Campus Crime Data Is a Separate Country: The Clery Act, the Daily Crime Log, and Why It Doesn't Merge With City Feeds

On October 1, every Title IV institution in the United States publishes a document full of crime counts. The numbers are federally mandated, publicly downloadable, and address-adjacent. They are also collected under different definitions, over different geography, by different reporters, on a different clock than every municipal incident feed you have ever used. Treating the two as one dataset is the single most common mistake made with campus crime data — and this year the schema changes again.

A Parallel Reporting System

The Jeanne Clery Act was passed in 1990, four years after Jeanne Clery was murdered in her dormitory room at Lehigh University. It requires every postsecondary institution participating in federal Title IV student aid to collect, publish, and submit campus crime statistics. That is roughly six thousand institutions, from research universities to single-campus cosmetology schools, all filing on the same template.

The output has three distinct artifacts, and conflating them causes most of the downstream confusion:

• The Annual Security Report (ASR)— published by October 1 each year, containing policy statements plus crime counts for the three most recent complete calendar years.
• The federal submission— the same counts filed to the Department of Education and exposed through the Campus Safety and Security Data Analysis Cutting Tool. This is the only nationally comparable campus crime file that exists.
• The daily crime log— a rolling record of reported incidents maintained by institutions with a campus police or security department, updated within two business days of a report. This is the only near-real-time artifact in the system, and it is the one almost nobody uses.

The ASR is the famous one. The daily crime log is the useful one. They contain overlapping but non-identical sets of events, which is the first thing worth internalizing.

Clery Geography Is Not a Boundary Polygon

Every crime data pipeline eventually resolves to a question of where. Clery answers that question with four categories rather than a shape, and the categories do not tile cleanly onto anything a mapping stack understands:

• On campus— institution-owned or controlled property within a reasonably contiguous area, used in direct support of educational purposes.
• On-campus student housing— a strict subset of the above, reported separately and therefore double-counted by anyone who sums the columns.
• Noncampus property— buildings owned or controlled by the institution or by a recognized student organization that are not within the contiguous campus. Fraternity houses, field stations, and leased space in another city all land here.
• Public property— streets, sidewalks, and adjacent parking immediately bordering or running through campus.

Three consequences follow directly. First, the on-campus housing column is nested inside the on-campus column, so a naive sum of the four fields overstates total incidents. Second, the public-property category means Clery counts events that also appear in the municipal police feed — the same robbery on the same sidewalk can legitimately appear in both datasets. Third, an apartment two blocks from the quad, where a large share of students actually live, is in none of the four categories and is invisible to Clery entirely.

Different Definitions, Different Counters

Clery reports a fixed list: criminal homicide, sexual assault (rape, fondling, incest, statutory rape), robbery, aggravated assault, burglary, motor vehicle theft, and arson; the VAWA offenses of domestic violence, dating violence, and stalking; hate crimes; and arrests and disciplinary referrals for liquor law, drug, and weapons violations. Larceny — the highest-volume property offense in every municipal feed in the country — is not on the list except as a hate crime.

Counting follows UCR conventions, including the hierarchy rule, so a multi-offense incident generally surfaces as its most serious component. Arson is the standing exception and is always counted. We have written before about why the same crime carries three different names across CAD, SRS, and NIBRS. Clery is a fourth taxonomy, and unlike the others it did not transition to NIBRS. That means a campus that reports to the FBI under NIBRS is simultaneously producing hierarchy-ruled summary counts for the Department of Education, from partially overlapping source events.

The reporters differ too. Municipal data comes from sworn officers filing records. Clery statistics come from campus police plus a much broader class called Campus Security Authorities — officials with significant responsibility for student activities, including residence life staff, athletics administrators, student conduct officers, and faculty leading off-campus trips. Professional and pastoral counselors are exempt. A resident assistant told about an assault creates a Clery-reportable statistic without a police report ever existing. This is a deliberate design choice, and it means Clery is not a subset of police data. It is a partially disjoint set.

What the National Numbers Show

The National Center for Education Statistics compiles the federal submissions. In 2021, degree-granting postsecondary institutions reported approximately 23,400 on-campus criminal incidents, a rate of 16.9 per 10,000 full-time-equivalent students, down from 20.0 per 10,000 in 2011. The composition:

Two features of that table deserve more skepticism than they usually get.

The first is the sex offense line. The reported rate rose from 2.2 per 10,000 FTE students in 2011 to 7.5 in 2021 — a factor of roughly 3.4. Read as a measurement of victimization, that would be an extraordinary finding. It almost certainly is not one. The 2013 VAWA reauthorization and its Campus SaVE provisions restructured this category outright, replacing the old forcible sex offense definitions with separate counts for rape, fondling, incest, and statutory rape, and the same period saw sustained federal enforcement pressure, mandatory CSA training, and Title IX attention aimed specifically at increasing reporting. When a definition change, a training regime, and an enforcement campaign all land on one category inside one decade, the resulting series measures the reporting apparatus at least as much as it measures the underlying events. This is the same interpretive trap as the dark figure problem in municipal data, running in the opposite direction: here the reported number rose because the funnel widened.

The second is that 2020 and 2021 are structurally broken years for campus data specifically. The rate fell 20% between 2019 and 2020 (18.8 to 15.0) and rebounded 12% in 2021. Campuses were closed or partially occupied. The denominator — FTE enrollment — barely moved, while the population physically present on the geography fell by a large and unmeasured fraction. Any per-capita campus crime rate spanning 2020 is comparing a normal denominator against an abnormal numerator.

The Undercount, and the Incentive Behind It

Clery statistics are widely understood to undercount. Longitudinal work published in the Journal of American College Healthin 2025 compared institutional Clery filings against the same institutions' Title IX office records and found Clery figures represent a considerable undercount of sexual misconduct, with the gap appearing whether or not the institution separately published Title IX data.

The enforcement structure explains part of why. The Department of Education can fine institutions per violation, with the maximum adjusted annually for inflation — $71,545 as of January 2025. Because violations are counted individually, totals compound fast. Liberty University agreed to a $14 million fine in March 2024, the largest ever levied under the Act, superseding Michigan State's $4.5 million in 2019 and Penn State's $2.4 million in 2016.

Note the direction of the incentive carefully, because it is not the simple one. Fines are levied for failure to disclose, not for reporting high numbers. In principle that pushes toward completeness. In practice, the institution reporting the statistic is also the institution marketing itself to prospective students, and the enforcement action arrives years after the reporting year it examines. The result is an unusually strong reason to be accurate and a persistent, quieter reason to be conservative about classification at the margin. Any dataset where the measured party controls the measurement deserves the same skepticism we apply to agency-reported clearance rates.

The Schema Changes This October

The Stop Campus Hazing Act, signed December 23, 2024 and effective January 1, 2025, amended the Clery Act and renamed it the Jeanne Clery Campus Safety Act. It added hazing as a reportable offense category. Institutions began compiling hazing statistics on January 1, 2025; the first Campus Hazing Transparency Reports were due December 23, 2025; and hazing counts appear for the first time in the Annual Security Reports published October 1, 2026 — covering calendar year 2025.

For anyone maintaining a campus dataset, this is a column addition with no history. The 2026 ASR will show hazing counts for 2025 and nothing before it. Expect the first-year figures to be low and highly variable across institutions, for the ordinary reason that a brand-new mandatory category takes several cycles for classification practice to converge. Year-one hazing numbers will describe institutional readiness to classify hazing far better than they describe hazing.

The Daily Crime Log Is the Real Feed

Everything above concerns an annual document with a nine-to-twenty-one-month lag. The daily crime log is the opposite: incidents reported to campus police or security must be entered within two business days, with date and time of occurrence, date and time reported, a description, general location, and disposition where known. It covers all criminal incidents in the patrol jurisdiction, not just the Clery-reportable categories, and the most recent sixty days must be open to the public without request.

That is a genuinely fast, genuinely public incident feed at roughly six thousand locations. It is also almost entirely unstructured. There is no mandated format, no mandated schema, and no mandated machine access. In practice institutions publish PDFs, HTML tables, embedded viewers, and occasionally a real data endpoint. The disposition field — open, closed, cleared by arrest, referred — is the most valuable and least standardized element in the whole record.

Ingesting these at scale is a normalization problem, not a collection problem, and the same discipline applies as anywhere else in this business: parse to a canonical taxonomy, retain the source string, geocode against the limits of the location precision actually provided, and never impute what the source did not say. The practical Python tooling crime analysts already use for municipal feeds transfers directly.

Rules for Building on Campus Data

• Never union Clery counts with municipal incidents. Different definitions, different geography, different reporters, guaranteed double-counting on public property. Keep them as separate layers with separate labels.
• Do not sum the geography columns. On-campus student housing is nested inside on campus.
• Watch the denominator. FTE enrollment is not the population at risk. Commuter schools, medical centers, and campuses with large non-student daytime populations are systematically distorted by per-student rates.
• Do not rank institutions. A higher Clery number is at least as likely to reflect a campus with a functioning reporting culture and broad CSA training as one with more crime. This is the inverse-signal problem in its purest form.
• Treat 2014 and 2025 as structural breaksin the sex offense and hazing series respectively, and 2020–21 as an occupancy anomaly. Do not fit a trend across them.
• Use the daily log for anything operational. An annual report published nine months after year-end cannot support a safety product. A two-business-day log can.
• Do not let a model narrate a count it did not compute. Language models remain poorly calibrated at expressing confidence, and campus data has enough definitional traps to generate confident nonsense at volume. Deterministic code owns the arithmetic.

Where This Actually Matters

The obvious application is student housing. A parent or a leasing platform evaluating an off-campus apartment is asking a question Clery structurally cannot answer, because most off-campus student housing sits outside all four Clery geography categories. The relevant data for that address is the municipal incident feed for the surrounding blocks, with the campus log as a supplementary layer for the walk between them. Presenting an institution's Clery numbers as though they describe the neighborhood a student will live in is a category error, and a fairly consequential one.

The second is institutional risk assessment — university general counsel, campus security directors, insurers writing student housing portfolios. Here Clery is genuinely useful, because the question really is about institutional control and institutional liability, which is precisely what Clery geography encodes.

The general lesson generalizes past campuses. Crime data in the United States is not one system. It is a set of overlapping mandatory reporting regimes, each with its own definitions, geography, reporters, latency, and enforcement incentives. NIBRS, the Real-Time Crime Index, municipal open data portals, and Clery all describe crime in America and none of them are substitutes for one another. The value in a crime data pipeline is not in having the feeds. It is in knowing exactly what each one refuses to tell you.