Numbeo’s Crime and Safety Index – More Accurate than People Magazine

You’ve probably seen news reports from “national” newspapers touting one or more of their cities’ high place in Numbeo’s “2020 Crime and Safety Index” or lamenting a city’s poor showing.

For example, “Abu Dhabi has been declared the most secure place on earth based on its low crime index of 11.33. Sharjah ranks fifth with a crime index of 16.48 while Dubai has taken the seventh spot with a crime index of 17.02.”

If you’re like AA, you wonder how Numbeo conducts its research, particularly because:

1. Numbeo is able to parse differences among cities to two places to the right of the decimal point. Precision like this would appear to be based on some very hefty methodology. Though as we saw with Transparency International’s Corruptions Perception Index, sometimes precision is more a matter of appearance than a real condition.
2. More importantly the results seem less than believable. Kuala Lumpur in Malaysia (ranking 35) is apparently less safe than Damascus (32) or Baghdad! But not as bad off as London which trails behind at 106.. Beating all of them is Kingston Jamaica in position 28.

What’s the answer?

Numbeo is a “crowd sourced” data base.

What does that mean in “small words”?

Let’s let Numbeo speak for itself.

This section is based on surveys from visitors of this website. Questions for these surveys are similar to many similar scientific and government surveys. Each entry in the survey is saved as the number in the range [-2, +2], with -2 having meaning of strongly negative and +2 meaning of strongly positive. We filter surveys to eliminate potential spam, like people entering a large amount of data which are differentiating from the median value.

What this means is that Numbeo doesn’t conduct any active data gathering or rely on third parties to do so as does Transparency International. Rather it relies on “visitors” to the website for its data.

Now those who believe that if a survey is not conducted using “proper” statistical methods, particularly those relating to sample selection, it’s not valid.

Clearly, Numbeo’s sampling/data collection doesn’t meet the statistical “test”.

So why then should we rely on its rating?

Again Numbeo has the answer:

Is this much less accurate than governmental statistics? In some countries, governments have a detailed statistics based on a number of reported crimes per capita. Those surveys are particular good in comparing crime between two cities in that country, but are not so good in cross country comparison for the following reasons: (1) people in some countries are much more likely to report a crime than in other countries (2) data could be forged by governmental institutions and (3) data are not available for most of the world

It would appear from the above that Numbeo believes that those folks who are shy in reporting crimes to their local authorities will be less reluctant to provide information to Numbeo. And that a statistically significant number of them will do so.

Sadly, the “wisdom” of crowds is generally turns out to be more an aspiration than a realistic assessment of the state of affairs.

Numbeo’s ratings then are nothing more than the perceptions of limited group of people who chose to respond (therefore a statisically “flawed” sample). And the results should be treated as being as valid as lists of the sexiest man or woman alive.