Passively collected "big" data sources are increasingly used to inform critical development policy decisions in low- and middle-income countries. While prior work highlights how such approaches may reveal sensitive information, enable surveillance, and centralize power, less is known about the corresponding privacy concerns, hopes, and fears of the people directly impacted by these policies --- people sometimes referred to as experiential experts. To understand the perspectives of experiential experts, we conducted semi-structured interviews with people living in rural villages in Togo shortly after an entirely digital cash transfer program was launched that used machine learning and mobile phone metadata to determine program eligibility. This paper documents participants' privacy concerns surrounding the introduction of big data approaches in development policy. We find that the privacy concerns of our experiential experts differ from those raised by privacy and development domain experts. To facilitate a more robust and constructive account of privacy, we discuss implications for policies and designs that take seriously the privacy concerns raised by both experiential experts and domain experts.

No abstract available.

Poverty maps derived from satellite imagery are increasingly used to inform high-stakes policy decisions, such as the allocation of humanitarian aid and the distribution of government resources. Such poverty maps are typically constructed by training machine learning algorithms on a relatively modest amount of ``ground truth" data from surveys, and then predicting poverty levels in areas where imagery exists but surveys do not. Using survey and satellite data from ten countries, this paper investigates disparities in representation, systematic biases in prediction errors, and fairness concerns in satellite-based poverty mapping across urban and rural lines, and shows how these phenomena affect the validity of policies based on predicted maps. Our findings highlight the importance of careful error and bias analysis before using satellite-based poverty maps in real-world policy decisions.

Phone sharing is pervasive in many low- and middle-income countries, affecting how millions of people interact with technology and each other. Yet there is very little quantitative evidence available on the extent or nature of phone sharing in resource-constrained contexts. This paper provides a comprehensive quantitative analysis of demographic variation in phone sharing patterns in Togo, and documents how a large cash transfer program during the COVID-19 pandemic impacted sharing. We analyze mobile phone records from the entire Togolese mobile network to measure the movement of SIM cards between SIM card slots (often on different mobile devices). By matching phone sharing measures derived from SIM reshuffling to demographic data from a government-run cash transfer program covering hundreds of thousands of individuals, we find that phone sharing is most common among women, young people, and people in rural areas. We also leverage randomization in the cash transfer program to find that the delivery of cash aid via mobile money significantly increases phone sharing among beneficiaries. We discuss the limitations of measuring phone sharing with mobile network data and the implications of our results for future aid programs delivered via mobile money.

Algorithms for home location inference from mobile phone data are frequently used to make high-stakes policy decisions, particularly when traditional sources of location data are unreliable or out of date. This paper documents analysis we performed in support of the government of Togo during the COVID-19 pandemic, using location information from mobile phone data to direct emergency humanitarian aid to individuals in specific geographic regions. This analysis, based on mobile phone records from millions of Togolese subscribers, highlights three main results. First, we show that a simple algorithm based on call frequencies performs reasonably well in identifying home locations, and may be suitable in contexts where machine learning methods are not feasible. Second, when machine learning algorithms can be trained with reliable and representative data, we find that they generally out-perform simpler frequency-based approaches. Third, we document considerable heterogeneity in the accuracy of home location inference algorithms across population subgroups, and discuss strategies to ensure that vulnerable mobile phone subscribers are not disadvantaged by home location inference algorithms.

Recent papers demonstrate that non-traditional data, from mobile phones and other digital sensors, can be used to roughly estimate the wealth of individual subscribers. This paper asks a question more directly relevant to development policy: Can non-traditional data be used to more efficiently target development aid? By combining rich survey data from a "big push" anti-poverty program in Afghanistan with detailed mobile phone logs from program beneficiaries, we study the extent to which machine learning methods can accurately differentiate ultra-poor households eligible for program benefits from other households deemed ineligible. We show that supervised learning methods leveraging mobile phone data can identify ultra-poor households as accurately as standard survey-based measures of poverty, including consumption and wealth; and that combining survey-based measures with mobile phone data produces classifications more accurate than those based on a single data source. We discuss the implications and limitations of these methods for targeting extreme poverty in marginalized populations.

While real-world decisions involve many competing objectives, algorithmic decisions are often evaluated with a single objective function. In this paper, we study algorithmic policies which explicitly trade off between a private objective (such as profit) and a public objective (such as social welfare). We analyze a natural class of policies which trace an empirical Pareto frontier based on learned scores, and focus on how such decisions can be made in noisy or data-limited regimes. Our theoretical results characterize the optimal strategies in this class, bound the Pareto errors due to inaccuracies in the scores, and show an equivalence between optimal strategies and a rich class of fairness-constrained profit-maximizing policies. We then present empirical results in two different contexts — online content recommendation and sustainable abalone fisheries — to underscore the generality of our approach to a wide range of practical decisions. Taken together, these results shed light on inherent trade-offs in using machine learning for decisions that impact social welfare.

What determines the success of community cellular networks? We leverage unique circumstances where all households in seven localities were interviewed before the launch of cellular networks. We observed substantial differences in network adoption across communities. Four communities displayed high and regular usage, while usage dissipated shortly after the network launch in three sites. Sixty-five percent of households made or received at least one call or text message. We find that a one standard deviation increase in household wealth is correlated with a three percentage point increase in network adoption and 43 additional cellular network transactions. Agricultural households were ten percentage points more likely to adopt the network than other households and female-headed households were five percentage points more likely to use the network at least once.

With the rapid expansion of mobile phone networks in developing countries, large-scale graph machine learning has gained sudden relevance in the study of global poverty. Recent applications range from humanitarian response and poverty estimation to urban planning and epidemic containment. Yet the vast majority of computational tools and algorithms used in these applications do not account for the multi-view nature of social networks: people are related in myriad ways, but most graph learning models treat relations as binary. In this paper, we develop a graph-based convolutional network for learning on multi-view networks. We show that this method outperforms state-of-the-art semi-supervised learning algorithms on three different prediction tasks using mobile phone datasets from three different developing countries. We also show that, while designed specifically for use in poverty research, the algorithm also outperforms existing benchmarks on a broader set of learning tasks on multi-view networks, including node labelling in citation networks.

Societal-scale data is playing an increasingly prominent role in social science research; examples from research on geopolitical events include questions on how emergency events impact the diffusion of information or how new policies change patterns of social interaction. Such research often draws critical inferences from observing how an exogenous event changes meaningful metrics like network degree or network entropy. However, as we show in this work, standard estimation methodologies make systematically incorrect inferences when the event also changes the sparsity of the data.

This paper presents eKichabi, a tool for retrieving contact information for agriculture-related enterprises in Tanzania. eKichabi is an Unstructured Supplementary Service Data (USSD) application which users can access through basic mobile phones. We describe our focus groups, a design iteration, deployment in four villages, and follow up interviews by phone. This work demonstrates the feasibility of USSD for information access applications that have the potential for deployment on a large scale in the developing world. From user interviews, we identified strong evidence of eKichabi fulfilling an unmet need for business related information, both in identifying business contacts in other villages, as well locating specific service providers. One of our key findings demonstrates that users access information through multiple modes, including text search, in addition to menu navigation organized by both business sector category and geographic area.

As a growing share of the world's population inhabits cities, a central focus of current development policy has been on building urban infrastructure to support increasing population density. An important component of such policy has been the construction of major roads and highways, which in principle can reduce congestion, promote development of peri-urban areas, and increase labor mobility. However, the actual impacts of such investments have been difficult to evaluate empirically. Here, we use a rich dataset capturing the mobility patterns of roughly 9 million individuals to study the impact of a new super-highway on travel patterns in and around Colombo, the capital of Sri Lanka. Our results indicate that this road had an immediate and pronounced impact on travel patterns: people changed their primary routes of travel, which reduced overall congestion, increased average travel speeds, and reduced the amount of time spent in transit. We further find that the super-highway led to a modest, but statistically significant, increase in the total amount of travel in and around Colombo. We discuss how such insights can inform future policymaking, and point to several promising areas for future research.

In many developing country environments, it is difficult or impossible to obtain recent, reliable estimates of human development. Nationally representative household surveys, which are the standard instrument for determining development policy and priorities, are typically too expensive to collect with any regularity. Recently, however, researchers have shown the potential for remote sensing technologies to provide a possible solution to this data constraint. In particular, recent work indicates that satellite imagery can be processed with deep neural networks to accurately estimate the sub-regional distribution of wealth in sub-Saharan Africa.

In the last decade, billions of people worldwide have upgraded from basic 2G feature phones to data-enabled 4G smartphones. In most cases, people upgrade in areas with 4G coverage (typically cities and large towns), but increasingly, people choose to upgrade in areas that only have 2G coverage or no cellular coverage at all. This counterintuitive behavior -- upgrading your phone despite living in an area that does not actively support many of the features of that new device -- is the focus of this work.

We explore the extent to which gender disparities in Pakistan are reflected in the anonymized mobile phone logs of millions of Pakistani residents. Our analysis uses data capturing the communications behavior of several million individuals, for whom we observe the gender, but no additional demographic or personally identifying information. Here, we focus on validating aggregate regional patterns, correlating metrics derived from the mobile phone logs with socioeconomic statistics collected from more traditional sources. In these preliminary results, we observe a statistically significant relationship between districts with relatively high rates of female mobile phone penetration and districts that report high levels of gender parity in traditional surveys. However, this relationship is not uniform, and less developed regions exhibit a weaker correlation. We interpret these findings as suggestive evidence that such data can provide a novel perspective on gender dynamics in developing countries.

Billions of people around the world live without access to banks or other formal financial institutions. In the past several years, many mobile operators have launched "Mobile Money" platforms that deliver basic financial services over the mobile phone network. While many believe that these services can improve the lives of the poor, in many countries adoption of Mobile Money still remains anemic. In this paper, we develop a predictive model of Mobile Money adoption that uses billions of mobile phone communications records to understand the behavioral determinants of adoption. We describe a novel approach to feature engineering that uses a Deterministic Finite Automaton to construct thousands of behavioral metrics of phone use from a concise set of recursive rules. These features provide the foundation for a predictive model that is tested on mobile phone operators logs from Ghana, Pakistan, and Zambia, three very different developing-country contexts. The results highlight the key correlates of Mobile Money use in each country, as well as the potential for such methods to predict and drive adoption. More generally, our analysis provides insight into the extent to which homogenized supervised learning methods can generalize across geographic contexts. We find that without careful tuning, a model that performs very well in one country frequently does not generalize to another.

Despite substantial interest in the potential for mobile money to positively impact the lives of the poor, little empirical evidence exists to substantiate these claims. In this paper, we present the results of a field experiment in Afghanistan that was designed to increase adoption of mobile money, and determine if such adoption led to measurable changes in the lives of the adopters. The specific intervention we evaluate is a mobile salary payment program, in which a random subset of individuals of a large firm were transitioned into receiving their regular salaries in mobile money rather than in cash.

Efficient health systems require reliable data. In developing countries the need for accurate data is particularly acute, as organizations are often forced to make decisions on a tight budget with limited capacity for data collection. In this note, we describe recent progress toward developing a set of algorithms that can help detect and classify anomalies in health worker data. Building on recent efforts to use unsupervised multinomial techniques for outlier detection, we outline the steps required to turn a set of statistical tests into a framework that can be implemented by health organizations, and calibrate these algorithms on a large dataset from a partner health organization. Here, we describe the core methods, present results from ongoing analyses, and outline our plan for future work, including plans to obtain labeled training data that will allow us to detect and classify different types of outlier in community health worker data.

We utilize disaggregated, transaction-level call records to explore differences in the communication patterns of men and women in Rwanda. Consistent with prior research, we find that in aggregate, men are significantly more active on their phones. However, by disaggregating usage by time of day and day of year, we show the male-dominated use of mobile phones is not uniform over time. Namely, while men are more active during the day, women become more active at night. We also observe striking differences in men and women's phone activity on Christmas, Valentine's Day, and on politically important days such as the Rwandan and Kenyan Election Days. This paper chronicles these differences, situating the results within the broader literature on how men and women in developing countries interact with mobile phones, as well as other information and communication technologies.

We combine data from a field survey with transaction log data from a mobile phone operator to provide new insight into daily patterns of mobile phone use in Rwanda. The analysis is divided into three parts. First, we present a statistical comparison of the general Rwandan population to the population of mobile phone owners in Rwanda. We find that phone owners are considerably wealthier, better educated, and more predominantly male than the general population. Second, we analyze patterns of phone use and access, based on self-reported survey data. We note statistically significant differences by gender; for instance, women are more likely to use shared phones than men. Third, we perform a quantitative analysis of calling patterns and social network structure using mobile operator billing logs. By these measures, the differences between men and women are more modest, but we observe vast differences in utilization between the relatively rich and the relatively poor. Taken together, the evidence in this paper suggests that phones are disproportionately owned and used by the privileged strata of Rwandan society.

Wikipedia, free encyclopedia, now contains over two million English articles, and is widely regarded as a high-quality, authoritative encyclopedia. Some Wikipedia articles, however, are of questionable quality, and it is not always apparent to the visitor which articles are good and which are bad. We propose a simple metric -- word count -- for measuring article quality. In spite of its striking simplicity, we show that this metric significantly outperforms the more complex methods described in related work.