WPS8278 Policy Research Working Paper 8278 Mobile Infrastructure and Rural Business Enterprises Evidence from Sim Registration Mandate in Niger Francis Annan Aly Sanoh Poverty and Equity Global Practice Group December 2017 Policy Research Working Paper 8278 Abstract Business enterprises and non-agricultural startups in rural prepaid SIMs and led to a remarkable decline in mobile economies play crucial roles in ending the vicious cycle use activity. The results suggest that there is economically of poverty. The propagation of business enterprises are, substantial effect of mobile infrastructure on propagation however, subject to a high degree of institutional frictions of business enterprises. Moving a household with mobile and vacuums e.g., information; but mobile infrastructure use activity to a no mobile use activity environment due which has the externality of flowing information can help to SIM deactivation results in roughly 33.1 percent points overcome most of these vacuums through reduced fixed decline in the likelihood of engaging in non-agricultural costs, lower cost of information or search, and outreach to business enterprises. Most of this effect come from three a broader customer base. This paper studies the effects of major sources: households’ ownership of a business service mobile infrastructure (“mobile use activity”) on propagation or center; ownership of small income generating activities; of rural business enterprises in Niger. Instrumental variable and involvement in the transformation of agricultural prod- estimates exploit the exogenous introduction of the 2013 ucts. There is suggestive evidence that the reform’s impact is national mandatory SIM registration reform which pro- disproportionately larger for women. With the expansion of vides a quasi-experimental set-up for mobile use and activity. mandatory SIM registration reforms in developing countries, The mandate deactivated about one-third of all existing the findings have extended implications in these contexts. This paper is a product of the Poverty and Equity Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at fannan@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team MOBILE INFRASTRUCTURE & RURAL BUSINESS ENTERPRISES: EVIDENCE FROM SIM REGISTRATION MANDATE IN NIGER Annan, Francis & Sanoh, Aly1 Poverty & Equity Global Practice World Bank Group Key words: mobile infrastructure; regulatory change; instrumental variable; business enterprises; poverty JEL classification: H54; O12; O18; I38; I32                                                         1 Annan: fa2316@columbia.edu & fannan@worldbank.org. Sanoh: asanoh@worldbank.org. For helpful discussions and suggestions, we thank Andrew Dabalen; discussant Tawanda Chingozha and seminar participants at the “Poverty and Inequality” session of 2017 CSAE Conference in Oxford. All remaining errors are ours. “These requirements present real ‘access’ difficulties for the most economically vulnerable individuals.”  Regulatory Affairs’ Head for Africa at Millicom International Cellular speaking about Africa’s SIM registration revolution. Donovan and Martin (2014). 1 INTRODUCTION Examining the impacts of infrastructure e.g., road network, mobile telephony, freight railways, and other public and private services on economic and development outcomes in general can be challenging (Hansen et al. 2013; Tompsett 2013) for many reasons. First, the location and roll out of many infrastructure services are nonrandom and largely driven by the implied costs, demand side aspects and in some cases potential gains to the provider. For instance, locations with higher “potential” demand for mobile infrastructure are more likely to receive first hand network coverage in the case of mobile infrastructure. Similarly, central planners are more likely to provide public services e.g., dams, to denser areas first conditional on the suitability of the underlying engineering processes. Second, once an infrastructure service is provided, the adjacent and nearby areas are more likely to receive the next service since expansion costs can be cheaper (Klonner and Nolen 2010). In rural economies, households’ engagement in business enterprises play crucial role in their ability to deal with unexpected shocks and poverty outcomes. For instance about 16% of the time, Nigerien households rely on non-agricultural business enterprises and incomes as a coping mechanism in the wake of shocks and extreme events that tend to shape households poverty status (World Bank 2016; Annan and Sanoh 2016). The propagation of business enterprises are, however, subject to a high degree of institutional frictions and vacuums2 e.g., information (Klonner and Nolen 2010) but mobile phones/infrastructure which has the externality of flowing information can help overcome most of these vacuums through reduced fixed costs, lower cost of information or search, and outreach to a broader customer base.                                                         2 The broader notion of frictions/imperfections in markets across various contexts and their consequences has received significant attention in the economics literature (see e.g., Akerlof 1970; Rothschild and Stiglitz 1976; Giné and Mazer 2016; Giné et al. 2008 and references therein); interested readers are directed to this rich and growing literature. ‐ 2 ‐    With these in mind, this paper takes advantage of the quasi-experimental variation created by the plausibly exogenous introduction of the 2013 mandatory SIM3 registration reform in Niger. This variation allows us to quantify mobile use and activity’s causal impact on the propagation of business enterprises and poverty among households. The SIM mandate deactivated about one- third of all existing prepaid SIMs and made it harder to use newly purchased mobile SIM cards, which led to a remarkable decline in mobile use activity over the period. Most mobile phones were rendered defunct and lesser incentive was created for households to acquire new phones due to the mandate. Mandatory SIM registration reforms are common in sub-Saharan Africa. But despite how widespread these policies are, there is scant bottom-up evidence about their actual effects on households. Our paper makes progress in first evaluating SIM mandates and second in understanding the induced-impact of access to mobile infrastructure on household behavior and welfare. Related Literature This paper contributes to several strands of inter-related literatures. First, a related body of research is concerned with investigating the economic impacts of information disseminated through mobile information technologies in poor and low-income economies. This literature examines impacts on economic growth (e.g., Waverman et al. 2005; Sridhar et al. 2004), agricultural markets (Aker 2010; Goyal 2010; Fafchamps and Minten 2012), fishing markets, (e.g., Jensen 2007), electoral participation and fraud (e.g., Aker et al. 2016; and Gonzalez 2015), migration and migrant remittances (e.g., Batista and Narciso 2016) and household savings and labor markets (e.g., Klonner and Nolen 2010; Aker et al. 2011). Second, there is a burgeoning literature investigating mobile innovation and infrastructure in development. This literature has appealed to low transaction costs of mobile financial services to document the impact of innovative services on mobile platforms on development. Fueled by the remarkable increase in mobile technology and adoption, innovative products/services have emerged including Mobile-Money and Mobile-Insurance. In theory, low transaction costs can foster development through various channels including the incentives to increase economic activity, and informal risk sharing and insurance mechanisms. Mbiti and Weil (2013), Jack and                                                         3 SIM: Subscriber Identity Module  ‐ 3 ‐    Suri (2014), and Suri and Jack (2016) document the revolution of Mobile-Money, M-PESA in Kenya. Jack and Suri (2014) estimate the effect of Mobile-Money on households’ consumption smoothing via risk sharing: based on variations in Mobile-Money agent networks and difference- in-differences strategy, the authors document significant effects on the ability of households to spread risk. More recently, Blumenstock et al. (2016) examined risk sharing on mobile phones: the authors provide empirical evidence that people used the mobile phone network to transfer airtime to those affected by an unexpected shock, 2008 Lake Kivu earthquake in Rwanda. Third, is the broader literature linking access to public infrastructure services and development. Studies have linked expansions in road infrastructure and connectivity on agricultural and labor markets (e.g., Dinkelman 2011; Banerjee, Duflo and Qian 2012), dams, local economy and agricultural production (e.g., Duflo and Pande 2007), and transport or engineering infrastructure and growth (e.g., Tompsett 2013; Banerjee, Duflo and Qian 2012; Ansar et al. 2016). Our paper can be placed at the intersection of these literatures. The combination of rich panel data sets and quasi-experimental variation from a SIM mandate enables us to quantify the impact of mobile infrastructure on two development outcomes which are important in thinking about the vicious cycle of poverty and how to deal with poverty. We ask whether the 2013 SIM registration mandate in Niger impacted mobile use and activity, and then how the exogenous variation in mobile use activity created by such mandate can be used to credibly identify mobile infrastructure’s impact on poverty and propagation of local business enterprises. Identification of effects come from the SIM mandate, so we are able to evaluate the (un)intended impacts the SIM mandate simultaneously. We find evidence of the following. First, our analysis indicate that the SIM mandate led to a significant distributional shift in mobile use activity over the period, indicating a substantial drop in mobile activity. Second, there is economically large effect of mobile infrastructure on the propagation of rural business enterprises. The drop in mobile use activity induced by the mandate resulted in about 33.1% points decline in the likelihood of households’ engaging in non-agricultural business enterprises. We explore the mechanisms underlying our results and find that most of the effect come from three major sources: households’ ownership of a business service or center; ownership of small income generating activities; and involvement in the transformation of agricultural products. In addition, there is suggestive evidence of gender differentiated impacts of the reform. The decline in the likelihood ‐ 4 ‐    of engaging in non-agricultural enterprises is larger (but statistically insignificant) for women compared to their male counterpart households. The results have extended impacts on poverty. Mandatory SIM Registration Revolution Figure 1: Distribution of SIM Registration Mandates Notes: Figure is due Donovan and Martin 2014. (see web–link-here); which is in turn based on data from multiple sources: research conducted by MobileActive (an advocacy group), Steve Song of Village Telco; Jentzsch 2012; and the authors’ own data collection. Where: red and green correspond to countries where SIM registration is required and not required, respectively, as of February 2014. The list of countries where SIM registration policy has not been introduced as of February 2014 include: Cape Verde, Lesotho, Mauritania, Namibia, Somalia and Swaziland. ‐ 5 ‐    AFRICA’S CASE: Mobile infrastructure has expanded rapidly across sub-Saharan Africa and other emerging countries. Africa’s mobile phone penetration rate is estimated to be around 67% of the total population (The Guardian 2015). The widespread penetration of mobile phones across the region has in turn been accompanied by the imposition of various communication regulations and practices such as data retention (Whitley and Hosein 2005); use of call and mobile device location by law enforcement agencies (Donovan and Martin 2014). Because mobile platforms have become a conduit for crime and major fraud issues, many governments have responded through the imposition of SIM registration mandates. In the past, such registration arrangements were not available. Figure 1 illustrates the spatial surge in SIM registration mandates across the region. None of the countries in sub-Saharan Africa had a SIM registration up until 2006. In such era, people or households across the region were able to buy a SIM card and anonymously use it. The first SIM registration mandate in sub-Saharan Africa was introduced in 2006; others followed thereafter. As displayed in Figure 1, SIM registration requirements are now in force in the majority of countries. Registration require users to provide details about their personal identification in order to purchase and use a SIM card. Existing SIMs are also required to be registered, similarly. In principle, these SIM registration reforms are justified to combat crime or fraud and to support anti-Money Laundering and the Combating of the Financing of Terrorism AML/CFT (Jentzsch 2012). But in practice, there is limited/no evidence regarding the efficacy of SIM registration exercises for security or commerce (Martin et al. 2009; Donovan and Martin 2014). NIGER’S CASE: In November 2013, Niger deactivated a third of its mobile phone connections as a result of lack of registration (AFP 2013; Donovan and Martin 2014). As highlighted in Figure A.1 of the Appendix, other media organizations and blogs gave similar account of the 1/3 cut in existing SIM lines in Niger to stop crime. Stories about implications of the mandatory SIM registration policies on mobile phone use and penetration have been told. For example, there are media reports suggesting that SIM registration exercises might negatively affect the growth in mobile subscriber base. In addition, the registration can lead to the disconnection of millions of existing SIM cards. Negative impacts on MNOs revenues have also been discussed within various circles. Mandatory registration leads to major temporal drop in the number of active ‐ 6 ‐    prepaid SIMs once the SIM registration deadlines passes or kicks in. Studies (e.g., Southwood 2011; Jentzsch 2012) have also shown the likely slowdown in the growth of mobile penetration due to SIM registration measures. Most arguments underlying these implications are tied to the lack of necessary formal identification documents among affected populations for the SIM registration process. A summary of the mandate’s timeline is provided in Figure 2. This paper will investigate the response of mobile use and activity to the 2013 registration in Niger that led to massive disconnection of mobile subscribers, and then the implied linkages with the local economy. Figure 2: Timelines of Policy Mandate Notes: This SIM deactivation exercise came to effect following earlier national registration campaigns. As in other countries, the goal was to curb criminal activity. The mandate is still in force, requiring that new SIMs be registered. These registration requirements uniquely bind in the case of Niger because most of the Nigerien poor lack formal identification proofs, thus limiting the effectiveness of SIM registration campaigns. 2 DATA AND SUMMARIES Our analyses rely on rich and new panel data sets collected in Niger between 2011/12–2014/15. One crucial aspect of the data is that it covers a period before and after the policy mandate, with no overlaps. The panel data come from joint data collection efforts between the World Bank WB and the National Statistical Office NSO of Niger, collected through sample surveys of households, and other business establishments over the period. The data is rich in content, where in each survey year two waves of data collection are carried out spanning three broader areas: individual households, agriculture and livestock holdings, and community level records. ‐ 7 ‐    The summary statistics of all relevant variables in our sample are reported in Table 1. The first two moments and order statistics of each variable are displayed.4 As shown, the data is made up of two sampling years: 2011/2012 and 2014/2015. Considerable variations exist among the variables which we shall exploit for identifying variation. Our main outcomes of interest are binary, denoted “Business Enterprise” and “Poor”. Business Enterprise is based on whether households are engaged in any form of non-agricultural business enterprise within the past 12 months. In our sample, about 22% of households are engaged in some form of business enterprises. Summaries about various aspects/dimensions of business enterprises are displayed in the bottom panel. Poor is an indication of whether a household is classified as either poor or not poor. Poverty classification of households uses Niger’s 2011 National Absolute Poverty Line NAPL of 189233 CFA. The sample suggest overall absolute poverty rate of about 40.2%. Table 1: Data and Summaries, 2011/12-2014/15 OBS (N) Mean SD Min Max Business Enterprise 7,590 0.220 0.414 0 1 Mobile use and activity 2,788 0.223 0.417 0 1 1 [Sex = female] 7,072 0.229 0.458 0 1 1 [Married = Yes] 7,072 0.629 0.483 0 1 Age (years) 7,072 35.30 20.57 18 95 Educational level+ 1,466 3.281 1.752 1 7 1 [Education = notNil] 4,399 0.999 0.037 0 1 Poor++ 6866 0.402 0.459 0 1 Year 7,781 - - 2011 2014 Transform Agr. Product 7,590 0.160 0.366 0 1 Make clothing 7,590 0.040 0.197 0 1 Own construction 7,590 0.039 0.195 0 1 Any profession for own 7,589 0.039 0.195 0 1 Own business service 7,589 0.047 0.212 0 1 Own commercial transport 7,589 0.024 0.154 0 1 Own restaurant /bar 7,589 0.015 0.123 0 1 Own small activity 7,589 0.160 0.366 0 1 Notes: Table reports summary statistics of sample variables including: household’s ownership of business enterprise; mobile use and activity in the sampling year; gender of respondents and other demographic information. +For education: 1=None, 2=primary, 3=secondary first cycle-general, 4=secondary first cycle- technical & professional, 5=secondary second cycle-general, 6=secondary second cycle-technical & professional, and 7=superior levels. ++While our sample suggests about 40.2% overall absolute poverty rate, rates reported elsewhere can be around 48%.                                                         4 Data is harmonized and matched across two survey years to obtain the panel. While the the main data frame houses many variables’ information; we restrict attention to those relevant to our empirical analysis. ‐ 8 ‐    The mobile infrastructure variable is denoted in Table 1 as “Mobile use and activity”. This derives from whether households utilized mobile phone within the past 30 days. This is regardless of personal ownership and hence allows for the possibility of sharing mobile phones, common in low-income environments. The overall mobile use activity is 22.3% with a standard deviation about 0.42. Table 1 also reports the gender of respondents/households and other demographic information. About 22.9% of respondents are females; 62.9% are married; and more than 99% had some form of education. The average age of respondents is approximately 35 years with a maximum of 95 years. Education is coded in 7 Likert levels, in increasing order. Average educational level is around the 3.2 Likert level. Empirical analysis to follow combine these variables along with plausible assumptions and the exogenous variation induced by the 2013 SIM mandate. 3 EMPIRICAL APPROACH In an ideal experiment designed to quantify the impact of mobile infrastructure: mobile use activity on economic outcomes: business enterprises and poverty, we would observe outcomes for two identically distributed households, then randomly introduce a mobile usage activity to the environment of one and compare outcomes across these “treatment” and “control” conditions. The 2013 national SIM mandate helps to mimic such conditions. The policy mandate deactivated some existing lines and made it harder to use newly purchased mobile SIM cards, which lead to a remarkable decline in mobile use activity over the period. In principle, the unaffected households after the policy can be considered a “control” for the same households just after the policy “treatment.” Here, the impact of the policy reform is identified as any response in household outcomes that occur due to the mandate via mobile use and activity. Our goal is exploit this policy and mobile use conditions to quantify the direction and extent of such responses. Constructing “Policy Instrument” and Model For household i in cluster c in year t, the simplest panel model of mobile use and activity that we estimate is: MOBILE Figure 3a displays the distribution of households’ mobile use and activity after household-level fixed effects are removed from the data (i.e., distribution of ). This is shown for the period before and after the 2013 SIM mandate. The figure demonstrates that the SIM mandate ‐ 9 ‐    had a large effect on mobile use and activity. Mobile use likelihoods dropped remarkably following the policy, as demonstrated in the distributional shifts. Exploiting the quasi-random nature of the policy, we construct an instrument for use and activity of mobile telephony in Niger as: PolicyInstrument 2013 MOBILE ∈ where 1[A] is an indicator function that is equal to one if the expression A is true, and equal to zero otherwise. The instrument essentially interacts the policy change with average mobile use and activity across all households in a cluster c. A given cluster c houses several households such that an individual household’s endogenous mobile use decision will not significantly matter (in the spirit of Duflo and Pande 2007; Klonner and Nolen 2010). A variation is created across clusters, which we combine with the policy change variable to construct the instrument. We exploit the induced policy variation to estimate a panel instrumental variable regression that links business enterprise and poverty outcomes to mobile use and activity MOBILE : βMOBILE where the PolicyInstrument is used to instrument for MOBILE controlling for both cluster and year fixed effects and respectively. All relevant control variables are housed in the vector : respondents level of education, gender, marital status, and age. First, it is crucial that the policy instrument be related to mobile use and activity (i.e., a relevance requirement). Even without any rigorous analysis, it is straightforward to discern relevance from Figure 3a. The figure demonstrates that access and use of mobile phones dropped remarkably in the sample years as result of the National reform.5 This is because the SIM reform made it harder to use new SIM cards and deactivated some existing SIM lines. Such observed changes in mobile use activity support from the data is consistent with discussions in SIM Registration section. Next, the identifying assumption underlying the research design is that the exact timing of the announcement of the mandatory SIM reform was not anticipated before it came to play. If households anticipated the announcement, they might have started to alter their SIM and mobile                                                         5 Figure 3b shows the growth/changes in average mobile use likelihoods across regions following the SIM deactivations, which suggests considerable variations that we shall exploit in our analysis to estimate the implied effects. ‐ 10 ‐    use activity prior to the 2013 mandate. But if this were true, it would likely cause us to under- estimate any true effect that the mandate might have had because pre-reform mobile use and activity would look more similar to post-reform mobile use behavior. Figure 3a: Effect of the 2013 SIM Figure 3b: Growth in average mobile use Mandate on mobile use and activity likelihood across regions following mandate Notes: Figure reflects the raw annual distribution of Notes: Figure displays the growth in average likelihoods the likelihood or probability that households’ use that households use mobile phones (i.e., MOBILE) by mobile phones (i.e., MOBILE) after household-level region after household-level fixed effects are removed fixed effects are removed from the data. The figure from the data. The sample includes all randomly selected is shown for the period before and after the National households. policy reform. The sample includes all randomly selected households. Boxes show inter-quartile ranges and whiskers contain inner 1.5×inter-quartile range of the observations (Turkey 1977). While households’ anticipation of the mandate is unclear, we provide partial evidence in favor of our assertion. Note that in the spirit of regression discontinuity RD set up, we consider time as a running variable and show that household characteristics are insignificantly different before and after the SIM reform (Lee and Lemieux 2010; Kothari and Warner 2006; Imbens and Lemieux 2008). Observable household characteristics reflect potential adjustments through which households might anticipate and respond to the SIM reform. ‐ 11 ‐    Table 2: Anticipation of reform Regime Statistic Educational 1 [Educ.,= 1 [Married = Age level notNil] Yes] Pre-reform Mean 3.400 0.991 0.853 45.445 SD (1.815) (0.180) (0.354) (14.537) Post-reform Mean 3.010 0.986 0.4101 44.826 SD (1.601) (0.117) (0.501) (19.861) P-value: Equal Means 0.999 0.142 <0.05 0.123 Density: # of Observations Pre-reform 4005 Pre-reform 3968 Notes: Table reports moments of the various observable household characteristics before and after the 2013 SIM mandate; and the density of time variable which in this case is the number of observations across the two regimes. The illustration in Table 2 partly supports the underlying argument. Changes in distributional moments of the various observable household characteristics before and after the 2013 SIM mandate are reported in Table 2. We expect to see significant changes in the relevant characteristics if households anticipated and reacted to the mandate before it came to effect. With the exception of the proportion of married respondents, the moment estimates in the table suggest no detectable evidence of reform anticipation as pre-reform moments are statistically similar to post-reform. Note that the analyses will directly control for all these observables to soak-up any potential adjustments to the reform. Thinking of time as a running variable, we would expect changes in the density of time if households significantly anticipated the un- intended effects of the mandate (e.g., in the spirit of McCrary Test; McCrary 2008). We do not find evidence of this as the number of observations in the sample across the two regimes are similar. Significant differences in the number of observations before and after the 2013 mandate could have threaten the internal validity of the policy instrument and research design since the instrument interacts the change in policy with average mobile use activity. 4 RESULTS This section reports and discusses the main empirical estimates that quantify the sign and magnitude of business enterprises and poverty responses to changes in mobile use activity. We estimate our main Equation using OLS and 2SLS. β is our main parameter of interest. This captures the sign and size of the above outcome responses to the mobile infrastructure. While the ‐ 12 ‐    2SLS estimates are based on the policy instrument, those from the OLS reflect endogenous linkages from mobile use activity. Table 3: OLS: Effect of mobile use and activity on business enterprise DV: Business (1) (2) (3) (4) (5) Enterprise MOBILE use 0.086***(0.023) 0.089***(0.023) 0.089***(0.023) 0.094***(0.023) 0.110***(0.028) activity 1 [Sex = female] 0.016(0.021) 0.021(0.021) 0.026(0.023) -0.055(0.036) 1 [Married = 0.051**(0.024) 0.060**(0.027) 0.060**(0.024) 0.018(0.034) Yes] Mobile X 1[Sex -0.018(0.046) 0.027(0.059) = female] Age in Quantiles No No Yes Yes Yes Any Education? Yes Constant 0.189***(0.014) 0.182***(0.030) 0.180***(0.048) 0.179***(0.048) 0.337*(0.198) R–squared (%) 1.07 1.44 1.74 1.74 2.58 # of 2,163 2,163 2,163 2,163 2,007 Households/OBS Fixed Effects hh+Time hh+Time hh+Time hh+Time hh+Time Notes: Table reports estimates from regressions of “engagement in non-agricultural business enterprises” on mobile use and activity. Columns differ based on the combination of model type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by ∗∗∗ 1%, ∗∗ 5%, ∗ 10%. Unit of Analysis: household-level Main estimates We begin with results from the OLS. Those pertaining to business enterprises are contained in Table 3. We expect positive relationship between mobile activity and propagation of business enterprises. Different columns correspond to different model specifications. All five specifications presented include household and year fixed effects. Column 1 excludes all control variables, while 2 adds gender and marital status. Column 3 flexibly adds age in quantiles to specification 2. Column 4 replicates 3 but includes an interaction between MOBILE and gender; while column 5 additionally control for whether respondents have some form of education. The coefficient on mobile activity MOBILE is positive and significant at conventional levels in all specifications. The OLS estimates range between 0.086 - 0.110; with an average estimate of about 0.101. An average OLS estimate of 0.101 implies that shifting a household with no mobile use activity to a mobile usage activity environment will result in about 10.1% points increase in the likelihood of engaging in some form of non-agricultural business enterprise. These estimates ‐ 13 ‐    may suffer from omitted variable bias and so should be interpreted with much caution and as mere correlations. Table 4: OLS: Effect of mobile use and activity on business enterprise DV: Business (1) (2) (3) (4) (5) Enterprise MOBILE use 0.064** 0.067**(0.022) 0.069**(0.022) 0.070**(0.025) 0.082**(0.028) activity (0.021) 1 [Sex = female] 0.012(0.021) 0.012(0.021) 0.014(0.023) -0.053(0.036) 1 [Married = Yes] 0.044*(0.023) 0.044*(0.025) 0.044*(0.025) 0.009(0.034) Mobile X 1[Sex = -0.007(0.044) 0.029(0.058) female] Age in Quantiles No No Yes Yes Yes Any Education? Yes Constant 0.200*** 0.182***(0.030) 0.172***(0.048) 0.172***(0.048) 0.392*(0.194) (0.014) R–squared (%) 1.05 1.44 1.70 1.70 2.50 # of Clusters/OBS 258 258 258 258 251 (1294) (1294) (1294) (1294) (1294) Fixed Effects Cluster+ Cluster+Time Cluster+Time Cluster+Time Cluster+Time Time Notes: Table reports estimates from regressions of “engagement in non-agricultural business enterprises” on mobile use and activity. Columns differ based on the combination of models type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by ∗∗∗ 1%, ∗∗ 5%, ∗ 10%. Unit of Analysis: cluster-level Next, results from 2SLS estimations that pertains to business enterprises are reported in Table 5.6 We expect negative and sizable effects of the SIM mandate since the reform resulted in much less mobile use and activity over the period. Similarly, results displayed in the different columns correspond to the different models. The estimate on mobile activity MOBILE is negative and significant across all model specifications. The average 2SLS estimate across specifications is about -0.331, which is about 3.3 times the size of the average OLS estimate. This suggests that the bias of the OLS estimation is downward, an attenuation. The 2SLS results imply that, on average households affected by the SIM mandate have lower likelihoods of engaging in some form on business enterprise all things being equal. A 2SLS estimate of -0.331 implies that moving a household with mobile use activity to a no mobile use activity due to the SIM reform                                                         6 Table 4 replicates OLS estimations reported in table 3 but with the unit of analysis now being cluster level. OLS results are similar across the two different unit of analysis. All 2SLS estimations are carried at the cluster level. ‐ 14 ‐    results in about 33.1% points decline in the chance of engaging in non-agricultural business enterprise. Table 5: 2SLS: Effect of mobile use and activity on business enterprise DV: Business (1) (2) (3) (4) (5) Enterprise MOBILE use activity -0.339** -0.351*** -0.3364*** -0.321** -0.318** (0.097) (0.099) (0.102) (0.129) (0.129) 1 [Sex = female] 0.041 0.046 0.056 0.060 (0.042) (0.042) (0.187) (0.069) 1 [Married = Yes] -0.018 -0.031 -0.032 -0.030 (0.035) (0.035) (0.034) (0.035) Mobile X 1[Sex = -0.040 -0.048 female] (0.187) (0.187) Age in Quantiles No No Yes Yes Yes Any Education? Yes Constant 0.256*** 0.239*** 0.228**(0.098) 0.225** 0.047 (0.039) (0.066) (0.102) (0.134) R–squared (%) 23.81 24.83 25.33 25.34 25.40 # of Clusters/OBS 231 229 229 229 229 (1337) (1294) (1294) (1294) (1294) Fixed Effects Cluster+ Time Cluster+Time Cluster+Time Cluster+Time Cluster+Time 1st Stage Estimates [-0.052,-0.125]** n X R-squared ≥ 51.6 √ No evidence of “weak instruments” Notes: Table reports estimates from regressions of “engagement in non-agricultural business enterprises” on mobile use and activity. Columns differ based on the combination of models type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by ∗∗∗ 1%, ∗∗ 5%, ∗ 10%. Unit of Analysis: cluster-level Most of the control variables included in the various regressions have their expected signs; but not all are precisely estimated. In the OLS, the gender variable which is equal to one if the survey respondent is a female and zero otherwise is positive throughout all specifications. This implies that female run households are more likely to engage in business enterprises compared to their male counterparts; highlighting the gender aspects and vital role that women play. But this estimate is not significant. Marital status which takes a value equal to one if the survey respondent is married is positive and statistically significant at 5% in all but the last specification. The implication is that the married are more likely to engage in non-agricultural enterprises than the unmarried households. Column 4 examines heterogeneous responses across gender by interacting MOBILE with the gender variable. The estimate of the interaction is negative but indistinguishable from zero, suggesting limited evidence in favor of heterogeneous ‐ 15 ‐    gender effects of mobile use activity on non-agricultural enterprises. In the 2SLS, gender is also positive and insignificant while marital status is negative and insignificant. Estimate for the interaction between MOBILE and gender is similar and insignificant. While the estimate for the interaction between MOBILE and gender (i.e., female) is not significant, the negative sign highlights that women have been more vulnerable to the mandatory SIM registration exercise. This can be explained by the fact that women tend to disproportionally work in non-farm self- employments. Table 6: OLS: Effect of mobile activity on aspects of business enterprise DV (1) (2) Make (3) Own (4) Any (5) Own (6) Own (7) Own (8) Own Transform clothing construction profession bus commercial restaurant small Agr. product for own service transport or bar activity MOBILE 0.0319* 0.00286 0.00313 -0.00180 0.0142 -0.00975 -0.00262 -0.0338* (0.0173) (0.0105) (0.0105) (0.00740) (0.00970) (0.00597) (0.00540) (0.0189) Constant 0.188*** 0.0348*** 0.0375*** 0.0329*** 0.0277*** 0.0181*** 0.0158*** 0.223*** (0.0150) (0.00588) (0.00555) (0.00482) (0.00462) (0.00440) (0.00424) (0.0167) # of OBS 2,785 2,785 2,785 2,785 2,785 2,785 2,785 2,784 # of 258 258 258 258 258 258 258 258 Clusters Controls No No No No No No No No Cluster FEs Yes Yes Yes Yes Yes Yes Yes Yes Year FESs Yes Yes Yes Yes Yes Yes Yes Yes Notes: Table reports estimates from regressions of the various aspects of engagement in non-agricultural business enterprises on mobile use and activity. Columns differ based on the combination of the dependent variable, model type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by *** 1%, ** 5%, *10%. Unit of Analysis: cluster-level. Table 7: OLS: Effect of mobile activity on aspects of business enterprise DV (1) Transform (2) Make (3) Own (4) Any (5) Own (6) Own (7) Own (8) Own Agr. product clothing construction profession bus commercial restaurant small for own service transport or bar activity MOBILE 0.0284* 0.00432 0.00302 -0.000682 0.0137 -0.0132** -0.00447 -0.0362* (0.0169) (0.0106) (0.0108) (0.00765) (0.00986) (0.00613) (0.00532) (0.0200) Constant 0.147*** 0.042*** 0.0454*** 0.0204** 0.0190 0.0138* 0.0197** 0.184*** (0.0227) (0.0112) (0.0113) (0.0103) (0.0154) (0.00723) (0.00832) (0.0289) # of OBS 2,677 2,677 2,677 2,677 2,677 2,677 2,677 2,676 # of Clusters 258 258 258 258 258 258 258 258 Controls Yes Yes Yes Yes Yes Yes Yes Yes Cluster FEs Yes Yes Yes Yes Yes Yes Yes Yes Year FESs Yes Yes Yes Yes Yes Yes Yes Yes Notes: Table reports estimates from regressions of the various aspects of engagement in non-agricultural business enterprises on mobile use and activity. Columns differ based on the combination of the dependent variable, model type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by *** 1%, ** 5%, *10%. Unit of Analysis: cluster- level. ‐ 16 ‐    Exploring Mechanisms So far, we have been working with a general definition of business enterprises i.e., households engagement in any form of non-agricultural business or startup over the period. This abstracts from multiple aspects, which we shall explore to explain the mechanisms underlying our baseline results. Engagement in business enterprises takes several forms including households’ ownership of a business service or center (e.g., a Mobile-Money center); ownership of a commercial transport business; ownership of a restaurant or bar; ownership of other small income generating activities; the ownership of a construction establishment; involvement in the making of clothing; involvement in the transformation of any agricultural products; and having any non-farm profession. Our panel surveys provide information about these aspects. Table 8: 2SLS: Effect of mobile activity on aspects of business enterprise DV (1) (2) Make (3) Own (4) Any (5) Own (6) Own (7) Own (8) Own Transform clothing construction profession bus commercial restaurant small Agr. for own service transport or bar activity product MOBILE -0.178* -0.0932* -0.105** 0.00611 -0.125** -0.0434 -0.00434 -0.147* (0.0954) (0.0550) (0.0501) (0.0431) (0.0633) (0.0440) (0.0409) (0.0763) Constant 0.437*** 0.286*** 0.0605** 0.374*** 0.0550** 0.0160 -0.00378 0.0668** (0.0378) (0.0218) (0.0279) (0.0175) (0.0255) (0.0175) (0.0164) (0.0325) # of OBS 1,337 1,337 1,337 1,337 1,337 1,337 1,337 1,337 R-squared (%) 25.20 18.50 18.50 19.00 19.80 18.90 21.70 31.90 Controls No No No No No No No No Cluster FEs Yes Yes Yes Yes Yes Yes Yes Yes Year FESs Yes Yes Yes Yes Yes Yes Yes Yes 1st Stage Estimates [-0.052,-0.125]** n X R-squared ≥ 51.6 √ No evidence of “weak instruments” Notes: Table reports estimates from regressions of the various aspects of engagement in non-agricultural business enterprises on mobile use and activity. Columns differ based on the combination of the dependent variable, model type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by *** 1%, ** 5%, *10%. Unit of Analysis: cluster-level. Tables 6-9 replicate previous estimations, but separately for each aspect of household business enterprises. All dependent variables are binary, and the columns reflect specified models. The OLS estimates are reported in Tables 6 and 7, while those from our preferred 2SLS are presented ‐ 17 ‐    in Tables 8 and 9.7 Note that results displayed in Tables 7 and 9 include all relevant control variables; Tables 6 and 8 omit the controls. Table 9: 2SLS: Effect of mobile activity on aspects of business enterprise DV (1) (2) Make (3) Own (4) Any (5) Own (6) Own (7) Own (8) Own Transform clothing construction profession bus commercial restaurant small Agr. for own service transport or bar activity product MOBILE -0.191* -0.105* -0.108** 0.0166 -0.141** -0.0553 -0.0114 -0.169** (0.0994) (0.0578) (0.0508) (0.0458) (0.0660) (0.0471) (0.0465) (0.0759) Constant 0.326*** 0.292*** 0.0665** 0.372*** 0.0487 0.0448* -0.00152 0.0457 (0.0601) (0.0294) (0.0299) (0.0247) (0.0339) (0.0265) (0.0218) (0.0487) # of OBS R-squared (%) 26.70 19.70 23.60 18.80 20.90 19.80 22.80 23.70 Controls Yes Yes Yes Yes Yes Yes Yes Yes Cluster FEs Yes Yes Yes Yes Yes Yes Yes Yes Year FESs Yes Yes Yes Yes Yes Yes Yes Yes 1st Stage Estimates [-0.052,-0.125]** n X R-squared ≥ 51.6 √ No evidence of “weak instruments” Notes: Table reports estimates from regressions of the various aspects of engagement in non-agricultural business enterprises on mobile use and activity. Columns differ based on the combination of the dependent variable, model type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by *** 1%, ** 5%, *10%. Unit of Analysis: cluster-level. The 2SLS estimates on mobile activity MOBILE are mostly significant and negative, as we would expect. Effects are large under the following 3 aspects: households’ ownership of a business service or center (-0.125); ownership of small income generating activities (-0.147); and involvement in the transformation of agricultural products (-0.178). These are then followed by households’ engagement in any sort of construction enterprises and the making of clothing for local sale. We do not find any effects for the other aspects, e.g., having own profession and ownership of a commercial transport business, which we do not find surprising. Most of the Nigerien households are agrarian, which helps to explain why the transformation of agricultural products will show the largest mobile activity impact. Next, households’ ownership of a business service or center is also expected to be largely affected by the SIM deactivation exercise since most of these service centers operate directly on the mobile infrastructures, e.g., the recent and                                                         7 The OLS results are included here for mere display, we shall focus discussions below on the results from our preferred 2SLS/Policy Instrument. ‐ 18 ‐    growing Mobile-Money services across sub-Saharan Africa (The Economist 2013; Jack and Suri 2014; Suri and Jack 2016). Business enterprises and household welfare Earlier, we emphasized that business enterprises and non-agricultural startups in rural economies shape poverty inclinations. This section descriptively analyze this belief by exploring whether households’ involvement in the various business enterprises play any role in observed household welfare/poverty. Table 10: Exploring linkages between household poverty and business enterprises DV (1) (2) (3) (4) (5) (6) Absolute Absolute Absolute Absolute Absolute Absolute poor poor poor poor poor poor Business Enterprise -0.0234* -0.0231* -0.0717** -0.0722** -0.0771*** -0.0767*** (0.0139) (0.0140) (0.0290) (0.0291) (0.0287) (0.0287) MOBILE use activity -0.0466 -0.0453 -0.0569* -0.0560* (0.0334) (0.0334) (0.0336) (0.0336) Bus. Enterprise X MOBILE 0.00856 0.00605 0.00622 0.00443 use (0.0502) (0.0503) (0.0502) (0.0501) Female -0.0522* -0.0450 (0.0273) (0.0276) Married 0.0240 0.0253 (0.0379) (0.0380) Constant 0.00351* 0.00145 1.3EXP-13 1.3EXP-16 0.0347 0.0554 (0.00208) (0.00551) (<0.001) (<0.001) (0.0645) (0.0691) # of Observations 6,859 6,859 2,471 2,471 2,367 2,367 R-squared (%) 27.00 27.00 25.20 25.30 27.10 27.20 Cluster FEs Yes Yes Yes Yes Yes Yes Year FEs No Yes No Yes Yes Yes Notes: Table reports estimates from regressions of household poverty status on their engagement in non-agricultural business enterprises and in some cases mobile use and activity. Columns differ based on the combination model type and fixed effects included. All standard errors are clustered at the region level and shown in parentheses. Significance levels are indicated by *** 1%, ** 5%, * 10%. Unit of Analysis: cluster-level. MOBILE use activity is hardly significant once households’ engagement in business enterprises is controlled for. Table 10 reports results based on OLS regressions of household poverty on business enterprises. Results from 6 models are presented; in some versions of the poverty–business relationship, we directly control for mobile use activity. But as expected, mobile use is hardly significant once households’ engagement in business enterprises are controlled for, highlighting the dominant role of business enterprises. The coefficient on “Business Enterprise is negative and significant across the wide range of models. These range between a -0.023 to -0.077; implying that households who engage in non-agricultural business enterprises are about 2.3% to 7.7% points ‐ 19 ‐    less likely to become absolute poor compared to households that do not engage in some form of business enterprise. While results from Table 10 document the vital role of business enterprises in household poverty, the estimates should be interpreted with caution in that they likely suffer omitted variable biases. 5 CONCLUSION The aim of this paper has been to exploit the quasi-experimental variation created by the exogenous introduction of the 2013 mandatory SIM registration reform in Niger to document the impact of mobile use activity on propagation of business enterprises and poverty among households. We leverage new rich data sets on various aspects of business enterprises, use of mobile telephony along with a policy instrument based on the SIM mandate/deactivation process. Our analysis document two important results. First, the SIM mandate lead to a remarkable decline in mobile use activity over the period. Second, there is economically large effect of mobile infrastructure on propagation of rural business enterprises. The drop in mobile use activity induced by the mandate resulted in about 33.1% points decline in the likelihood of households’ engaging in non-agricultural business enterprises, which has extended impacts on poverty. There is suggestive evidence that the reform’s impact is disproportionally larger for women. Exploring the potential mechanisms underlying our results, we find that most of the effect come from three major sources: households’ ownership of a business service or center; ownership of small income generating activities; and involvement in the transformation of agricultural products. These results have important policy implications. Mandatory SIM policies meant to curb crime have become popular across sub-Saharan Africa. First, the soundness of these policies require much broader Benefit-Cost assessments, which in turn have led to various policy discussions across various circles. The findings provide evidence in favor of ongoing policy debates and arguments suggesting that mandatory SIM policies can generate un-intended consequences on the local economy among poor populations e.g., Niger. Second, given the large documented un- intended negative effects of Niger’s SIM mandate, we will need fuller assessment of such registration exercises across the spectrum for security or commerce moving forward to justify them. ‐ 20 ‐    REFERENCES CITED 1) Abhijit Banerjee, Esther Duflo and Nancy Qian. 2012. “On the Road: Access to Transportation Infrastructure and Economic Growth in China.” Mimeo MIT. 2) Aker, Jenny. 2010. “Information from Markets Near and Far: The Impact of Mobile Phones on Agricultural Markets in Niger.” American Economic Journal: Applied Economics, 2 (3) : 46–59. 3) Aker, Jenny, Michael Clemens and Christopher Ksoll. 2011. “Mobiles and Mobility: The Effect of Mobile Phones on Migration in Niger.” Unpublished Manuscript. 4) Aker, Jenny, Paul Collier and Pedro Vicente. 2016. “Is Information Power? Using Mobile Phones and Free Newspapers during an Election in Mozambique.” Review of Economics and Statistics (Forthcoming). 5) AFP, 2013. “Niger Cuts off third of Mobile Phones to Stop Crime.” (27 November), Accessed 29 January 2014. 6) Atif Ansar, Bent Flyvbjerg, Alexander Budzier, and Daniel Lunn. 2016. “Does infrastructure investment lead to economic growth or economic fragility? Evidence from China.” Oxford Review of Economic Policy, 32(3): 360–390. 7) Annan, Francis and Aly Sanoh. 2016. “Social protection in Niger: What has Shocks and Time Got To Say?” Mimeo World Bank. 8) Batista, Catia and Gaia Narciso. 2016. “Migrant Remittances and Information Flows: Evidence from a Field Experiment.” World Bank Economic Review (Forthcoming). 9) Blumenstock, Joshua, Nathan Eagle and Marcel Fafchamps. 2016. “Airtime Transfers and Mobile Communications: Evidence in the Aftermath of a Natural Disaster” Journal of Development Economics (Forthcoming). 10) Dinkelman, Taryn. 2011. “The Effects of Rural Electrification on Employment: New Evidence from South Africa.” American Economic Review, 101(7): 3078–3108. 11) Duflo, Esther, and Pande, R. 2007. “Dams.” Quarterly Journal of Economics, 122(2), 601–46. 12) Fafchamps, Marcel and Bart Minten. 2012. “Impact of SMS-Based Agricultural Information on Indian Farmers.” World Bank Economic Review, 26(3): 383–414. 13) Gonzalez, Robert. 2015. “Social Monitoring and Electoral Fraud: Evidence from a Spatial Regression Design in Afghanistan.” Mimeo University of North Calorina, Chapel-Hill. 14) Goyal, Aparajita. 2010. “Information, Direct Access to Farmers and Rural Market Performance in Central India.” American Economic Journal: Applied Economics, 2(3): 22–45. 15) Hansen, H., Andersen, O. W., and White, H. 2013. Impact Evaluation of Infrastructure Interventions, London, Routledge. 16) Imbens, G. W., and T. Lemieux. 2008. “Regression Discontinuity Designs: A Guide to Practice.” Journal of Econometrics, 142 (2): 615–635. 17) Jack, William and Tavneet Suri. 2014. “Risk Sharing and Transactions Costs: Evidence from Kenya’s Mobile Money Revolution.” American Economic Review, 104(1): 183– 223. 18) Jensen, Robert. 2007. “The Digital Provide: Information (Technology), Market Performance and Welfare in the South Indian Fisheries Sector.” Quarterly Journal of Economics, 122(3): 879–924. ‐ 21 ‐    19) Jentzsch, Nicola. 2012. “Implications of Mandatory Registration of Mobile Phone Users in Africa.” Deutsches Institut für Wirtschaftsforschung, Disussion papers #1192. 20) Klonner, Stefan and Patrick Nolen. 2010 “Cell Phones and Rural Labor Markets: Evidence from South Africa.” Unpublished Manuscript. 21) Lee, D., and T. Lemieux. 2010. “Regression Discontinuity Designs in Economics.” Journal of Economic Literature, 48: 281–355. 22) Martin, Aaron K., Rosamunde E. van Brakel and Daniel J. Bernhard. 2009. “Understanding Resistance to Digital Surveillance: Towards a Multi–disciplinary, Multi– actor Framework.” Surveillance and Society, Volume 6, Number 3, pp. 213–232, Accessed 21 January 2014. 23) Martin, Aaron K., and Kevin P. Donovan. 2014. “The Rise of African SIM Registration: The Emerging Dynamics of Regulatory Change.” First Monday, Volume 19, Number 2- 3, Accessed 29 September 2016. 24) Mbiti, Isaac and David Weil. 2013. “The Home Economics of E-Money: Velocity, Cash Management, and Discount Rates of M-Pesa Users.” American Economic Review, 103(3): 369–74. 25) McCrary, J. 2008. “Manipulation of the Running Variable in the Regression Discontinuity Design: A Density Test.” Journal of Econometrics, 142 (2): 698–714. 26) Southwood, Russell. 2011. “SIM registration in Africa: Subscribers numbers down but what about revenue and ARPU?” Balancing Act, Number 554. Accessed 29 January 2013. 27) Sridhar, Kala and Varadharajan Sridhar. “Telecommunications Infrastructure and Economic Growth: Evidence from Developing Countries.” National Institute of Public Finance and Policy (New Delhi, India) Working Paper No. 14, 2004. 28) Suri, Tavneet and William Jack. 2016. “The long-run poverty and gender impacts of mobile money.” Science, 354(6317): 1288-1292. 29) Tompsett Anna. 2013. “Bridges: Transport Infrastructure and Economic Geography on the Mississippi and Ohio 1960–2000.” Mimeo Columbia University. 30) The Economist. 2013. “Why does Kenya lead the World in Mobile Money?” Accessed July 27, 2016. 31) The Guardian. 2015. “Africa’s mobile phone penetration now 67%.” Accessed July 27, 2016. 32) Waverman, Leonard, Meloria Meschi and Melvyn Fuss. 2005. “The Impact of Telecoms on Economic Growth in Developing Countries.” Vodafone Policy Paper Series, 3. 33) Whitley, Edgar A., and Ian Hosein. 2005. “Policy discourse and data retention: The technology politics of surveillance in the United Kingdom.” Telecommunications Policy, Volume 29, Number 11:857–874. Accessed 21 January 2014. 34) World Bank. 2016. Republic of Niger, “Priorities for Ending Poverty and Boosting Shared Prosperity.” Systematic Country Diagnostic. ‐ 22 ‐    6 APPENDIX Figure A.1: Extract about Niger’s SIM deactivation in the news ‐ 23 ‐