New York City's mayoral turnout nearly doubled in 2025, to 39 in every hundred adult citizens, the most for a mayor since 1993

• “A historic turnout of more than two million voters—the highest participation in a mayoral election in over half a century, dating back to 1969” (the Board’s 2025 annual report). Holds as a count: 2,218,647 ballots, the most in a mayoral race since 2,458,203 in 1969. Per adult citizen it was the most since 1993, for a mayor.
• The Board’s own percentage for last November, 45%, is ballots over the active voters on its rolls “as of 10/26/2025”; the table doesn’t say whether its earlier figures count the same way, so they don’t line up across the decades.

Because a percentage of registered voters can rise or fall when the rolls change, even if the ballot count doesn’t. The census’s estimate of citizens 18 and over is a steadier base, though it is wider than the eligible: the Board’s rules also ask, among other things, that a voter “Be a New York City resident for at least 30 days” and “Not be in prison for a felony conviction”, which the estimate can’t apply.

The ballots are the Board’s own, read from its reports, and the two printings of its table agree on every count of ballots and registered voters in every year they share (the percentages it prints are not used). The census’s published count of the city’s citizens 18 and over for 2024 comes out of the microdata within its margin, for the city and for each borough; the figures are in the check below.

The ballots are the Board’s counts, not a sample, so the sampling error is all in the census’s estimate of adult citizens. Last year’s rise from 2021, 18.7 ballots per hundred adult citizens, carries a margin of about 0.2 either way at the 90% level the Census Bureau uses, the two years’ estimates treated as independent samples and 2024’s taken for 2025. Sampling error in the survey is not a plausible explanation for a rise that size.

The yearly survey has a second limit. Each year’s file is weighted to the Census Bureau’s population estimates of that year, and the 2020 census revised the city’s upward: the 2018 and 2019 files sit about 5% below the estimates as since revised, so a rate before 2020 stands on a lower base than one after.

Put on the revised series, the 2018 governor’s race comes out at 37.5 per hundred and last year’s mayoral race at 38.8. We print the files’ own values and checked the comparisons the story makes across 2020 this way: the rebound of 2018 and the fall to 2022, last year’s race against 2018 and against the mayoral races of the 2010s, and the slide from 1994 (whose denominator comes from the censuses). All hold either way; “level with 2018” does not, so we don’t say it.

That margin is not the range given for 2025 in the story, which asks what the rate would be had the survey’s count of adult citizens moved in 2025 by as much as it did in any year since 2006.

The 1990 and 2000 censuses and the yearly survey from 2006 are the files we use; the years between take a straight line between the nearest measured years, an assumption. 2005 has a survey file, but the Census Bureau’s guidance says group quarters were “not included in the 2005 ACS”, so it is left out.

2020’s file is the Census Bureau’s experimental release, with no replicate margins. It sets the top of the presidential band, 56 per hundred; without it the band tops out at about 51, in 2008 and 2016 alike, and comparisons with 2020, such as the drop to the 2022 governor’s race, are point estimates with no margin.

2025 takes the 2024 estimate, also an assumption, because the Census Bureau says “the release date for the 2025 ACS 1-year estimates is being determined”, and its microdata folder still listed 2024 as the latest year when we checked on 18 October 2026. The largest one-year moves in the survey’s count since 2006, a rise of about 181,000 and a fall of about 113,000, give the range in the story.

Every ballot cast: scanned at a poll site on Election Day or in early voting, mailed, military and overseas, and affidavit ballots that were accepted. A ballot that leaves the top race blank still counts, so the votes for mayor, governor or president are lower than the turnout here, by amounts these totals don’t give. Years whose top race was not for mayor, governor or president are left out of every comparison.

The Board’s 2025 report says that in 2026 “voters will elect members of the New York State Legislature, the Governor”, among other offices. The City adds five questions from “the 2026 New York City Charter Revision Commission”: “there are five proposals on the ballot” (the Mayor’s office). The Board set 24 October as the last day to register for the election.

The Board’s table. The 2022 annual report’s “General Elections Turnout” table gives the registered voters, the turnout and a percentage for the mayoral elections of 1953, 1957, 1961 and 1965 and every November from 1969 to 2022; the odd years without a race for mayor, governor or president, which print no count to 2007 and a partial one after, are left out. It is read from the saved report and checked row by row against the 2006 report’s printing of the same table, whose counts agree in every year both carry; two typesetting slips in the 2022 printing (a period for a comma in 2014’s turnout, spaces inside 2012’s figures) are read as whole numbers. 2023, 2024 and 2025 come from each report’s November summary, whose borough figures add to the city’s and whose ballot kinds add to each borough’s turnout. The 2026 early-voting check-ins are the Board’s page as saved at the close of each day (this version reads the copy of 24 October, named in the results), and the earlier elections’ the Internet Archive’s copies of it, each registered with its date and checksum; each day’s boroughs add to the printed total. In the data file they sit in their own columns, cumulative and unofficial, never to be summed across days.

The rates. Ballots over adult citizens of the same year: citizens 18 and over from the 1990 and 2000 censuses (public-use samples, no margin calculated) and the American Community Survey’s one-year files from 2006 (margins from the replicate weights, which 2020’s experimental release lacks), for the city and each borough. The estimates keep the Census Bureau’s edited and filled-in citizenship answers, which the sampling margins don’t cover. The change from 2021 is the difference of two such rates, its margin the two estimates’ margins combined as independent samples. The one-basis check multiplies each year’s adult-citizen estimate by the ratio of the Census Bureau’s revised 1 July estimate of the city’s population to the file’s own total, which is to divide its rate by that ratio (nycdata.elections.population_controls): a sensitivity calculation that assumes the file’s share of adult citizens is unaffected, not a new estimate.

Code and checks. One script, analysis.py (linked below), produces every figure and chart, and checks each quoted sentence against the saved reports, the two printings of the Board’s table against each other, the summaries’ sums, the check-ins’ sums and the census counts against the published table; the build stops if any check fails. The dataset cards behind it (nyc_boe_turnout, census_history, acs1_pums) list the traps.