Financial Inclusion, Shocks, and Poverty

Notes: Number of observations: 2,338 households. In Panel A, we present information on the distribution of mobile money agents across communities over the two waves. Agent availability is an indicator variable for the presence of an agent within the enumeration area. Cost to nearest agent is calculated based on travel cost given in the LSMS-ISA survey. Agent availability is also presented for different radiuses around the village center. Panel B of the table reports summary statistics of mobile money accounts used by the households across the two surveys. The first entry reports the fraction of households with at least one mobile money account. The second entry reports number of mobile money accounts per capita, and lastly we report the different service providers adopted by households (MM provider Tigo was not yet operational in 2010). Panel C presents the frequency of use of mobile money services as a fraction of adopter households by year. Panel D reports the most frequent uses of mobile money services. This shows the overall most important uses of mobile money services by users as a fraction of all adopter households by year. In the 2010 LSMS-ISA survey wave “large purchase” was not listed as possible answer.

↵Notes: The poverty index takes a value of one for daily real per capita expenditure above 1.25 USD, and zero otherwise. Mobile money denotes the propensity to adopt mobile money account (see Online Appendix A3 for details) at the household level. Rainfall shock denotes the deviation from long-term average rainfall, such that a negative value denotes less than the average rainfall. Each column reports the estimates from a separate regression for 3,448 observations (1,724 households). All regressions include household and year fixed effects. The entries of Columns 1 and 2 of the table report the DiD coefficients from a linear probability model of mobile money, rainfall shock, and their interaction term on a poverty indicator. In Column 3, the variable Mobile money adoption is instrumented by the presence of and distance to the nearest mobile money agent such that Mobile money (MM) [interaction] is instrumented by agent availability in the village and distance to nearest agent (interaction of agent availability in the village with rainfall shocks and distance to nearest agent with rainfall shocks). First-stage results are presented in Online Appendix Table A6. The controls used in the estimation of Column 2 and 3 include an array of household-level covariates (gender of household head, education and occupation categories of household head, household size, average household age, rural dummy, household asset value, number of mobile phones in the household, indicator variables for household membership of a SACCO group, household membership of any other credit and savings society, household access to loan facilities and bank account ownership, and the interaction of the financial inclusion variables with the shock variable). Robust standard errors, clustered at the enumeration area, are reported in parentheses. Robust standard errors, clustered at the district level, are reported in square brackets. Significance: p 0.10, **p < 0.05, ***p < 0.01.

↵Notes: The entries present the coefficients from the DiD coefficients of mobile money, rainfall shock, and their interaction term on the log amount per capita expenditure by wealth quintiles. We use asset-holding details from the 2012 wave. The 2012 survey questionnaire reports two measures for each household asset, the purchase price (when it was bought) and the market price during the time of the interview. We construct current nonagricultural wealth across households by weighing each household asset using the average price between the two asset prices. We then proceed to sum up the worth of each asset holding to measure nonagricultural asset index of the household and produce quintiles of household asset wealth. See notes in Table 3 (Column 2) for the specification and the set of controls used in the estimation. Robust standard errors, clustered at the enumeration area, are reported in parentheses. Significance: p < 0.10, **p < 0.05, ***p < 0.01.

↵Notes: The entries of the table report the DiD coefficients of mobile money, rainfall shock, and their interaction term on the log amount of welfare receipts from government and NGOs over the past 12 months. The question in the LSMS-ISA questionnaire is: “How much money did your household receive from government or NGOs in the last 12 months?” See notes in Table 3 (Column 2) for the specification and the set of controls used in the estimation. Robust standard errors, clustered at the enumeration area, are reported in parentheses. Significance: p<0.10, **p<0.05, ***p<0.01.

↵Notes: The entries of the table report the DiD coefficients of mobile money, rainfall shock, and their interaction term on health expenditure and bed net use. The entries in Column 1 present the coefficients from a linear probability model on an indicator variable for preventative health expenditure; entries in Column 2 are from a linear regression on log preventative healthcare expenditure. The preventative health expenditure indicator in Column 1 takes a value of one if an individual spends a positive amount on preventative health in the four weeks prior to the survey, and zero otherwise. Preventative health expenditure in Column 2 is calculated as the natural logarithm of real preventative health expenditure (in thousand Tanzanian shillings). Results in Columns 3 and 4 represent estimated coefficients for indicators of bed net use and treated bed net use. The bed net use question refers to sleeping under bed net the night before the survey. See notes in Table 3 (Column 2) for the specification and the set of controls used in the estimations. Robust standard errors, clustered at the enumeration area, are reported in parentheses. Significance: p < 0.10, **p < 0.05, ***p < 0.01.

↵Notes: The entries of the table report the DiD coefficients of mobile money, rainfall shock, and their interaction term on a number of educational inputs. The outcome variable in Column 1 is log real per capita school expenditure. The outcome variable in Column 2 is an indicator for (current) school enrollment that takes a value of one if the child is currently enrolled at school, and zero otherwise. The outcome variable in Column 3 is an indicator variable that takes a value of one if the child has missed school in the two weeks prior to the survey, and zero otherwise. The outcome variable in Column 4 is the number of hours that a child spends per day on homework and studying during the week prior to the survey. See notes in Table 3 (Column 2) for the specification and the set of controls used in the estimations. In addition to household-level controls, age and gender of individuals are used as additional individual controls in all regressions. The number of observations varies across outcomes, as the information on absenteeism and homework is not available in all household questionnaires. Robust standard errors, clustered at the enumeration area, are reported in parentheses. Significance: p < 0.10, **p < 0.05, ***p < 0.01.