THE POLITICAL ECONOMY AND NETWORK SCIENCE OF PUBLIC SERVICE DELIVERY IN DEVELOPING COUNTRIES

A growing political economy literature examines the persistence of low levels of public service delivery in new and emerging democracies. Much of this literature addresses the incentive structure facing politicians and civil servants. The 2004 World Development Report: Making Services Work for the Poor describes the long route of accountability where citizens as voters use voice to keep politicians accountable and a short route of accountability where citizens as monitors can hold frontline service providers accountable (The World Bank 2004).

In the long route to accountability, access to information is a critical prerequisite to active citizenship. In the seminal work on modernization theory, Lipset argues that education and access to information makes people more tolerant and helps them make more informed decisions (Lipset Reference Lipset1959). A more recent finding is that voting behavior is malleable, and so voters choose candidates who serve them more effectively provided they have the information that meaningfully identify performance levels (Banerjee et al. Reference Banerjee, Kumar, Pande and Su2011). In the short route to accountability, citizens directly monitor the behavior of local providers (Björkman and Svensson Reference Björkman and Svensson2010).

Not noted in this political economy work is the existence of a vast cognitive science and behavioral economics literature demonstrating that incentives are but one of several important factors considered when individuals make decisions. This literature articulates a framing problem: a wide variety of factors determine behavior, and therefore the underlying drivers of decision making are difficult to deduce (Gigerenzer, Todd, and the ABC Research Group Reference Gigerenzer and Todd1999; Gilovich, Griffin, and Kahneman Reference Gilovich, Griffin and Kahneman2002). Some of the factors that researchers have added to the list of determinants include defaults, when the easier choice to make becomes more likely (Goldstein, Cialdini, and Griskevicius Reference Goldstein, Cialdini and Griskevicius2008; Thaler and Sunstein Reference Thaler and Sunstein2008); priming, when certain words or stimuli can affect behavior (Bargh, Chen, and Burrows Reference Bargh, Chen and Burrows1996; Mandel and Johnson Reference Mandel and Johnson2002); or confirmation bias, when individuals are more likely to favor information that confirms their prior beliefs (Bond et al. Reference Bond, Gibbs, Shipton and Jones2007).

It may be impossible to determine all of the inputs to individual decision making, however, physicists and social scientists are exploring the network science of how behavior spreads among individuals. A large part of this literature explores positional inequality; or how the relationship to the core of any network determines behavior. This work has shown, for example, that the risk of dying of a heart attack increases after shocks to an individual's social network, such as after the death of a spouse (Elwert and Christakis Reference Elwert and Christakis2008). This network science literature also explores how ideas and information flows across the network itself (Christakis and Fowler Reference Christakis and Fowler2009; Watts Reference Watts2003). Economists have found that criminal behavior is contagious through social interactions, increasing the likelihood that more crimes will occur in particular neighborhood (Glaeser, Sacerdote, and Scheinkman Reference Glaeser, Sacerdote and Scheinkman1996) and that charity spreads with the same characteristics; when people move from a less-generous to a more-generous neighborhood, their likelihood of giving increases (Carman Reference Carman2003).

PERVASIVE AND MOBILE COMPUTING

Are high levels of quality in service delivery contagious in the same way as crime and charity? More broadly, what can a better understanding of network structure tell us about the decisions of individuals at the front lines of public service delivery. Recently, computer scientists, physicists, and social scientists have developed approaches to using network structure data to answer important social science questions (Lazer et al. Reference Lazer, Pentland, Adamic, Aral, Barabasi, Brewer, Christakis, Contractor, Fowler, Gutmann, Jebara, King, Macy, Roy and Van Alstyne2009). The mobile phone has been called “perhaps the biggest source of data in the world” for these kinds of studies (Pentland Reference Pentland2011).

An early attempt to use mobile phones to collect behavioral data was made in 2005, when researchers at the MIT Media Lab conducted experimental ethnographic research by tracking movements using cellphone tower location and tracking voice, text, and data usage among subjects. Building on this work, researchers in 2011 developed smartphone software to collect multidimensional datasets about the lives of individuals and social groups. They labeled this approach Social fMRI, drawing parallels to functional magnetic resonance imaging (fMRI), a leading tool in the medical world for measuring and studying the brain's response to stimuli. Like fMRI, Social fMRI seeks to sense the social world through the use of technology and test a number of interventions in the real world without disrupting the lives of participants.

The Friends and Family Study put Social fMRI to the test with a group of MIT graduate students mapping how networks affect patterns of decision-making and conducting experiments related to increasing frequency and levels of exercise (Aharony et al. Reference Aharony, Pan, Ip, Khayel and Pentland2011). This study collected a variety of measures of social networks (face-to-face, phone usage, and self-reported), sleep patterns (through alarm usage), patterns of physical activity, and movements around campus. Analyzing these data alongside background survey information, researchers found that individual preferences and changes in behavior were strongly influenced by those around them. For example, face-to-face interaction drives diffusion of mobile application more than perceived friendship ties, and behavior change (in this case level of exercise) can be influenced by social support systems. These behaviors are in line with the observable characteristics of the contagion hypothesis described earlier. Following the publication of the MIT Friends and Family Study, the research team launched Behavio, an open-source Android platform that offers a wide range of possibilities for researchers interested in understanding a wide range of behaviors.

TANZANIA'S COMMITMENT TO IMPROVING GOVERNANCE

Tanzania offers a unique opportunity to examine the applicability and utility of pervasive computing data in testing hypotheses about the link between networks and governance outcomes. Tanzania is an emerging leader in the field of open government, and its efforts to lead in this area have generated some of the first data that could be used to rigorously test this hypothesis.

The Tanzanian Minister of Justice and Constitutional Affairs, Mathias Chikawe, is a member of the Steering Committee, and Tanzania's central OGP commitments focus on improving core services such as water, education, and health. The Ministry of Health and Social Welfare (MoHSW), for example, has committed to building a grievance redressal mechanism that will publicly track the reporting and resolution of service delivery complaints (Government of Tanzania 2012). It has also committed to publicly tracking the availability of 15 essential medicines, which the government has mandated must be stocked in every clinic at all times.

Tanzania's approach suggests a causal logic similar to that outlined in the literature described above, wherein citizens with more access to information on the performance of clinics in their community will be likely to demand improved service delivery from politicians and frontline providers. At the same time, however, emerging evidence suggests that this causal logic may be either flawed or, at least, incomplete. One recent study in East Africa found that a program called Uwezo, which focused on empowering citizens through sharing information about local education performance, had no statistically significant effect on levels of active citizenship (Lieberman, Posner, and Tsai Reference Lieberman, Posner and Tsai2012).

What can social network data tell us about the quality of essential medicine provision in Tanzania? The remainder of this article explores how pervasive computing can help identify network data for individuals involved in the essential medicine supply chain in Tanzania, and how these data could provide a new way to experimentally test the role of these claims.

RISKS AND LIMITATIONS

The preceding sections have demonstrated the promise of data generated by pervasive computing for social science research—particularly their utility for generating a wealth of detailed, empirical data on the incentives and behavior of civil servants. As a result, this approach presents new opportunities to explore the micro-foundations of public service delivery and bureaucratic performance at an unprecedented level of rigor and detail.

There are, however, important limitations and risks to this approach, both technical and social. These limits and risks vary across economic and political contexts. On the technical side, programs like Behavio operate on smartphones running the Android mobile platform. This makes the software easily accessible and customizable, but also requires, in many developing country settings, that researchers distribute phones to the target population. This requires a larger research budget, careful thinking about how research subjects are selected, and careful consideration of whether the introduction of phones themselves might introduce changes in network shape and behavior and therefore exert an independent causal effect. Other logistical issues in developing countries involve ensuring that subjects have the ability to charge phones—particularly important given that Behavio and other data usage functions quickly draw down the batteries of smartphones—and a nuanced understanding of the connectivity available in a particular research setting, since data collection from smartphones requires at least a semiregular Internet connection to transfer data remotely.

The use of pervasive computing technology also raises important concerns about privacy. In the realm of global policy, pervasive computing has led to an extensive conversation about who owns the massive amounts of information that these technologies consistently generate (Pentland Reference Pentland2011). In the United States, the Data Bill of Rights has taken important steps to break these data out of their silos and make sure the citizens who generates them has as much control as possible (The White House 2012). This could be a challenge in countries without the full protection of civil liberties. Research ethics require that scholars carefully think about the potential for these technologies to become surveillance tools, and the political and ethical risks associated with distribution of a technology they may no longer be able to control. Best practice in the field is to remove all human readable text from the information collected, such as phone numbers, text messages, and other unique identifiers, but this does not guarantee that personal data cannot be reconstituted. As these technologies become more extensive and are used in new ways, scholars and institutional ethical review boards will have to consider carefully the ethical issues involved.