# Campus safety: primary-source research and methodological review

Reviewed 25 September 2026. This is a targeted review of official guidance, original research and institutional documents, not a systematic literature review or human peer review. It informs the frozen 2022–2024, 42-institution descriptive protocol. It does not establish that any institution is safer, explain a particular campus's changes, or estimate unreported victimization. The final publication requires the separate audit in [CITATION_AUDIT_PLAN.md](CITATION_AUDIT_PLAN.md).

## Measures and interpretation

The requested residential measure is **reported offenses in on-campus student housing per 1,000 documented fall student residents**. Show the offense, report year, count, occupancy and source beside each rate. If the housing-property boundaries cannot be reconciled, call this an **approximate residential normalization** in the view itself. It is not the percentage of residents victimized, a probability, or an incidence rate based on measured person-time.

For institution i and report year y, calculate `1,000 × housing_count[i,y] / fall_occupancy[i,y]`. Pool years or institutions as `1,000 × sum(count) / sum(occupancy)` over the same eligible observations. This is a denominator-weighted aggregate. An unweighted mean of institution rates answers a different question; it must not silently replace the pooled ratio. The denominator sum across years is a sum of annual snapshots, not unique people or measured resident-years.

The separate enrolled-population view uses on-campus offenses and matching fall enrollment. NCES provides a normalization precedent, but publishes rates per **10,000 full-time-equivalent students**; headcount, FTE and housing residents are different denominators. NCES also states that its on-campus statistics include staff and guests, omit students' off-campus incidents, and can contain some duplicate reporting across nearby institutions. Do not compare this study's ratios directly with NCES rates without reconciling these definitions. [NCES (2024), figure 1 and notes](https://nces.ed.gov/programs/coe/indicator/a21).

## Counting rules that affect the comparison

| Rule | Consequence | Primary authority |
|---|---|---|
| Calendar **report** year | A 2024 count can include an earlier occurrence. | [34 CFR 668.46(c)(3)](https://www.ecfr.gov/current/title-34/section-668.46#p-668.46(c)(3)) |
| Housing is an on-campus subset | Never add housing to on-campus totals. | [Section (c)(5)](https://www.ecfr.gov/current/title-34/section-668.46#p-668.46(c)(5)) |
| Reports to police or campus security authorities | These are reported allegations within specified geography, not convictions. Investigation is not required for inclusion. | [Sections (c)(1)–(2)](https://www.ecfr.gov/current/title-34/section-668.46#p-668.46(c)(1)) |
| Restricted removal as unfounded | No arrest, noncooperation or nonprosecution does not itself remove a report. | [Section (c)(2)(iii)](https://www.ecfr.gov/current/title-34/section-668.46#p-668.46(c)(2)(iii)) |
| Hierarchy and exceptions | Several categories are not a deduplicated incident total; arson and murder/sexual-offense exceptions apply. | [Section (c)(9)](https://www.ecfr.gov/current/title-34/section-668.46#p-668.46(c)(9)) |
| Stalking and confidential counseling rules | Some stalking courses generate multiple report-year statistics; counselor disclosures need not enter these statistics. | [Sections (c)(6), (c)(8)](https://www.ecfr.gov/current/title-34/section-668.46#p-668.46(c)(6)) |

Keep criminal offenses, VAWA offenses, hate-crime classifications, arrests and disciplinary referrals separate. UCSC explicitly explains that VAWA and hate-crime counting can overlap criminal categories; its faculty/staff housing also differs from student housing. This supplies an institutional illustration, not a substitute for another institution's geography. [UCSC 2025 ASR, printed pp. 8–10](https://bpb-us-w2.wpmucdn.com/wordpress.ucsc.edu/dist/d/120/files/2025/09/2025-UC-Santa-Cruz-ASFSR_online-3.pdf#page=9).

The 2016 federal handbook was rescinded in October 2020. Use current regulation and the Federal Student Aid handbook for current guidance; historical documents remain evidence of their own periods. Do not add newly required hazing data retrospectively to a 2022–2024 outcome. [ED campus-security guidance](https://www.ed.gov/teaching-and-administration/safe-learning-environments/school-safety-and-security/campus-security); [2026–2027 FSA Handbook, volume 2, chapter 6, Campus Crime and Safety Information](https://fsapartners.ed.gov/knowledge-center/fsa-handbook/2026-2027/vol2/ch6-consumer-information-and-school-reporting).

## Denominator and source-quality requirements

California State Auditor report 2024-111 separates **occupancy** (Appendix A, tables A.1/A.2, printed pp. 56–59; PDF pp. 62–65) from **bed capacity** (Appendix B). It reports fall occupancy for academic years 2019–20 through 2024–25, drawing on UC and CSU housing reports. These are usable documented counts, but that table does not establish a property-by-property crosswalk to Clery geography. [Official report](https://www.auditor.ca.gov/wp-content/uploads/2025/10/2024-111-Report.pdf#page=62).

For each residential denominator, retain: year, census timing if known, actual occupancy versus capacity, student versus family/staff population, property scope, and match quality. University housing can differ from the Clery on-campus subset. A count of all university-managed residents is not automatically a matched denominator. Do not substitute beds, an undergraduate housing percentage, or a current-year estimate for observed annual occupancy. If occupancy is unavailable, preserve the count and mark the rate unavailable.

Even a matched fall population does not identify the people involved in the numerator. Guests, nonresident students and nonstudents may be involved, residents may move, and disclosure can occur after the offense. These limitations preclude interpreting the ratio as personal victimization risk. They also make adjustments based only on enrollment size insufficient for comparisons between commuter, residential, medical-center and urban institutions.

Use a single frozen federal collection for the primary cohort, with clearly versioned institutional-ASR comparisons. A disagreement is a source discrepancy, not proof that the lower value is false or that the ASR necessarily postdates the federal extract. Preserve both; do not selectively replace unfavorable values. Match federal campus identifiers to enrollment units, document branches and combined institutions, and prevent duplicate enrollment when aggregating several reporting campuses.

Missing, withheld, inapplicable and zero are distinct. Read each source dictionary before interpreting codes. Never turn an absent row or blank into zero. A zero denominator gives an undefined rate. A reported zero is not evidence that no victimization occurred. No correction multiplier is justified by the studies below.

## Evidence from comparable original work

| Source and exact scope | What it supports | Limits and accurate use |
|---|---|---|
| **Krebs et al. (2016), BJS Campus Climate Survey Validation Study**, nine pilot campuses; printed p. 110 / PDF p. 131, comparison box. [Report](https://bjs.ojp.gov/content/pub/pdf/ccsvsftr.pdf#page=131) | Surveys and Clery counts answer different questions. The study's survey estimate of 60 Clery-comparable on-campus rape incidents was not statistically distinguishable from 40 official Clery incidents across the pilot campuses. | The survey and Clery comparison differed in time period and population; the survey covered undergraduates. It neither establishes universal reporting completeness nor licenses calling the larger survey–Clery gap concealment. Do not turn the proportion meeting Clery criteria into a national underreporting adjustment. |
| **Cantor et al., AAU 2019 survey, revised 17 January 2020**, 33 schools, 181,752 respondents; table 2, printed p. 6; resource contact pp. 28–31; nonresponse appendix 4. [Report](https://www.aau.edu/uploads/images/general/Revised-Aggregate-report-and-appendices-1-7_01-16-2020_FINAL.pdf) | Survey prevalence and official reporting are distinct measurements. The response rate was 21.9%; the report investigates nonresponse and uses weighting. Campus-resource contact encompasses more than police reporting. | Participating schools are not a probability sample of all colleges. Survey definitions, since-enrollment and academic-year reference periods do not match annual Clery rape statistics. Resource-contact percentages cannot be described as police-reporting rates. No survey-to-Clery conversion factor. |
| **Richards, Branscum, Goldsmith & Ali (2024)**, *Victims & Offenders* 19(7), 1301–1319; 42 Maryland institutions, 2018–2020. Author postprint, Method, Tables 1–3, Limitations. [Full author text](https://www.crimrxiv.com/pub/5ab2b5jp/release/1) | The study reports associations between institutional residentiality/reporting knowledge and sexual-misconduct reporting. It compares Title IX and Clery/VAWA measures using enrollment-normalized rates. | Exploratory observational analysis, small sample, heterogeneous campus surveys and missing climate responses. Title IX and ASR outcomes differ in scope. The postprint's Table 1 total-misconduct mean (6.46) differs from the narrative (5.32), so do not quote a precise ratio or transplant the coefficients. Associations do not prove that better disclosure explains any named campus's counts. |
| **Nobles, Fox, Khey & Lizotte (2013)**, *Crime & Delinquency* 59(8), 1131–1156, DOI 10.1177/0011128710372188. One southeastern university, geospatial offending/arrest analysis. [Publisher](https://journals.sagepub.com/doi/abs/10.1177/0011128710372188); [author abstract reproduced by OJP](https://www.ojp.gov/ncjrs/virtual-library/abstracts/community-and-campus-crime-geospatial-examination-clery-act) | Demonstrates the relevance of location and student/nonstudent distinctions in campus crime research. | Abstract reviewed; no numerical effect used. One campus and arrest/offending measures cannot establish victimization differences across this 42-institution cohort. |
| **California State Auditor, report 2017-032**, 2016 statistics at four institutions; table 2. [Audit results](https://information.auditor.ca.gov/reports/2017-032/auditresults.html) | Direct record checks found 42 errors: 11 underreporting, 25 overreporting and six classification errors. Source quality can fail in either direction. | The 212 tested crimes were selected for audit purposes, not a national probability sample. Do not apply an error percentage or correction factor to current campuses. The audit's references to the then-current 2016 handbook are historical. |

An additional often-cited paper is Yung (2015), *Concealing campus sexual assault: An empirical examination*, DOI [10.1037/law0000037](https://doi.org/10.1037/law0000037). Publisher/author full-text retrieval was blocked in this review. It is an identified lead, **not a verified quantitative source for this publication**. No percentage from a press release is used. This access limitation is preferable to treating a headline as the study's estimand.

The reviewed evidence supports cautious comparison of administrative reporting, plus separate climate-survey evidence when available. It does not determine how much a given campus's count reflects victimization, reporting access, classification, population composition or geographic coverage. A rise may be consistent with several explanations; an institutional causal explanation needs its own evidence.

## Institution-specific checks

The two table pages below were checked against the downloaded PDFs and visually rendered pages, not OCR alone. File hashes and retrieval timestamps are in [document_retrieval_metadata.json](document_retrieval_metadata.json); machine-readable values are in [verified_asr_rape_counts.csv](verified_asr_rape_counts.csv).

**UC Santa Cruz.** The 2025 ASR, printed p. 11 (PDF p. 12), reports on-campus rape counts of 25, 24 and 46 for 2022–2024; housing counts are 24, 21 and 38. Footnote 6 explains that 21 separate offenses in the 2024 housing count came from one disclosure concerning two parties over three weeks. Those offenses also enter the campus total. Keep the reported count; do not collapse the disclosure to one offense or relabel the count as 38 victims. [UCSC table and footnote](https://bpb-us-w2.wpmucdn.com/wordpress.ucsc.edu/dist/d/120/files/2025/09/2025-UC-Santa-Cruz-ASFSR_online-3.pdf#page=12).

**UC San Diego.** The current 2025 ASR, PDF p. 142, reports on-campus rape counts of 15, 19 and 28 and housing counts of 12, 11 and 20 for 2022–2024. Its footnotes identify rape revisions for 2022 and 2023. The 2023 changes include two housing offenses, one additional on-campus offense and four noncampus offenses; 2022 includes one additional on-campus and one noncampus offense. The electric-scooter explanation is attached to motor vehicle theft, not rape. No explanation for a rape trend should be inferred from that footnote. [UCSD table and footnotes](https://www.police.ucsd.edu/docs/annualclery.pdf#page=142).

Independent direct reads of the archived federal workbooks verify UCSD 2022 on-campus rape as 14, compared with 15 in this ASR; noncampus rape is 4 versus 5. The other checked rape cells match. Keep the discrepancies visible. A report bearing a 2025 title and an extract prepared in 2026 do not, by themselves, establish which underlying revision was available first. [Cell comparisons](asr_federal_rape_comparison.csv); [verification](asr_comparison_verification.json).

**San Diego State.** The 2025 ASR main-campus table gives housing rape counts of 9, 7 and 1 and on-campus counts of 12, 11 and 1 for 2022–2024. These match the federal main-campus cells. The ASR lists zero rape offenses for Imperial Valley and Georgia. Their federal housing cells are blank, however, and are not ordinary numeric-zero observations in this audit. Resolve structural applicability from screening metadata before aggregating housing. No causal explanation for the main-campus change is asserted. [ASR PDF p. 7](https://police.sdsu.edu/_resources/files/annual-security-reports/2025-annual-security-report-finalized-08-18-25.pdf#page=7); [official HTML report](https://police.sdsu.edu/public-information/clery-act-compliance/2025-annual-security-report).

## Uncertainty and presentation decisions

These are descriptive ratios of observed administrative totals, not estimates from a random sample of campuses. Default to counts, denominators, source notes and rounded ratios, without significance tests or confidence-interval league tables. Display one or two decimal places; calculate from unrounded inputs.

A Poisson interval could only describe uncertainty under an explicit event-process model with an appropriate exposure and independence/variance assumptions. It would not measure uncertainty from underreporting, geography mismatch, source revisions or fall snapshots. Repeated offenses within one disclosure can violate simple independence assumptions. A binomial interval is inappropriate because these numerators are not unique victim counts among the denominator population. Zero events would not prove zero underlying risk under either model.

Use alphabetical defaults, neutral colors and explicit geography/year labels. If the user sorts rates, label the result as reported rates among included, eligible institutions. Avoid safest, most dangerous, likelihood of rape, percent victimized, national rank and causal explanations. Preserve missing rows and disclose exclusions from pooled denominators. These are interpretation safeguards, not reasons to suppress correctly sourced counts.

## Open items for the final audit

1. Verify all federal values, campus-ID mappings and denominator joins from the frozen raw files independently of the production calculation.
2. Quantify property-scope mismatch where possible; otherwise retain the approximate-residential label for the State Auditor denominator.
3. Verify the full UCSD federal/ASR discrepancy matrix, not only the known rape cell.
4. Inspect every substantive claim in the actual webpage, downloads and PDF under the separate audit plan. This review is not that final clearance.