At Teleport, we collect a lot of data on cities all over the world. We help users slice and dice this data and figure out which city is optimal based on their individual preferences. We also see what Teleport users care about when making their choices.

Previously, we had looked at how similar cities are based on city data. In this post we’ll take a look at whether geographic origin plays a role in what our users are looking for.

On Teleport Cities, every user can choose from hundreds of criteria when ranking cities—from income taxes and quality of universities to mountain access and LGBT friendliness. In order to keep the analysis manageable, we reduced the number of distinct preferences to around 50. For example, instead of looking at all the various pollution factors a user can choose from, we simply register whether a user cares about low pollution or not.

By simply tabulating the results from this, we can see that the five most important preferences are:

  1. Low pollution – 72% of users
  2. Low living costs – 61%
  3. Safety – 57%
  4. Tolerance towards minorities – 51%
  5. Rent – 48%

More interesting than popularity numbers, though, would be to see what are the differences in prevalent preferences across geographies, and which locations are similar to one another. To achieve this, we do some further number crunching, resulting in a graph that shows how users in various countries differ.


The graph is most useful for estimating how alike users from various locations are on average—countries where average user preferences are similar clusters nearby each other. For example, English speaking countries are all grouped in the top left corner, Latin American countries in the bottom right, and so on.

The position itself indicates how far from the global average Teleport users from a particular country are. For example, users from Russia care much more than average about the availability of fast internet and much less than average about how tolerant a city is. As a result, they end up on the right side of the horizontal axis.

Alternatively, the US ends up in the top part of the graph because users there are very interested in good airport connections and not too keen on how developed the host country is economically. However, the axes are combinations of various factors, and the position depends on how they all come together.
We can also look at the similarity of users from individual cities. The graph below shows the top 15 US cities that Teleport users hail from:


For comparison, a few countries are shown as well. The cities cluster around the US average, which is to be expected, considering that the same users make up the stats for the US as a whole.

One issue with analyzing users in this manner is that we are working with what could be considered as an “average” user. Of course, there is no such thing. This can be illustrated with an extreme example of four users from two cities:


In the first city, Chandra cares about low pollution and being by the sea, whereas Bob cares for neither of these. On the average, therefore, the preferences have 50% popularity in city 1. In the second city, both users are interested in only one preference, resulting in identical average preference stats to the first city.

If we were to analyze these averages in a similar manner to what we did above, the cities would seem exactly the same and even squinting very, very hard wouldn’t allow us to distinguish these on the plot. Which would be misleading, as all four users are totally different, as is the user makeup in each city.

Instead of looking at how popular preferences are on average, we now seek to find patterns within preferences. Combining these patterns in different ways would produce preferences for different users. You can think of it as putting parts on Mr. Potato Head to get new characters.

In our case, this would mean finding sets of preferences that users tend to consistently enable together. After processing the numbers (there will be a more detailed post on how this is done) we were able to find 5 such sets describing our users. These were:

  • Life quality—composed of low pollution, safety, healthcare, tolerance, low corruption, good education
  • Entrepreneurshipgrowing economy, business freedom, startup scene, wealthy country, hiring people, low income taxes, good airport connections
  • No cargood public transport, walkability, bike routes, low pollution
  • Remote workingspeed of internet, working remotely
  • Cost of livinglow rent, low living costs

A single user’s preference profile could be reproduced approximately by adding up one or more of these sets. From the five sets we found, 32 different preference profiles could be assembled. For example, a user could have the individual preferences from the sets life quality, no car and cost of living all combined together to produce their individual set of preferences.

However, it turns out that the 6 most popular combinations account for more than 60% of all users. The most popular being just the life quality set, followed by life quality combined with entrepreneurship, and so on.


Digging deeper into country specific details, we can plot the popularity of particular sets across countries:


Here, countries are ordered vertically according to what percentage of users use the life quality set in their preferences. This appears to be very popular in some non English speaking western countries. Less so for users in Asia and Eastern Europe. Interestingly, both remote work and cost of living sets are negatively correlated with the life quality set.

It is worth keeping in mind here that Teleport users from a given country are not necessarily a representative sample of the entire population. The numbers we show do not intend to portray a country as a whole but simply such residents that happen to be Teleport users.

Looking at the clustering based on preference set popularity we can see that the doomsday scenario, proposed above when looking at Chandra, Bob, Jin and Jacqueline and relying entirely on average popularities of single preferences, has not materialized. Country clusters obtained from analyzing preference sets are quite similar to those we got from average preferences.

English speaking countries are still together, as are Western European ones. Notably, Brazil has moved closer to the English speaking countries, indicating that when looking at entire profiles instead of the averages, the Brazilian users are more akin to those in English speaking countries.


As a result of analyzing Teleport users across various countries, we found that countries in a similar cultural space also tend to look for similar features in cities. This was illustrated by analysis of user profiles with two complementary methods. Also, we were able to determine five sets of preferences which tend to be enabled by users as a whole. This allows us to look at user profiles in a more compact way and makes it easier to compare them across geographies.

Want to add data parts to your own personal Mr. Potato Head? Check out Teleport Cities to play around with preferences and find out which cities match you best.