Can Mobile Money Induce Financial and Entrepreneurial Behaviors of Members of Village Savings and Loan Associations in Rural Malawi?

VSLAs members in Malawi will be trained on how to use mobile money services to conduct business transactions and other aspects of financial literacy. The training aims at giving the participants basic knowledge and skills on how the mobile money platform can be used for business purposes as well as some elements of financial literacy. The training will be followed by SMSs that will remind the participants about the important contents of the training they have undergone As other studies related to saving behavior in similar contexts have already shown that training alone does not lead to significant outcomes unless it is combined with SMSs that remind the participants to save, the two elements of our intervention (training and SMSs) will be promoted as a package. The experiment will therefore have two treatment branches: i) the control arm, which will not receive the intervention; and ii) the treatment arm, which will receive face to face, field based, training on mobile money services usage for business purposes. Treated individuals will also be sent SMS reminders on the use of mobile money for business transactions. Both the treatment group and control group members will be participants in VSLAs, henceforth the positive impacts of VSLAs on financial inclusion and business outcomes that were established by other studies will be assumed. The project will train each member of the treated group on the use mobile money services for business purposes.

2018-11-01

2019-04-30

Proportion of VSLA members that use mobile money for business purposes
Level of savings measured as percentage

Proportion of VSLA members with a bank account,
Level of capital injected into business
Demand for credit measured as loans that have been obtained

In our intervention context, several VSLAs are grouped together to form a Group Village Headman (VGVH). As required by our partner NGO in the intervention area, all VSLAs within a given GVH must be assigned to the same treatment status (treatment or control); and if a given VSLA is assigned to the treatment group, all its members must be treated. Taking this context into account, we decided to work with a cluster design. To implement this design, we will first randomly chose one VSLA per GVH, and then randomly assign half of these VSLAs to the treatment group and the other half to the control group.

In our intervention context, several villages (six to ten) are grouped together to form a Group Village Headman (GVH). In each of the villages, we will have at least one VSLA. As recommended by our partner NGO in the intervention area, all VSLAs within a given GVH must be assigned to the same treatment status (treatment or control); and if a given VSLA is assigned to the treatment group, all its members must be treated. Taking this context into account, we decided to work with a cluster design. To implement this design, we will first randomly chose one VSLA per GVH, and then randomly assign half of these VSLAs to the treatment group and the other half to the control group.
We will base our power calculations considering a binary outcome: usage of mobile money for business transactions. Assuming that usage in the control group will be around 5% (with a 95% CI between 1% and15%) and considering that our project will evaluate usage in the very short term (less than six months after the intervention), we expect our treatment to increase usage by at least 8 percentage points. We also consider that on average we will be able to collect information from 18 individual per cluster (according to our partner NGO, the average VSLA size is 20, and the minimum size is 15).
The parameters that have been used to determine the required number of clusters in our design are then as follows: average cluster size = 18; probability of success in the treatment group=13%; probability of success in the control group=5%; and 95% C.I. for usage in the control group between 1% to 15%. With these parameters, it is found that a total of 42 clusters will allow us to detect our expected treatment effect with power of 80%. Considering 42 clusters, the total number of individuals that must be included in the baseline is 756.
Taking into account the number of clusters determined in the binary outcome calculations (42), we now estimate the minimum effect that we will be able to detect in the continuous case with power 80%. Two additional assumptions need to be made for this purpose. In first place, we assume that the Inter Cluster Correlation (ICC) is about 0.05; and in second place we assume that the baseline covariates explain fifteen percent of the observed variation in the model (R2=0.15). This analysis shows that we can observe that with power 80% we will be able to detect a minimum effect of 0.27 s.d. (which is higher, but relatively close, to the standard 0.20 s.d. effect generally recommended in the impact evaluation study literature).
Finally, we need to adjust the sample for potential individual attrition as well as for potential individual noncompliance, which are expected to be about 3% and 25%, respectively. The resulting sample after these adjustments contains 973 individuals in 42 clusters. This sample size will be attained by interviewing a minimum of 18 individuals per cluster.

Randomisation will be done in office on a computer with representatives of Emmanuel International in attendance.

Group village headman (cluster)

Yes

42 group village headmen (clusters).

973 individuals that are members of VSLAs. Half of these individuals will be in the treatment group and half will be in the control group

Participants from 21 group village headmen will not receive any intervention (control) and Participants from 21 group village headmen will receive training in use of mobile money and SMSs to remind them of the training content.

ICC is assumed at 0.05, minimum detectable effect (difference in proportion of savings to total income) is assumed at 0.20, and R-squared=0.15.