Social Network Effects on Mobile Money Adoption in Uganda

Abstract

This study analyses social network effects on the adoption of mobile money among rural households in Uganda. We estimate conditional logistic regressions controlling for correlated effects and other information sources. Results show that mobile money adoption is positively influenced by the size of the social network with which information is exchanged. We further find that this effect is particularly pronounced for non-poor households. Thus, while social networks represent an important target for policy-makers aiming to promote mobile money technology, the poorest households are likely to be excluded and require more tailored policy programmes and assistance.

Acknowledgements

This research was financially supported by German Research Foundation (DFG) and German Academic Exchange Service (DAAD). We are also grateful to Grameen Foundation for support in fieldwork coordination. The authors commit to providing the data and Stata codes on request.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. This includes individuals, households, and institutions.

2. This encompasses all forms of communication, for example word of mouth, text messages or voice calls, and so forth.

3. Households may of course have indirect access to information from weak ties through their strong tie contacts. The opportunities for such indirect access likely increase with the absolute number of strong ties in the household’s social network.

4. Mobile money user and adopter are used interchangeably in this article.

5. We also tried a different definition of weak ties based on the type of relationship, but this did not change the results of our models.

6. For example, in social networks that are characterised by highly educated members more information may be available and this may positively influence household i’s adoption decision. Note that it is the availability of high-quality information within the network, rather than the actual adoption decision of other network members, that is here assumed to influence household i.

7. For example, the education level of household i may have an effect on his/her adoption decision. If households with higher education levels tend to create networks with other highly educated households, observed adoption within the network may be high due to generally high education levels, rather than imitation or available information within the network. Similarly, network members may be exposed to similar village-level institutions, extension agents, mobile money agents and so forth that shape their adoption decisions in similar ways.

8. In the research area, extension is usually provided by community knowledge workers. These are locally recruited peer farmers who are trained by Grameen Foundation to use Android smartphones to disseminate agricultural and market information to fellow farmers in their respective villages.

9. The incidental parameters problem we refer to is really a theoretical problem. Note that a difference between linear and non-linear estimators is that in general fixed effects (in this context, at the village level) cannot be differenced out. Therefore, if fixed effects are introduced into a binary choice model, each one actually has to be estimated. Since the estimation of requires the estimation of all N fixed effects parameters (assuming the number of observations in each village is fixed at M), the problem is that theoretically root N consistency requires that N approaches infinity, so necessarily the estimate of is inconsistent (for example Wooldridge, 2002).

10. While we consider religion as an important control variable in our analysis, our sample (consisting of a very small share of Muslim households) is not adequate to derive any general conclusions on the influence of religion and religious beliefs on mobile money adoption.

Additional information

Funding

This work was supported by the Deutsche Forschungsgemeinschaft (DFG);Deutscher Akademischer Austauschdienst.