Page 48 - ITU Journal - ICT Discoveries - Volume 1, No. 2, December 2018 - Second special issue on Data for Good

ITU JOURNAL: ICT Discoveries, Vol. 1(2), December 2018

In this article, we describe this type of data, which
has complementary strengths and weaknesses to
both traditional data and other big data sources, in
detail and show how it can be used to improve
global development statistics.

In the next sections, we first describe this rarely
used data source. Next, we summarize case studies
that use this type of data to (i) nowcast Internet
access gender gaps around the globe, and (ii)
improve international migration statistics. After
presenting these encouraging findings, we then
discuss important limitations and challenges,
including the issues of selection bias and privacy
concerns.

2. ONLINE ADVERTISING AUDIENCE
ESTIMATES
Fig. 1 –Screenshot of Facebook’s Ads Manager. The potential
Online advertising is the main revenue for reach (1,100) is an estimate of the number of monthly active
companies such as Facebook, Twitter and Google. users (MAUs) living in Geneva who are men and expats from
To support potential advertisers in planning Germany. The estimate of MAUs can be obtained
budgeting requirements for an advertising programmatically via Facebook’s Graph API. The screenshot
has been edited to fit into a single column.
campaign, all of these platforms provide services to
estimate the potential audience reach for a given set These estimates are available in near real time and
of targeting criteria. As an example, Fig. 1 shows a can be disaggregated by age and gender, as well as
screenshot of Facebook’s Ads Manager . The other characteristics. The spatial resolution differs
2
screenshot shows on the right the “Potential reach”, between platforms and the country targeted and
1,100 people, when targeting Facebook users living ranges from state level to ZIP codes at the sub-city
in Geneva who are men and expats from Germany. level. However, to avoid re-identification of
Similar interfaces are provided by the advertising individual users, platforms apply techniques
platforms of Twitter, LinkedIn and Snapchat among related to k-anonymity [16]. This means that the
others, though the targeting criteria supported smallest audience estimates that are presented
differ. Google’s advertising interface differs in that have to include 100-1000 users, depending on the
it shows estimates for ad impressions, rather than platform. Furthermore, the estimates are rounded
users, where more active users will create more ad to further limit potential re-identification attempts.
impressions. Most platforms provide an
3
application programming interface (API) to
facilitate collecting audience reach estimates
programmatically.

2 Any reader with a Facebook account can access this interface 3 See https://support.google.com/google-
following the steps at ads/answer/6320?hl=en for details on the definition of an
https://www.facebook.com/adsmanager/creation until “impression”.
reaching the audience selection step.

26 © International Telecommunication Union, 2018