What City Crime Index Numbers Really Measure

How is a city crime index calculated?

Most of the crime index numbers you see on travel blogs, relocation sites and ranking pages come from one source: Numbeo, a crowdsourced database run out of Belgrade that lets anyone with an internet connection answer survey questions about any city. There are no interviewers, no sampling frames, no police reports behind the numbers. The index is a running average of what users voluntarily submitted.

Numbeo asks respondents 15 questions on a scale from very negative (-2) to very positive (+2). The questions cover a general perception of crime, whether crime has gotten worse in the last three years, how safe the respondent feels walking alone during the day and at night, worries about mugging, robbery, car theft, physical attack, verbal harassment, bias-motivated attacks, property crime severity and violent crime severity (Numbeo methodology page). Each answer is rescaled to 0-100 using the formula (response + 2) × 25. The 15 values are then averaged, with the "general level of crime" question weighted three times. Scores above 20 get an exponential adjustment that pushes bad perceptions further up the scale.

The published thresholds are: 20 or below is very low, 20 to 40 is low, 40 to 60 is moderate, 60 to 80 is high, and over 80 is very high. The Safety Index is just 100 minus the Crime Index, which is why a city with a crime score of 25 has a safety score of 75. They are not two separate measurements.

Is a crime index of 58 dangerous?

Our best-cities ranking lists Medellín with a crime index of 58, which Numbeo's own scale puts at the top of the "moderate" band, just shy of "high." Read that in isolation and the obvious conclusion is that moving to Medellín is dangerous. The police and prosecutor data tell a more complicated story.

Medellín's mayor's office reported 300 homicides in 2024, a rate of 11.04 per 100,000 residents, the lowest the city has recorded since records began in the 1940s (Colombia One). That figure ticked up to 13.13 per 100,000 in 2025 (Medellín Rainbow summary of city data). For comparison, Nashville, which carries a crime index of 52 in our data and sits in the same "moderate" band, recorded about 16.3 homicides per 100,000 in 2024, higher than Medellín's rate that year (WSMV coverage of FBI data).

What the 58 captures is something real, but it is not raw danger. It is the composite worry of the users who answered Numbeo's survey about Medellín: concern about theft on the metro, bag-snatching in El Poblado, scopolamine incidents, and the city's reputation among foreigners who remember the 1990s. Medellín Advisors' review of city statistics notes that homicide is only one axis, and robbery and extortion remain concerns (Medellín Advisors crime statistics). The index signals that people feel unsettled. It does not tell you your probability of being victimized in a given year.

Why does Medellín score so much worse than Copenhagen on crime?

The gap between Medellín's 58 and Copenhagen's 25 feels enormous, and the underlying reality is that Copenhagen is safer by almost every measurable indicator. Denmark's national homicide rate was 0.8 per 100,000 in 2024, about a fourteenth of Medellín's that year (Macrotrends homicide data). The Economist Intelligence Unit's Safe Cities Index, which ranked 60 cities across 76 indicators spanning digital, health, infrastructure, environmental and personal security, placed Copenhagen first in its most recent edition in 2021 (Cities Today).

Part of the size of the gap on Numbeo is a methodology artifact. The exponential adjustment above 20 widens the distance between moderate-crime cities and high-crime cities on the index, so a city whose respondents feel moderately worried ends up looking much worse than its raw averages would suggest. The questions also compound. If a city has a visible homeless population, aggressive panhandling, or well-publicized stories about gang violence, nearly every one of the 15 questions will drift negative at once, even if the actual homicide rate is modest. Nordic cities benefit from the opposite dynamic: Copenhagen's visible normalcy pulls all 15 questions toward the positive end.

So the index captures a genuine signal, Copenhagen users do not feel unsafe and Medellín users do, but it compresses two different things into one number: how much crime is happening, and how much the respondent pool worries about it. A tourist shaken by one aggressive street encounter in Medellín can file a survey that looks identical to an answer from a resident whose apartment was burgled, and both get the same weight.

What is the difference between a crime rate and a crime index?

A crime rate is a count of recorded offenses divided by population, almost always expressed per 100,000 people. In the United States this traces back to the FBI's Uniform Crime Reporting program, which began in 1929 and is now carried forward by the National Incident-Based Reporting System (BJS overview). A crime rate is reproducible: if you know the number of murders reported in a jurisdiction and the population, you can calculate it yourself and check it against the agency's filing.

The historical FBI "Crime Index" was the sum of eight Part I offenses (murder, forcible rape, robbery, aggravated assault, burglary, larceny-theft, motor vehicle theft, arson). The FBI discontinued that specific term because the sum was dominated by larceny and distorted comparisons. The official statistical community dropped "Crime Index" as too crude, and the version you see now on travel sites is a different animal built from user surveys.

The practical difference is epistemic. A crime rate is a measurement with a defined denominator and a clear data source. A perception-based crime index is a poll with a self-selected sample. Both can be informative, but they answer different questions, and treating them as interchangeable is how readers end up believing that one city is three times safer than another on a precise numeric basis when the underlying data does not support that precision.

Are Numbeo crime scores reliable?

The honest answer is: partially, and only if you know what you are looking at. Numbeo is useful because no one else gathers continuous, comparable, free data across thousands of cities. It is also unreliable in specific, documented ways.

The sample is self-selected. Only people who happen to visit numbeo.com, read English well enough to parse the survey, and feel motivated to answer take the poll. That skews toward expats, digital nomads, relocation researchers and travelers. VERA Files' review of Manila's rankings noted that as of April 2024 Manila's data came from 417 users accumulated over five years, far below any survey research threshold (FactRakers summary of VERA findings). Users can rate cities they have never visited, and the platform does not verify who answered.

The system has been shown to be manipulable. In January 2017, a Swedish developer named Linus Trulsson exploited a flaw that let one IP address submit unlimited negative ratings for Lund, a university town of 12 prior reviews with a crime rate below the Swedish national average, and in less than a day pushed it to the top of Numbeo's "most dangerous cities" list (The Local, Sweden). Numbeo closed the exploit within 24 hours, marked the submissions as spam, and has since published a page on how it addresses manipulation (Numbeo, addressing data manipulation). The incident illustrated how small a push is needed to swing a city with few contributors.

Fact-checkers have repeatedly flagged the resulting rankings as misleading. Africa Check examined the 2019 claim that six South African cities ranked among the world's twenty most dangerous and concluded it was not supported by official statistics. The pattern is consistent: perception rankings for cities with vocal online communities drift away from the police data over time.

Can I trust crime index numbers when choosing where to live?

A crime index is useful for one specific thing: comparing how people who bother to answer the survey feel about safety across cities. That is not nothing. If you are choosing between Krakow at 28 and Charlotte at 45 for a two-year contract, the fact that Krakow respondents report feeling safer walking at night is a real data point, especially if Krakow has several thousand contributors behind its score.

For a decision as large as where to live, lean on three layers, not one. First, pull the official homicide and violent crime rate from the national statistics office or police report; they are published for most major cities. Second, read neighborhood-level accounts in local newspapers, which tell you whether the risk is concentrated in specific areas you can simply not live in, which is almost always the case. Third, if the city appears on composite indices like the EIU Safe Cities Index or Mercer's Quality of Living Ranking (Wikipedia summary of indices), use those as a sanity check because they draw on 39 to 76 weighted indicators rather than one poll.

The pattern you want is agreement between sources. Our data gives Zurich and Singapore a crime index of 15. Singapore ranked third in the EIU's 2021 Safe Cities Index, and Zurich placed first in Mercer's 2024 Quality of Living Ranking (Mercer). Numbeo, the composite indices and the official violent crime data all point the same way for both cities. When they disagree, as they often do for cities like Medellín, Marseille or Johannesburg, the disagreement itself is the information: the city is complicated, the risk is uneven, and a single number cannot summarize it.

When the index misleads, and when it is useful

The index misleads most when a city has a famous past that lingers in travel writing, when its contributor pool is small or skewed to one demographic, or when reporters covering the rankings treat the number as hard crime data. Medellín is the textbook case of the first, Manila of the second, and almost every "world's most dangerous cities" listicle of the third.

The index is most useful as a rough sort across cities you already have a shortlist for, as a way to check how foreign residents feel about day-to-day life, and as a signal for what to ask locals about. A moderate score with a lot of contributors is a prompt to investigate, not a verdict. If you are picking between Valencia, Tallinn and Dublin for a nomad base, the crime index differences are small enough (32, 25, 32) that they should not drive the decision. The more important variables are your neighborhood, your routine, and what the local reporting actually says.