How we source & present this data

The four questions behind every number

Before any figure appears on a city page, it must be able to answer four questions. If it can’t, we don’t present it as a confident value.

1. How was it counted? The definition and collection method used.
2. By whom? The named source, and where it sits in our hierarchy.
3. When? The reference year — the year the data describes, not when it was published.
4. For what exact place? The precise geographic unit — city, municipality, metro area, state, or country.

Every figure on the site is drillable: expand “What’s behind this number” on any cell to see all of this, for each contributing source.

Where our numbers come from

We work top-down through a source hierarchy and only fall to a lower tier when a higher one doesn’t cover the place or metric.

1. Primary government statistical agencies — the bodies legally responsible for counting (e.g. national statistics offices, city police departments). Least filtered, most traceable.
2. Intergovernmental bodies — UNODC, PAHO, UN Women. They standardize across countries but add a layer of processing.
3. Academic & peer-reviewed — analytically rigorous, but often lags two to three years.
4. NGOs with transparent methodology — used only when they publish their raw data and methods.
5. Aggregators — used for discovery and cross-checking only; we always trace back to the cited primary source before storing a figure.

What we exclude: figures sourced from travel blogs, tourism boards, real-estate sites, security-company marketing, or news articles that don’t cite a primary source. These have incentives to skew, or no documented method at all.

Every figure carries its provenance

For each source behind a figure we store and display:

• Source name & tier
• Link to the source (and an archived copy)
• Access date
• Reference year
• Exact geographic unit & place name
• The definition used
• Collection method
• The dataset version we pulled
• The verbatim text the number came from

We store both the raw count and the rate per 100,000 wherever a source provides them, so a population-normalized figure can always be checked against the underlying count.

Some sources publish only a count (e.g. a city’s police records), which isn’t comparable between places of different sizes. In those cases we compute the rate ourselves — count divided by an official population for the exact same area — and label it as a computed figure. The drill-down shows both sources (the count and the population) and the math, and we only do this when the population is for the same geographic unit; we never divide by a different area’s population to manufacture a rate.

When sources disagree, we show the disagreement

Crime is measured differently by different institutions — a criminal-justice source and a public-health source can count different things. When two credible sources give different values for the same figure, we do not silently pick a winner. We show the range and flag it as a conflict, because the disagreement is often more informative than a single tidy number. You can open the drill-down to see each source’s value side by side.

Comparing cities fairly

Most cross-city comparisons fail in predictable ways. We build a guard against each:

Geographic-unit mismatch

A “city” figure and a “municipality” or “metro-area” figure aren’t the same thing. We store the exact unit with every figure and surface it in the drill-down. When several figures exist for a place we show the one closest to the city itself; many official sources only report a wider area (the surrounding municipality, metro, or whole country), so when that’s all we have we still show it but badge it as covering a larger area — a figure to read with that in mind, especially when comparing against a true city-level number.

Reference-year mismatch

Comparing a 2024 figure to a 2020 one is a false comparison. When two compared cities’ figures are more than one year apart, the comparison row is visibly flagged.

Definitional mismatch

“Femicide,” “sexual assault,” and even “homicide” are defined differently across legal systems. We record the definition each source used and align to the UNODC ICCS standard as a reference point.

Reporting-rate differences

Police-recorded crime and victimization surveys measure different populations of incidents. A low reported rate in a place with weak institutions may mean low reporting, not low crime. We tag each figure’s collection method so these are never treated as interchangeable.

What we deliberately don’t show

• Unsourced numbers. If a figure can’t be traced to a retrievable source and the exact passage it came from, it doesn’t appear as a value.
• “No reliable data” is shown as itself — not as a zero, a blank, or a guess. Missing data is information; many cities genuinely lack vetted figures for some categories, and we say so plainly.
• False precision. We never present a number as more exact than its source supports. Where methodology is weak or definitions differ, we show a caveat alongside the figure rather than a clean-looking but misleading value.
• Resolved conflicts. We don’t average away or hide disagreement between sources.

How current the data is

Different metrics go stale at different rates, so we refresh them on different schedules. Travel advisories can change quickly and are checked frequently. Homicide, robbery, and femicide statistics are released roughly annually, so they’re refreshed on a slower cycle. Government datasets are sometimes revised after the fact, so we record exactly which release we pulled, and every city page shows when its data was last generated.