the margins; the suburbs took about as many as the other states combined in 2024, not more; in 2019 only the degree share of arrivals from abroad was higher, not their employment share; the rise in departures reached most bands, not every one; the pandemic exception ran from 2020 to 2022; the ledger's domestic figure usually, not always, runs below the census's. Charts: 2020 is drawn without a margin, because the Bureau warns that the experimental replicate weights for that year understate the uncertainty; Figure 9's caption no longer describes a year the chart does not draw, and Figure 13's describes 2025 correctly. Small rounding fixes.
4. 2026-09-27 · v1.4 · Correction — Correction: Florida's 22,000 leavers in 2024 are no longer described as fewer than in 2015; the difference from 2015's 28,000 is inside the two years' combined margin, so we now say much the same. The arrivals-from-abroad caption now says South America and Africa were about a sixth of arrivals in 2015 (16%), not a seventh. Charts: Figures 3, 4 and 6 (destinations, net migration by group, arrivals from abroad) are now one row per group with a dot for each year, so each fits one screen; the bar charts moved into the questions at the end.
5. 2026-09-27 · v1.3 · Update — Narrative, after an editorial review of how the series makes its point; no figure changed. The opening now says what the flows add up to before laying out the evidence: less an exodus than a steady exchange that ends in a loss to the rest of the country every year, so that births and above all immigration decide whether the city grows. The section headings state what each section finds, and the conclusion follows the Census Bureau's ledger to that point: the city grew through the year to July 2016 and in the years to July 2023 and 2024, when births and immigration outweighed the domestic loss, and shrank in the other years. Two new charts: the ledger drawn as one column per year, the components above and below zero with the year's change marked, in place of the chart that showed one component at a time; and arrivals, leavers and people who did not move drawn as dots on one line per trait, in place of the 48-bar profile chart. Both bar charts moved into the questions at the end.
6. 2026-09-26 · v1.2 · Correction — Corrections and wording, after an outside review of the whole site. Residents' household incomes, which the census files already hold in 2024 dollars, were adjusted for inflation a second time while leavers' were adjusted once, overstating residents' incomes by about 28% in 2015, 21% in 2019 and 1.5% in 2024 and tilting every comparison of arrivals or stayers with leavers by household income. Corrected, the median domestic arrival's household earns about $148,000 rather than $150,000 ($148,125, not $150,384), the median stayer's $94,418 rather than $95,858 and the median arrival from abroad's $79,697 rather than $80,912; the net domestic flow up the seven household-income bands is −14,000, −21,000, −30,000, −23,000, −5,000, −9,000 and −13,000, not −14,000, −21,000, −31,000, −22,000, −6,000, −8,000 and −13,000; and the out-migration rate among people now in households above $300,000 is 4.6%, not 4.5%. The 2015 and 2019 income views in the charts and the dashboard moved more, since those years had been overstated by a quarter and a fifth. No count of people, and no figure by age, degree, race or nativity, changed. We had said the city gains 18-to-24-year-olds from the rest of the country; their net balance is inside its margin of error in every benchmark year (+1,900 ± 7,200 in 2024, +7,500 ± 7,540 in 2019, −3,100 ± 8,100 in 2015), so they are now described as about even, with the margins in the FAQ, and the 2019 figure is no longer given as a gain of 8,000. The FAQ said an average staying tax return reported about $100,000 in 2021–22; it was $118,000, so the gap with leaving returns had narrowed to about $22,000 before 2022–23. That last year is the first on a revised IRS matching method, which the IRS calls a new series, and we now say so wherever we compare it with earlier years; the gap is no longer said to have mostly closed. The tax returns' income is described as the income of the year of the move, from the second return, not income from before the move. The monthly survey's share of residents born abroad and the shelter census are reported as separate measures, not as two readings of the foreign-born population. Departures rose by half during the pandemic rather than the pandemic raising them, and the conclusion no longer says the exchange makes the city younger and more graduate.
7. 2026-09-25 · v1.1 · Update — Brought up to date to September 2026. The summary and the In-brief box now open with the latest year the Census Bureau has estimated, the year to July 2025, and say what the two series that run into 2026 show. The shelter census runs to 24 September 2026 rather than 9 September (about 84,000 people in September, unchanged). The note on the census's 2025 file now says why it is late: the Bureau postponed the release in August 2026 while it adapts it to a Commerce Department order on disclosure protection. The FAQ on the March supplement now covers the 2026 file, released on 15 September (about thirty movers among 2,200 city residents, still too few), and gives the 2025 count as about twenty-five rather than about twenty. No census figure changed; the IRS flows for 2023-24 have still not been posted.
8. 2026-09-17 · v1.0 · Published — Published. Before release the narrative was compressed to five sections with about two figures a paragraph; the year-by-year flows, the origin and destination detail, the departure rates with their chart, and the ledger detail moved to the FAQ.

Read the article →

1. 2026-10-05 · v1.12 · Update — Brought up to date to 30 September 2026. The MTA counts now run to 30 September rather than 23 September. In 2026 the subway's Tuesday-to-Thursday days run at 77 to 80% of 2019 rather than 77 to 79%, weekends at 88% rather than 88 to 89%, and a midweek day carries about 4.22 million trips rather than 4.21 million (3.87 million on a Friday rather than 3.86 million); Metro-North is at 85 to 89% midweek rather than 85 to 88%, and at 75% on Mondays rather than 74%. The weekday subway share (75%), Monday (69%), Friday (71%) and the midweek plateau (78%) are unchanged. The file's newest day, 1 October, arrived with its subway and bus counts still blank and is left out. The office-week chart names the last day of data rather than its month.
2. 2026-10-04 · v1.11 · Correction — Corrections, from a review of the published articles on the rebuilt code. The monthly survey's groups by employer had counted the unincorporated self-employed as private employees and left the self-employed row empty: private employees are now 25% at home in 2026, not 26%, and the self-employed, 29%, are back in Figures 8 and 15. Claims the margins do not support are withdrawn: the racial gap stays at about eight or nine points once the degree or the industry is held fixed, where it was said to narrow; the job mix explains part of the Bronx's gap and of the gradients by pay and borough, not most of them; the Bronx and Staten Island are the lowest boroughs together; the bus and, in the office-based industries, walking held about the same; the number of resident workers was about the same in 2019 and 2024, not slightly higher; three occupation groups rose by about 17 points each, among the largest rises rather than the largest. The turnstile counts gained about six points of their 2019 level from 2024 to 2025 and about two in the same months of 2026, where a part year had been set against whole years as five and one. Also corrected: the 2024 earnings bands run from 9.2%, not 10.4%; the monthly survey's samples are about 650 people at work, not 1,500, and its shares are of New Yorkers at work in the survey week; the turnstile figures are trips, not riders; Suffolk is inside the census study area; the employer surveys are dated (March and December 2025); the reproduction covers the years checked, 2019 and 2021 to 2024; the telework question has been asked since October 2022 and in the public files since June 2024. The pay chart's value axis now has a title. The note on rerunning the figures names the steps that make the template charts and the dashboard.
3. 2026-09-28 · v1.10 · Correction — Corrections. The census share at home fell by 0.9 points from 2023 to 2024, which is beyond the margin, not inside it: the share has all but levelled off rather than held flat. The monthly survey's approximate margins were too narrow, because neighbouring months share most of their sample; they now carry a floor for that, and with it no group's 2025-to-2026 change, government employees' included, is beyond the margin, so 'only government workers moved' is withdrawn. Inside the office-based industries the borough gap closes for Brooklyn and Queens but not for Staten Island (16%) as well as the Bronx. White workers are at home at 17%, not 18%; the gaps are 8.9 and 7.9 points, not 9.0 and 7.8. The degree gap between remote workers and commuters already existed in 2019 and has widened; only the pay gap is new. The subway accounts for about three quarters of the fall in commuting, not nearly all of it. The employer survey and the badge index measure different things. 2020 is drawn without a margin: the Bureau's experimental replicate weights for that year understate the uncertainty. Smaller wording fixes throughout.
4. 2026-09-28 · v1.9 · Update — Charts: Figure 9 is now a calendar of every day since January 2023, each against the 2019 average for its day of the week, for the subway and the two commuter railroads; the bar chart by day of the week moved into the questions at the end as Figure 16. No figure changed.
5. 2026-09-27 · v1.8 · Update — Charts: Figures 3, 7 and 8 redrawn to fit one screen. Working from home by occupation is now one row per group from 2019 to 2024; the monthly survey is now lines for any, some and all hours at home with the national share beside them; the 2025-to-2026 comparison by group is now one row per group, grey where the change is inside the approximate margin. The bar charts moved into the questions at the end. No figure changed.
6. 2026-09-27 · v1.7 · Update — Narrative, after an editorial review of how the series makes its point; no figure changed. The opening now says what the census shows before laying out the evidence: working from home has settled rather than faded, and it has settled on the better paid. The section headings state what each section finds and the sections are linked. Two new charts: the pay gradient drawn as a slope from 2019 to 2024, in which every earnings band starts near one worker in twenty and the lines fan out; and a map of working from home, working in Manhattan and median earnings across the city's 55 census neighbourhoods in 2024, from the 2024 one-year records with their margins. The paired-bar earnings chart and the chart of who works from home by race, degree, sex, age, nativity and class of worker moved into the questions at the end.
7. 2026-09-26 · v1.6 · Correction — Correction and wording, after an outside review of the whole site. Correction: the census files we read were already in 2024 dollars, and the analysis adjusted earnings for inflation a second time, which overstated 2019 earnings by about a fifth and 2024 earnings by 1.5%. Corrected, the median worker at home in 2024 earns about $73,000 rather than $74,000 and the median commuter $53,000 rather than $54,000 ($73,098 and $52,793 in the methods note, not $74,213 and $53,598); in 2019 the three medians were about $55,000, not $66,000 ($54,600, not $66,300); in 2019 about a quarter of remote workers and of commuters were paid $100,000 or more, not 35% and 33%; the 2019 shares at home up the earnings bands are 5.5%, 4.0%, 4.5%, 4.6% and 5.0%, not 5.5%, 4.1%, 4.1%, 4.9% and 4.9%, within 1.5 points of each other rather than 1.4; 37% of remote workers earn $100,000 or more in 2024, not 38%; a Bronx resident earning $100,000 to $150,000 is at home at 11%, not 12%; and the office-industry bands are 20.3%, 16.3%, 22.5%, 24.1% and 20.8%, not 20.2%, 16.3%, 22.3%, 24.3% and 20.9%. The finding stands: in 2019 working from home did not rise with pay, and in 2024 it does. Wording: the census and the monthly survey are fresh samples each year and follow no one, so sentences that described particular people moving home or stopping going to the office now describe who works from home in each year and the groups where it rose most: 'Who moved home is not random' became 'Who works from home now is not random', the second section's heading and the people chart's title changed with it, and the conclusion describes the workers now most likely to work from home rather than 'the people who stopped going'. The opening no longer says the gap between the census's one in eight and the monthly survey's quarter is the hybrid worker; much of it is consistent with hybrid workers, alongside differences between the two surveys. Four FAQ questions were reworded the same way.
8. 2026-09-25 · v1.5 · Correction — Brought up to date to September 2026. The summary and the In-brief box now open with the 2026 monthly-survey figures (a quarter with some work at home, one in ten with all of it) and then the census's 2024 one in eight. The MTA counts now run to 23 September 2026: the midweek plateau and the Monday and Friday figures are unchanged at the rounding shown; in the methods note the midweek count is 4.21 million rather than 4.19 million, the Long Island Rail Road 92% rather than 91% on Wednesdays and Thursdays, and bridge and tunnel crossings 98 to 104% rather than 95 to 105% (the MTA has revised its bridge and tunnel counts back to 2020). Correction: the 2022 and 2023 midweek plateau figures in the methods note, computed as 60.6% and 69.3%, had been printed as 60% and 70%; they are now 61% and 69%. Manhattan office vacancy is given at the end of June 2026 (13.1%, unchanged from March). The conclusion said the turnstile counts had crept up a point or two a year; weekday ridership in fact rose about five points of its 2019 level from 2024 to 2025 and about one from 2025 to 2026, and it now says so. The monthly survey still ends in August 2026, the latest month released, and no census figure changed: the Census Bureau has postponed its 2025 one-year release.
9. 2026-09-17 · v1.4 · Update — Tightened for readability: the narrative is told in five sections, with fewer figures in each paragraph; the borough chart, the band-by-band and occupation detail, the Manhattan commuter breakdown, office vacancy and the employer surveys moved into the FAQ. The summary shown on the homepage was shortened the same way. No number changed.
10. 2026-09-11 · v1.3 · Update — After a fourth outside review: the two headline measures (12.5% usually at home, about a quarter with any hours at home) are set side by side in the opening; the Manhattan commuter change is described as a change within the census study area rather than a complete count; the turnstile week is described as consistent with a hybrid office week rather than a measurement of it.
11. 2026-09-10 · v1.2 · Update — After a third outside review: four sentences that read as causes are now descriptions (which occupations moved home, the Bronx's jobs, pay and remote work, the subway's share). A reconciliation table against the published Census tables is shown in the Reproduce box.
12. 2026-09-10 · v1.1 · Update — After an outside audit: the conclusion presents the Manhattan commuter count, office vacancy and the turnstile week as three separate measurements rather than one causal chain; 'stopped falling' became 'levelled off' (the 2023 to 2024 change is inside the margin); the residual racial gap is no longer attributed to 'the jobs'; the glossary tooltip for office-based industries said a fifth of workers where the story says 30%. Evidence-tier badges on every chart and an approximate margin on the dashboard.
13. 2026-09-09 · v1.0 · Published — Published.

Read the article →

1. 2026-10-05 · v1.13 · Update — Brought up to date to 30 September 2026. The MTA ridership counts now run to 30 September rather than 23 September, and Figure 8's small multiples add September. Weekend subway ridership is 88% of 2019 rather than 89%; the weekday figures are unchanged (subway 75%, buses 61%, about 4.0 million trips a day). The file's newest day, 1 October, arrived with its subway and bus counts still blank and is left out. Figure 15 and its 2026 bar name the last day of data rather than its month.
2. 2026-10-04 · v1.12 · Correction — Corrections, from a review of the published articles on the rebuilt code. The MTA's ridership figures count trips, not people, so the 2026 subway figure is about 4.0 million weekday trips against 5.4 million, and the weekday transit trip is made about a quarter less often than in 2019, rather than a quarter fewer riders. The bus speeds and subway delays run to August 2026, not September. In 2017 the subway carried close to its largest share of commuters, not a record share: 46.5%, against 46.7% in 2018. Buses have been back at about their 2019 speed since 2023, not 2024. The subway riders' share over the hour edged down from 1990, rather than barely moving. At 45 minutes the boroughs' order changes, so the claim that the 90-minute order holds there is withdrawn; the MTA's figures are no longer called the only public data on the time since congestion pricing; the dashboard does carry an approximate margin on its headline share. Figure 6 now has its own image for readers without scripts, where it showed Figure 5's, and Figure 8's value axis has a title. The note on rerunning the figures names the steps that make the template charts and the dashboard, and the analysis script's notes give the 1990 and 2000 samples' real size, 120,000 to 130,000 records.
3. 2026-10-04 · v1.11 · Correction — Correction: the 1990 and 2000 census points are recomputed after a fix to how those two censuses' means-of-transportation codes were read. People who worked at home had been counted as commuters with a zero-minute trip, and the few who commuted by other means had been counted as working from home. The hour-plus share was 22.8% in 1990 and 24.5% in 2000, not 22.3% and 23.8% (23% and 25% where rounded, rather than 22% and 24%); the average trip was 36.5 minutes in 1990 and 40.0 in 2000, not 35.7 and 38.8, so today's trip is four minutes longer than in 1990 rather than five; working from home was 2.5% of workers in 1990 and 2.9% in 2000, not 0.5%, about one worker in forty rather than almost nothing; the subway carried 37.5% of commuters in 1990 and 38.8% in 2000, not 36.7% and 37.9%, and the car 33.6% and 33.9%, not 33.0% and 33.1%. In Figure 7 the boroughs' 1990 and 2000 points move by less than a point, and their average trips by up to two minutes. The hour-plus shares by mode, every year from 2005 on and the conclusions are unchanged.
4. 2026-09-28 · v1.10 · Correction — Corrections: filtered to one mode, the racial gap widens on the subway, reverses for bus riders and disappears among drivers, and inside a borough it mostly stays (not on Staten Island), where Figure 5's caption said it narrows but never closes; the longest commutes are construction, cleaning, protective service and production, and the shortest law, education and the arts, where Figure 12's caption named health support and management; drivers are over the hour at one in five or fewer in every borough (Brooklyn is 20%), not fewer than one in five; in the adjusted model legal work (about −5 points) and the arts (about −4) sit further from office work than 'a couple of points', and the $100,000 to $150,000 band keeps about 2 points, so earnings sit beside race rather than disappearing; the share over the hour already dips at $100,000 to $150,000, not only above $150,000; the census series runs thirty-four years, 1990 to 2024, not thirty-six; the five-year file holds 151,386 person records, not 155,257. Charts: 2020 is drawn without a margin, because the Bureau warns that the experimental replicate weights for that year understate the uncertainty. Smaller wording and rounding fixes, and a self-reference removed.
5. 2026-09-28 · v1.9 · Update — Charts: Figure 8 now shows every weekday month since March 2020 against the same month of 2019, in five small panels, one per MTA service, on one scale; the bar chart by year, with its weekend view, moved into the questions at the end as Figure 15. No figure changed.
6. 2026-09-27 · v1.8 · Update — Charts: Figure 7, thirty-six years of the hour-plus share, is now one line per borough and one for the whole city, so the series and the boroughs' order fit one screen; in every year Manhattan's residents were over the hour about one time in ten and the other four boroughs' one time in four or more. The bar chart moved into the questions at the end, with working from home and the mode shares by year. No figure changed.
7. 2026-09-27 · v1.7 · Update — Narrative, after an editorial review of how the series makes its point; no figure changed. The opening now says what the census shows before laying out the evidence: the hour-long commute is a matter of address and vehicle far more than pay, and has been for a generation. The section headings state what each section finds, the sections are linked, and the conclusion ends on what the pattern implies rather than restating the findings. New Figure 1, a map of the city's 55 census neighbourhoods showing the share of commuters over the hour, the average trip, the share riding the subway or a bus and the share working in Manhattan, from the same 2020-24 records with their margins: from about one commuter in twenty over the hour in lower and midtown Manhattan to nearly four in ten in Canarsie and Flatlands. The chart of travel-time bands moved into the questions at the end.
8. 2026-09-26 · v1.6 · Update — Wording, after an outside review of the whole site; no number changed. The summary and the model section no longer say the racial gap survives the adjustments 'the census variables allow' or 'everything the census records': the model holds fixed the characteristics it includes, and about half the gap remains after them. The In-brief box says the MTA figures show which modes and days have recovered rather than which riders came back, since they count trips, not people. Sentences that read as following the same people over time (short-trip commuters taken out of the count, the people at the ends of the lines who kept riding) now describe who commutes in each year. The FAQ on working from home no longer suggests a more punctual subway shortened the remaining trips: the MTA's delay measure was about the same in 2019 and 2024.
9. 2026-09-25 · v1.5 · Update — Brought up to date to September 2026. The MTA ridership counts now run to 23 September 2026 (weekday buses at 61% of 2019 rather than 60%, the weekend subway at 89% rather than 88%), and two MTA operating series are added to the 2026 section: weekday bus speeds from 2019 and the share of subway trips more than five minutes behind schedule from 2017, both to August 2026, so the section says whether the trip got any quicker as well as how many ride. No census figure changed: the Census Bureau has postponed its 2025 one-year release, so 2024 is still the latest census year.
10. 2026-09-17 · v1.4 · Update — Tightened for readability: the narrative is told in five sections (the adjusted model now sits with the gaps it adjusts), with fewer figures in each paragraph; the occupation chart and the band-by-band, borough-by-borough detail moved into the FAQ. The summary shown on the homepage was shortened the same way. No number changed.
11. 2026-09-11 · v1.3 · Update — After a fourth outside review: the model section is titled by what it does, adjusting for the variables the census records, rather than 'holding everything else fixed'; the one reproduction miss (the five-year commuter count) is explained beside the reconciliation table; the FAQ reports the borough, race and mode patterns at 90 minutes as well as 60, since answers heap on the hour.
12. 2026-09-10 · v1.2 · Update — The summary no longer says the racial gap survives 'every filter the census allows'; it survives the adjustments the census variables allow here. A reconciliation table against the published Census tables is shown in the Reproduce box.
13. 2026-09-10 · v1.1 · Update — After an outside audit: the adjusted model's attenuation is described as a descriptive adjustment, not a decomposition of the racial gap into causes; bridge-and-tunnel figures are described as vehicle crossings, not returning drivers. Evidence-tier badges on every chart and an approximate margin on the dashboard.
14. 2026-09-07 · v1.0 · Published — Published.

Read the article →

1. 2026-10-08 · v1.10.1 · Correction — Correction: the notes under the two rent-and-pay index charts called the indexes administrative; they are BLS survey series, the consumer price index for rent and the payroll survey's average hourly earnings, as the charts' badges say, published without margins. No number changed.
2. 2026-10-04 · v1.10 · Correction — Corrections, from a review of the published articles on the rebuilt code. The burdened six-figure household's effective tax rate averages 28%, not 29%. Under the official measure one in nine six-figure families and people living alone is over the 30% line (11.6%), not one in eight, and the group is named as that, families and people living alone, where it had been called families. The move-in staircase falls in order except its last two steps, which are within the margin of each other, and holds in every income band and in each borough with records enough to tell (Staten Island has too few). The move-in gradient is called one of the steepest, as the methods say, not the largest; owners with a mortgage cross the line about twice as often as renters, or more, in every band. The tax estimate gives the child tax credit to related children under 18, a year above the credit's own limit, and does not claim the credit for other dependents, where it had said it did. The rent and pay indexes come from surveys, not administrative records, and carry that badge. The reproduction table is the synthetic population's check, not the published-table reconciliation, which is in the Reproduce box. The 2023 survey's rent medians are credited to the reports they come from, and the proxy for the stabilised stock no longer counts a boat, RV or van as a building of five or more units, which moves no printed figure. The note on rerunning the figures names the steps that make the template charts and the dashboard.
3. 2026-09-28 · v1.9 · Correction — Corrections: the newest arrivals do not have the highest median income of the move-in steps (households of one to two years do, $178,986 against $168,138), so we now say incomes are highest in the first four years; the median householder of families and singles is 42, not 45; 62.7% of city renters have less left after tax and rent than the median burdened six-figure family, not 62.8%, and the single person's figure is $40,100 a head, not $40,200; the complex households fall short of the budget because they more often have children and lower incomes against larger family budgets, not because each extra adult draws a full adult's budget, which the budget does only for roommates. Wording: the worked tax example is described as medians that need not add up; the 2024 check is said to hold at a higher overall rate rather than to be sharper; owners cross the line more than twice as often as renters in every band; it is the census's 2025 release that was postponed; the population check covers fourteen statistics, eleven of them held-out burden shares; the move-in gradient is described as one of the strongest, as steep as income's across the bands used here. The analysis script's method notes were brought up to date, and it now stores the split by total income and the householder ages the text quotes.
4. 2026-09-28 · v1.8 · Update — Charts: Figure 2 now draws rent and pay as lines indexed to a base year the reader chooses, 2024 or 2019; the paired bar chart moved into the questions at the end as Figure 9. No figure changed.
5. 2026-09-27 · v1.7 · Correction — Narrative, after an editorial review of how the series makes its point, and one correction. The opening now says what the census records show before laying out the evidence: the salary is a poor guide to who is stretched; the six-figure households most often over the line are roommates adding up smaller incomes and people who moved in recently; and once tax and a basic budget are counted, families with children join them, most of them from under the official line. The sections are linked, and the conclusion makes that one point instead of listing three findings. Correction: the budget section said nearly all of the households the budget flags but the official count does not have children; among families and people living alone about two in three do, and more than half with the budget moved ten percent either way (about half across all six-figure households), so it now says most. No chart changed.
6. 2026-09-26 · v1.6 · Update — Wording, after an outside review of the whole site; no number changed. The claim that children are 'the biggest dividing line of all' now carries its frame, the living-cost budget derived here from the City's measure, and says one of the sharpest, since roommate households fall short about as often as families with children. The FAQ on why six-figure households rent no longer says the salary does not buy a home here; a citywide price-to-income ratio is context, and whether a household can buy depends on its savings, debts, the mortgage rate and the home. The reproduction is described as landing well inside the published margin rather than within a few hundred households, since the difference is 855. The summary and the homepage text no longer say three things explain the burden; the answer turns on them.
7. 2026-09-25 · v1.5 · Update — Brought up to date to September 2026. The census's latest year is still 2024, because the Census Bureau has postponed its 2025 one-year release, so the counts are unchanged; what is new is how the two sides of the ratio have moved since. A new chart and paragraph set the rents New York-area tenants pay (the consumer price index, up about 9% between the first eight months of 2024 and the same months of 2026) against hourly pay in the city's private-sector jobs (up about 7%), and the move-in section adds the Rent Guidelines Board's freeze of stabilised renewals from October 2026, after 3% and 4.5% the year before. A new FAQ entry gives the detail; the Comptroller's asking rent is dated January 2024.
8. 2026-09-17 · v1.4 · Update — Tightened for readability: the narrative is told in five sections (the after-tax and budget sections are now one), with fewer figures in each paragraph; the sensitivity ranges and supporting detail moved into the FAQ. The summary shown on the homepage was shortened the same way. No number changed.
9. 2026-09-11 · v1.3 · Update — After a fourth outside review: the headline count is introduced as a 2020–24 five-year figure, with the 2024 single-year figure pointed to; the budget result is phrased as falling short of the TCOL-derived budget used here rather than being unable to cover an ordinary budget, in the executive summary, the story and the chart caption.
10. 2026-09-10 · v1.2 · Correction — Correction: the Housing and Vacancy Survey comparison used the survey's contract-rent medians ($1,641 citywide, $1,500 stabilised, $2,000 market) against the census's gross rent; it now uses the HVS gross-rent medians ($1,695, $1,570, $2,115) from the Rent Guidelines Board's 2024 study, and names both. The analysis script's description of shared households now matches the code, which splits them with the census's HHT2 item. A reconciliation table against the published Census table is shown in the Reproduce box.
11. 2026-09-10 · v1.1 · Update — New cover illustration; chart value labels moved past the margin ticks.
12. 2026-09-01 · v1.0 · Published — Published.

Read the article →