The World Bank Economic Review, 37(2), 2023, 235–256 https://doi.org10.1093/wber/lhad003 Article Mobile Broadband, Poverty, and Labor Outcomes in Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Tanzania Kalvin Bahia, Pau Castells, Genaro Cruz, Takaaki Masaki , Carlos Rodríguez-Castelán, and Viviane Sanfelice Abstract What are the impacts of expanding mobile broadband coverage on poverty, household consumption, and labor- market outcomes in developing countries? Who benefits from improved coverage of mobile internet? To respond to these questions, this paper applies a difference-in-differences estimation using panel household survey data combined with geospatial information on the rollout of mobile broadband coverage in Tanzania. The results reveal that being covered by 3G networks has a large positive effect on total household consumption and poverty reduction, driven by positive impacts on labor-market outcomes. Working-age individuals living in areas covered by mobile internet witnessed an increase in labor-force participation, wage employment, and non-farm self-employment, and a decline in farm employment. These effects vary by age, gender, and skill level. Younger and more skilled men benefit the most through higher labor-force participation and wage employment, while high-skilled women benefit from transitions from self-employed farm work into non-farm employment. JEL classification: F63, I31, L86, O12 Keywords: Africa, consumption, labor-force participation, welfare, Tanzania 1. Introduction Enabling universal access to the internet is deemed a critical step towards achieving prosperity in devel- oping countries. In line with this target, the digital landscape in Sub-Saharan Africa has been changing Kalvin Bahia is Principal Economist at GSMA, London, UK; his email address is kbahia@gsma.com. Pau Castells is Head of Economic Analysis at GSMA, London, UK; his email address is pcastells@gsma.com. Genaro Cruz is Managing Partner and Co-founder of Tenolli, London, UK; his email address is cruzgenaro@gmail.com. Takaaki Masaki (corresponding author) is Senior Economist at the World Bank Poverty & Equity Global Practice and the World Bank/UNHCR Joint Data Center on Forced Displacement, Copenhagen, Denmark; his email address is tmasaki@worldbank.org. Carlos Rodríguez-Castelán is Lead Economist at the World Bank Poverty & Equity Global Practice, Washington DC, USA, and also Research Fellow at IZA, Bonn, Germany; his email address is crodriguezc@worldbank.org. Viviane Sanfelice is Assistant Professor at the Department of Economics, Temple University, Philadelphia PA, USA; her email address is viviane.sanfelice@temple.edu. The authors wish to thank Tim Kelly, Dhiraj Sharma, Maurice Kugler, Siobhan Murray, Arden Finn, Nadia Belhaj Hassine, and Jonathan Kastelic for their comments and advice. This study received funding from the Digital Development Partnership Trust Fund and from the UK’s Foreign, Commonwealth and Development Office. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent, or the UK government’s official policies, or those of the GSMA and GSMA Intelligence. A supplementary online appendix is available with this article at The World Bank Economic Review website. © 2023 International Bank for Reconstruction and Development / The World Bank. Published by Oxford University Press. 236 Bahia et al. drastically, with fast-growing mobile broadband networks. Mobile broadband coverage in the region in- creased from 24 percent of the population in 2010 to 81 percent in 2020 (GSMA 2021). However, despite enthusiasm around the potential role that the internet plays in spurring growth and tackling poverty across developing countries, there is limited evidence on the welfare effects of mobile broadband – particularly among the poor and vulnerable in Sub-Saharan Africa.1 Given this knowledge gap, this paper seeks to answer the following questions: What are the impacts Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 of expanding mobile broadband (the broad range of communications protocols made feasible by 3G coverage) on poverty, household consumption and labor-market outcomes in developing countries? And who benefits from improved coverage of mobile broadband?2 These questions are particularly important in Africa, where a predominant share of people have access to the internet only through their mobile phones. In particular, Tanzania is among the top few countries where notable increases in mobile phone and smartphone penetration are expected in the coming years.3 To answer these questions, we leverage three waves of a nationally representative longitudinal household survey on living standards in Tanzania with geospatial information on the rollout of 3G coverage between 2008 and 2013. By matching the location of each household in the panel survey with coverage maps of mobile broadband, we can determine with precision the time when individual households began receiving mobile broadband coverage, and we can empirically test whether the staggered rollout of mobile broadband networks has contributed to changes in welfare and poverty reduction. We implement a difference-in-differences (DID) estimation method, which follows an intention-to- treat framework. Specifically, our analysis focuses on the effects of mobile broadband network coverage (or availability) instead of usage (or access) since the former is independent of individual household con- sumption decisions. Moreover, we perform additional validity checks to circumvent the potential concern regarding our empirical strategy that 3G coverage has been gradually rolled out in a non-random manner, since operators tend to target more prosperous areas first. In these alternative specifications, we control for local-level time trends and exploit the timing of the treatment to find more compatible control groups. We also conduct conditional pre-trend tests. Based on these methods, we conclude that the assumption of similar pre-treatment trends is likely to hold in our context. Finally, we also apply a doubly robust DID as suggested by Callaway and Sant’Anna (2020) to deal with potential bias that arises from the staggered DID setup where households were treated at different points in time. Our main results show significant positive effects of 3G coverage on household per capita consumption. Households that resided in areas covered by 3G experienced an 7–11 percent increase in total per capita consumption. These digital dividends materialize over time and become statistically significant after more than one year of exposure to 3G coverage. Mobile broadband coverage also reduces the proportion of households below the national poverty line by 5–7 percentage points. These results are consistent and similar in magnitude to recent studies for Nigeria (Bahia et al. 2020), the Philippines (Blumenstock et al. 2020), and Senegal (Masaki, Granguillhome Ochoa, and Rodriguez-Castelan 2020).4 These effects are 1 Most hitherto studies focus almost exclusively on cell phone access – that is, second-generation (2G) technologies, those that enable voice, SMS, and limited internet access, while third-generation (3G) technologies enable more rapid internet browsing and data downloading (Jensen 2007; Labonne and Chase 2009; Muto and Yamano 2009; Aker 2011; Beuermann, McKelvey, and Vakis 2012; Ky, Rugemintwari, and Sauviat 2018; Blumenstock et al. 2020) – and fixed broadband internet (Atasoy 2013; Akerman, Gaarder, and Mogstad 2015; Hjort and Poulsen 2019). 2 Coverage of mobile broadband is defined as the provision of 3G coverage, which enables high-speed access to the internet and convenient use of multimedia content such as video. However, 2G coverage is excluded from this definition as it only provides for limited internet browsing and applications. 3 Smartphone usage in Tanzania is expected to grow at an annual rate of 19 percent, reaching more than 30 million mobile subscriptions by the end of 2024 (Chen et al. 2020). 4 Bahia et al. (2020) show that having at least one year of mobile broadband coverage increases total consumption by 5.8 percent, and up to 9.2 percent after three years or more of coverage. Masaki, Granguillhome Ochoa, and Rodriguez- Castelan (2020) find that total consumption among households covered by 3G technology is about 14 percent greater The World Bank Economic Review 237 heterogeneous, with higher welfare gains in urban areas and among households headed by female, poorer, or less-educated household heads. Our analysis also shows that an important mechanism through which 3G coverage translates into higher household consumption and moving out of poverty is its impact on labor-market outcomes, in line with previous studies (Hjort and Poulsen 2019; Bahia et al. 2020). Working-age individuals living in areas covered by mobile broadband witnessed an increase in labor-force participation, wage employment, Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 and non-farm self-employment by 3–8 percentage points after 1–2 years of 3G exposure. Living in areas covered by mobile broadband also reduced farm employment by 4–9 percentage points. These results indicate that 3G coverage has facilitated a transition out of farm jobs into wage employment and non- farm self-employment. This is an important finding since farm employment in developing countries has been typically characterized by lower productivity and labor earnings (Schultz 1956; Restuccia, Yang, and Zhu 2008; McMillan and Rodrik 2011). These positive labor-market outcomes of 3G coverage vary by age, gender, and skill level. We find that younger and more educated men benefit the most through higher labor-force participation and wage employment. Results also show that skilled women, with at least completed primary education, benefit from transitions from self-employed farm work into non-farm employment. But, differently to men, we do not find evidence of changes in labor-force participation or changes in wage employment for women. These findings are consistent with previous studies which suggest that, while women can benefit from digital technologies, they often face greater difficulties to leverage them due to a mix of social norms, intra-household dynamics, lack of access to productive assets, and being less likely than men to use the internet (Bimber 2000; Klonner and Nolen 2010; Zhao 2020). Our findings are relevant to the growing literature on the positive economic impact of digital technolo- gies such as cellular services (Jensen 2007; Labonne and Chase 2009; Muto and Yamano 2009; Aker 2011; Beuermann, McKelvey, and Vakis 2012; Ky, Rugemintwari, and Sauviat 2018; Blumenstock et al. 2020) or fixed broadband internet (Atasoy 2013; Akerman, Gaarder, and Mogstad 2015; Hjort and Poulsen 2019). In our analysis, we focus on the role of mobile broadband – more specifically, 3G technology–while con- trolling for other technologies such as 2G coverage. The focus on mobile broadband is important because, for most people in Africa, the only means to access the internet remains through mobile phones.5 Internet connectivity can directly affect workers’ productivity, internet-induced human capital accumulation, and internet-induced changes in firm–worker matching. The internet may also influence firm-level productiv- ity by facilitating adoption of (other) technologies or market access. Moreover, internet connectivity may reduce information frictions (for a complete discussion in the context of developing economies see Hjort and Tian 2021). Our study also contributes to the broader literature studying the development effects of digital tech- nologies. Previous studies have focused on the impact of digital technology on various development out- comes, including, but not limited to, the labor market (De los Rios 2010; Klonner and Nolen 2010; Marandino and Wunnava 2014; Guerrero and Ritter 2014; Paunov and Rollo 2015; Chun and Tang 2018; Hjort and Poulsen 2019; Fernandes et al. 2019; Viollaz and Winkler 2020), firm productivity (Abreha et al. 2021), input and output price and rural markets (Aker and Mbiti 2010; Aker 2011; Aker and Fafchamps 2015; Tadesse and Bahiigwa 2015; Kaila and Tarp 2019), financial inclusion (Aker and Wilson 2013; Ky, Rugemintwari, and Sauviat 2018), and access to capital markets (Alibhai et al. 2018; Hasbi and Dubus 2020). Other studies have also shed light on the role that mobile-based finan- cial services play in poverty reduction and facilitating consumption smoothing against exogenous risks (Jack and Suri 2014; Suri and Jack 2016; Blumenstock, Eagle, and Fafchamps 2016; Abiona and Koppensteiner 2022). This study is also similar to Hübler and Hartje (2016) which shows that the than the total consumption of households not covered by 3G. Blumenstock et al. (2020) find that the introduction of a new phone tower in rural areas in the Philippines led to an increase in household income of 17 percent, and increased household expenditures by 10 percent. 5 In 2020, mobile accounted for more than 98 percent of broadband connections in Africa (ITU 2020). 238 Bahia et al. ownership of smartphones has positive impact on household income. But it is important to note that our focus is the coverage of mobile broadband infrastructure instead of the ownership of smartphones per se. Our work builds upon these existing studies attesting to the potential positive impact of mobile broadband on household welfare and poverty. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 2. Background and Data Our analysis draws upon two data sets: geographical coverage of mobile networks and three rounds of the Tanzania National Panel Survey (NPS). By linking the location and the chronology of these data sets, we are able to determine the availability of 3G technology at the exact household location. 2.1. Mobile Broadband Coverage Data In this study we look at mobile broadband network coverage (or availability). Coverage is distinct from usage – or access – which is when an individual has an active SIM card that can be used in a mobile phone to access the internet. We look at coverage because it is not determined at the household level and because coverage captures not only the direct impact of individuals accessing the internet but also spillover effects. The latter includes, for example, internet users sharing information with non-users, job creation and productivity gains among domestic firms due to more technology use and access to national and international markets, or more financial capital accumulation and utilization due to increased use of digital payment platforms and mobile banking. Mobile networks can be divided into two components. The first is the core network, which ensures the intelligence of the network, such as switching user calls or routing user data to and from the inter- net. The second is the radio access network, which is the collection of relay sites (i.e. towers hosting base stations and radio equipment) that connects the user terminal (e.g. mobile phones) to the core of the network. Relay sites communicate with mobile phones in their vicinity using electromagnetic signals. The quality and availability of this communication link can be affected by several factors, such as dis- tance between the relay site and the mobile phone or the presence of obstacles (e.g. hills or buildings). A geographical area is considered covered when the signal of any relay site is strong enough for mobile phones in that area to establish a usable connection link with that relay site. The aggregated coverage of a mobile network is calculated by adding up the coverage of all the relay sites in its radio access network. To produce the aggregate coverage data for Tanzania, we collected network infrastructure data directly from three Mobile Network Operators (MNOs) that accounted for more than 90 percent of the mobile market during the period of analysis. For each individual relay site we collected the following parameters: (a) location in geographical coordinates, (b) height of the tower hosting the antennas, (c) signal emitting power, (d) antenna parameters such as the gain, azimuth, and tilt, (e) frequency band used, (f) type of technology available (2G, 3G, or 4G), and (g) date of deployment. We calculated the coverage of each relay site using a Radio Propagation Model (RPM). RPMs are empirical mathematical models widely used by MNOs for planning the setup of their networks, allowing them to plan the location and characteristics of each relay site so as to maximize coverage and decrease costs. There are several RPMs available that are optimized for specific settings or technologies. We used an Irregular Terrain Model (ITM), also known as the Longley Rice model (Rice and Longley 1968), which is optimized to deliver accurate results in rural and peri-urban areas.6 6 We estimated coverage using the ITM model and MNO infrastructure inputs instead of collecting the coverage footprint estimated by MNOs themselves. This approach ensured consistency in the coverage modeling for each MNO network and across all the period of interest. The World Bank Economic Review 239 The ITM uses two sets of input variables. The first are the technical parameters of each individual relay site that we collected from MNOs. The second are the characteristics of the transmission medium, such as the terrain profile7 and the type of vegetation in the area. The output of the ITM model is a geocoded image showing the area covered with signal strength above a predefined threshold.8 , 9 During the period of analysis, from 2008/2009 to 2012/2013, 3G population coverage in Tanzania increased from 13 percent to 30 percent, while 2G coverage increased from 81 percent to 85 percent. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 None of the operators started to deploy 4G until after the period of analysis. 2.2. Household Survey Data Data on household welfare (e.g. consumption and poverty) and other household-specific characteristics are sourced from the first three rounds of NPS, conducted in 2008/2009, 2010/2011, and 2012/2013.10 The NPS collected information on a wide range of topics including agricultural production, non-farm income generating activities, consumption expenditures, and a range of socioeconomic characteristics. The panel structure of the survey data – tracking the same sample of households and individuals over time – allows us to explicitly control for household-specific and individual-specific characteristics. The NPS maintains a highly successful recapture rate (roughly 96 percent retention at the household level), thereby minimizing potential bias introduced by attrition. The original sample of the 2008/2009 survey consisted of 3,265 households – the second round tracked 97 percent of these original households and the third round tracked 96 percent of the second round households.11 The core outcome variables of our interest derived from the household survey include total consump- tion (which is the sum of all food consumption and purchases, including meals outside home, as well as non-food expenditure such as education, housing rents, clothing, fuel, utilities, transportation, communi- cation, reaction, and other services), food consumption, and non-food consumption. The poverty status of households is calculated based on three different poverty lines: the basic needs poverty line defined on the basic needs approach – which measures the cost of acquiring enough food to provide adequate daily nutrition per person (food line) plus the cost of some non-food essentials (non-food component)12 – as well as the international poverty lines of US$1.90 and US$3.20 per day (2011 PPP). All the mone- tary values in the surveys have been deflated to convert nominal values in real/constant values, using the Consumer Price Index (CPI) for Tanzania. Figure 1 presents the location of the enumeration areas surveyed in NPS and shows which had access to 3G coverage in each round of NPS. We consider a household to have mobile broadband coverage when 7 We used the SRTM 90m Digital Elevation Database created by NASA. See https://cgiarcsi.community/data/srtm-90m- digital-elevation-database-v4-1/. 8 The predefined signal strength thresholds that we used are −85 dBm (medium signal strength) and −73 dBm (strong signal strength) for 2G and −91 dBm (medium signal strength) to −83 dBm (strong signal strength) for 3G. 9 The ITM model, as with many other empirical radio propagation models, is inherently statistical. One of the inputs of the ITM is a “reliability” parameter, which can be interpreted as the probability that true signal strength is at least as strong as the predicted one. In our modeling we used a reliability value of 95 percent, meaning that signal strength should be at least as strong as the value predicted with 95 percent confidence. Furthermore, for our study we used the medium signal strength to determine coverage (see footnote 9), while mobile operators tend to use weaker signal values to calculate service coverage. These two factors compound to make our coverage predictions conservative. 10 Although the first three rounds of NPS are real panel data, the last round (2014/2015) was implemented as a cross- sectional survey based on a new redrawn sample. It is therefore excluded from this analysis. 11 See more details on the tracking performance of NPS in the final report produced based on Wave 3 of NPS: https://microdata.worldbank.org/index.php/catalog/2252/download/34054. 12 More specifically, for NPS, the poverty line was derived based on the cost of buying 2,200 calories per adult per day, according to the food consumption patterns prevailing in a population whose per adult real consumption is below the median during a period of 28 days, valued at prices faced by the reference population. The non-food component of the basic needs poverty line uses the average non-food consumption share of the population whose total consumption per 240 Bahia et al. Figure 1. Coverage for Surveyed Locations in 2008 and 2013 (a) (b) Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: These maps show the locations of enumeration areas with access to 3G technology. Black dots mean covered enumeration areas whereas light grey ones indicate no coverage. it is covered by a medium or strong signal in 3G. The two data sets were linked by matching the coverage footprint for each radio bearer with the almost exact locations of households based on GPS coordinates, which are much more precise than georeferencing data used in previous studies.13 2.3. Summary Statistics Although our sample is not balanced across survey waves, it only includes households and individuals that have data for at least two points in time. Table 1 provides averages for the main variables in our study by survey wave. The first panel presents household-level information. The main outcome variables – namely, log consumption and poverty rates – do not change considerably over time. By contrast, the treatment variable, 3G coverage, increases by more than half from the first to the third survey wave, going from 16 percent of the household being under covered areas in 2008–2009 to 35 percent in 2012–2013. In terms of control variables and household profile, we see that the majority of households own their dwelling, do not have access to electricity, are located in rural areas, and have a male and literate household head. Panel B of table 1 displays averages for key variables in our individual-level sample. Around three- quarters of individuals in the sample (adults aged 15 to 64 years old) are in the labor force. Types of employment are also among our dependent variables and consist of four categories: wage employment, non-farm self-employment, farm employment, and other employment.14 These categories are not mutually exclusive, as individuals could report working in more than one type of job. Most workers are employed in farming jobs. Employment rates in wage jobs and non-farm self-employment are of similar magnitude, with about 15 percent of individuals in our sample employed in these two categories. Around half of individuals have completed primary school and the vast majority are literate. adult is in the bottom 25 percent. See more details on the basic needs approach to derive the poverty line for NPS in World Bank (2020). 13 More specifically, we rely on GPS coordinates of households with the maximum offset of 45 m to assign to each house- hold the initial date of coverage (if any) by any combination of radio bearer and signal strength. In contrast, other studies have relied on data integration using buffer sizes from 1 to 5 km in urban areas, and 1 to 20 km in rural areas (see for instance DHS – Demographic and Health Surveys). 14 Other employment is defined as an indicator variable equal to 1 when an individual responded that he/she works but did not report working in any of the other categories, i.e. on wage employment, non-farm self-employment, and farm self-employment. The World Bank Economic Review 241 Table 1. Summary Statistics Wave 1, 2008–09 Wave 2, 2010–11 Wave 3, 2012–13 A. Households Outcome variables Total consumption 72,398.34 71,392.32 73,536.83 Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Food consumption 42,322.51 48,892.47 64,132.79 Non-food consumption 18,260.08 22,517.49 28,940.96 Poor (basic needs poverty line) 0.19 0.19 0.19 Extreme poor ($1.9 PPP poverty line) 0.39 0.38 0.38 Poor ($3.2 PPP poverty line) 0.65 0.65 0.64 Treatment variables 3G coverage 0.16 0.31 0.35 Less than 1 year 3G coverage 0.01 0.07 0.02 1–2 years 3G coverage 0.02 0.08 0.05 2–3 years 3G coverage 0.08 0.01 0.07 More than 3 years 3G coverage 0.05 0.15 0.21 Controls and demographics 2G coverage 0.76 0.77 0.79 Ownership of dwelling unit 0.79 0.74 0.76 Access to electricity 0.21 0.24 0.27 Own at least one cell phone 0.45 0.62 0.72 Dwelling located in rural area 0.65 0.68 0.66 Household size 5.13 5.29 5.32 Female head of household 0.24 0.24 0.23 More primary educ. head household 0.21 0.22 0.22 Literacy status head of household 0.78 0.75 0.76 Age head of household 46.26 46.57 48.26 Observations 2,752 3,672 3,465 B. Individuals Outcome variables Labor-force participation 0.75 0.78 0.77 Wage employment 0.15 0.16 0.18 Self-employed non-farm 0.16 0.16 0.19 Self-employed farm 0.44 0.45 0.45 Other employment 0.11 0.09 0.06 Treatment variables 3G coverage 0.16 0.32 0.37 Less than 1 year 3G coverage 0.01 0.07 0.01 1–2 years 3G coverage 0.02 0.10 0.05 2–3 years 3G coverage 0.08 0.01 0.07 More than 3 years 3G coverage 0.05 0.15 0.22 Controls and demographics 2G coverage 0.76 0.77 0.79 Female 0.52 0.52 0.52 More primary educ. 0.51 0.50 0.47 Literacy status 0.84 0.81 0.82 Age 31.41 31.36 33.02 Observations 7,403 10,376 9,891 Source: National Panel Survey (NPS) Waves 1, 2, and 3. Note: This table reports the mean values of each of the variables listed for each wave of NPS. Note that 2G and 3G coverage areas are not mutually exclusive and there is a significant overlap between them. 242 Bahia et al. 3. Empirical Strategy We are interested in assessing how exposure to mobile broadband, measured as coverage by 3G tech- nology, affects household and individual welfare and whether this impact may differ by some individual characteristics. However, identifying the impact of 3G coverage on welfare is not trivial because expo- sure to the treatment is not random. Households residing in areas with access to mobile broadband are Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 likely to be distinct in several dimensions from households with no access. Mobile broadband services are provided by profit-maximizing firms that supply the service where economic benefits are expected. This means that households which receive coverage are more likely to be economically prosperous, and therefore have means to consume the product. In order to overcome this endogeneity we take advantage of the temporal and spatial variation in exposure to 3G by applying a DID approach that compares outcomes of households in treated and non- treated areas before and after mobile broadband expansion. It is worth noting that our estimates will reflect the effect of switching from 2G to 3G coverage because most of the variation in our treatment comes from this shift.15 To retrieve the effect of 3G coverage on household welfare and labor-market outcomes, here denoted by β , we consider the following equation: yit = β coverageit + Xit θ + υi + ωt + it , (1) in which i is a household or individual during period (survey wave) t. The variable y denotes an outcome variable such as consumption or labor-market participation. The variable coverage is the variable of interest and is defined by an indicator variable equal to 1 if household or individual i is covered by a 3G network in time t and 0 otherwise. The vector X contains time-varying control variables. These include household size, access to electricity, and whether the household dwelling is owned, which are broadly in line with the controls used in the literature. Further, we add access to 2G coverage as a control to ensure that the analysis isolates the impact of upgrading coverage to 3G and does not combine the impact of gaining 2G coverage. The variable υ i denotes household fixed effects (or individual fixed effects depending on the outcome variable), which capture time-invariant characteristics that could correlate to coverage and the dependent variable. The variable ωt represents survey wave fixed effects and accounts for aggregate trends over time. To account for spatial autocorrelation, standard errors are clustered at the survey enumeration area (EA) level. The EA matches most closely the area covered by a mobile site. Our empirical strategy follows an intention-to-treat framework. That is, we look at mobile broadband network coverage (or availability) instead of usage (or access). This is because the former is external to household decisions and usage is endogenous to household welfare. Additionally, having 3G coverage as the variable of interest will not only capture the direct impact of households accessing mobile broadband but also its potential spillover effects. The identification assumption in equation (1) is that conditional on time fixed effects, household fixed effects, and household time-variant characteristics contained in X, coverage timing is orthogonal to unob- served characteristics related to economic development. A key assumption in the identification strategy is that all differences across treated and non-treated are accounted for, either by the household fixed effects (time-invariant characteristics) or by the variables in X (time-variant characteristics). We also disaggregate the 3G coverage effect by time of exposure by adjusting equation (1) and adding time exposure intervals, as stated below. It is important to assess how the varying lengths of treatment exposure have differential impacts on welfare, because, plausibly, the welfare impact of mobile broadband 15 In our sample we only observe 24 households changing their coverage status from no coverage to 3G coverage. Thus, the parameters of interest are identified by comparing outcomes of treated units before and after their coverage status changes from 2G to 3G coverage, using as comparison group units whose coverage status remained unchanged over the sample period. When we run our basic specification dropping households with no coverage (i.e. only comparison of 3G versus 2G), our results remain very similar to the results using the full sample. The World Bank Economic Review 243 (if any) may not necessarily materialize immediately but take some time before it manifests any meaningful welfare effect. This exercise allows us to differentiate short- and medium-term effects, yit = β1 less than 1 yearit + β2 1 to 2 yearsit + β3 2 to 3 yearsit +β4 more than 3 yearsit + Xit θ + υi + ωt + it . (2) Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Despite coverage not being determined at the household level, it could still be endogenous, because mobile broadband coverage is not rolled out randomly. This is because operators tend to target more prosperous areas first, which in turn are likely to have higher consumption levels and lower poverty rates. Thus, even if longitudinal data is available, regional trends can correlate to 3G coverage and the outcome variables. This means that households receiving mobile broadband coverage may not be comparable to non-treated households. To address this concern, we implement two robustness checks to account for location time trends. The first check adds to equations (1) and (2) non-linear time trends for each region in Tanzania.16 The second validity check is based on the method discussed by Abadie (2005) and allows for non-linear time trends based on observed characteristics of the household location. Since operators target certain areas with higher economic development (or expected development), allowing for regional trends or trends by observed characteristics of the household location in addition to household fixed effects should mitigate potential concerns about the identification strategy. We also carry out a conditional pre-trend test that provides evidence on whether the assumption of similar pre-treatment trends for the treated and non-treated holds for our analysis. Finally, we imple- ment an additional analysis in which the sample is reduced to only include comparable observations. The idea is to explore the timing of the 3G coverage rollout to find a control group more similar to treated units. Specifically, we use households/individuals treated later on in wave 3 as a comparison group for households/individuals treated earlier.17 In our context, treatment is not stagnated, as households are covered at different points in time. Recent advancement in the DID literature has shown that the two-way fixed effect (TWFE) DID estimation can yield biased estimates. This bias may result from the variance weighting implicit in ordinary least squares, and more importantly due to the embedded use of past treated units as effective controls for later-treated units (Baker, Larcker, and Wang 2022). This bias can be relevant particularly to our study given that our treatment was rolled out gradually to households across multiple time periods. In this sense, our sample deviates from the so-called canonical DID setup in which all the units in the treatment group receive the treatment at the same point in time. To address this estimation challenge, we take advantage of a recent method introduced by Callaway and Sant’Anna (2020) (C&S hereafter) that establishes a procedure18 to (a) flexibly incorporate covariates into the staggered DID setup with multiple groups and multiple periods, (b) test pre-trends conditioned on those covariates, and (c) estimate group-time average treatment effects and aggregate them in flexible ways. Their proposed approach can also be implemented using doubly robust standard estimators, which rely on less stringent assumptions than the TWFE models (Sant’Anna and Zhao 2020). Finally, we explore potential heterogeneity in the effects of digital technology along several key demo- graphic characteristics to better understand who stood to benefit from the expansion of mobile broadband. In the majority of low- and middle-income countries, there remains a significant and persistent digital di- vide between men and women (Broadband Commission 2013; Alozie and Akpan-Obong 2017; Fatehkia, Kashyap, and Weber 2018; Granguillhome Ochoa et al. 2022) and rural and urban populations (Grazzi and Vergara 2014; Granguillhome Ochoa et al. 2022). Women are less likely to use mobile broadband 16 Households in our sample reside in 30 regions. 17 Results of this analysis and the pre-trend test are provided in the “Robustness Checks” section when implementing the Callaway and Sant’Anna (2020) method. 18 This procedure can be implemented in the R “did” package. 244 Bahia et al. than men, and rural inhabitants are less likely to use it than urban residents, while adoption is also lower for the poorest populations. Furthermore, evidence has shown that younger and more educated individu- als are more likely to use the internet than the elderly and less educated (GSMA 2021). We therefore test whether the effect of 3G coverage has different impacts based on gender, geography, age, and education. To implement heterogeneous effects analysis we include an interaction between the variable of interest in equation (1) and the characteristics of individuals. This approach allows us to separately estimate the Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 effect of 3G coverage for groups in the population, and to also test whether these effects are statistically different across groups.19 The heterogeneous effects analysis also looks at the impact of mobile broadband coverage on individ- uals that have access to a mobile phone and those who do not. This serves two purposes. The first is that populations that never acquired or accessed a mobile device are very unlikely to be those targeted by oper- ators for 3G coverage. Therefore, if we do find an effect on these households, it is much more likely to be the case that mobile broadband increases welfare. The second is that if we find that 3G coverage affects la- bor outcomes for those without a phone, this provides strong evidence of the existence of spillover effects. This could include, for example, non-phone users benefiting from information received from phone users and/or benefiting from the wider job creation and productivity gains that mobile broadband can generate. 4. Results 4.1. Baseline Results Table 2 displays estimates of parameters in equations (1) and (2) for household outcomes such as consumption and poverty status. In Panel A, which displays estimates for an indicator variable about 3G coverage, all coefficients are economically meaningful. In particular, based on the specification that includes fixed effects and control variables, we find that being covered is associated with a 7 percent increase in consumption, 6 percent increase in food consumption, and 9 percent increase in non-food consumption. We also observe a significant reduction of 5 percentage points in the basic need poverty rate (in wave 1, 19 percent of the households in our sample are classified as poor under the basic need poverty line). Poverty rates at the US$1.9 and US$3.2 PPP poverty lines also decline, though the coefficients are not statistically significant (in wave 1, 39 percent and 65 percent of the households in our sample (unweighted) are classified as poor under the US$1.9 and US$3.2 PPP poverty lines, respectively). These results on poverty indicate that the effect of mobile broadband is particularly strong for households that are poorer and teetering on the edge of the basic poverty line (the point at which they are not able to consume sufficiently to keep their daily calorie intake requirement). Panel B presents estimates broken down by coverage exposure intervals. Significant effects on house- hold consumption and poverty status are observed for households being covered for at least one year. Although the estimates change for coverage intervals greater than one year, with large standard errors we cannot reject that they are statistically equal to each other. Being covered for less than a year has a small and statistically insignificant effect on all of the household outcomes. It is also worth noting that if coverage expansion occurred in areas that were already growing (in terms of economic development and consumption), our estimates would overestimate the true effect of mobile broadband on household wel- fare. The fact that we do not observe statistically significant estimates for less than one year of coverage suggests that confounding factors are not overstating our estimates. Table 3 shows estimates using individual-level data and dependent variables on labor-market outcomes. We can see that 3G coverage is positively associated with a 3 percentage points increase in labor-force participation. Most of the impacts on the labor market come from transitions across employment types. 19 Another potential identification strategy that we could pursue is a Spatial Regression Discontinuity design. In addition to the 3G coverage status of each household, we have data on distance to the coverage boundary, which can be used to compare welfare levels for households near the boundary formed between coverage and non-coverage areas (Keele and Titiunik 2015). However, this approach has not been used for our main analysis because some of the identifying Table 2. DID Results for Household Outcomes Food Non-food Basic need Extreme poor Poor Dep. variable: Consumption consumption consumption poor ($1.9 PPP) ($3.2 PPP) A. Exposure effect The World Bank Economic Review 3G coverage 0.08*** 0.07*** 0.07** 0.06** 0.10*** 0.09*** −0.05*** −0.05*** −0.03* −0.03 −0.02 −0.02 (0.02) (0.02) (0.03) (0.03) (0.03) (0.03) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Observations 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 R2 0.79 0.81 0.73 0.74 0.83 0.83 0.57 0.58 0.65 0.66 0.72 0.73 B. Effect by time of 3G exposure <1 year 0.01 0.00 −0.02 −0.02 0.07 0.06 −0.02 −0.03 −0.01 −0.01 0.01 0.01 (0.03) (0.03) (0.04) (0.03) (0.05) (0.05) (0.02) (0.02) (0.02) (0.02) (0.03) (0.03) 1–2 years 0.11*** 0.11*** 0.11*** 0.10*** 0.11*** 0.10** −0.06*** −0.06*** −0.05** −0.05** −0.04 −0.03 (0.03) (0.03) (0.03) (0.03) (0.04) (0.04) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) 2–3 years 0.08** 0.07** 0.09** 0.07* 0.11** 0.09** −0.04** −0.04** −0.03 −0.02 −0.02 −0.01 (0.03) (0.03) (0.04) (0.04) (0.04) (0.04) (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) >3 years 0.11*** 0.10*** 0.11*** 0.10*** 0.13*** 0.11*** −0.06*** −0.07*** −0.02 −0.02 −0.03 −0.03 (0.03) (0.03) (0.03) (0.03) (0.04) (0.04) (0.02) (0.02) (0.02) (0.02) (0.03) (0.02) Observations 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 9,887 R2 0.79 0.81 0.73 0.74 0.83 0.83 0.57 0.58 0.65 0.66 0.72 0.73 Controls No Yes No Yes No Yes No Yes No Yes No Yes Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: All regressions include household and wave fixed effects. Controls refer to 2G coverage, access to electricity, house ownership, wealth index, and household size. Panels A and B use the same specifications but present results of separate regressions. In Panel A the treatment variable is an indicator variable equal to 1 if the household is exposed to 3G in time t. In Panel B the treatment variables are mutually exclusive dummies equal to 1 if the household in time t has been exposed to 3G for that time interval. Consumption variables are in log. Robust standard errors clustered at the EA level (410 clusters) are shown in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. 245 Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 246 Table 3. DID Results for Individual Outcomes Labor-force Wage Self-employed Self-employed Other Number of Dep. variable: participation employment non-farm farm employment employments A. Exposure effect 3G coverage 0.03** 0.03** 0.03*** 0.03** 0.03** 0.02** −0.08*** −0.07*** 0.03** 0.03** 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) Observations 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 R2 0.65 0.65 0.60 0.61 0.61 0.61 0.67 0.68 0.39 0.39 0.63 0.63 B. Effect by coverage time <1 year 0.01 0.01 0.02 0.02 −0.03* −0.03* −0.10*** −0.09*** 0.07*** 0.06*** −0.04 −0.04 (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.03) (0.03) 1–2 years 0.03* 0.02* 0.02 0.01 0.05*** 0.04*** −0.08*** −0.07*** 0.02 0.02 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) (0.01) (0.01) (0.02) (0.02) 2–3 years 0.03 0.03 0.06*** 0.05*** 0.04** 0.03** −0.05*** −0.04** −0.03 −0.03 0.02 0.02 (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) >3 years 0.08*** 0.08*** 0.06*** 0.04*** 0.04*** 0.04** −0.09*** −0.07*** 0.05*** 0.05*** 0.06** 0.06** (0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Observations 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 27,604 R2 0.65 0.65 0.60 0.61 0.61 0.61 0.67 0.68 0.39 0.39 0.63 0.63 Controls No Yes No Yes No Yes No Yes No Yes No Yes Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: All regressions include individuals and wave fixed effects. Controls refer to 2G coverage, access to electricity, house ownership, wealth index, and household size. Panels A and B use the same specifications but present results of separate regressions. In Panel A the treatment variable is an indicator variable equal to 1 if the household is exposed to 3G in time t. In Panel B the treatment variables are mutually exclusive dummies equal to 1 if the individual in time t has been exposed to 3G for that time interval. Robust standard errors clustered at the EA level (410 clusters) are shown in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. Bahia et al. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 The World Bank Economic Review 247 Specifically, we find that individuals in areas gaining 3G coverage leave farm jobs and increase wage employment and non-farm self-employment. Wage employment and non-farm self-employment increase by 2 and 3 percentage points respectively, while farm employment decreases by 7 percentage points. These results are revealing, as non-farm jobs can be perceived as a sign of prosperity for providing better working arrangements and more stable cash flow. It is notable that while farm self-employment reduced strongly for individuals treated with 3G coverage, participation in farm self-employment in Tanzania throughout Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 the period remained fairly stable (table 1). When analyzing the estimates separately by time of coverage, we see a substantial and statistically significant effect in the medium term on labor-force participation, which expands by 8 percentage points when individuals have been exposed to mobile broadband for more than three years. The results also show that while there is an immediate and persistent reduction in farm employment from the first time of exposure, estimates for non-farm self-employment are only statistically significant after the first year of coverage, and for wage employment only after the second year of exposure to the technology. This suggests that transitions to higher-paying jobs take time to materialize once individuals gain coverage. 4.2. Robustness Checks Our first robustness check controls for location time trends in order to separate the impact of mobile coverage arrival from other potential ongoing trends in regional outcomes. Tables with estimates can be found in columns 2, 6, and 10 of tables 4 and 5. While some estimates reduce in magnitude and in statistical significance, the findings for 3G coverage effect on household consumption and poverty status hold, especially after one year of coverage. A similar picture can be seen in terms of individuals’ labor- market outcomes, with the findings on employment broadly consistent, especially after one or two years of coverage. We conclude that in general these tests corroborate our findings. The second robustness check adopts the semi-parametric method proposed by Abadie (2005), which accounts for non-linear time trends based on observed characteristics of the household location. Specifi- cally we add as controls to equations (1) and (2) dummy variables for each wave interacted with variables about the household: (a) distance to major road, (b) distance to nearest population center, (c) distance to nearest border crossing, and (d) distance to headquarters of district of residence). Since operators are more likely to target areas with higher economic development (or expected development), allowing for regional trends or trends by observed characteristics of the household location in addition to household fixed effects should mitigate potential concerns about the identification strategy. The results are presented in columns 3, 7, and 11 of tables 4 and 5. The main results stand fairly robust to our baseline results, although the magnitude and significance of some estimates reduce. The third robustness check is to use the C&S approach, the results of which are presented in columns 4, 8, and 12 of tables 4 and 5.20 Note that in these tables, the control group is a group of households that were never exposed to the treatment throughout Waves 1–3, whereas we also report in supplementary online appendix S1 the results from using the “not yet treated,” which include both the never treated as well as those households/individuals that, for a particular point in time, have not been treated yet (though they eventually became treated). The results do not significantly alter when the “not yet treated” is used as the control group (see tables S1.3 and S1.4). In this analysis, we examine how the welfare effect of exposure to 3G treatment may also vary depending on the length of exposure: the contemporaneous effect of 3G coverage, one or more years of coverage (≥1 year), two or more years (≥2 years), and three or more years (≥3 years). It is important to note that when using the C&S approach, exposure is based on whether households were treated at the time of the relevant wave, rather than the time they assumptions do not seem to hold in our case, and there could be potential spillover effects that could mask the treatment of 3G coverage in the spatial discontinuity design. Further discussion is provided in supplementary online appendix S2. 20 More detailed results are also presented in tables S1.1 and S1.2 in the supplementary online appendix. Table 4. Robustness Checks for DID Results for Household Outcomes, Alternative Specifications 248 Dep. Variable: Consumption Food consumption Non-food consumption Specification: Baseline Region x Abadie C&S Baseline Region x Abadie C&S Baseline Region x Abadie C&S Wave FE (2005) Wave FE (2005) Wave FE (2005) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) A. Exposure effect 3G coverage 0.07*** 0.05* 0.05** 0.09** 0.06** 0.03 0.05 0.08 0.09*** 0.11*** 0.06* 0.08 (0.02) (0.03) (0.03) (0.05) (0.03) (0.03) (0.03) (0.05) (0.03) (0.04) (0.03) (0.06) B. Effect by time of 3G exposure <1 year 0.00 −0.01 −0.01 — −0.02 −0.04 −0.03 — 0.06 0.06 0.04 — (0.03) (0.03) (0.03) (0.03) (0.04) (0.03) (0.05) (0.05) (0.05) 1–2 years 0.11*** 0.10*** 0.09*** 0.11** 0.10*** 0.08** 0.09*** 0.11** 0.10** 0.16*** 0.08* 0.08 (≥1 year for C&S) (0.03) (0.03) (0.03) (0.05) (0.03) (0.04) (0.03) (0.05) (0.04) (0.05) (0.04) (0.07) 2–3 years 0.07** 0.04 0.04 −0.02 0.07* 0.04 0.05 −0.00 0.09** 0.09* 0.06 −0.01 (≥2 years for C&S) (0.03) (0.04) (0.04) (0.07) (0.04) (0.04) (0.04) (0.07) (0.04) (0.05) (0.05) (0.09) >3 years 0.10*** 0.09** 0.06* 0.06 0.10*** 0.07 0.07* 0.02 0.11*** 0.18*** 0.06 0.10 (≥3 years for C&S) (0.03) (0.04) (0.03) (0.06) (0.03) (0.04) (0.04) (0.05) (0.04) (0.05) (0.05) (0.07) Dep. variable: Basic need poor Extreme poor ($1.9 PPP) Poor ($3.2 PPP) Specification: Baseline Region x Abadie C&S Baseline Region x Abadie C&S Baseline Region x Abadie C&S Wave FE (2005) Wave FE (2005) Wave FE (2005) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) A. Exposure effect 3G coverage −0.05*** −0.05*** −0.04** −0.06*** −0.03 −0.04* −0.02 −0.07 −0.02 −0.03 0.00 −0.04 (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.04) (0.02) (0.02) (0.02) (0.04) B. Effect by time of 3G exposure <1 year −0.03 −0.02 −0.02 — −0.01 −0.03 0.00 — 0.01 −0.00 0.03 — (0.02) (0.02) (0.02) (0.02) (0.03) (0.02) (0.03) (0.03) (0.03) 1–2 years −0.06*** −0.06*** −0.05** −0.07** −0.05** −0.06** −0.03 −0.05 −0.03 −0.05** −0.01 −0.00 (≥1 year for C&S) (0.02) (0.02) (0.02) (0.04) (0.02) (0.03) (0.02) (0.03) (0.02) (0.02) (0.02) (0.04) 2–3 years −0.04** −0.05** −0.03 −0.01 −0.02 −0.04 −0.00 −0.03 −0.01 −0.03 0.00 −0.13** (≥2 years for C&S) (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) (0.03) (0.02) (0.03) (0.03) (0.03) (0.04) >3 years −0.07*** −0.06** −0.04* −0.01 −0.02 −0.04 0.01 −0.04 −0.03 −0.07** 0.01 −0.09** (≥3 years for C&S) (0.02) (0.03) (0.02) (0.02) (0.02) (0.03) (0.03) (0.04) (0.02) (0.03) (0.03) (0.03) Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: All regressions include individual fixed effects, dummies for household wave and controls variables. Robust standard errors clustered at the EA level (410 clusters) are shown in parentheses. Abadie (2005) refers to estimations including wave dummies times four variables at the household level: distance to nearest major road, distance to nearest population center with +20,000, distance to nearest border crossing, and distance to nearest headquarters of district of residence. C&S refers to aggregated treatment effect parameters under the conditional parallel trends assumptions following Callaway and Sant’Anna (2020). More specifically, the coefficients in the C&S columns correspond to the single average effects of the treatment defined as exposure to greater than 1 year, 2 years, or 3 years of exposure to 3G coverage. All the estimates from C&S use clustered bootstrapped standard errors at the cluster level and the doubly robust estimators. ***p < 0.01, **p < 0.05, *p < 0.1. Bahia et al. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Table 5. Robustness Checks for DID Results for Individuals Outcomes, Alternative Specifications Dep. variable: Labor-force participation Wage employment Self-employed non-farm Specification: Baseline Region x Abadie C&S Baseline Region x Abadie C&S Baseline Region x Abadie C&S Wave FE (2005) Wave FE (2005) Wave FE (2005) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) A. Exposure effect 3G coverage 0.03** 0.02 0.02 0.08*** 0.03** 0.02* 0.01 0.02 0.02** 0.02* 0.02 0.00 (0.01) (0.01) (0.01) (0.02) (0.01) (0.01) (0.01) (0.02) (0.01) (0.01) (0.01) (0.02) B. Effect by time of 3G exposure <1 year 0.01 0.01 0.01 — 0.02 0.01 0.01 — −0.03* −0.03* −0.03* — (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) The World Bank Economic Review 1–2 years 0.02* 0.02 0.02 0.09*** 0.01 0.01 0.00 0.12*** 0.04*** 0.05*** 0.04*** −0.03 (≥1 year for C&S) (0.01) (0.02) (0.02) (0.03) (0.01) (0.02) (0.01) (0.03) (0.01) (0.02) (0.02) (0.04) 2–3 years 0.03 0.03* 0.02 0.10** 0.05*** 0.05*** 0.03* 0.02 0.03** 0.04** 0.02 −0.02 (≥2 years for C&S) (0.02) (0.02) (0.02) (0.05) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) >3 years 0.08*** 0.03 0.05*** 0.05** 0.04*** 0.03* 0.02 −0.01 0.04** 0.04** 0.02 0.01 (≥3 years for C&S) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.03) Dep. variable: Self-employed farm Other employment Number of employments Specification: Baseline Region x Abadie C&S Baseline Region x Abadie C&S Baseline Region x Abadie C&S Wave FE (2005) Wave FE (2005) Wave FE (2005) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) A. Exposure effect 3G coverage −0.07*** −0.07*** −0.04*** 0.01 0.03** 0.03** 0.01 0.04*** 0.01 0.00 0.00 0.07** (0.01) (0.02) (0.01) (0.03) (0.01) (0.02) (0.01) (0.01) (0.02) (0.02) (0.02) (0.03) B. Effect by time of 3G exposure <1 year −0.09*** −0.10*** −0.06*** — 0.06*** 0.08*** 0.05** — −0.04 −0.04 −0.03 — (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) 1–2 years −0.07*** −0.06*** −0.04** 0.00 0.02 0.02 0.01 0.05*** 0.01 0.02 0.01 0.07 (≥1 year for C&S) (0.02) (0.02) (0.02) (0.03) (0.01) (0.02) (0.02) (0.01) (0.02) (0.03) (0.02) (0.04) 2–3 years −0.04** −0.06*** −0.02 −0.07 −0.03 0.00 −0.02 0.08** 0.02 0.03 0.01 0.10** (≥2 years for C&S) (0.02) (0.02) (0.02) (0.04) (0.02) (0.02) (0.02) (0.03) (0.02) (0.03) (0.02) (0.05) >3 years −0.07*** −0.06*** −0.05** −0.03 0.05*** 0.01 0.05** 0.07*** 0.06** 0.02 0.04 0.04 (≥3 years for C&S) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) (0.03) (0.03) (0.03) Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: All regressions include individual fixed effects, dummies for household wave and control variables. Robust standard errors clustered at the EA level (410 clusters) are shown in parentheses. Abadie (2005) refers to estimations including wave dummies times four variables at the household level: distance to nearest major road, distance to nearest population center with +20,000, distance to nearest border crossing, and distance to nearest headquarters of district of residence. C&S refers to aggregated treatment effect parameters under the conditional parallel trends assumptions following Callaway and Sant’Anna (2020). More specifically, the coefficients in the C&S columns correspond to the single average effects of the treatment defined as exposure to greater than 1 year, 2 years, or 3 years of exposure to 3G coverage. All the estimates from C&S use clustered bootstrapped standard errors at the cluster level and the doubly robust estimators. ***p < 0.01, **p < 0.05, *p < 0.1. 249 Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 250 Bahia et al. initially received coverage. In terms of the C&S test for the conditional parallel trends assumption, our results suggest little evidence to reject this assumption. The “group-time average treatment effects” row in tables S1.1 and S1.2 report both group-time average treatment effects identified in periods when t ≥ g (i.e. post-treatment periods for each group), as well as pseudo group-time average treatment effects when t < g (i.e. pre-treatment periods for group g). The latter can be used as a pre-test for the parallel trends assumption (as long as we assume that the no-anticipation assumption indeed holds). None of the effects Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 identified when t < g are statistically significant, except for two instances – the effects of ≥2 years of coverage for labor-force participation and other employment.21 Most of the overall treatment effects of 3G coverage on consumption are positive, while these effects turn negative for poverty. Although group-time average treatment effects individually are not statistically significant, the event study (with balanced groups) aggregate effects are statistically significant (<0.05) for total consumption (only after one year of coverage), food consumption (after one year of coverage), and basic needs poverty (after one year of coverage), as well as poverty at the US$3.2 poverty line (after two years of coverage). The effects are not significant for non-food consumption and poverty at the US$1.9 poverty line. In terms of labor outcomes, the results are also consistent with our TWFE DID results. More specifically, the effects based on the C&S approach show statistically significant positive effects on labor-force participation, wage employment (only after one year of coverage), and other employment, as well as the number of employments, while significant negative effects are found for non-farm jobs, again providing evidence that 3G coverage enables individuals to transition out of farm jobs to other types of jobs. As seen in our baseline results (tables 2 and 3), some of the estimated effects are stronger for longer years of treatment exposure – particularly after one or more years of coverage for total consumption, food consumption, basic needs poverty, as well as wage employment. The results of the C&S approach reported above use the never treated group – or the group of house- holds/individuals that did not ever participate in the treatment across all the three waves – as the control group. However, we can also test how our results may alter once the control group includes the not-yet- treated households/individuals – which include the never treated as well as those households/individuals that, for a particular point in time, had not been treated yet (though they eventually became treated). The overall results largely stand robust when the not yet treated group is used as the control group instead of the never treated group (see tables S1.3 and S1.4). 4.3. Heterogeneous Effects In order to understand the mechanisms behind our main finding that 3G coverage increases welfare and brings households out of poverty, we further investigate its impact on labor-market outcomes. Based on results presented above, we conclude that 3G coverage is associated with an increase in wage employ- ment and a reduction in farm employment. However, these effects on labor productivity are unlikely to be homogeneous across populations. In this section we explore heterogeneous effects by individuals’ demographics, i.e. gender, geography, education, and age. Heterogeneous results on labor-market outcomes are displayed in table 6. For some outcomes we observe quite different results by gender. Gaining 3G coverage is associated with a 6 percentage points increase in labor-force participation for men, while it has zero impact on women’s participation. The same pattern is observed for wage employment. 3G coverage is predicted to increase non-farm self-employment for males and females by 2 to 3 percentage points; however, the effect is only statistically significant for 21 Along with the C&S test for the conditional parallel trend assumption, we also conducted the conventional test for the common trend assumption by evaluating whether changes in the outcome of interest between Waves 1 and 2 are predicted by getting 3G coverage between Waves 2 and 3, as suggested by Angrist and Pischke (2009) and following Bahia et al. (2020). In this test, we find evidence that future treatment (3G coverage at t + 1) significantly impacts consumption at t, although no such effect is observed for poverty and labor outcomes. This casts some doubt on our results specifically on consumption in the TWFE framework, but positive effects observed on consumption in the C&S framework suggest that these effects are not an artifact of pre-trends. The World Bank Economic Review 251 Table 6. Heterogeneous Results – Individuals Female Locality type Primary educ. Age < 30 Has a cell phone No Yes Urban Rural Less More No Yes No Yes Dep. variable: Labor-force participation Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 3G exposure 0.06*** 0.00 0.05*** −0.00 −0.01 0.06*** 0.01 0.05*** 0.01 0.03** (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.02) (0.03) (0.01) Dep. variable: Wage employment 3G exposure 0.06*** −0.00 0.03** 0.02 0.03** 0.02* −0.00 0.05*** −0.03 0.03*** (0.02) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.01) Dep. variable: Self-employed non-farm 3G exposure 0.03** 0.02 0.02 0.03** 0.01 0.04*** 0.03** 0.02* 0.03 0.02** (0.01) (0.01) (0.02) (0.02) (0.01) (0.01) (0.02) (0.01) (0.02) (0.01) Dep. variable: Self-employed farm 3G exposure −0.07*** −0.07*** −0.04** −0.10*** −0.09*** −0.05*** −0.04** −0.09*** −0.08*** −0.07*** (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.03) (0.01) Dep. variable: Other employment 3G exposure 0.02 0.03** 0.01 0.04** 0.00 0.04*** 0.02 0.03** 0.02 0.03** (0.01) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) (0.01) (0.02) (0.01) Dep. variable: Number of employments 3G exposure 0.04* −0.02 0.02 −0.01 −0.04* 0.05*** 0.01 0.01 −0.05 0.02 (0.02) (0.02) (0.02) (0.03) (0.02) (0.02) (0.02) (0.02) (0.04) (0.02) Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: The table presents results separate regressions by heterogeneous category (every two columns) and by dependent variable. All regressions include individual fixed effects, dummies for waves and controls. Robust standard errors clustered at the EA level (410 clusters) are shown in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. males. The impact on farm employment is the same for men and women: for both genders 3G coverage is predicted to reduce farm self-employment by 7 percentage points. This suggests that household welfare gains are likely driven by improved labor-market outcomes for men. The table also reveals some heterogeneous results on labor-market outcomes by individuals’ locality, education, and age. Labor-force participation for individuals in urban areas or with primary education increases by 5 and 6 percentage points respectively. These are in contrast to participation for individuals in rural areas or with less than primary education, which remains unchanged. 3G coverage is associated with an increase in wage employment for male and younger adults. The impact on non-farm self-employment is even across demographics, but stronger for more educated individuals. All groups experience a statistically significant decrease in farm employment, especially individuals that are younger, less educated, or live in rural areas. Finally, the last two columns of table 6 display estimates by cell phone ownership status. Interestingly, the positive and statistically significant association of having 3G coverage and participation in the labor force and wage employment only holds for individuals whose household reported owning a cell phone. On the other hand, the reduction effect on farm employment is regardless of cell phone ownership. We interpret this overall decrease in farm employment as spillover effects beyond direct access to mobile broadband, likely due to demand side effects on firms and productivity. Table 7 presents results disaggregated to a more granular level that separates the estimates by demo- graphic groups as presented in table 6, broken down by gender. While 3G exposure is associated with an increase of labor-market participation for younger and more educated men, it is associated with a reduction in female labor-market participation in rural areas. The increase in wage jobs for male workers is totally driven by younger individuals. The increase in non-farm self-employment, for both men and 252 Bahia et al. Table 7. Heterogeneous Results by Gender – Individuals Dep. variable: Labor-force participation Wage employment Subgroup: Urban Primary educ. Age < 30 Urban Primary educ. Age < 30 For males Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 3G exposure × no 0.05*** 0.02 0.00 0.06** 0.06** 0.00 (0.02) (0.02) (0.01) (0.02) (0.03) (0.02) 3G exposure × yes 0.06*** 0.09*** 0.11*** 0.06** 0.05*** 0.11*** (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) For females 3G exposure × no −0.05** −0.03 0.00 −0.01 0.01 −0.01 (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) 3G exposure × yes 0.04* 0.03* 0.00 0.01 −0.01 0.00 (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) Dep. variable: Self-employed non-farm Self-employed farm Subgroup: Urban Primary educ. Age < 30 Urban Primary educ. Age < 30 For males 3G exposure × no 0.04* 0.03 0.03 −0.08*** −0.08*** −0.04* (0.02) (0.02) (0.02) (0.03) (0.02) (0.02) 3G exposure × yes 0.03 0.04** 0.04** −0.06*** −0.06*** −0.10*** (0.02) (0.01) (0.02) (0.02) (0.02) (0.02) For females 3G exposure × no 0.03 −0.01 0.01 −0.12*** −0.09*** −0.03 (0.02) (0.02) (0.02) (0.03) (0.02) (0.02) 3G exposure × yes 0.00 0.04*** 0.01 −0.03 −0.05*** −0.10*** (0.02) (0.01) (0.01) (0.02) (0.02) (0.02) Dep. variable: Other employment Number of employments Subgroup: Urban Primary educ. Age < 30 Urban Primary educ. Age < 30 For males 3G exposure × no 0.03 −0.01 0.00 0.04 −0.00 −0.00 (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) 3G exposure × yes 0.01 0.04*** 0.03* 0.04 0.07*** 0.08*** (0.02) (0.01) (0.02) (0.03) (0.03) (0.03) For females 3G exposure × no 0.06** 0.02 0.03* −0.05 −0.08** 0.00 (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) 3G exposure × yes 0.01 0.05*** 0.04** −0.00 0.03 −0.04 (0.02) (0.02) (0.02) (0.03) (0.02) (0.02) Source: National Panel Survey (NPS) Waves 1, 2, and 3 combined with mobile coverage maps provided by operators. Note: In each column the table presents results of separate regressions by category, by gender, and by dependent variable. “Yes” and “No” refer to the subgroup category. All regressions include individual fixed effects, dummies for household wave and controls. Robust standard errors clustered at the EA level (410 clusters) are shown in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. women, is driven by more educated individuals. Finally, we observe a reduction in farm employment across all types of individuals. Bringing these results together, we can conclude that mobile broadband drove significant welfare gains in Tanzania, which is consistent with previous studies. This was due in large part to improved labor outcomes, most notably for younger and more educated males that moved out of self-employed farm work to non-farm work and wage employment. This finding is perhaps partly explained by the fact that this group was more likely to use mobile broadband once they received coverage. While the NPS does not include a question on individuals’ use of mobile broadband, a Global System for Mobile Communications The World Bank Economic Review 253 Association (GSMA) consumer survey carried out in 2018 in Tanzania found that women were 52 percent less likely to use mobile broadband than men, while individuals in rural areas were 48 percent less likely to use mobile broadband than urban residents. Another recent study for West Africa finds that women have a lower probability of adoption of mobile broadband of about 6 percentage points less than the average adoption rate (Granguillhome Ochoa et al. 2022). On the other hand, the results also show that women who received 3G coverage moved out of self- Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 employed farm work. Those women who were skilled (had at least a primary education) were able to transition to non-farm self-employment or other employment (which corresponds mostly to non-farm family work).22 These findings are consistent with the broader literature, which suggests that while women can benefit from digital technologies, they often face greater difficulties to leverage mobile phones due to a mix of social norms, intra-household dynamics, lack of access of productive assets, and being less likely than men to use the internet. This suggests that while the provision of mobile broadband can enable improvements in welfare, it is not always sufficient in itself, and further steps to reduce the gender gap in adoption and digital skills are necessary to ensure that all individuals can equally benefit from technology. 5. Conclusion Over the past decade, investments in digital technology have been brought to the forefront of the global agenda for sustainable development. A lack of digital infrastructure is deemed one of the key impedi- ments to shared growth and prosperity in Africa, where a significant share of people remain unconnected. This paper sheds light on the welfare effects of mobile broadband infrastructure in Tanzania, one of the most populous countries in the Sub-Saharan Africa region. It leverages a data set that is nationally rep- resentative, with detailed data on consumption and living standards, which tracks households over time and matches them with precise 3G coverage data from mobile operators. The results show that mobile broadband coverage has positive and significant impacts on consumption and poverty reduction, a find- ing that is consistent with similar studies in other African countries (e.g. Nigeria and Senegal). This gives further assurance to policy makers about the positive effects that digital technology can deliver. A key mechanism through which mobile broadband coverage translates into welfare gains is through its positive impact on labor outcomes. These positive labor outcomes are seen mostly among men–particularly younger and educated men. While educated women who received 3G coverage moved out of farm-work to other types of employment, they do not benefit to the same extent as men. This shows that while mobile technology can enable individuals to improve their welfare, some socioeconomic and demographic groups still face significant barriers to materialize the potential gains of being connected. This highlights the importance of ensuring that under-served groups, particularly women, but also older individuals, those living in rural areas, or less educated, have the skills and resources to leverage the economic opportunities brought by digital technologies. Data Availability The data underlying this article will be shared on reasonable request to the corresponding author. References Abadie, A. 2005. “Semiparametric Difference-in-Differences Estimators.” Review of Economic Studies 72(1): 1–19. 22 Using information on primary activity in Waves 2 and 3, we identify that about 55 percent of workers in the category “Other employment” responded to be working as “unpaid family helper (non-agric),” 21 percent responded to be a paid employee, and 17 percent responded to be non-farm self-employed without employees. We conclude that other employment mostly means helping family in a non-agriculture job. 254 Bahia et al. Abiona, O., and M. F. Koppensteiner. 2022. “Financial Inclusion, Shocks, and Poverty Evidence from the Expansion of Mobile Money in Tanzania.” Journal of Human Resources 57(2): 435–64. Abreha, K., J. Choi, W. Kassa, H. J. Kim, and M. Kugler. 2021. Mobile Access Expansion and Price Information Diffu- sion: Firm Performance after Ethiopia’s Transition to 3G in 2008. Policy Research Working Paper No. WPS 9752, Washington, DC: World Bank Group. http://documents.worldbank.org/curated/en/230161629725057008/Mobile- Access-Expansion-and-Price-Information-Diffusion-Firm-Performance-after-Ethiopia-s-Transition-to-3G-in-2008. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Aker, J. C., and K. Wilson. 2013. “Can Mobile Money Be Used to Promote Savings? Evidence from Northern Ghana.” WIFT Institute Working Paper No. 2012-003. Available at SSRN: https://ssrn.com/abstract=2217554 or http://dx.doi.org/10.2139/ssrn.2217554. Aker, J. C. 2011. “Dial “A” for Agriculture: A Review of Information and Communication Technologies for Agricul- tural Extension in Developing Countries.” Agricultural Economics 42(6): 631–47. Aker, J. C., and M. Fafchamps. 2015. “Mobile Phone Coverage and Producer Markets: Evidence from West Africa.” World Bank Economic Review 29(2): 262–92. Aker, J. C., and I. M. Mbiti. 2010. “Mobile Phones and Economic Development in Africa.” Journal of Economic Perspectives 24(3): 207–32. Akerman, A., I. Gaarder, and M. Mogstad. 2015. “The Skill Complementarity of Broadband Internet.” Quarterly Journal of Economics 130(4): 1781–824. Alibhai, S., N. Buehren, R. Coleman, M. Goldstein, and F. Strobbe. 2018. “Disruptive Finance : Using Psychometrics to Overcome Collateral Constraints in Ethiopia.” Policy Research Working Paper 9386, World Bank, Washington, DC. Alozie, N. O., and P. Akpan-Obong. 2017. “The Digital Gender Divide: Confronting Obstacles to Women’s Develop- ment in Africa.” Development Policy Review 35(2): 137–60. Angrist, J., and J.-S. Pischke. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion (1 ed.). Princeton University Press. Atasoy, H. 2013. “The Effects of Broadband Internet Expansion on Labor Market Outcomes.” ILR Review 66(2): 315–45. Bahia, K., P. Castells, G. Cruz, T. Masaki, X. P. Pedros, T. Pfutze, C. Rodriguez-Castelan, and H. Winkler 2020. “The Welfare Effects of Mobile Broadband Internet: Evidence from Nigeria.” Policy Research Working Paper No. 9230. World Bank, Washington, DC: © World Bank. https://openknowledge.worldbank.org/handle/10986/33712. License: CC BY 3.0 IGO. Baker, A. C., D. F. Larcker, and C. C. Y. Wang. 2022. “How Much Should We Trust Staggered Difference-in-Differences Estimates?” Journal of Financial Economics 144(2): 370–95. Beuermann, D. W., C. McKelvey, and R. Vakis. 2012. “Mobile Phones and Economic Development in Rural Peru.” Journal of Development Studies 48(11): 1617–28. Bimber, B. 2000. “Measuring the Gender Gap on the Internet.” Social Science Quarterly 81(3): 868–76. Blumenstock, J. E., N. Eagle, and M. Fafchamps. 2016. “Airtime Transfers and Mobile Communications: Evidence in the Aftermath of Natural Disasters.” Journal of Development Economics 120: 157–81. Blumenstock, J., N. Keleher, A. Rezaee, and E. Troland. 2020. “The Impact of Mobile Phones: Ex- perimental Evidence from the Random Assignment of New Cell Towers”. Unpublished. Available at https://www.jblumenstock.com/files/papers/jblumenstock_2020_ccn.pdf. Broadband Commission. 2013. Doubling Digital Opportunities: Enhancing the Inclusion of Women and Girls in the Information Society. Geneva: UNDP/ITU. Available at http://www. broadbandcommission.org/documents/working-groups/bb-doubling-digital-2013.pdf. Callaway, B., and P. H. Sant’Anna. 2020. “Difference-in-Differences with Multiple Time Periods.” Journal of Econo- metrics 255(2): 200–30. Calonico, S., M. D. Cattaneo, and R. Titiunik 2014. “Robust Nonparametric Confidence Intervals for Regression- Discontinuity Designs.” Econometrica 82(6): 2295–326. Chen, X., F. Fatima, N. Halewood, S. Malgioglio, and N. Yoshida. 2020. Has the Digital Revolution Come to Tanzania? Household Survey Findings on Mobile Phone Adoption. World Bank. Chun, N., and H. Tang. 2018. “Do Information and Communication Technologies Empower Female Workers? Firm- Level Evidence from Viet Nam.” ADB Economics Working Paper Series 545, Asian Development Bank. De los Rios, C. 2010. Welfare Impact of Internet Use on Peruvian Households. Lima: Instituto de Estudios Peruanos. Fatehkia, M., R. Kashyap, and I. Weber. 2018. “Using Facebook Ad Data to Track the Global Digital Gender Gap.” World Development 107: 189–209. The World Bank Economic Review 255 Fernandes, A. M., A. Mattoo, H. Nguyen, and M. Schiffbauer 2019. “The Internet and Chinese Exports in the Pre-Ali Baba Era.” Journal of Development Economics 138: 57–76. Gonzalez, R. M. 2021. “Cell Phone Access and Election Fraud: Evidence from a Spatial Regression Discontinuity Design in Afghanistan.” American Economic Journal: Applied Economics 13(2): 1–51. Granguillhome Ochoa, R., S. Lach, T. Masaki, and C. Rodríguez-Castelán. 2022. “Mobile Internet Adoption in West Africa.” Technology in Society 68: 101845. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Grazzi, M., and S. Vergara. 2014. “Internet in Latin America: Who Uses It?... and for what?” Economics of Innovation and New Technology 23(4): 327–52. GSMA. 2021. The State of Mobile Internet Connectivity 2021. GSM Association, London. Available at https://www. gsma.com/r/wp-content/uploads/2021/09/The-State-of-Mobile-Internet-Connectivity-Report-2021.pdf. Guerrero, M. Y., and P. Ritter. 2014. “The Effect of Internet and Cell Phones on Employment and Agricultural Pro- duction in Rural Villages in Peru.” Tesis de Pregrado en Economía. Universidad de Piura. Facultad de Ciencias Económicas y Empresariales, Programa Académico de Economía y Finanzas, Piura, Perú. Hasbi, M., and A. Dubus. 2020. “Determinants of Mobile Broadband Use in Developing Economies: Evidence from Sub-Saharan Africa.” Telecommunications Policy 44(5): 101944. Hjort, J., and J. Poulsen. 2019, March. “The Arrival of Fast Internet and Employment in Africa.” American Economic Review 109 (3): 1032–79. Hjort, J., and L. Tian. 2021. “The Economic Impact of Internet Connectivity in Developing Countries (November 16, 2021).” INSEAD Working Paper No. 2021/68/EPS. Available at SSRN: https://ssrn.com/abstract=3964618 or http://dx.doi.org/10.2139/ssrn.3964618. Hübler, M., and R. Hartje. 2016. “Are Smartphones Smart for Economic Development?” Economics Letters 141: 130–33. ITU. 2020. Measuring Digital Development: Facts and Figures 2019. International Telecommunications Union, Geneva, https://www.itu.int/en/ITU-D/Statistics/Documents/facts/FactsFigures2020pdf. Jack, W., and T. Suri. 2014. “Risk Sharing and Transactions Costs: Evidence from Kenya’s Mobile Money Revolution.” American Economic Review 104(1): 183–223. Jensen, R. 2007. “The Digital Provide: Information (Technology), Market Performance, and Welfare in the South Indian Fisheries Sector.” Quarterly Journal of Economics 122(3): 879–924. Kaila, H., and F. Tarp. 2019. “Can the Internet Improve Agricultural Production? Evidence from Viet Nam.” Agricul- tural Economics 50(6): 675–91. Keele, L. J., and R. Titiunik. 2015. “Geographic Boundaries as Regression Discontinuities.” Political Analysis 23(1): 127–55. Klonner, S., and P. Nolen. “Cell Phones and Rural Labor Markets: Evidence from South Africa.” Working Paper 56, Verein für Socialpolitik, Ausschuss für Entwicklungsländer. ———. 2010. “Cell Phones and Rural Labor Markets: Evidence from South Africa.” Proceedings of the German De- velopment Economics Conference, Hannover 2010 56, Research Committee Development Economics. Hannover, 14 September 2010. Ky, S., C. Rugemintwari, and A. Sauviat. 2018. “Does Mobile Money Affect Saving Behaviour? Evidence from a Developing Country.” Journal of African Economies 27(3): 285–20. Labonne, J., and R. S. Chase. 2009. “The Power of Information: The Impact of Mobile Phones on Farmers’ Welfare in the Philippines.” Policy Research Working Paper Series 4996, The World Bank. https://elibrary.worldbank.org/doi/abs/10.1596/1813-9450-4996. Marandino, J., and P. V. Wunnava. 2014. “The Effect of Access to Information and Communication Technology on Household Labor Income: Evidence from One Laptop per Child in Uruguay.” IZA Discussion Papers 8415, Institute of Labor Economics (IZA). Masaki, T., R. O. Granguillhome, and C. Rodriguez-Castelan. 2020. “Broadband Internet and House- hold Welfare in Senegal.” Policy Research Working Paper No 9386, World Bank, Washington, DC. https://openknowledge.worldbank.org/handle/10986/34472. License: CC BY 3.0 IGO. McMillan, M. S., and D. Rodrik. 2011. “Globalization, Structural Change and Productivity Growth.” Working Paper Series National Bureau of Economic Research. http://www.nber.org/papers/w17143. Muto, M., and T. Yamano. 2009. “The Impact of Mobile Phone Coverage Expansion on Market Participation: Panel Data Evidence from Uganda.” World Development 37(12): 1887–96. 256 Bahia et al. Paunov, C., and V. Rollo. 2015, 04. “Overcoming Obstacles: The Internet’s Contribution to Firm Development.” World Bank Economic Review 29 (1): 192–204. Restuccia, D., D. T. Yang, and X. Zhu. 2008. “Agriculture and Aggregate Productivity: A Quantitative Cross-Country Analysis.” Journal of Monetary Economics 55(2): 234–50. Rice, P. L., and A. G. Longley. 1968. “Prediction of Tropospheric Radio Transmission Loss over Irregular Terrain: A Computer Method–1968” Institute for telecommunication sciences, Boulder - Col. Available at http://www. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 visuallmr.com/documentation/pathlossmodels/ntis.longleyrice.676874.pdf. Sant’Anna, P. H., and J. Zhao. 2020. “Doubly Robust Difference-in-Differences Estimators.” Journal of Econometrics 219(1): 101–22. Schultz, T. W. 1956. “Reflections on Agricultural Production, Output and Supply.” Journal of Farm Economics 38(3): 748–62. Suri, T., and W. Jack. 2016. “The Long-Run Poverty and Gender Impacts of Mobile Money.” Science 354(6317): 1288–92. Tadesse, G., and G. Bahiigwa. 2015. “Mobile Phones and Farmers’ Marketing Decisions in Ethiopia.” World Devel- opment 68: 296–307. Viollaz, M., and H. J. Winkler. 2020. “Does the Internet Reduce Gender Gaps?: The Case of Jordan.” World Bank Policy Research Working Paper 9183s, World Bank. World Bank. 2020. Tanzania Mainland Poverty Assessment: Part 1 - Path to Poverty Reduction and Pro-Poor Growth. Washington, DC: World Bank. https://openknowledge.worldbank.org/handle/10986/33542. License: CC BY 3.0 IGO. Zhao, J. 2020. “Internet Usage and Rural Self-Employment in China.” Asian Perspective 44(1): 77–101. Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 257 The World Bank Economic Review Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Bahia et al. 258 Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 259 The World Bank Economic Review Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Bahia et al. 260 Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 261 The World Bank Economic Review Downloaded from https://academic.oup.com/wber/article/37/2/235/7028405 by Joint Bank-Fund library user on 04 September 2023 Bahia et al. 262