Methodology

The photos were downloaded from the Instagram API in two sets of five days: from Jan. 28-Feb.2, 2014 and Mar. 3-7, 2014.Each day’s data consists of the 24-hour period from midnight to midnight to account for all time zones equally. During that time, 402,197 Instagram photos tagged “selfie” that included geographic coordinates. That only accounts for a fraction of all selfies uploaded during that time since the majority of photos do not include a location. Of this sample, 28 percent came from the United States.

For every city in the world of at least 250,000 residents, we then counted the number of selfies taken within 5 miles and divided by the population of that city.

While the metric of “people who take selfies per capita” is far from a perfect measure, it was far and away the most comprehensive means of comparing the 459 world cities that turned up at least 25 individual users in the database. Of course, this measure bakes in more variables than just a penchant for selfies, such as smart phone adoption rates and median income. As such, this list tends to favor regions just outside major urban areas.

Not all photos tagged as “selfie” are in fact selfies. Informal tests that we ran on our dataset using out-of-the-box facial detection algorithms found that the vast majority of the photos were of a single person. We also made several attempts to find international versions of the word “selfie,” but none of the suggested translations showed up in any appreciable volume.

The list of world cities and their populations comes from the GeoNames database. In almost all cases, cities were limited to municipalities as opposed to culturally defined population centers.

Photos were then matched to cities using a simple “nearest neighbor” algorithm that found the nearest city with at least 250,000 residents. This means that, even within the fairly tight radius of 5 miles from the center of a city, some photos are counted which were not physically taken in that city. (See the map of Manhattan, for instance, which includes a small sliver of New Jersey.) All of our tests suggested this was an acceptable tradeoff for the speed of this method of geocoding, which also serves to group the innermost suburbs with the city that they surround.

The data is not being permanently stored on TIME’s servers, in accordance with Instagram’s terms of service.