Modelling Internal Migration in Vanuatu with Mobile Network Data


Understanding the movements of individuals at a national scale is important given its potential to support planning for services, infrastructure and policy. However, modelling such movement at this scale has been challenging in the past due to high costs related to generating national statistics. With the global proliferation of mobile phones and the data sets that, in turn have been generated, these fine-grained representations of movement can lead to a better understanding of internal migration. This includes both temporary migration, which could be induced by natural hazards, seasonal trends, or educational opportunities, as well as permanent migration, which could be spurred by access to work or other natural causes such as droughts.

Vanuatu is an archipelagic nation in the Pacific Ocean with four main islands and 79 smaller islands, of which 65 are permanently inhabited. The two largest phone operators in Vanuatu, Telecom Vanuatu Limited and Digicel provide coverage to most parts of the country. The research used mobile network data in the form of Call Detail Records (CDRs), which is the pseudonymous information contained in records produced by mobile network operators for billing purposes and summarising mobile subscribers’ activities. With CDR data from Digicel that spans the entire year of 2017 (1st January to 31st December 2017), the team examined trends in the aggregated movements of pseudonymous users over the course of the year. More specifically, the research analysed how individuals move between different districts and how these trends vary. It is important to note that our analysis of mobile network data throughout this project complies with the GSM Association privacy guidelines for the use of mobile phone data.

Methods employed

The team used an open-source toolbox called bandicoot to extract and analyse hundreds of behavioural indicators from the raw CDR data. The main indicators extracted were transaction type (Call or Text), transaction direction (In or Out), caller ID (pseudonymous), correspondent ID (pseudonymous), Date and time and antenna ID. All these transactions were aggregated at a weekly level (a week starts on Monday and ends on Sunday). For each week, we developed behavioural information on each pseudonymous subscriber that made a transaction during that week.

To map migration flows within the country, first it was important to locate the ‘home’ of the subscribers. The home location was defined as the most frequent cell tower used by each pseudonymous subscriber, based on frequent calls through a particular cell tower on a weekly basis.

What we found

The annual trends of movement in Vanuatu were inferred by aggregating weekly movements in 2017 and calculating the total number of transitions between each administrative district in the country. We found that there was a large inflow into the capital city, Port Vila, from a number of different districts including North Efate, Mele, Erakor and Eratap. We also saw inflow into Luganville from East Santo and South Santo. These trends demonstrate that Port Vila and Luganville are two of the most attractive destinations compared to other locations in Vanuatu.

Visualisation of aggregated movements between different administrative districts in 2017. The thickness of each line represents the relative magnitude of transitions from the origin district and the colour of each line represents the destination district.

To better understand the dynamics between different districts, we calculated the standard deviation in the percentage of transitions between each pair of districts. Standard deviation is a measure of how spread out the values in a particular data set are. Therefore, a group with higher standard deviation shows that the values span a larger range and one with a lower standard deviation are clustered closer to the mean (average). Analysing the standard deviation, we are able to point to which pairs of districts were highly volatile in their numbers of inflow or outflow over the year. We see a high standard deviation of transitions into Port Vila, suggesting the inflow into the capital city may be seasonal and the same effect is observed to a lesser degree for Luganville. We also note the high standard deviations out of North Ambae are likely due to the volcanic eruption on the island which resulted in mandatory evacuations.

Standard deviation of transitions between districts in Vanuatu

Going forward

The research shows the immense potential of using new sources of data such as mobile network data as an alternative to conventional methods of collecting national level statistics. We hope to further this research and to make new interpretations of the movements, such as whether the communications, social and economic relationships between districts influence internal migration.

We extend our thanks to our partner Digicel for making its pseudonymised data available for this research, and to the National Statistics Office of the Government of Vanuatu for requesting the insights.

Pulse Lab Jakarta is grateful for the generous support from the Government of Australia.



UN Global Pulse Asia Pacific
United Nations Global Pulse Asia Pacific

UN Global Pulse Asia Pacific is a regional hub that aims to drive data innovation and sustainable development to ensure that no one is left behind.