Blog

Digital Smoke Signals

Robert Kirkpatrick
Apr 21, 2011

Over the years, governments, development organizations and the private sector have invested untold billions in creating mobile phone-powered programs and services with the potential to help people lift themselves out of poverty and realize their dreams.

Wherever people are using mobile phones to access digital services and participate in development programs, they are leaving trails behind in the data. We call this "data exhaust," and people everywhere are contributing to an ever-widening ocean of it, for free, merely by going about their daily lives. The private sector is now using innovative technologies to analyze data exhaust from commercial services to understand their customers, identify new markets, and make investment decisions. The time has come for policymakers to begin using these new kinds of data, tools and approaches to protect communities from multiple slow-onset crises that threaten to reverse hard-won progress in human development.

Mobile Services as Crisis Monitoring Networks

It's our working hypothesis here at Global Pulse that when people begin to be impacted by slow-onset crises, they change how they use services and participate in programs.  We think of these changing patterns of usage as potential "digital smoke signals" that could alert policy makers that vulnerable populations are in trouble. Once we figure out a way to anonymize and aggregate this data to protect both the individuals and organizations involved, the implications are profound.  What if we could find a way to transform all of these existing programs and services that people use to better their lives into dual-purpose sensor networks capable of detecting the early impacts of slow-onset crises?

The Invisible Descent into Poverty

When a crisis hits, affected populations cope by changing their collective behavior in response. Thanks to decades of research, this process is a fairly well understood phenomenon.  At the risk of vastly oversimplifying a complex and non-linear process, it might be conceived of in three phases:

  1. Those who can afford to, start out by reducing unnecessary expenses while also seeking opportunities to earn additional income. They often reach out to family, friends and neighbors for help.
  2. If the situation fails to improve, the next step for households is usually to begin spending whatever savings they have accumulated, taking on debt, and selling off property to raise money.  Here is where the harm begins, as they erode their last remaining financial resilience against the impacts of future crises.
  3. In the end, as options dwindle, the real sacrifices begin:  do they send their children off too school too hungry to pay attention in class, or pull them out of school entirely to earn money in the market?  The damage done here cuts deep, and those affected may never fully recover.

Real-time tracking of the coping strategies adopted during a crisis is notoriously difficult. Governments may be able to get access to real-time contextual information such as food prices, fuel prices, and rainfall.  Yet  data on behavioral changes households adopt in response to shocks is gathered through rapid impact assessments that are too expensive to implement for large-scale routine monitoring. Often, we don't find out about the nature of what is happening until real harm has already been done. To catch these changes earlier will require a radically different approach.

An Hypothesis

Imagine how a slowly unfolding crisis might have looked through the eyes of a typical mobile phone carrier in sub-Saharan Africa offering services for mobile banking, livestock trading, and access to health and agricultural hotlines. Let us suppose that fuel and food prices have been on the rise for some time, and rainfall has been below average in many areas. As affected subscribers begin to cope with these shocks, they begin using their mobile phones in different ways, and in some cases, mobile phones play a critical role in helping them cope. All the while, they leave behind characteristic patterns in data.  What could mobile carriers and services providers know about the impacts of crises on their customers?

  1. As food and fuel prices rose across the country, the carrier might see subscribers in the western part of the country begin to shift from adding their average $10 once per month on their SIM cards to a pattern of only topping off with 50 cents every few days.  Perhaps they also see a drop in voice calls and an increase in texting.
  2. Soon, subscribers begin emptying their "mobile money" savings accounts, with the average dropping ominously week after week.
  3. A month later, people in the affected areas begin to default on mobile repayments of microloans in large numbers.
  4. As the months pass, carriers see a significant increase in attempts to sell livestock though a mobile trading network. Many of these attempts are originating in communities where livestock sales are rare at that time of year.
  5. Soon carriers see that subscribers who purchased their mobile accounts in drought-stricken rural areas have been connecting through cell towers in urban areas and other areas not affected by drought.
  6. Inevitably, calls to health hotlines from rural areas begin to see increased volumes of calls reporting symptoms consistent with the health impacts of malnutrition and unsafe water sources.

Making sense of digital smoke signals

The scenario above is purely hypothetical.  We do not yet know how crisis-related coping strategies play out in the use of mobile services. Yet it's not difficult to imagine how the descent into poverty might be mirrored in "data exhaust" -- the unfolding story of people coping with a crisis that ultimately touches many aspects of their lives.  

In isolation, the first few digital smoke signals to appear may reveal nothing more than generalized economic stress. Where there is smoke, there is fire, but understanding quickly whether the school, the factory, the farm, or the clinic is likely on fire requires additional analysis of context. Those smoke signals that appear later might yield clearer indications of impacts in specific sectors such as agriculture and health, but by then harm is already underway.  What is needed here is a process for integrating these signals into our existing monitoring systems.

Lessons from Public Health

As it happens, a model for this process has been used for years in the field of public health to monitor for outbreaks of infectious diseases. Detecting the appearance of a combination of changes in collective behavior such as failing to show up at work or school, buying increased quantities of certain drugs, and presenting certain symptoms (e.g., fever, headache) at clinics is a reliable harbinger of the beginning of flu season.  When this combination of anomalies appears, it is used to trigger a process of investigation, collection of samples, laboratory analysis, verification, and response.

More recently, Google made headlines with a project called Flu Trends. As it turns out, one of the first changes in collective behavior among people in households affected by flu is an increase in the number of online Google searches that include certain keywords such as "fever." By monitoring for changes in how people use its search services, Google is able to detect the onset of seasonal influenza weeks earlier than is possible using traditional methods of outbreak surveillance.

Turning this into a Science

Global Pulse is looking to extend this approach beyond public health, looking for the smoke signals that could reveal collective changes in behavior related to incipient harm in many areas of human life.  Our Pulse Labs -- beginning with Pulse Lab Kampala late this summer -- will allow governments, development organizations, academe and the private sector to come together, experiment with new data, tools and approaches, and develop a formal methodology whereby the earliest smoke signals are used to trigger a process of investigation, verification, and response.  We will likely need to analyze data gathered through remote sensing, social network analysis, news media mining, and crowdsourcing of citizen reports, and combine it with traditional indicators already used in crisis monitoring.  Only then might we be in a position to understand the underlying causes and risks of future impacts on health, nutrition, livelihoods, or education, and to use this information to improve our capacity to protect vulnerable populations from harm.

For more about UN Global Pulse's research into real-time monitoring, open source technology platform, and network of Pulse Labs, please see our About page.

Add comment