Crime statistics get revised because they are assembled, not measured in one pass: police agencies amend their submissions to the FBI after release, the Bureau of Justice Statistics re-estimates survey figures with new weights, and methodological redesigns reset whole series. Per FBI published documentation, offense counts for a calendar year can shift for up to two years after the initial release, sometimes by percentages large enough to change the direction of a trend story.
Readers who quote crime data should treat each figure as dated to its release and revision history; this site publishes information, not legal advice.
Why do police agencies change numbers after release?
Police records are administrative files, and administrative files get amended. A reported shooting initially classified as an aggravated assault may be reclassified as murder after a victim dies — a reclassification the FBI's rules require, since homicide counts by the offense a victim ultimately died of. A batch of thefts may be consolidated once investigators link them. Late forensic results, witness recantations, and data-entry corrections all flow back into the record, and agencies resubmit updated files to the FBI's Uniform Crime Reporting program, which republishes the tables.
Per FBI published documentation, this is expected behavior rather than error: the program's technical notes explicitly tell data users that submissions continue to be updated and that the most recent file supersedes earlier ones. The practical consequence is that a January headline number and the same number two years later are often different figures, both official.
How does the FBI's estimation process add revisions?
When participating agencies are missing or partial, the FBI does not simply leave a hole in the national table. In recent years the program has published both raw counts from reporting agencies and statistical estimates that impute data for non-participating ones. Estimates depend on modeling assumptions, and they change as coverage improves: per the FBI's published technical notes for 2021 and 2022, national estimates were computed while a large share of agencies were still completing the NIBRS transition, and the imputation weights were refined in later releases.
This matters most at turning points. An estimate for a year with thin coverage carries wider uncertainty, and a preliminary decline or spike announced on release day can shrink — or reverse — in the final tables. Analysts who track the data routinely wait for the second or third publication before calling a trend.
What did the 2021 transition do to the series?
The NIBRS transition created the clearest recent revision episode. When the FBI retired the Summary Reporting System on January 1, 2021, many large agencies had not converted, and the 2021 national tables rested on an incomplete base. Per the FBI's own technical documentation, later releases supplied imputed national estimates that filled the gaps, and academic and think-tank analysts published alternative estimates of their own, some of which differed from the FBI's first figures by amounts large enough to reach headlines.
The episode illustrates the general rule: when measurement infrastructure changes, the first published numbers carry the most uncertainty, and the honest comparisons are the ones that note the break. Per BJS and FBI guidance alike, 2021 comparisons that span the transition are labeled with caveats by the agencies themselves.
Related stories: Why single-cause explanations of the crime drop rarely survive scrutiny · How hate crime counts are collected, and why underreporting persists.
Do survey numbers get revised too?
Yes, by design. The National Crime Victimization Survey publishes estimates with confidence intervals, and the figures move as new interview rounds come in. BJS also revises the survey's instruments and weights: the major 1992 redesign raised measured rape and assault levels permanently, and the 2016 redesign period produced a documented break that BJS published with explicit instructions not to compare across it. These are methodological revisions, not corrections of mistakes — the definition of what counts changed, so the series changed.
Survey revisions also occur at the estimate level: weight adjustments after census updates alter historical figures slightly, which is why a table downloaded today can differ marginally from the one in a 2019 report. BJS marks superseded tables in its archive rather than quietly replacing them.
Which famous figures moved after publication?
The pattern recurs across decades. Homicide totals for a given year have grown in later FBI releases when pending deaths were reclassified and late hospital or medical examiner information arrived. City-level figures have moved in both directions after audits: several departments have resubmitted entire years of offense data after internal reviews found systematic misclassification, and the FBI's files record the resubmissions. Crime analysts at universities and at organizations such as the Council on Criminal Justice have documented cases where a city's published trend line changed sign between the initial release and the revised tables.
The direction of revision is not systematically upward or downward. Some corrections raise counts, some lower them, and the aggregate national effect has generally been small — but for a single city in a single year, a revision can be the difference between a spike story and a flat one.
Why do revisions attract so much suspicion?
Because the political stakes are high and the revision mechanics are obscure. A preliminary national estimate published weeks after year's end is quoted in campaigns and policy debates long before the final tables arrive; the revised figures rarely get the same coverage. Researchers who study the data emphasize that this asymmetry is a media problem more than a statistical one — the agencies publish both versions, archive their corrections, and document their imputation methods, but the revision rarely travels as far as the first number.
What should a careful reader check before quoting a crime number?
- The release date of the figure, and whether a later revision exists.
- Whether the number is a raw count from reporting agencies or a modeled national estimate.
- The participation rate for the year and program involved.
- Whether the series crosses a documented methodological break, such as 2016 or 2021.
- Whether the offense classification could plausibly change — deaths pending outcome, for instance.
None of this means the data cannot be trusted; it means the data has a version history, like any record built from thousands of local files. The FBI and BJS publish their revision practices openly, which is exactly what makes skeptical reading possible.
For more context, read How clearance rates are calculated, and why they often mislead.
For more context, read hate crime statistics.