UN World Food Programme (WFP), Global Pulse, Université Catholique de Louvain in Belgium and Real Impact Analytics recently collaborated on a project to analyse the potential use of mobile phone data as a proxy for food security and poverty indicators. Ahead of the Netmob 2015 Conference on Mobile Phone Data Analysis, we caught up with Amit Wadhwa (Head of Food and Nutrition Security Analysis Unit at WFP Indonesia) and Adeline Decuyper (lead researcher on the project and former Global Pulse fellow who is presenting the project at Netmob), to ask them about their experience of collaborating and the project findings.
Q: Tell us about the project you collaborated on:
Amit: Food security is a complex phenomenon with multiple causal factors. Measuring food security is equally complex. We often describe food security through a handful of proxy indicators. Some of the most common proxy indicators are measurements of food consumption, which describe food access – one of the key dimensions of food security. Food consumption varies considerably over time with household patterns adapting to seasonal changes in food availability and variability in food prices. Capturing those changes is difficult given the nature of data collection through household surveys. We were interested to understand if a data source that is more quickly obtainable, such as mobile phone data, can be used as a proxy for our food consumption indicators. If we can demonstrate the ability of mobile phone data to adequately describe consumption patterns in a population, it opens up a number of possibilities for use of this data for monitoring food security more quickly than we are currently able and in locations that we may not be able to access.
Adeline: As Amit says, this project – Using mobile phone data and airtime credit purchases to estimate food security – is an exploration of big data generated by mobile phone use and airtime credit top ups compared against official surveys to see how closely correlated different aspects of these two areas are. Broadly speaking, the big data revolution is changing how people face problems, in many different ways. Big data is relatively easy and cheap to collect in comparison to costly and time consuming surveys, big data is also updated in real time compared to annual or bi annual surveys that only capture a single point in time. Big data analysis has been shown to be of potential use for global development in other areas – so we focused on whether big data can be used to provide updates on the food security landscape in a specific geographic area.
Q: Why mobile phone data in particular?
Adeline: Mobile phone data in the form of Call Detail Records (CDRs) are collected for billing purposes by each mobile phone provider. These data are very rich in that they capture many aspects of human behavior, ranging from mobility to economics or social behavior. They have been used in the past by researchers, at first for marketing purposes or social behavior analysis. A couple of years ago, we’ve seen the rise of uses of data for development purposes, as the mobile phone market has reached less developed countries. Researchers have since shown how valuable these data could be to policy makers or governments to make the decisions on the next steps towards development.
Amit: Food security assessments usually rely on primary data collected through household surveys. These assessments can be rapid and light for emergency situations or more in-depth and comprehensive for baseline assessments. The data source we used in this research is from a baseline assessment with multiple modules including demography, asset ownership, income and livelihoods, water and sanitation, household expenditures, and food consumption. These assessments are called Comprehensive Food Security and Vulnerability Assessments, or CFSVAs. CFSVAs are implemented in close partnership with government, NGOs and other UN agencies. They offer an in-depth view of the food security situation in a given country, and are conducted every 3-5 years.
Due to their broad scope, CFSVAs are valuable instruments for conducting research on food security. Typically, these assessments cover an entire country and highlight geographic pockets of food insecurity. They also provide a baseline for policies and programs that may follow their implementation.
The implementation of a CFSVA is long – from start to finish, these assessments often take a year or more to plan, implement and publish results. While this is typical for large, national surveys, it often diminishes the value of results, particularly for time-sensitive information.
Q: How did you hone in on the research questions?
Adeline: The question of mapping socio-economic levels in a country from mobile phone data has been addressed by several research teams in the past few years. Building on their experiments and findings, we wanted to investigate whether mobile phone data could help us monitor food security. As food security is closely related to poverty, we thought if mobile phones can capture socio-economic levels, then maybe they can also help us locate populations that are most vulnerable in terms of access to food.
Amit: As Adeline noted, food insecurity is very often a function of poverty. While the relationship between mobile phone usage and poverty is fairly clear, the specific relationship between food consumption and mobile phones was not yet explored fully. For WFP, understanding changes in food consumption patterns over time is incredibly useful for food security monitoring. For example, if we can capture a change in consumption of a staple food, like rice, wheat or cassava, in near real-time, WFP can take action to prevent malnutrition related to poor food energy intake. Though a mobile phone won’t capture the change in food consumption itself, it can be a powerful proxy that can trigger further investigation by WFP.
While CFSVAs provide the baseline information, we often follow it with a Food Security Monitoring System (FSMS) in the most problematic areas identified in the CFSVA. FSMS make use of multiple data sources but are still largely based on household surveys implemented on a quarterly basis. Though these monitoring systems are lighter and more rapid than a CFSVA, they still face the same challenges as other household surveys. Ultimately, we want to see how new data sources can be used to help address the weaknesses in monitoring systems that depend on household surveys. The faster we can detect a change in the food security situation, the faster we can take action to protect vulnerable households.
Q: When you worked together were there any surprises? How was the process of working together, in an interdisciplinary team, from beginning to end?
Adeline: It all started from the NetMob 2013 conference in Boston, held together with a day on the Data for Development (D4D) Orange challenge. I was on the organizing team of the D4D challenge, and met the Global Pulse team at that conference. During a social event, we started talking about how great it would be to start a project together on the follow-up of the D4D challenge and spirit, and do data science for development purposes. Given its position at the UN, Global Pulse’s contacts would be an invaluable asset in bringing together a multi-disciplinary team to conduct a project directly thinking of its applicability.
Thanks to Vincent Blondel’s contacts with Real Impact Analytics, they supported us throughout the process in accessing African data, and providing means of analyzing the data efficiently.
A few months later, I landed in New York in the Global Pulse lab, where I was welcomed by a very nice team of motivated data scientists. During my stay with the Global Pulse team, I had many opportunities to meet people in other instances of the UN, but also to discuss research on different levels. Coming from a department of mathematical engineering, I wasn’t used to talking with, for example, people who fight hunger in developing countries and are in contact with the reality of the field. This started a few days after my arrival, as we began designing the project in more details in partnership with the World Food Programme. The overall picture of the project and collaboration is a very positive one.
Amit: The biggest surprise for WFP was the actual availability of a large mobile phone dataset that covered the same point in time as our CFSVA. This was the type of scenario that researchers dream of and we were excited to be a part of it. Adeline and the Global Pulse team took the lead in bringing us all together and thinking about what types of questions we might be able to answer through this data. Our local WFP office knows the country well and was able to give some key insights into what was happening on the ground when the data was collected. We were surprised at how strong some of the correlations between consumption of individual food items and air-time credits were. We expected a relationship, but the findings were even stronger than expected.
Q: Do you think new sources of data like mobile data could replace 'ground truth' data like traditional statistics and surveys?
Amit: I don’t see this data replacing traditional statistics and surveys anytime soon. While we recognize that surveys are imperfect, they are still critical for the development and humanitarian community. Mobile data and the like can fill gaps cost-effectively and rapidly. That’s huge for WFP. But we also recognize that proxies can only tell you so much and we will always need to ground truth information before we make a programmatic decision.
I’m also not so sure the floodgates will be open in the near future to liberate mobile data in real-time. Global Pulse’s efforts in data philanthropy are an important part of that liberation process. If we can demonstrate to the private sector the value of their data for humanitarian efforts, we remain positive that they’ll find a way to make it available. This research is an important contribution to the way forward, but we’ve got a way to go still.
Adeline: Like Amit I don't think mobile data can replace 'ground truth' methods, or at least not in the near future. The purpose of our study was to evaluate how mobile phone data could complement the census data that is collected by national surveys. There will always be a need for actual surveys to confirm the accuracy of mathematical models and to calibrate them. However, surveys give a picture of the situation at a given point in time, whereas mobile phone data could serve as a real-time proxy for several food security indicators. The next steps of the research in that area aim towards evaluating how fast mobile phone indicators respond to specific situations such as sudden changes in food stocks or weather conditions.
Q: After completing this project what's next for each of you? What did you take away from this project?
Amit: We’ve got a large amount of data that we are opening up and we want to engage more researchers and data scientists to explore what we have and to contribute to the broader food security research community. Ultimately, we’d like to operationalize the use of big data for food security monitoring. To get there, we need to build more evidence and we’d like to develop tools that can help us harness big data more effectively. WFP has a number of analysts and researchers globally but the big data world is still new territory for us. We’ve participated in a few hackathons over the past couple of years and we’ve been increasingly contributing our data to the Humanitarian Data Exchange. If anyone reading this is interested in data for development and has some time and skills to contribute to WFP’s mission, do get in touch or leave a comment below!
Adeline: The next steps for the research community are to extend these analyses to more countries, and more projects of this type of research for development. I think research in data mining could help save lives in many different ways, from monitoring humanitarian action in developing countries to helping respond to an emergency situation such as an earthquake or tsunami. It's an exciting and expanding field.
Top image: Produce is sold at a roadside market in East Africa, Lori Howe via Creative Commons