# Campus safety: analysis protocol

Frozen 2026-09-25, before cohort rates were calculated. Public study: `/writing/data-analysis/campus-safety/`.

## Question and scope

How do recorded Clery offenses compare after accounting for the size of the enrolled population, and where defensible, the resident population? These are administrative reporting rates, not victimization probabilities or a ranking of safety.

Primary window: calendar report years 2022–2024, from the latest public federal collection (2025). Each year's fall population is a snapshot denominator. The pooled rate is the sum of counts divided by the sum of annual populations, multiplied by 1,000. No unweighted mean of campus or annual rates. No causal models or significance-based league tables. Historical data outside this window may be retained as source material, but no expanded outcome window without an explicit amendment.

## Fixed comparison cohort

All ten UC institutions; all eight Ivy League institutions; San Diego State (requested in the originating research); and 23 additional widely recognized public/private research universities. This is a purposive comparison set, not a representative national sample or a prestige ranking. Institutions are included irrespective of their crime counts. Main-campus and institution-wide units must remain explicit.

- UC: Berkeley, Davis, Irvine, Los Angeles, Merced, Riverside, San Diego, San Francisco, Santa Barbara, Santa Cruz.
- Ivy League: Brown, Columbia, Cornell, Dartmouth, Harvard, Pennsylvania, Princeton, Yale.
- Additional public: San Diego State, Georgia Tech, Illinois Urbana-Champaign, Michigan Ann Arbor, North Carolina Chapel Hill, Ohio State Columbus, Penn State University Park, Texas Austin, Florida, Virginia, Washington Seattle, Wisconsin Madison.
- Additional private: Caltech, Carnegie Mellon, Chicago, Duke, Georgetown, Johns Hopkins, MIT, Northwestern, Notre Dame, NYU, Stanford, USC.

Total planned: 42 institutions. If an exact unit cannot be matched, retain an explicit unavailable row; do not replace it based on outcomes.

## Measures

1. On-campus reported criminal offenses per 1,000 enrolled students: enrollment must match the institution/campus aggregation as closely as the source permits. Published institutional totals across campuses require an explicit scope note; no institution-wide denominator silently attached to one branch.
2. Residential-facility reported offenses per 1,000 documented residents: only actual occupancy/headcount with year and population scope stated. Beds, capacity, undergraduate housing percentages and estimated residents cannot silently substitute for occupancy. State-audit occupancy may provide an approximate normalization where property boundaries cannot be reconciled; label that limitation in the view itself.
3. Show numerator and denominator alongside every rate; missing/withheld/unavailable values are not zero. Fall enrollment includes graduate/professional students where source scope requires it. Distance-only enrollment and medical-center context are disclosed; no invented correction factors.
4. Keep primary criminal offenses, VAWA offenses, hate-crime classifications, arrests and disciplinary referrals distinct. Housing is a subset of on-campus geography. Never add it to the on-campus total. Do not add overlapping outcome categories into an undocumented total.
5. Clery reports may concern nonstudents, repeated offenses, and occurrences before the report year. Reported offenses are not unique victims or survey prevalence. Explain institution-specific corrections or clustered disclosures supported by primary records.
6. Offer individual categories, years, cohort/search filters and campus detail; default alphabetical presentation. Sorting a rate is a descriptive convenience, not a safety ranking. Do not infer that a zero count means no crime occurred.

## Verification and publication gates

- Preserve source URLs, retrieval dates, versions, hashes, tables/fields/pages and coverage notes. Resolve discrepancies against clearly identified source vintages; retain a discrepancy ledger.
- Independently recompute all counts, denominator joins, annual/pooled rates and displayed examples. Verify missingness, duplicates, geographic subsets and source-year alignment.
- Review prior official analyses and primary research for compatible definitions and interpretation; do not mix FTE, enrollment headcount, residents and survey prevalence as equivalent quantities.
- Separate final claim-by-claim citation audit from statistical checks. Check that every substantive source-dependent claim has a precise, appropriate citation and that linked evidence supports the wording.
- Render equations with accessible mathematical markup and review desktop/mobile layouts and PDF pages visually. Test keyboard access, empty states, downloads, filtering, sorting and live deployment.
- Publish only reviewed aggregate data, reproducible calculations, methods, limitations and audit findings. Preserve the completed Crime and Heat study.