What is being compared?
The opening vehicle-theft comparison was added on September 29, 2026. It uses 2024 on-campus reports divided by the same year’s total fall enrollment. The initial peer group comprises the nine undergraduate-serving UC institutions; all ten UC institutions and all 42 study universities are also selectable. This is an enrollment-adjusted reporting measure, not vehicle-owner exposure or a nationwide safety ranking. UC San Diego’s new 2026 report confirms its 2024 vehicle-theft counts; its newer 2025 count is described separately. The full-study datasets and report below remain the September 25 snapshot. Opening comparison and calculations, dated source review.
The unit is a reported Clery offense, divided by a documented population denominator. The analysis covers the ten UC institutions, eight Ivy League institutions, San Diego State, and 23 additional public and private research universities. This fixed comparison set is purposive; it does not represent all United States (U.S.) higher education. Institutions were selected before this study calculated their rates. The originating shared research had already identified several UC and San Diego State results. [109].
The default uses current institutional annual reports checked on 25 September 2026, Pacific time, retaining each cell’s report edition, calendar report year, branch, geography, raw value and source locator. Report files have now been recovered for all 42 institutions. Retrieval does not establish complete or internally consistent reporting: missing, ambiguous or unverified cells remain withheld from affected calculations. The current source inventory identifies the editions reviewed and their remaining limitations. [13], [16].
The original 2025 federal Campus Safety and Security collection, covering 2022–2024, remains an explicitly labeled archived option. Its binary hash was rechecked and is unchanged; the federal catalog still lists 2025 as its latest bulk collection. It is not silently mixed into the current institutional series. Older shared explorer links retain their federal source, while new links record the selected source explicitly. An older institutional edition is used only with its attribution, as in San Diego State’s 2022 worked example. [110], [111].
Clery assigns offenses to the year they were reported to a campus security authority or local police, which need not be the year they occurred. [112]. Reports may concern students, staff or visitors. A count is neither a unique-victim count nor a count of report forms. Consequently, these rates do not estimate the proportion of students victimized. [4], [105].
Geography and categories
On campus is a legal reporting geography, not a synonym for a university’s city or every place its students live. On-campus student housing is a subset of on-campus geography. It is never added a second time. Noncampus property and adjacent public property are distinct reporting areas; neither is included in the explorer’s on-campus rate. The geographic explorer displays each reporting campus separately. The combined geographic total is campus total plus noncampus plus public property. Other on-campus reports are campus total minus housing, only where both values are known and internally consistent. This total still excludes ordinary off-campus crime outside Clery geography; it is not all crime experienced by students. [12].
The default category sums eleven listed criminal-offense categories: murder/nonnegligent manslaughter, negligent manslaughter, rape, fondling, incest, statutory rape, robbery, aggravated assault, burglary, motor vehicle theft and arson. This is a sum of reported offenses, not deduplicated incidents. Domestic violence, dating violence and stalking are available separately under the federal VAWA categories. They can overlap other criminal classifications and must not be added to the default total. Arrests, disciplinary referrals and hate-crime classifications are not combined with these measures. [113], [114].
The institution is identified by its unique institution identifier in the Integrated Postsecondary Education Data System (UNITID). Reported branches are aggregated within the selected source’s documented institutional scope; an institution’s enrollment is counted once, not once per reporting campus. New branches retain source-specific identifiers where a federal crosswalk has not been verified. Missing required branches or unresolved scope changes withhold institution-wide comparisons. This can include medical facilities, satellite locations and international reporting campuses. The explorer lists every included campus. Institution-wide results should not be described as the crime rate at a single main campus. [115], [116].
Population denominators
The default residential view uses actual reported student occupancy, not available beds or a percentage of undergraduates living in housing. California State Auditor Report 2024-111 supplies fall 2022, 2023 and 2024 occupancy for all ten UCs and San Diego State. Its housing inventory has not been reconciled property by property with Clery’s residential geography. These ratios are therefore an approximate housing normalization, even though the occupancy values themselves are documented. [19].
The expanded population review adds Stanford’s autumn 2023–2025 student-housing counts, Carnegie Mellon’s 2022–2024 housing census and Yale’s 2022–2025 fall student-housing census. These sources retain their actual reference periods and scope qualifications. Stanford’s 2024 count comes from a university-authored fact book recovered from a publisher-hosted copy; identical bytes from the former official URL could not be authenticated. Carnegie Mellon’s count combines separately listed university-housing and fraternity/sorority residents; its housing office states that on-campus housing is unavailable to graduate students. The source’s Pittsburgh label and its 2024 all-location undergraduate total do not fully reconcile. None of these housing footprints has been matched property by property to every Clery location. [117], [118], [119], [120], [121].
The 26 September review recovered Stanford’s 2024 fact book from an Internet Archive capture of the original university file. It documents 14,137 students in university-provided housing in autumn 2023; the archived file’s digest agrees with the capture index. Yale’s current housing report supplies 6,255, 6,064, 6,011 and 6,082 student residents for fall 2022 through fall 2025. Its columns distinguish students living singly or with partners/dependents; the latter are student classifications, not additional family-member headcounts. Yale’s 2025 population does not create a 2025 rate because the current crime sources do not supply that year. [122], [123], [124].
A documented population does not by itself make a rate available. Stanford has a 2025 resident count, for example, but its overseas crime reports stop at 2024. Texas Austin’s housing report identifies 10,018 residents in spring 2024, but does not establish that the total excludes nonstudent residents; that candidate is not adopted as a student-population denominator. Its branch crime tables also leave some residential cells blank. Counts and populations must both pass the selected calculation’s coverage checks. Partial student groups, rounded estimates, bed capacity and residents from another year are not substituted for a missing whole-population count. The population ledger records the date, field, source hash and scope of every adopted value; the interface exposes those details for the selected institution. [125], [3], [11].
The housing evidence disclosure separately retains 37 partial, approximate or definition-unresolved observations across 14 further institutions. For example, a residence-hall count may omit apartments or graduate housing, and an annual service total may include dependents. Academic and fiscal spans remain labeled as such. These observations provide context; they never supply a rate denominator. A remaining gap means no compatible complete population has been verified for this comparison, not that the institution publishes no housing figures. [126].
The broader optional enrollment view uses fall enrolled headcount, including undergraduate and graduate/professional students, full time and part time, as reported to IPEDS. Each report year uses the fall population from that same calendar year. A fall snapshot approximates institutional size; it is not measured time physically present on campus. Exclusively distance-enrolled students remain in the institutional denominator, and their counts are retained in the data. Removing them would require assumptions about which students were exposed to each reporting location. [15], [127].
The federal enrollment files for fall 2022–2024 were re-downloaded and their hashes remain unchanged. No verified fall 2025 enrollment series is adopted. The release schedule distinguishes fall-collection components from the separate spring enrollment component; a 2025–2026 collection label does not itself establish fall 2025 headcounts. [128], [129].
The alternative housing reports per enrolled student uses the housing numerator with total enrollment. It permits a consistent enrollment-based comparison of reported housing offenses across institutions, but it is not a measure of risk among residents. The interface labels each numerator/population combination explicitly.
Rate calculations
For institution i and report year t, the annual descriptive rate is
Rit=kNitCit.
Here, Cit is the selected reported-offense count, Nit is the documented population snapshot assigned to that calendar year, and k is 1,000 by default, or 10,000 when selected. Enrollment uses fall headcount; resident populations retain their stated census season. A population snapshot is not measured person-time or an annual average. Changing the scale changes the numerical unit, not the comparison. [3].
The pooled annual rate for 2022–2024 is
Ri,pooled=k∑t=20222024Nit∑t=20222024Cit.
The denominator is the sum of three annual population snapshots. It does not count unique students across three years. Pooling is equivalent to an enrollment- or occupancy-weighted mean of the three annual rates; it is not their unweighted average. The pooled result remains a rate per year, rather than a three-year cumulative victimization probability. Counts and population sums accompany every displayed rate.
Calculations retain full precision until display. An observed zero remains zero. Missing, inapplicable and unavailable source entries retain distinct processing statuses; an unknown cell is never replaced by zero. A selected rate is withheld when any required count or population is unavailable. The source-cell ledger preserves reported values separately from calculation eligibility. Explicitly absent geographic areas can contribute zero to a derived total while remaining marked not applicable in the raw record. An omitted column alone, unavailable agency return or ambiguous year label cannot support that rule. A latest report that no longer covers 2022 does not reaffirm that year’s earlier count. [130], [131].
For the archived federal 2024 residential totals only, the frozen count files are supplemented by official federal application programming interface (API) declarations for survey year 2024: 103 reporting campuses provide numeric housing counts; 109 explicitly report no housing. The latter are excluded as an absent geography, while their original cells remain blank. This rule is not applied to earlier years. The API snapshot may include updates after the closed bulk file; its role and retrieval metadata are separately identified. [132], [133].
Interpreting differences
A larger reporting rate can reflect more offending, more disclosure, differences in geography or recording, or a combination of these. These data cannot separate those mechanisms. Conversely, a small or zero count does not establish that no victimization occurred. No institution is labeled “safest” or “most dangerous,” and no causal interpretation is attached to differences.
The comparison is descriptive. No hypothesis tests, confidence bands or significance-based rankings are presented. An assumed Poisson process would not account for unmeasured reporting or population-boundary error; a binomial interpretation would incorrectly treat offenses as unique victimized students. Multiple offenses disclosed together also undermine a simple independent-events interpretation. [134], [105].
Comparison with other research
National Center for Education Statistics (NCES)’s national campus-crime indicator also places administrative counts over a population denominator, but uses full-time equivalent (FTE) enrollment and reports per 10,000 FTE students. This study uses enrolled headcount or documented residents. The resulting numbers are not interchangeable; simply changing the scale to 10,000 does not turn headcount into FTE. [4].
Campus-climate surveys address a different question: respondents’ experiences of victimization within a survey’s stated population and recall period. The Bureau of Justice Statistics (BJS)’ Campus Climate Survey Validation Study compared survey estimates with Clery reports across nine pilot schools. Its analysis illustrates why population, geography and reporting to authorities must be aligned before comparing survey and administrative figures. Neither source can be treated as a direct substitute for the other. [135].
These precedents inform the definitions and limits here. They do not validate a causal claim about any institution or supply a universal correction for underreporting. The separate review records the design, population, numerator and denominator of each study considered. [134].
Reproducibility and independent checks
Public files include the fixed cohort, reporting-campus crosswalk, source-derived counts and population records, exact source URLs and hashes, category definitions, annual rates, pooled calculations, and calculation code. Source-specific extraction scripts retain field and row references. The current archive reproduces the revised rates from its published aggregate inputs and decisions. The federal archive remains unchanged. Re-extraction requires the separately identified institutional or national files and their verified versions; mutable URLs alone do not identify immutable evidence.
Current-source calculation checks and the separate citation review accompany this revision. The earlier methods and related-research review remains identified as the original review; its existence is not presented as a new independent certification. [130], [134], [136]. Formula and page-layout checks cover the website and report. [137]. These are computational and source-verification audits, not external human peer review. Remaining source discrepancies, population-boundary limitations and missing cells are disclosed rather than statistically “corrected.” Audit files and reproducibility package.