Notes

mobile money and of policies that foster interconnectivity between banks, reduce transaction costs, and increase the supply of mobile money agents in rural areas. Subsequent research can shed light on the implications of mobile broadband as a cornerstone for private and public platforms, including mobile money and e-government applications, which so far tend to be based mostly around 2g technology in the region.

NOTES

1. This chapter is based on the results and analysis from the following background papers:

Atiyas and Doğanoğlu (2020); Enamorado et al. (2020); Masaki, granguillhome ochoa, and rodríguez-Castelán (2020); and rodríguez-Castelán et al. (2021a, 2021b). The policy discussion incorporates inputs provided by izak Atiyas. The chapter benefited from comments from Mark Dutz. 2. The literature identifies socioeconomic characteristics such as education, gender, income, age, digital literacy, or household size as drivers related to the adoption and use of the internet (Coelho, Silva, and Ehrl 2019; goldfarb and Prince 2008; grazzi and vergara 2014;

Kongaut and Bohlin 2016; Martínez-Domínguez and Mora-rivera 2020; and nishijima, ivanauskas, and Mori Sarti 2017). Existing studies from SSA reach similar findings (Penard et al. 2012, 2015; Birba and Diagne 2012; gillwald, Milek, and Stork 2010). Most of this research, however, concentrates on fixed rather than mobile broadband. This distinction is important in the context of African countries, where most people access the internet through mobile phones rather than through fixed broadband (iTU 2019). Country-level characteristics can also play a role on internet coverage and adoption. Electricity access appears to be a key driver of internet adoption in poor countries (Armey and Hosman 2016). And increased competition among digital service providers can reduce prices and allow new entrants to adopt internet services (Decoster et al. 2019; rodríguez-Castelán et al. 2019). indeed, competition appears to have had a positive and significant effect on cell phone coverage systems in SSA (Buys et al. 2009). other research considers the role of prices and variable and fixed costs as factors behind internet adoption. Studies from the

United States show that consumers tend to place a premium on unlimited data and unmetered pricing, and a lower value on bandwidth (liu, Prince, and Wallsten 2018; varian 2002). others focus on the surplus generated from residential broadband usage and the degree of capture of internet service providers (nevo, Turner, and Williams 2016). A cross-country oECD study finds that the demand for internet services is price-inelastic— where people continue demanding it even as its price increases—while its income elasticity is greater than one, suggesting that relative to changes in income, the internet is more of a luxury good (goel et al. 2006). 3. This section draws results from the background paper “Mobile internet Adoption in West

Africa” (rodríguez-Castelán et al. 2021). 4. The number of active mobile broadband subscriptions per 100 inhabitants in Africa in 2019 was 34, compared with 0.4 for fixed broadband subscriptions (iTU 2019). 5. These variables are defined at the household level. Cell phone age corresponds to the question “for how long have you owned the following item?” 6. All variables pertain to characteristics of individuals above 15 years of age. Each tabulation is carried out at the individual level adjusting for household weights. Connectivity should thus be interpreted as the percentage of individuals who access the internet through their mobile phone. 7. The results presented in this paragraph and the next one regarding the difference of means between connected and not-connected individuals compared with sociodemographic characteristics are all significant at the 1 percent level, except where marked otherwise. 8. The analysis models adoption as a two-stage process, implementing a Heckman-corrected probit model to account for selection bias. it is assumed that there is a fundamental relationship between households that are covered by 3g (first stage) and those that decide to adopt mobile internet (second stage). in addition to a welfare metric (household expenditures), the analysis incorporates a human capital dimension (education, sector of employment, language), access to different internet modalities and household assets (computer,

television, and tablet ownership), and a price proxy for mobile internet. The analysis also controls for selection bias by truncating the sample conditional to those households with 3g coverage. Coverage data were obtained directly from Collins Bartholomew, a digital mapping provider, and the three major mobile operators in Senegal: Expresso, orange (or its local subsidiary, Sonatel), and Tigo. Mobile internet price is obtained by calculating the median expenditure of prepaid mobile phone cards and airtime/data transfers among mobile internet users in each country’s geographic area for which the survey is representative. This value is then computed as a share of total consumption at the same geographic level to adjust for cost of living. This value is then imputed to each individual observed in the microdata.  9. interpretations of the model are based on average marginal effects (AME). The results presented here differ slightly from previous estimates reported by Zeufack et al. (2020) as the model now controls for 3g coverage by implementing a Heckman-corrected probit model (though the parsimonious model yields similar results) and incorporates the price proxy derived from individual-level expenditures on prepaid mobile phone cards and airtime/data transfers. 10. This change is equivalent to a one unit increase in the standard deviation of the log of per capita expenditures. Currency conversion based on an exchange rate of US$1.00 = CfAf 555.45 in 2018. international financial Statistics (database), international Monetary fund,

Washington, DC, https://data.imf.org/?sk=4c514d48-b6ba-49ed-8ab9-52b0c1a0179b. 11. This change is equivalent to a 1-unit increase in the standard deviation of the mobile data price proxy derived from individual-level expenditures on prepaid mobile phone cards and airtime/data transfers. 12. This section draws results from the background paper “internet Adoption in Senegal,”

Atiyas and Doğanoğlu (2020). 13. This section draws results from the background paper “Broadband internet and Household

Welfare in Senegal” (Masaki, granguillhome ochoa, and rodríguez-Castelán 2020). 14. Data on terrestrial backbone networks were obtained from http://www.africaband widthmaps.com. internet traffic in countries runs first through national “backbone” or fiber-optic networks, which are then connected to end users through last-mile infrastructure such as fiber cables, copper cables, wireless transmission, or cell phone towers among others. See Hjort and Poulsen (2019) for a detailed discussion. 15. results are robust after controlling for household demographics and other spatial characteristics (for example, region-fixed effects, road density, nighttime lights, or elevation) as well as for access to complementary digital infrastructures, such as 2g coverage or fixed broadband internet. results are also robust to an iv approach, using distance to 3g coverage in neighboring areas as an instrument. 16. Taking advantage of longitudinal data in nigeria, Bahia et al. (2020) prove that the parallel-trends hypothesis holds for their difference-in-difference analysis. in contrast, the results for Senegal likely overestimate the real effect of internet access, because the digital infrastructure is almost surely located in more affluent areas with lower poverty rates and higher household consumption. 17. Employment is defined as working-age (15–64 years) individuals who worked at least one hour in the past seven days. Salaried/wage employment includes those employees who work in a place that is not their own farm or in a business not run by their own household. 18. Socioeconomic characteristics—such as age, education, and income—appear to influence the adoption of DTs for different uses. A study using data from Cameroon finds that young people tend to use the internet for leisure purposes while users who are older, more educated, and computer savvy are more likely to use it to search for information (Penard et al. 2015). Analogously, a study of rural users in Mexico finds that young people are more likely to use the internet for entertainment, while working-age people go online for information, communication, and e-commerce activities (Martínez-Domínguez and Mora-rivera 2020). The study also finds that higher education increases the types of uses. in the United

States, goldfarb and Prince (2008) show that low-income people are more likely to do time-consuming, inexpensive activities online. However, the authors find that more hours using the internet are related to an increase in the use of more “valuable” activities, such as for e-government, researching purchases, telemedicine, and news. 19. The literature suggests that the internet may have heterogeneous effects by location, benefiting some geographic areas over others, as its effect in reducing frictions and costs varies