Inferring patterns of internal migration from mobile phone call records: evidence from Rwanda

Abstract

Understanding the causes and effects of internal migration is critical to the effective design and implementation of policies that promote human development. However, a major impediment to deepening this understanding is the lack of reliable data on the movement of individuals within a country. Government censuses and household surveys, from which most migration statistics are derived, are difficult to coordinate and costly to implement, and typically do not capture the patterns of temporary and circular migration that are prevalent in developing economies. In this paper, we describe how new information and communications technologies (ICTs), and mobile phones in particular, can provide a new source of data on internal migration. As these technologies quickly proliferate throughout the developing world, billions of individuals are now carrying devices from which it is possible to reconstruct detailed trajectories through time and space. Using Rwanda as a case study, we demonstrate how such data can be used in practice. We develop and formalize the concept of inferred mobility, and compute this and other metrics on a large data set containing the phone records of 1.5 million Rwandans over four years. Our empirical results corroborate the findings of a recent government survey that notes relatively low levels of permanent migration in Rwanda. However, our analysis reveals more subtle patterns that were not detected in the government survey. Namely, we observe high levels of temporary and circular migration, and note significant heterogeneity in mobility within the Rwandan population. Our goals in this research are thus twofold. First, we intend to provide a new quantitative perspective on certain patterns of internal migration in Rwanda that are unobservable using standard survey techniques. Second, we seek to contribute to the broader literature by illustrating how new forms of ICT can be used to better understand the behavior of individuals in developing countries.

Keywords:

• ICTD
• internal migration
• migration
• mobility
• development

Acknowledgements

The authors gratefully acknowledge financial support from the National Science Foundation and the International Growth Centre. The authors would also like to thank Yian Shang and Sergey Goder for providing excellent research assistance.

Notes

Thomas Molony is the accepting Guest Editor for this article.

We focus our attention primarily on the effect of internal migration. For surveys of the much larger literature on the labor market effects of international migration, see Friedberg and Hunt (1995) and Borjas (1999).

A vibrant literature is being developed to describe the ways in which ICTs can have a positive impact on the lives of people and their communities, and on their social development (Aker, 2008; Jensen, 2007; Qureshi, 2009). While this paper is intimately related to that literature, the goal is different. Our intent, rather, is to emphasize how ICTs can help researchers and policy-makers better measure and evaluate processes of development, rather than assess the causal impact of the interventions themselves (be they ICT-based or not).

Bilsborrow (1997) provides an excellent overview of the strengths and weaknesses of the different sources of migration data.

A very closely related body of research addresses the ways in which ICTs can be used to better understand other aspects of human behavior beyond mobility and migration, including the structure of friendship networks (Onnela et al., 2007), the spread of innovations and products (Szabo & Barabási 2006), and patterns of economic growth (Eagle, Macy, & Claxton, 2010). See Kwok (2009) and Hazas Scott, and Krumm (2004) for brief overviews of this literature.

http://www.itu.int/ITU-D/ict/statistics/at_glance/KeyTelecom.html, Accessed July 2011. It should be noted, however, that mobile penetration in Africa is the lowest worldwide at 41%, and that in certain countries the uptake is much lower.

During the window of time we examine, the operator we focus on maintained over 90% market share of the mobile market. The company's primary competitor did not gain traction in the market until the end of 2008, and only more recently has the market become competitive. The number of landlines in Rwanda is insignificant (roughly 0.25% penetration).

All notation remains as before, except that we allow for i's ROG and COG to vary by month (i.e. ROG _i is i's total ROG, whereas ROG _ik is i's ROG during the month k).

In an effort to make our statistics more comparable with those collected by the Rwandan government, we count migrations that occurred during the 3-month period from December 2007 through February 2008, which is exactly one year before the 3-month window queried in the CFSVANS survey. Unfortunately we do not have data from December 2008 through February 2009.

Among the primary sample of 901 mobile subscribers that we use in most of our analysis, over half used their phone for the first time in 2008, so it is not possible to compute as rich a set of longitudinal metrics for this group of individuals.

The fact that the migration rate among this sample of long-term subscribers is lower than the rate reported in Table 1 among all subscribers is further evidence that mobile phone users (in this case, early adopters) are different (in this case, less likely to migrate) from the at-large population.

The statistics in Table 2 differ from those in Table 1 because they are computed on a different sample (people active over 4 years vs. people contacted in the phone survey), and because Table 2 includes migrations over the entire 4-year interval, whereas Table 1 enumerates migrations in the 3-month window prior to March 2008.

This finding, which contradicts much of the prior literature on the subject, presents a mystery that we cannot explain without further evidence. However, we suspect that it may result from the fact that men and women who own phones may be more similar than men and women who do not own phones.

More generally, we interpret the fact that the migration statistics are not perfectly correlated with the mobility statistics as a validation of the quantitative instruments we employ. For instance, we note that the overall (4-year) radius of gyration for people who do not migrate is not significantly different from that of those who do, which suggests that the definition of inferred migration proposed in (2a)–(2c) is not merely a by-product of the fact that people who move a lot (but do not migrate) are more likely to be inadvertently classified as movers.