Page 51 - 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

Data on migration, especially flows, are often
unavailable for a number of countries. When the
data does exist, it is typically outdated or
inconsistent across countries. This is because some
countries rely on survey data, others on census
data, registration systems or various other
administrative sources. Such traditional data has
various degrees of under-reporting issues and the
statistics obtained from the data is often published

(a) Internet access gender gaps according to 2015 ITU data. with a substantial delay.
In this context, data from Facebook’s advertisement
platform offers a new source of information that is
timely and that can be considered as a continuous
census of a large portion of the world population.
The main challenge is that Facebook users are not
necessarily representative of the underlying
populations. In order to understand how the
selection bias changes across demographic groups,
Zagheni et al. [23] used the following linear
regression model:
(b) Internet access gender gaps according to 2017 Facebook
data.
log(ACS foreign-born pop ij) = β0 +
z
β1log(Facebook expats ij) + β2 I(origin 1) + …
z
Fig. 2 – Two world maps showing the ratio of (percentage of
women with Internet access)/(percentage of men with + β30 I (origin 30) + β31 I(age group 1) +
z
Internet access) on a per-country basis. ITU data from β38 I(age group 8) + ε ij (1)
2015(top) is compared to model predictions of the online
model (see Table 1) using Facebook data from 2017 (bottom). where ACS foreign-born pop i is the number of people
z
The model manages to largely reproduce ITU ground truth in the age-sex group z born in country i and living in
data while substantially improving global coverage. See [19] US state j, according to the American Community
for additional details.
Survey. Facebook expats ij is the respective quantity
z
4. CASE STUDY 2: INTERNATIONAL from Facebook data. The indicator variables add
MIGRATION STATISTICS controls for different age groups and countries of
origin. The mean absolute percentage error (MAPE)
for the ‘naive’ model that does not include indicator
Migration is not included directly among the
Sustainable Development Goals. However, variables, is on 56 percent. The MAPE for the
migration is one of the driving forces behind age-origin model, that explicitly models biases
demographic changes among the globe and SDG related to differences in countries of origin or in age
#10, Reduce Inequalities, includes the target to groups, is significantly lower at 37 percent. This
“facilitate orderly, safe, regular and responsible indicates that biases show regularities along the
migration and mobility of people, including through dimensions of age and country of origin. These
the implementation of planned and well-managed regularities can be modeled and the biases can be
migration policies”. The implementation and corrected, at least partially.
monitoring of such policies is tightly linked to the Facebook data offers timely, but possibly biased
availability of accurate and timely data on data. Traditional surveys offer time series of
migration, as emphasized by the Global Compact for representative statistics, but often outdated or from
Safe and Orderly Migration , an intergovernmental relatively small samples. Combining the two types
7
agreement negotiated with the support of the of sources holds promise for generating timely and
United Nations, and designed to cover migration in reliable migration statistics. [24]
a comprehensive way.

7 Details at https://www.iom.int/global-compact-migration .

© International Telecommunication Union, 2018 29