Why the actual state of crime might be worse than initially trumpeted
Each month, in a familiar ritual, the police commissioner announces the latest crime statistics — the standard way the NYPD has reported data since the de Blasio administration ended monthly in-person CompStat press conferences. On Monday, the department announced that major crime fell about 8% in February, with record lows for murders and shootings. The accompanying press release used restated February 2025 figures as its year-ago baseline — but those figures look nothing like the ones the department originally published. (New York calls these "major" or "index" crimes, echoing but not exactly mirroring the FBI's legacy classification of seven serious offenses: murder, rape, robbery, felony assault, burglary, grand larceny and grand larceny auto).
Those numbers were accurate as of March 3, 2025, when they were posted. But compare them to yesterday's press release, when the NYPD used that same month as its comparison baseline, and the numbers are different: Index crimes are now at 8,250, not 7,821; subway crimes grew from 135 to 162; and the month's murder count rose from 19 to 24. The February 2025 numbers increased by 5.5% overall and 20.0% for transit between the time they were first announced and their appearance in yesterday's release. A year ago, Commissioner Jessica Tisch celebrated "the third consecutive month of double-digit declines in nearly every major crime category" and credited the department’s subway safety plan with a 15.1% reduction in subway crime. What was initially reported as a “double-digit” decrease in subway crime is now a slight increase after the numbers settled.
This isn't a mistake, nor is it an artifact of delayed reporting or fraud. The NYPD classifies every crime by the date it was first reported, not the date it occurred. Report a burglary today that happened a month ago, and it lands on today's ledger.
What changes is the classification of those crimes — for example, whether the crime is a felony or misdemeanor assault, an attempted murder or a murder. Seriousness upgrades, downgrades or lateral reclassifications occur as complaints move through the department’s extensive auditing process. A hospital exam reveals a victim’s broken jaw, upgrading a misdemeanor assault to a felony; someone who reported a car stolen learns it was actually towed, removing the crime from the ledger.
A Vital City analysis of eight years of revisions reveals a consistent and one-directional pattern. New Yorkers, journalists and policymakers should treat the monthly figures that drive the city's public safety hot takes as a first draft, not a final verdict.
This is a national phenomenon. New York City just lets us measure it.
The reliability of crime statistics has become a national and partisan issue. During the 2024 presidential campaign, the two federal pillars of crime measurement— the FBI's Uniform Crime Reporting Program, which compiles crimes reported to police, and the Bureau of Justice Statistics' National Crime Victimization Survey, which asks households directly whether they've been victimized— told contradictory stories about whether violent crime was rising or falling. When the FBI quietly revised its 2022 figures, changing a reported 2.1% decline in violent crime to a 4.5% increase, it deepened suspicion on all sides. The Council on Criminal Justice noted that the FBI revised its data "without explanation," and called the absence of transparency a threat to public trust.
The FBI has since moved to monthly data releases with a built-in three-month lag, acknowledging that data needs time to “stabilize.” In effect, the FBI is trading timeliness for greater stability — a reasonable choice, but one that sacrifices the real-time figures that drive policy and politics. Cities like New York don’t wait three months. They announce crime numbers within days of a month closing, and those numbers are treated as facts by reporters, elected officials and the public, much the way the Bureau of Labor Statistics’ monthly jobs report drives headlines and markets.
New York gives us an unusual opportunity to study the phenomenon. The NYPD releases monthly crime data through press releases and updates its numbers continuously through weekly CompStat reports, the CompStat 2.0 website, and quarterly open data releases. Each new monthly release includes updated year-ago comparison numbers. By archiving these releases over time, it is possible to track how the same month’s crime count changes as it is reported, re-reported and eventually finalized.
Why the numbers don’t always match
Astute readers of Vital City’s year-end crime reports may notice that the figures cited there sometimes differ from the NYPD’s own published totals. This is not an error. It reflects the fact that the NYPD effectively publishes crime data through multiple channels that do not always agree: monthly press releases (the first and most widely reported), weekly CompStat pages, the CompStat 2.0 website, the City’s Open Data portal (updated quarterly) and historical CompStat tables (the numbers used in historical reporting).
Because each source captures data at a different point in the revision cycle, discrepancies are inevitable. The latest available press release for a given month will typically report higher crime counts than the historical end-of-year tables, because the press releases reflect the most recent reclassifications while the year-end tables appear to freeze counts closer to when the “books are closed” — meaning that late-arriving reclassifications for December, in particular, may not be captured.
Vital City’s analyses draw on whichever data source is most appropriate for the question at hand, and differences from NYPD-published totals generally reflect these timing gaps rather than any disagreement about the underlying facts.
The numbers almost always go up
Of the 95 monthly major crime totals examined, spanning 2018 through November 2025, every one was later revised upward between the initial press release and subsequent available data. None were revised downward. The average upward revision for total index crime was 2.7%, ranging from less than 1% to just over 6%. This is not random noise; it is a one-directional measurement artifact that systematically understates crime at the moment it is announced.
That translates to roughly 3,000 additional major crimes absent from the initial annual count. But recent months have run more than double that average, with revisions reaching 6%, suggesting the gap between what's announced and what actually happened is widening. And unlike polling error, which fluctuates randomly in both directions, these revisions moved only one way — upward — in every month examined. The initial numbers are the ones that make headlines; the corrections arrive quietly, if they are noticed at all.
This pattern has an institutional logic. There are asymmetric pressures at work in police crime classification. On one side, individual officers and precinct commanders face well-documented incentives to classify crimes at lower severity levels: Lower crime numbers mean better performance metrics, and felonies generate more paperwork, mandatory investigations and supervisory scrutiny than misdemeanors. On the other side, the department’s auditing apparatus exists precisely to catch and correct underclassification, because undercounting the seriousness of crime carries real consequences: delayed detective involvement, misallocated resources, and victims who don’t receive the investigative attention their cases warrant.
The auditing process tends to run in one direction because underestimating a crime's severity carries real costs, such as delayed detective response and misallocated resources, while overestimation is the less consequential error. The result is a system that reliably undercounts crime on first release and reliably adjusts upward over time.
What does this mean in practical terms? Under the more conservative historical rate, a reported 3% decline is likely to finish nearly flat; a 'flat' month is probably an increase of about 3%. Under the recent rate, the adjustments are larger: a reported 5% decline may land closer to flat, and a 'flat' month likely finishes up 4 to 5%. Either way, the direction of the revision is the same, and any announced decline of less than 5% is at risk of shrinking substantially or flipping signs entirely.
The problem occurs when the public conversation treats first-release numbers as final, when elected officials cite them at press conferences without caveat and nobody reads the 'preliminary' fine print at the bottom of the press release.
How a crime gets counted — and recounted
Better understanding this phenomenon requires a brief primer on the NYPD’s complaint process. When someone calls 911, walks into a precinct or even mails a letter (yes, it happens), a responding officer conducts an initial investigation and determines whether a crime has occurred. That officer classifies the offense, assigning both a crime type (such as murder, robbery, or larceny) and a severity level (such as felony or misdemeanor), and files a complaint report.
A supervisor reviews and signs off. The report enters the system within hours.
The process that follows is governed by the Crime Complaint Reporting System Guide, which sets out the rules for categorizing offenses." From there, the precinct's Crime Analysis Unit, a small team of civilian and sworn staff responsible for reviewing complaint reports, may upgrade or downgrade the classification, sometimes calling complainants back for additional information, usually within a day or two.
If the case warrants investigation, detectives take over. As new evidence or witness statements come in, detectives can request reclassifications through the precinct. In the ensuing weeks and months, the department’s Data Integrity Unit, which sits outside precinct command structures and is generally indifferent to local pressures, samples complaints for quality and sends them back to precincts for reclassification when errors are found. The Quality Assurance Division investigates specific complaints of misclassification.
It is sometimes assumed that prosecutors play a role in how crimes are classified in police statistics. They do not. Under the FBI’s Uniform Crime Reporting, crime statistics are collected and reported exclusively by law enforcement agencies. The system was designed to capture offenses known to the police, not charging decisions or court outcomes. The NYPD is no exception: District attorneys have no say in how the department classifies a crime complaint.
Auditing is what separates serious crime statistics from wish lists. The problem is that the original numbers (the ones that make headlines, shape narratives and inform policy decisions) carry an asterisk that nobody reads. There is also a practical cost: When a crime is initially classified at a lower severity than it deserves, it can delay the investigative response. Detectives may be notified late, witness memories fade and cases that warranted immediate attention get a slower start.
Not all crimes revise the same way
The direction and magnitude of revision varies dramatically by crime type, and the pattern reveals something about the mechanics of each offense.
Murder has the largest average revision: +13.5%. This is almost entirely driven by the Medical Examiner’s office. When someone is shot or seriously injured, the initial complaint is often classified as a felony assault or as “investigate aided,” NYPD shorthand for a case involving a person requiring medical attention whose outcome is still uncertain. If the victim dies days or weeks later, the crime is reclassified as a homicide, often too late for the initial CompStat report. Some death investigations take months to determine the cause. A month that initially reports 16 murders routinely closes at 18 to 20. The absolute numbers are small, but the percentage shift is large enough to change the story about whether murder is rising or falling in any given period.
Year-over-year murder comparisons can swing by 8 percentage points as the data matures, wide enough to flip the sign of the trend when comparing to prior years. When the department announces that “murder is down 10% this month,” the honest answer is often: We don’t know yet. (The department does track these pending cases internally and generally knows which deaths are likely to be reclassified, but that information doesn’t make it into the early press releases.)
A January 2025 press release illustrates the problem. At the time, the NYPD reported 25 murders for the month, a 24.1% drop from January 2024’s 33. By January 2026, the January 2025 figure had been revised to 30. The actual decline was 9%, not 24.1%. The headline number was directionally correct but dramatically overstated.
Felony assault revises upward by an average of 4.4%, and it does so with remarkable consistency. This is predictable undercounting, and it makes mechanical sense. The line turns on the victim's injuries: broken bones, loss of consciousness, or hospitalization can upgrade a misdemeanor to a felony once medical records arrive. And while 4.4% may sound modest, applied to the tens of thousands of assaults recorded annually, it represents a substantial number of violent crimes missing from the initial counts. Felony assault is the single largest contributor to total revision in 56 of the 95 months studied.
Grand larceny revises by a more modest 2.2%. But because it is the city's highest-volume index crime (roughly 48,000 cases a year, compared with 309 murders), it accounts for about a third of the total revision in absolute terms. Some of the reasons for this volatility include the nature of the crime: When a credit card is used fraudulently, the evidence of that fraud may arrive after the initial report. Grand larceny and felony assault together drive approximately two-thirds of the total revision each month.
Burglary revises upward by 5.0%. Some of this may be driven by retail theft cases where an arrestee turns out to have been previously banned from a store. Under New York law, burglary requires entering or remaining unlawfully in a building with intent to commit a crime. A shoplifter with no prior ban is simply trespassing or committing larceny, but once a store has formally excluded someone, any subsequent entry becomes unlawful, satisfying the burglary statute and upgrading the charge. What began as a larceny becomes a burglary, but only after investigators confirm the prior ban, a fact rarely known at the time of the initial report.
And then there is grand larceny auto, the only index crime that consistently revises downward, by about 1.0%. The reason is straightforward: Initial reports of stolen vehicles are sometimes reclassified once it turns out the car was towed by the city, the owner misremembered where it was parked or the complaint involves insurance fraud rather than theft.
The revision rate is increasing
Across all index crimes, the annual revision rate was relatively stable from 2018 to 2023, hovering between 1.7% and 3.3%. In 2024, the rate averaged 3.2%, but with a clear acceleration in the second half of the year, when monthly revisions regularly exceeded 3.5% and climbed as high as 4.4%. In 2025, the average monthly revision rate jumped to 4.5% — and because most of those months are still maturing, the final figure will be higher.
The monthly time series tells a more granular story. The revision rate was remarkably stable under most commissioners, averaging between 1.9% and 2.7% during the tenures of O'Neill, Shea, Sewell, Caban and Donlon. Under Tisch, the average has jumped to 4.7%, and because those months continue to mature, the final figure will climb further.
Transit crime is revising more than anything else
Transit crime numbers deserve special attention. Since 2022, subway crime has been a central focus of the department’s public messaging. Gov. Hochul secured $77 million last year for enhanced transit patrols, and transit crime figures have become among the most closely watched numbers the department publishes.
The NYPD is revising transit crime upward at rates that far outpace revisions for the city as a whole, and the gap is widening. Between 2018 and 2022, the department revised transit crime upward by a stable 2.2% per year on average, consistent with citywide crime numbers. In 2023, the revision rate started to diverge from other crime revisions, jumping to 4.6% while the overall revision in citywide index crimes was 2.3%. In 2024, transit crime revisions reached 8.7% while citywide index crimes were at 3.2%. For the 2025 months analyzed, the average transit crime revision has hit 11.2%, compared to 4.6% citywide.
This means that a headline claiming transit crime is “down 10%” in a given month, based on initial data, could finish flat or even up once the numbers stabilize. The February 2025 transit numbers are a case in point: reported as 135 crimes last February, it has since grown to 162, a 20.0% increase.
This analysis cannot say definitively why the transit revision rate has climbed so sharply. The pattern is consistent with several possible explanations: heightened political attention to subway safety may have increased auditing scrutiny on transit complaints; the influx of officers on transit patrols may mean more initial reports filed by personnel less familiar with transit crime classification or the same downclassification pressures documented in the Hyland case may be more widespread than a single district, with the department's auditing apparatus catching and correcting them after the fact. The data alone cannot distinguish among these explanations, and they are not mutually exclusive. What is clear is the practical implication: Transit crime numbers are now the least settled figures in the department's monthly releases.
A note on transit data integrity
One data point that may bear on the 2025 transit revision numbers: in March 2025, the NYPD stripped Captain Steven Hyland from his command at Transit District 20, which covers central and northern Queens, after an Internal Affairs audit found that he had allegedly directed subordinates to forge signatures and reclassify legitimate complaints to make his district's numbers look better. The department confirmed that a “handful” of reports had been “intentionally misclassified.” The department said the misclassified reports did not significantly affect Transit District 20’s overall numbers, in spite of the overall increase in transit revisions in 2025.
The damage premature crime stat announcements can do – 2023 example
The changing crime data is not an abstract concern. At the end of 2023, Mayor Eric Adams and the NYPD announced that crime had declined for the year. As data continued to mature early into 2024, the numbers told a different story. By spring, the New York Post reported that crime had in fact risen and that the celebrated decline was a product of preliminary numbers that had not yet been revised upward. But by then the damage is done. The initial numbers have already shaped coverage and public perception, leaving a false impression and further eroding trust in official data.
This is the scenario the data predicts. With an average upward revision of nearly 3%, any initially reported decline of less than 3% is at serious risk of flipping signs. And because the revision is systematic, not random, the “preliminary” caveat buried at the bottom of each press release is not just boilerplate. It is one of the most important sentences in the document.
What should change
Four things can improve how the city communicates and consumes crime data.
First, the NYPD should publish revision tables alongside each monthly release. Include a sidebar in each press release showing how the comparison month's numbers have changed since they were first reported. If January 2025's crime numbers were initially reported as 8,433 and are now 8,940, say so. Let the reader calibrate. The department should also timestamp all data releases, as it already does for its historical Compstat tables. Clear labeling of different versions would reduce unnecessary confusion when observers notice discrepancies between sources.
Second, the mayor’s office and the department should be cautious about making definitive trend claims based on single-month preliminary data. A month’s crime report is a useful data point, not a verdict. Claiming “eight straight months of decline” when each month’s number is still maturing is an invitation to the kind of embarrassment that followed the 2023 end-of-year announcement — and to the broader erosion of trust that follows when official numbers don’t hold up. Confidence intervals belong in the conversation, at least informally.
Third, journalists should routinely compare the prior year’s numbers in the current press release against the original press release for that period. This is a five-minute check that would immediately contextualize every monthly crime story. The raw materials are public; the NYPD archives its press releases online.
Fourth, the NYPD should invest in better front-end classification. The revision pattern is not just an accounting problem. It reflects thousands of individual moments where an officer or supervisor got the severity wrong on the first pass. The goal isn't to eliminate revisions; some reclassification is inevitable as facts develop. It is to reduce the share driven by errors that better training could have prevented.
How to read the next headline
When the next monthly crime report drops, and it announces that major crime is up or down by some percentage compared to the same period last year, observers should mentally add 4 to 5 percentage points to any reported change in total index crime. Add more for burglary, felony assault and especially murder. Subtract for grand larceny auto. Treat transit numbers, for the moment, as provisional sketches rather than finished portraits.
The NYPD’s crime data is among the most granular and well-audited in the country. That auditing process is a strength. But the public conversation about crime — in the press, at the podium and in politics — runs on first drafts. And first drafts, it turns out, are systematically wrong.
