Measuring social inequality based on social connectivity data

1 Context, objectives and potential impact. One popular inequality measure is the
Gini Index for income inequality. But income is just one social dimension amongst
many others and it is not clear how the Gini Index can be generalized to other social
dimensions like education or ethnicity and how different dimensions can be included
into one single inequality measure. We propose to use social connectivity data for the
construction of such a measure, which we call Social Gini Index. In particular, our Social
Gini Index quantifies the surprisal of a connection between any two individuals with
given socio-demographic background compared to a world where people form friendships
independent of their socio-demographic background.
We use the Social Connectivity Index data from Facebook [1], which consists of
the aggregated number of Facebook links (friendships) between different regions. We
plan to compute the Social Gini Index for a variety of countries to compare inequalities
across countries. Furthermore, we want to investigate the relation between the Social
Gini Index and several health outcomes, our hypothesis being that social inequalities
have a detrimental effect on health outcomes. We hope that this novel social inequality
measure can help in understanding the implications of social inequalities, specifically for
health outcomes.
2 Novelty of the research methodology. We develop a generative model of network
formation that allows us to learn connection probabilities between individuals with different social demographic backgrounds from aggregated network data. Aggregated network data consists of the number of the aggregated links between regions rather than
data consisting of links between individuals. Our model builds on similar approaches
[2]. An important part of the model is a careful consideration of how physical distance
between individuals impacts the probability that they connect. To do so, we develop
a model based on an intervening opportunity approach [3]. We combine different data
sets (Facebook's Social Connectivity Index, micro-census data and data collected from
the Facebook Marketing API) and take care that the uncertainty with respect to the
different data sources and the various modeling steps gets propagated through to the
computation of the Social Gini Index.
3 Alignment to EPSRC's strategic themes and research areas. The project aligns with
two of the EPSRC's research areas: Artificial intelligence, digitisation and data: driving
value and security and Mathematical sciences. It also aligns with the EPSRC's
strategic theme Transforming health and healthcare.
4 Collaborators. This is joint work with Sumeet Agarwal, Till Hoffmann, Nick Jones
and Sahil Loomba and builds on previous work by Till Hoffmann and Nick Jones [4].