Chile Safety Map
EN ES

Methodology — How We Calculate Crime Rates

This page explains in full how we collect, process, and present the crime statistics on this site: the source of the data, the rate calculation, the underreporting caveat, how we define trend directions, the criteria we use for comparisons, and — critically — what these figures cannot tell you. We use the first-person "we" to refer to the ischilesafe.com editorial team throughout.

Data source: CEAD

All quantitative crime data on this site comes from CEAD — Centro de Estudios y Análisis del Delito, the Chilean government's official crime statistics body, which operates under the Ministerio del Interior y Seguridad Pública. Official CEAD statistics are published at cead.minsegpublica.gob.cl/estadisticas-delictuales/.

CEAD compiles police-recorded incident data from two law enforcement agencies:

  • Carabineros de Chile — the uniformed national police force, which handles the majority of recorded incidents.
  • Policía de Investigaciones (PDI) — the plainclothes investigative police, which handles more complex criminal cases.

CEAD publishes annual series from 2005 onward at the commune, regional, and national level. We access this data via CEAD's official public endpoints at cead.minsegpublica.gob.cl. The raw data is processed by our open-source pipeline, validated against a strict schema, and stored as versioned JSON files in the project's public GitHub repository. The full scraper and validation code is available for inspection.

What "casos policiales" means: We use the CEAD measure casos policiales, which combines two sub-measures additively at the offence level: denuncias (tipoVal=1 — formal complaints filed with the police) and detenciones flagrantes (tipoVal=2 — arrests in the act). These two figures are added together to give the total cases for each offence type. We do not use aprehendidos (tipoVal=3), which counts persons apprehended and would double-count events already captured by the other two measures. We use only the annual series (serie anual); we do not use quarterly or monthly figures.

Partial-year data: The current calendar year is excluded from all rates and rankings because CEAD has not yet closed its annual series for it — figures for a partial year would be incomparable to a full annual count. The most recent past year displayed (currently 2025) is treated as a complete annual series, but note that CEAD may publish minor revisions to recent years after their initial release; the figures shown reflect the data available at the time of our last pipeline run.

Drug offences scope: The CEAD category Drogas covers criminal offences under Ley 20.000 (grupo 401 in the CEAD catalog — trafficking, supply, and related crimes). Personal drug consumption is a separate category recorded under Incivilidades (grupo 702) and is not included in the Drogas figure shown on this site. As a result, Drogas rates reflect enforcement of drug-supply laws, not the prevalence of drug use.

Reference sources for specific crime types: For certain crime categories, authoritative external sources provide independent cross-checks on CEAD figures. The displayed incidence metric on this site remains CEAD casos policiales in all cases; these references are cited for transparency and context:

  • Homicides: the Subsecretaría de Prevención del Delito (SPD) publishes a dedicated homicide database (Violencia Homicida en Chile) with commune-level counts from 2018 onward, available at prevenciondehomicidios.cl. This is the authoritative government source for homicide counts and is used as a cross-reference against CEAD homicide data.
  • Economic exposure (empresas / empleados): commune-level business and employment data from the Servicio de Impuestos Internos (SII) is used as a proxy for daytime working population. SII statistics are available at sii.cl/sobre_el_sii/estadisticas_de_empresas.html.
  • Kidnappings (secuestro): this offence is not tracked in CEAD's standard statistical tables. Regional-level figures are published by the Fiscalía de Chile (Ministerio Público) in their annual prosecution statistics, available at fiscaliadechile.cl/persecucion-penal/estadisticas.

Map boundaries: chilemapas

The commune boundary polygons displayed on the interactive map are sourced from chilemapas (github.com/pachadotdev/chilemapas / pacha.dev/chilemapas), a project by Mauricio Vargas Sepúlveda licensed under the GNU General Public License v3 (GPL-3). The chilemapas boundaries are derived from official Chilean government sources: INE (Instituto Nacional de Estadísticas), SUBDERE (Subsecretaría de Desarrollo Regional y Administrativo), and BCN (Biblioteca del Congreso Nacional de Chile).

The map geometry (commune shapes) comes from chilemapas; the crime statistics overlaid on those shapes come from CEAD (described above). These are separate, independent sources.

How the rate is calculated

The primary metric on this site is the rate per 100,000 inhabitants. We use the rate as published directly by CEAD (medida = tasa por 100.000 habitantes), which applies INE (Instituto Nacional de Estadísticas) population projections as the denominator for each commune, region, and national aggregate.

Using CEAD's own published rate — rather than computing our own from raw counts and INE population figures — ensures consistency with the official source cited on every page.

Floating-population denominator caveat: The INE denominator is based on registered resident population. In business-hub communes such as Providencia and Las Condes, the daytime working population substantially exceeds the resident figure, because many people commute in from other communes. As a result, property and other crime rates in these communes may appear higher than their resident population would suggest. The rate is mathematically correct on the official INE figure; users should be aware that high rates in commercial centres partly reflect visitor and worker exposure, not solely resident risk.

Low-count volatility caveat: In communes with small absolute case counts, rates can be highly volatile year to year. A single additional homicide, for example, can cause a large swing in a small commune's per-100k figure. For this reason, we exclude low-population communes (fewer than 10,000 inhabitants) from national and regional rankings, and display individual volatility caveats on those commune pages.

Important note on national and regional figures: The national and regional totals that CEAD publishes as tasa por 100.000 are sum totals — that is, the sum of all commune rates in the territory, not a weighted per-capita average. We do not display these sum figures as averages. When we show a national average for comparison purposes (e.g., on commune pages and the lowest-reported-incidence ranking), we compute the unweighted mean of non-low-population commune rates using each commune's individual rate at its latest complete year. This is a typical-commune figure — it is not a true population-weighted per-capita average.

Communes with populations below 10,000 inhabitants are excluded from national and regional averages and from the lowest-reported-incidence ranking. Small populations produce highly volatile per-capita rates because a single incident can produce a large per-100k figure. Their individual commune pages still display the rate with a volatility caveat.

Underreporting (cifra negra)

All figures on this site represent police-reported incidents only. They do not represent the total number of crimes committed. The gap between actual crime occurrence and reported crime is called the cifra negra (literally "dark figure") or underreporting rate.

Underreporting is not uniform across crime types or localities. Key patterns documented in Chilean crime surveys (Encuesta Nacional Urbana de Seguridad Ciudadana — ENUSC):

  • Property crime (theft, robbery) has relatively high reporting rates when insurance or bureaucratic requirements (e.g., police reports for insurance claims) incentivise victims to file. But opportunistic petty theft is widely underreported.
  • Intra-family violence is significantly underreported across all socioeconomic groups. Survey-based prevalence estimates are consistently higher than police-recorded figures.
  • Homicides have the lowest cifra negra of any crime category because deaths are discovered and recorded regardless of victim reporting. Homicide rates are therefore the most reliable indicator in CEAD data for inter-commune and inter-year comparisons.

A commune showing a low reported rate may have low actual crime prevalence, high underreporting, or both. We cannot determine the proportion from police data alone. Where appropriate, individual commune pages note these caveats.

News-only crime category: sexuales

The news incident layer classifies recent incidents into 8 crime families — one more than the 7 CEAD statistical families used elsewhere on this site — the 8th being sexuales (sexual offenses). This news-only family has no CEAD quantitative counterpart: there is no per-100,000 rate for it, it is not shown on the choropleth map, and it is not an input to the composite crime index. It exists only as a qualitative classification label applied to individual items on the recent-incidents news layer, and it is never merged into the CEAD-derived statistics described elsewhere on this page.

Household Victimization Indicator (ENUSC 2024 SAE)

The ENUSC 2024 communal household victimization rate (VHDV) is an experimental statistic produced by INE using Small Area Estimation (SAE) methods. It measures the proportion of households reporting at least one violent crime, covering 136 of 346 comunas. INE labels this an "estadística experimental, etapa inicial de madurez." This indicator is displayed as an additive module on covered commune pages — it is NOT merged with the CEAD-based crime index (which reports police-recorded cases — casos policiales — per 100,000 population, a distinct measurement). Source: INE Estadísticas Experimentales, Seguridad Ciudadana.

Trend formula

Each commune page and map pin displays a trend indicator: rising, falling, or stable. We compute the trend as follows:

  1. Take the latest complete year rate (year N) and the rate three years prior (year N−3), both from the CEAD annual series.
  2. Compute the percentage change: ((rate_N − rate_N3) / rate_N3) × 100.
  3. Apply a 5 % threshold: if the change is more than +5 %, the trend is rising; if less than −5 %, it is falling; if within ±5 %, it is stable.

We use a three-year window rather than a single year-over-year comparison to reduce the influence of anomalous single-year fluctuations (e.g., pandemic-era reporting disruptions). If a commune has fewer than four complete years of data, the trend is not computed and is omitted from display.

The 5 % threshold was chosen to avoid labelling statistically marginal changes as meaningful trends in communes with small absolute case counts. For communes with large populations and stable series, the threshold provides a conservative signal; for smaller communes, year-to-year variation can exceed 5 % due to sample noise rather than genuine change. Both the 5 % threshold and the 3-year window are editorial parameters — they are documented as editorial choices in the project's SOURCES registry, not derived from an external standard.

Weighting note (F2 vs F5): The national and regional time-series rates shown in trend charts use population-weighted commune means — correct for tracking genuine per-capita change across time. By contrast, the national and per-family aggregate averages shown on commune pages and ranking heroes are unweighted means of eligible commune rates (the typical-commune figure described above). These two approaches answer different questions: unweighted = "what is the typical commune like?"; population-weighted = "what does the average resident experience?". Both figures are clearly labelled on the pages where they appear.

Comparison criteria

When we compare communes to each other or to a national average, we apply the following criteria consistently:

  • Metric: rate per 100,000 inhabitants (CEAD-published figure, as described above).
  • Year: the latest complete year for which CEAD has published a closed annual series for all communes. This is the same year for all communes in a given build. The current calendar year is excluded as partial (see partial-year note above).
  • Exclusion of low-population communes: communes with fewer than 10,000 inhabitants are excluded from national and regional rankings due to statistical volatility (DATA-04 rule). Their pages remain accessible but note the exclusion.
  • Ranking basis: the national rank shown on commune pages is computed across all non-low-population communes in Chile (currently 346 total communes, of which a subset meet the population threshold). Rank 1 = highest reported rate in Chile for that year — rank 1 identifies the commune with the most reported crime, not the one with the least.
  • National average definition: the unweighted mean of non-low-population commune rates (see rate calculation section above). Not the CEAD national sum.

We compare communes only to the national mean or to regional peers — not to international figures, because reporting standards, legal definitions, and demographic structures differ substantially across countries, making cross-country rate comparisons unreliable.

Per-family national mean (F3)

On crime-ranking pages, the National mean hero figure shows the unweighted mean of eligible commune rates for that specific crime family. It uses the same typical-commune basis as the overall national aggregate described above: non-low-population communes only, latest complete year, unweighted (not population-weighted). It answers "what is the typical Chilean commune's rate for this crime type?" — it is not a national per-capita average.

Per-region breakdown mean (F4)

On crime-ranking pages, the regional breakdown bars show the unweighted mean of eligible commune rates within each region for the displayed crime family. Communes with fewer than 10,000 inhabitants are excluded from this regional mean for the same volatility reasons that apply to the national figure. This is a typical-commune-within-a-region figure, not a population-weighted regional rate.

Incidence level classification — LevelChip 1–5 (F6)

Each commune page displays an incidence level chip rated 1 to 5. This is a relative classification derived at build time from the distribution of non-low-population commune rates: eligible communes (those with 10,000 or more inhabitants) are divided into five approximately equal tiers from lowest (1) to highest (5) reported incidence for the latest complete year. A level 3 commune sits near the national median for reported incidence; a level 5 commune ranks in the top tier.

This is a relative ranking, not an absolute threshold. The tier boundaries shift each build year as the rate distribution changes. It is not a safety verdict — it describes where a commune falls within the distribution of police-recorded incidence, subject to all the caveats about underreporting, population denominators, and crime-type scope described on this page. See also the choropleth map colour scale, which uses the same 5-tier classification.

Composite crime index (F13 + F14)

In addition to the per-family CEAD rates described above, this site displays a composite crime index — a single normalised score that combines seven per-100,000 metrics into one comparable figure for each commune. The composite is used only for relative comparisons across communes; it is not an absolute safety verdict.

Formula

The composite score is a weighted sum of seven winsorized metrics (reference year 2024): 

composite_score = sum(w ₂₃ × normalized ₂₃)  for each metric

The locked weight vector (C-02) is:

  • spd_homicide_rate — 0.30 (M1)
  • cead_robos_rate — 0.20
  • cead_vida_rate — 0.15
  • cead_propiedad_rate — 0.12
  • cead_vif_rate — 0.10
  • cead_drogas_rate — 0.08
  • cead_armas_rate — 0.05

Normalization method

Each metric is first winsorized (limits [0.01, 0.01] — the top and bottom 1 % of commune values are clipped) before rescaling to [0, 1] via min-max normalization. Winsorization prevents extreme outliers (micro-communes with very small populations can produce per-100k rates in the tens of thousands) from collapsing the colour scale for the remaining 99 % of communes.

M1 — homicide: SPD VHC switch (F13)

The homicide metric used in the composite index (M1, spd_homicide_rate) comes from the SPD VHC dataset (Subsecretaría de Prevención del Delito / Centro para la Prevención de Homicidios — Violencia Homicida en Chile), not from CEAD grupo 101. CEAD grupo 101 ("Homicidios y femicidios") cannot isolate homicide from femicide at the subgroup level, making SPD VHC the authoritative source for the index. The CEAD-sourced homicide figure is preserved for the per-family display on commune pages (F1).

SII exposure caveat (F14)

One of the seven metrics is an economic exposure proxy derived from the Servicio de Impuestos Internos (SII): the number of dependent workers (trabajadores dependientes informados) registered in each commune. This figure uses the firm's registered domicile (HQ), not physical establishment location, which concentrates counts in business-hub communes. To prevent this distortion from dominating the composite, when the ratio of SII workers to INE resident population exceeds 5.0, the metric falls back to the INE resident population as the denominator (5.0 is a fallback trigger threshold, not a multiplier applied to INE). Communes without SII data have their available metric count decremented by one.

All composite methodology parameters are documented in the project's SOURCES registry.

What this site does NOT say

The following statements are not made anywhere on this site, and if you see language that implies them, please report it as a bug:

  • We do not classify any commune, city, or region as having an absolute level of risk. The map and ranking pages present relative reported incidence — lower or higher than other communes — not absolute risk levels. A commune that ranks low for reported crime is not declared free of crime; a commune that ranks high is not declared to have unacceptable risk.
  • We do not publish probability-of-victimisation figures. Reported rates are not individual risk probabilities. A rate of 1,000 per 100,000 does not mean a 1 % annual probability of becoming a crime victim — it means 1,000 police-recorded incidents per 100,000 registered residents, with all the underreporting and denominator caveats described above.
  • We do not make predictive or forward-looking claims. Trend indicators describe what CEAD data shows for past years; they do not predict future conditions. A "falling" trend does not mean a commune will continue to decrease.
  • We do not rank neighbourhoods within communes. CEAD data is available at the commune level only. We cannot show which barrio or street within a commune has higher or lower reported incidence.
  • We do not compare Chile to other countries. International crime rate comparisons require harmonised definitions and reporting standards that do not currently exist for the countries most relevant to our audience.

Our goal is to make official CEAD data accessible and clearly presented — not to produce verdicts about territories. If you have questions about the data or methodology, use the Contact page.