Intro

In the past few weeks, many academic research groups (including our own) have been working on providing information about mobility in many countries. This is an incredible effort, carried out during extraordinary times and by really talented people. There are many methodologies (using different mobility metrics: radius of gyration, number of visited places), datasets (gps points, data detail records), and analysis (trips, activity) going around, so maybe trying to keep a centralized repository of them, with some commentary, would be worthwhile both to keep updated and personal edification, and also as a service to the community. We have done something similar for mobile stream data research per country. [EDIT 2020-04-28 10:42:16 -0400: A large group of researchers have written a fantastic essay on the topic that should be used to frame the discussion of the issues here.]

In this instance, I will only look at “in the wild” mobile phone data, either by telco data (e.g., xdr or cdr), or by other companies that provide “app” data (e.g., gps data). In the table below, the field date is the date the report was made publically available, countryis the country that the report applies to (not necessarily where the research was carried), the paper field contains a URL link to the paper (preferably in pdf format) and finally the tags field gives an idea of the general issue being addressed. So far it’s only been mobility, plus one study analyzing “social mixing” as well.

Perhaps a small methodological clarification is needed here: xdr data comes from “meaningful” (i.e., long enough to be billed by the telco) internet interactions with a cellphone antenna. The gps data comes from apps in phones (like Facebook, Google, and many (most?) smaller ones) and usually companies that aggregate them. The gps data is of “higher quality”, very fine grained, both spatially and temporally, while xdr data is not (data points at 15 mins intervals or xMB downloaded, for example, and at the level of antennas, not individuals). However, xdr usually reports unique devices in the millions , while gps will depend on the level of adoption of the different apps being aggregated, roughly in the order of 100,000s unique devices per country. A new “player” in all this is the madid value for the dtype field. I assume the authors are talking about the mobile advertising ID feature in Android and Apple phones. I did not know about this until yesterday [2020-04-11]. Here’s some information about it from Google, and an article from WIRED.

Countries with mobility studies using mobile phone datasets, according to the table below. Map is courtesy of mapchart.net. Please help expand this table by sending me an email at lferres@udd.cl, through my Twitter @leoferres, or posting in the comment section below.

  • The table with the data, in CSV format is HERE

NOTE: There’s a special mention here to the Google Mobility effort, since it covers the whole the world! Although a little bit “underexplained”, its simplicty and informational density make them quite extraordinary. There’s more information in their blog post.

[2020-04-13 08:07:01 -0400]: There’s a memo by the GSMA outlining some suggestions for the handling of mobile phone data through the COVID19 crisis. Thanks to Ciro Cattuto for bringing it to my attention.

[2020-04-26 10:24:05 -0400]: There was a big effort in to attempt a large scale coverage of mobility in Latin America and the Caribbean by the IADB using gps. There’s a study by the University of Texas as well in the United States to using gps to measure social distancing and projections of deaths. The New York Times ran a similar story also using gps to see which counties stayed at home. (Interesting, since it’s been so critical about this methodology before, see here and here).