Skip to content

DAAF-Contribution-Community/daaf / .claude/skills/education-data-source-campus-safety

education-data-source-campus-safety

Verified

CSS — annual Clery Act crime/fire safety for Title IV institutions. Portal: hate crimes only (2005-2021); primary offenses, VAWA, arrests, fire safety need ope.ed.gov directly. Use for campus crime analysis. Identified by IPEDS unitid.

Campus Safety Data Source Reference

Campus Safety and Security (CSS) — annual Clery Act crime and fire safety reporting for all Title IV postsecondary institutions. Full dataset covers primary offenses, VAWA offenses, hate crimes, arrests/referrals, and fire safety. Portal mirror provides only hate crimes data (2005-2021); primary offenses, VAWA, arrests, and fire safety require direct access to ope.ed.gov/campussafety. Use when analyzing campus hate crimes via Portal or understanding Clery Act reporting structure. Identified by IPEDS unitid. Hazing statistics were added by Pub. L. 118-173 (applies to annual security reports from calendar year 2026 onward, per the Act's 2-years-after-enactment effective-date provision); they are not in the Portal mirror.

Guide to understanding and using campus crime and fire safety data collected under the Clery Act.

CRITICAL: Value Encoding

The Education Data Portal mirrors use integer codes for categorical variables, not string labels:

VariableExample CodeMeaning
bias1Race-based bias
crime_type13Intimidation
fips6California

See ./references/variable-definitions.md for complete code mappings.

Truth Hierarchy: When interpreting variable values, apply this priority:

  1. Actual data file (what you observe in the parquet/CSV) -- this IS the truth
  2. Live codebook (.xls in mirror) -- authoritative documentation, may lag
  3. This skill documentation -- convenient summary, may drift from codebook

If this documentation contradicts the codebook, trust the codebook. If the codebook contradicts observed data, trust the data and investigate. Use get_codebook_url() from fetch-patterns.md to download codebooks.

Codebook coverage gap (verified 2026-08-07): the mirrored csafety .xls codebook documents only 7 of the 13 data columns — the bias category and the 6 geographic offense counts — and omits crime_type, year, unitid, inst_name, opeid, and fips. For those 6 omitted columns the codebook is silent, so this skill's own data-verified variable documentation (see ./references/variable-definitions.md) is the better authority; the "trust the codebook" step above applies only to columns the codebook actually covers.

What is Campus Safety and Security Data?

The Campus Safety and Security (CSS) data comes from the annual survey required by the Jeanne Clery Campus Safety Act (renamed by Pub. L. 118-173, Dec. 23, 2024; formerly the Jeanne Clery Disclosure of Campus Security Policy and Campus Crime Statistics Act):

  • Federal mandate: All Title IV institutions must report crime and fire statistics annually
  • Consumer protection: Designed to inform students, parents, and employees about campus safety
  • Primary source: U.S. Department of Education Office of Postsecondary Education
  • Coverage: All postsecondary institutions receiving federal financial aid
  • Reporting period: Calendar year (January 1 - December 31) in which the crime was reported (34 CFR 668.46(c)(3)); the Portal year is that calendar year
  • Available years: 2005-2021 (hate crimes in Portal mirrors)
  • Primary identifier: unitid (IPEDS institution ID, 6-digit integer)
  • Official portal: https://ope.ed.gov/campussafety/
  • Portal mirror coverage: Hate crimes only. For primary offenses, VAWA, arrests, and fire safety, access the Department of Education portal directly.

Reference File Structure

FilePurposeWhen to Read
clery-act.mdLegal framework, history, reporting requirementsUnderstanding compliance context
crime-categories.mdCriminal offenses, definitions, classification rulesInterpreting crime statistics
vawa-offenses.mdDating violence, domestic violence, stalkingAnalyzing gender-based violence data
hate-crimes.mdBias categories, additional offenses, classificationWorking with hate crime statistics
campus-geography.mdOn-campus, residence halls, public property, noncampusUnderstanding location categories
fire-safety.mdFire statistics, safety systems, HEOA requirementsAnalyzing residential fire data
arrests-referrals.mdDrug, alcohol, weapons violationsWorking with disciplinary data
variable-definitions.mdKey variables, data structure, identifiersBuilding queries
limitations.mdUnderreporting, comparability issues, interpretationCritical data analysis

Decision Trees

What type of crime data do I need?

Crime data type?
├─ Violent crimes (murder, assault, robbery, sex offenses)
│   └─ See ./references/crime-categories.md
├─ Property crimes (burglary, motor vehicle theft, arson)
│   └─ See ./references/crime-categories.md
├─ Sex offenses (rape, fondling, incest, statutory rape)
│   └─ See ./references/crime-categories.md
├─ Dating/domestic violence, stalking
│   └─ See ./references/vawa-offenses.md
├─ Hate crimes (bias-motivated)
│   └─ See ./references/hate-crimes.md
├─ Drug/alcohol/weapons violations
│   └─ See ./references/arrests-referrals.md
└─ Fire incidents in residence halls
    └─ See ./references/fire-safety.md

Where did the crime occur?

Understanding crime location?
├─ On-campus (main campus buildings/grounds)
│   └─ See ./references/campus-geography.md
├─ On-campus student housing (subset of on-campus)
│   └─ See ./references/campus-geography.md
├─ Public property (streets, sidewalks adjacent to campus)
│   └─ See ./references/campus-geography.md
├─ Noncampus (off-site but institution-controlled)
│   └─ See ./references/campus-geography.md
└─ Not sure which category
    └─ See ./references/campus-geography.md (definitions section)

Can I compare schools?

Planning institutional comparisons?
├─ Want to rank schools by safety
│   └─ See ./references/limitations.md (CRITICAL: comparability issues)
├─ Comparing similar institution types
│   └─ See ./references/limitations.md (controlling for factors)
├─ Analyzing trends over time
│   └─ See ./references/limitations.md (reporting changes)
└─ Understanding what statistics mean
    └─ See ./references/limitations.md (interpretation guidance)

Quick Reference: Crime Categories

Crime Type Codes (Portal Integer Encoding)

Labels are the Portal variable-metadata labels for endpoint 130 (fetched 2026-10-19). The era column records which codes were observed in the UC Berkeley (unitid 110635) 2010/2016 Portal probes; it is an observation, not a documented rule.

CodePortal labelCategoryObserved in UC Berkeley 2010/2016 Portal probes
1MurderCriminal offense2010 and 2016
2RapeSex offense (2014+ category)2016 only
3FondlingSex offense (2014+ category)2016 only
4Aggravated AssaultCriminal offense2010 and 2016
5IncestSex offense (2014+ category)2016 only
6Statutory RapeSex offense (2014+ category)2016 only
7RobberyCriminal offense2010 and 2016
8BurglaryCriminal offense2010 and 2016
9Motor Vehicle TheftCriminal offense2010 and 2016
10ArsonCriminal offense2010 and 2016
11Simple AssaultHate crime only2010 and 2016
12LarcenyHate crime only2010 and 2016
13IntimidationHate crime only2010 and 2016
14Destruction, Damage, or VandalismHate crime only2010 and 2016
15Forcible Sex OffenseSex offense (pre-2014 category)2010 only
16Neglient Manslaughter [sic] (negligent manslaughter)Criminal offense2010 only
17Nonforcible Sex OffenseSex offense (pre-2014 category)2010 only
18Body InjuryUndocumentedNot observed

Note: Crime type code 99 (Total) is documented in some references but does NOT appear in the actual Portal mirror data. The data contains individual crime types 1-18 only. VAWA offenses (domestic violence, dating violence, stalking) are not in the hate-crimes file and have no crime_type code. crime_type == 16 is negligent manslaughter; it appeared in the 2010 probe but not the 2016 probe. The 2019 CSS hate-crime form has no negligent-manslaughter item, although 34 CFR 668.46(c)(1)(iii)(A) includes negligent manslaughter among hate-crime-reportable offenses. On the pinned mirror (HF revision 0ad00ce0, probed 2026-10-28), code 16 rows occur only in years 2005-2013 and in no year from 2014 on, and no code-16 row in any year has a positive total_hate_crimes.

Bias Category Codes (Portal Integer Encoding)

Portal labels as above; the parenthesized meanings come from matching Portal 2017 rows to UC Berkeley's CY2017 Annual Security Report hate-crime table. Code order 1-8 matches the column order of ED's CSS hate-crime form (Race, Religion, Sexual orientation, Gender, Gender Identity, Disability, Ethnicity, National Origin; 2019 CSS form), and code 2 = Religion also matched UC Riverside CY2015 and CY2016 ASR rows. Row-level Portal/ASR disagreements occur (UC Berkeley CY2016: two bias codes; LSU CY2017: one on-campus intimidation hate crime absent from the Portal); treat these as institution submission or revision differences, not a code change.

CodePortal labelMeaningObserved in UC Berkeley 2010/2016 Portal probes
1RaceRace2010 and 2016
2ReligionReligion2010 and 2016
3SexSexual orientation (ASR match)2010 and 2016
4GenderGender2010 and 2016
5Gender IdentityGender identity2016 only
6DisabilityDisability2010 and 2016
7EthnicityEthnicity (ASR match)2016 only
8NationalityNational origin (ASR match)2016 only
9Ethnicity (Pre 2014)Pre-2014 ethnicity category2010 only
99TotalAll bias categories combined2010 and 2016

VAWA 2013 (Pub. L. 113-4) added national origin and gender identity to the statutory bias categories (ethnicity was already listed). The Portal carries pre-2014 ethnicity as code 9 and post-2014 ethnicity/national origin as codes 7/8 — do not splice 9 with 7 without documenting the break.

VAWA Offenses (full CSS data only; not in the Portal mirror)

OffenseDefinition
Domestic ViolenceViolence by current/former spouse, cohabitant, or similar
Dating ViolenceViolence by person in romantic/intimate relationship
StalkingCourse of conduct causing fear for safety

VAWA 2013 (Pub. L. 113-4) required good-faith reporting of these offenses starting with calendar-year 2013 statistics in the October 1, 2014 ASR; the final regulations (79 FR 62752, Oct. 20, 2014) took effect July 1, 2015. See ./references/vawa-offenses.md.

Hate Crime-Only Offenses

Codes 11-14 (Simple Assault, Larceny, Intimidation, Vandalism) are hate-crime-only offenses. They are not standalone Clery crimes unless bias-motivated.

Arrests and Disciplinary Referrals

CategoryWhat's Counted
Liquor Law ViolationsArrests + Disciplinary referrals
Drug Law ViolationsArrests + Disciplinary referrals
Weapons ViolationsArrests + Disciplinary referrals

Key Identifiers (Hate Crimes Dataset)

IDFormatLevelExampleNotes
unitid6-digit integerInstitution100654IPEDS institution ID
opeid8-character StringInstitution00100200OPE institution ID (126 nulls in data)
inst_nameStringInstitutionUniversity of AlabamaInstitution name
fipsIntegerState6State FIPS code

Note: The hate crimes dataset uses inst_name (not instnm). Columns like branch, address, city, zip, state, sector, enrollment are NOT present in the Portal mirror data. Join with IPEDS directory on unitid to obtain institutional characteristics.

Geographic Categories

LocationDefinitionReported Separately?
On-CampusBuildings/property owned/controlled by institution, reasonably contiguousYes
On-Campus Student HousingResidence halls (subset of on-campus)Yes (as subset)
NoncampusInstitution-owned/controlled property not contiguous; student org propertiesYes
Public PropertyStreets, sidewalks, parking within or immediately adjacent to campusYes

Fire Safety Data

Fire data is reported only for on-campus student housing facilities:

Data ElementDescription
Number of firesTotal fires per building per year
Cause of fireIntentional, unintentional, undetermined
InjuriesFire-related injuries requiring treatment
DeathsFire-related fatalities
Property damageEstimated dollar value
Fire safety systemsSprinklers, alarms, smoke detectors, etc.

Missing Data Codes

CodeMeaningWhen Used
nullMissing/not reportedData not submitted or not applicable (e.g., 352k null values in bias column)
0Zero incidentsExplicitly reported as zero (distinct from missing)

Empirically verified: The hate crimes dataset in the Portal mirror does NOT use -1, -2, or -3 coded values. All missing data appears as null. This differs from some other Portal datasets (e.g., CCD) that use negative integer codes. Always verify against the actual data.

Data Access

Datasets for Campus Safety are available via the mirror system. See datasets-reference.md for canonical paths, mirrors.yaml for mirror configuration, and fetch-patterns.md for fetch code patterns.

DatasetTypeYearsPathCodebook
Hate CrimesSingle2005-2021csafety/colleges_csafety_hate_crimescsafety/codebook_colleges_csafety_hate_crimes

CRITICAL: Limited Data in Portal Mirrors -- Only hate crimes data is available in the Portal mirrors (1 dataset). For other campus safety data (primary offenses, VAWA, arrests/referrals, fire safety), access the Department of Education directly at https://ope.ed.gov/campussafety/. Consult datasets-reference.md for the authoritative list of available mirror datasets.

Codebooks are .xls files co-located with data in all mirrors. Use get_codebook_url() from fetch-patterns.md to construct download URLs:

url = get_codebook_url("csafety/codebook_colleges_csafety_hate_crimes")
# get_codebook_url() is a Python helper; in R, build the URL from the mirror root.
# Mirror failover: see `education-data-query/references/fetch-patterns.md` (R pattern)
mirror <- yaml::read_yaml("mirrors.yaml")$mirrors[[1]]
url <- paste0(mirror$root_url, "/", mirror$vintage$hf_revision, "/", "csafety/codebook_colleges_csafety_hate_crimes", ".xls")

Fetching Data

Use the fetch_from_mirrors() pattern from fetch-patterns.md:

import polars as pl

# Fetch hate crimes data (all years, single-file dataset)
df = fetch_from_mirrors("csafety/colleges_csafety_hate_crimes")

# Filter by year
df_2021 = df.filter(pl.col("year") == 2021)

# Filter by bias category (1 = Race)
race_crimes = df.filter(pl.col("bias") == 1)

# Filter by crime type (13 = Intimidation)
intimidation = df.filter(pl.col("crime_type") == 13)

# Filter by institution
institution = df.filter(pl.col("unitid") == 100654)

# Filter by state FIPS code (6 = California)
california = df.filter(pl.col("fips") == 6)
# R equivalent
library(arrow)
library(dplyr)

# fetch_from_mirrors() is a Python helper; in R, build the URL from the mirror
# root — the canonical path is NOT a local file. Mirror failover: see
# `education-data-query/references/fetch-patterns.md` (R pattern)
mirror <- yaml::read_yaml("mirrors.yaml")$mirrors[[1]]
# NOTE: illustrative only — mirror parquet files are Polars-written and may
# declare string_view columns, so a plain read can fail under R arrow
# ("cannot handle Array of type <utf8_view>"). Real fetch scripts must use the
# view-safe parquet read from `education-data-query/references/fetch-patterns.md`.
df <- read_parquet(paste0(mirror$root_url, "/", mirror$vintage$hf_revision, "/",
                          "csafety/colleges_csafety_hate_crimes", ".", mirror$format))

# Filter by year
df_2021 <- df |> filter(year == 2021)

# Filter by bias category (1 = Race)
race_crimes <- df |> filter(bias == 1)

# Filter by crime type (13 = Intimidation)
intimidation <- df |> filter(crime_type == 13)

# Filter by institution
institution <- df |> filter(unitid == 100654)

# Filter by state FIPS code (6 = California)
california <- df |> filter(fips == 6)

Direct from Department of Education (Non-Mirror Data)

For data NOT available in Portal mirrors (primary offenses, VAWA, arrests, fire safety):

Common Pitfalls

PitfallIssueSolution
Using string codesPortal uses integer codes for bias, crime_type, etc.Always map integers via ./references/variable-definitions.md
Ranking schools by crime countsHigher numbers may indicate better reporting, not worse safetyNormalize by enrollment; note reporting culture differences in ./references/limitations.md
Comparing across years naivelyCrime definitions changed over time (VAWA 2013: CY2013–2014 statistics transitional, final-rule definitions from CY2015; Portal crime_type/bias codes differ between pre-2014 and 2014+ rows)Restrict comparisons to consistent-definition periods; document breaks
Summing total_hate_crimesAppears to add the residence-hall subset to on-campus (double-counting housing incidents — verified against UC Berkeley CY2016–2017 and UC Riverside CY2015 ASRs)Recompute as on_campus + non_campus + public_property (+ other)
Treating zero as missingZero means explicitly reported zero incidents; missing means not reportedCheck for null (missing) vs 0 (zero incidents) before analysis
Comparing CSS to FBI UCR dataDifferent definitions, scope, and reporting requirementsNever equate CSS and FBI statistics; see ./references/limitations.md
Assuming all crime data in mirrorMirror only has hate crimesUse Department of Education portal for primary offenses, VAWA, arrests, fire safety
Using instnm as column nameHate crimes data uses inst_name, not instnm (which is the IPEDS convention)Check actual column names with df.columns
Expecting -1/-2 missing codesHate crimes data uses null for missing, not negative integer codesCheck for nulls with pl.col("bias").is_null(), not pl.col("bias") == -1
Assuming one row per unitid × year × crime_type × biasObserved on the pinned mirror (Portal v0.26.1, HF revision 0ad00ce0, probed 2026-10-27): 5,994,330 of the file's 15,230,698 rows are exact duplicates, with up to 128 copies of one key. After dropping exact duplicates, 7,530 keys at 795 institutions still have 2-4 rows with differing counts (e.g., UC Riverside, unitid 110671, 2017-2020; its 2021 duplicates are all exact). opeid and inst_name do not distinguish them, and there is no branch or campus column. Before dedup, the number of duplicated keys rises about tenfold between 2008 (8,860) and 2009 (84,903). The Portal API showed the same multi-row keys for UC Berkeley 2016 and UC Riverside 2017Drop exact duplicate rows first (this would also merge two genuine rows that happen to report identical counts). For the multi-row keys that remain, do not apply a sum or max by default: the mirror cannot tell a second campus from a duplicated record. Inspect the affected institutions (Annual Security Reports, or the full CSS download per ./references/campus-geography.md), decide, and state the rule as an analytic assumption. These counts are a mirror observation, not documented Portal behavior
Ignoring underreportingMany crimes go unreported; CSS captures only reported incidentsAcknowledge underreporting as a limitation in all analyses

Reporting Timeline

DateAction
Calendar yearData collection period
October 1Annual Security Report (ASR) due
Date set annually by EDCSS web-survey submission (34 CFR 668.41(e)(5))
Following springData available publicly

Related Data Sources

SourceRelationshipWhen to Use
education-data-source-ipedsInstitutional characteristicsJoining with IPEDS directory for enrollment, sector, control
education-data-explorerParent discovery skillRouting questions to mirror dataset files
education-data-queryData fetchingDownloading parquet/CSV files

Topic Index

TopicReference File
Clery Act history and requirements./references/clery-act.md
Criminal offense definitions./references/crime-categories.md
Sex offense classification./references/crime-categories.md
VAWA amendments./references/vawa-offenses.md
Dating violence definition./references/vawa-offenses.md
Domestic violence definition./references/vawa-offenses.md
Stalking definition./references/vawa-offenses.md
Hate crime bias categories./references/hate-crimes.md
Hate crime offenses./references/hate-crimes.md
On-campus definition./references/campus-geography.md
Residence hall reporting./references/campus-geography.md
Public property definition./references/campus-geography.md
Noncampus property./references/campus-geography.md
Fire statistics./references/fire-safety.md
Fire safety systems./references/fire-safety.md
Arrests vs referrals./references/arrests-referrals.md
Drug/alcohol violations./references/arrests-referrals.md
Weapons violations./references/arrests-referrals.md
Variable names and codes./references/variable-definitions.md
Underreporting issues./references/limitations.md
Institutional comparisons./references/limitations.md
Trend analysis caveats./references/limitations.md
Data interpretation./references/limitations.md
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 321kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 197kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 197kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 197kUpdated Claude CodeCodexVerifiedInstall