Measuring Spillover Effects from an Entrepreneurship Programme: Evidence from a Field Experiment in Tanzania

Abstract

In this paper, we identify treatment spillovers between microfinance clients in Tanzania using data from a partial population experiment where only a subset of loan group members was offered treatment in the form of business training, a business grant, or both. Our results show large and significant spillover effects from indirect exposure to treatment through group peers. In particular, we find that male microfinance clients with peers receiving both business training and the business grant experience to have significantly higher sales than those not receiving any treatment. Moreover, microfinance clients with positive spillovers make higher investments and borrow more. In addition, the treatment impacts are higher for group members with smaller loans (which is consistent with higher marginal rates of return to capital), and for members in groups with a greater share of women. Our findings illustrate that loan groups may be an important arena for the sharing of entrepreneurial resources, and that standard treatment–control analyses of similar interventions may underestimate impacts in settings with close social interactions.

KEWORDS:

• Microfinance
• entrepreneurship
• field experiment
• spillovers

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplementary Materials are available for this article which can be accessed via the online version of this journal available at https://doi.org/10.1080/00220388.2021.1945039

Notes

1. Several studies have shown that female entrepreneurial outcomes are more difficult to raise than men entrepreneurial outcomes (for a review of this literature see Berge & Garcia Pires, 2020). This is one of the main puzzles in this type of development interventions, for which to the best of our knowledge there are only partial explanations. Berge and Garcia Pires (2020) point out that female entrepreneurs are more time constrained than male entrepreneurs (in part due to family obligations, domestic work and taking care of relatives). Berge, Bjorvatn, Garcia Pires, and Tungodden (2015b), in turn find that male entrepreneurs are more competitive than female entrepreneurs.

2. Angelucci and Di Maro (2015) categorise four types of spillovers that are particularly relevant in development literature. These are: (i) externalities, such as when deworming drugs also benefit the non-treated by reducing disease transmission (Miguel & Kremer, 2004); (ii) social interactions, where, for example, the treated share resources with non-treated; (iii) context equilibrium effects, where behaviour or norms are changed and (iv) general equilibrium effects, where interventions affect equilibrium prices through changes in supply and demand. However, in practice it is often difficult to distinguish formally between these different types of spillovers.

3. Our analysis also relates to the literature on the direct impact of business training and other interventions aimed at improving entrepreneurial outcomes, including de Mel et al. (2008), Field, Jayachandran, and Pande (2010), Karlan and Valdivia (2011), Klinger and Schündeln (2011), de Mel et al. (2014), McKenzie and Woodruff (2014), Berge et al. (2015a) and Bulte, Lensink, van Velzen, and Vu (2017). While many of these studies conclude that it is often more difficult to promote female entrepreneurship, Field et al. (2010) conclude that business counselling can actually improve female entrepreneurial outcomes, but only in the treatment condition where entrepreneurs are invited to attend with a friend. This suggests that spillover effects can catalyse entrepreneurship.

4. In nondevelopment settings, recent studies using partial population experiments include Duflo and Saez (2003) (examining retirement plan decisions), Crepon, Duflo, Gurgand, Rathelot, and Zamora (2013) (investigating whether labour market policies have displacement effects) and Sinclair, McConnell, and Green (2012) (analysing voting behaviour).

5. Many studies have shown that female entrepreneurs are more business constrained than men entrepreneurs because of social norms. This might explain, at least partially, why these type of programmes have less impact on female entrepreneurs. See also endnote 1 (for a review of this literature see Berge & Garcia Pires, 2020). In this context, Berge and Garcia Pires (2020) show evidence that indeed women entrepreneurs are more business constrained than men entrepreneurs and this results at least in part, from less time dedicated to their business due to family obligations.

6. Our experiment provides no exogenous variation in treatment saturation, so we cannot estimate to what extent the percentage of individuals treated in treatment clusters drive the spillovers. Crepon et al. (2013), Sinclair et al. (2012) and Baird et al. (2014) are examples randomising treatment saturation.

7. Note that except for MFI and the loan group related variables, we collected the data after the assignment of the training and business grants. However, we believe that it is very unlikely that the treatment spillovers influence education, age and the other demographic characteristics. Nonetheless, we concede that these may affect programme variables like sales and work hours, suggesting that the treatments may influence the responses.

8. In Appendix A1 (see supplementary material), we also provide the treatment–control balances for male and female entrepreneurs. The general picture is once again that the control and treatment groups are similar. For males, and in terms of individual characteristics, the treatment and control groups are unbalanced for ‘born in Dar es Salam’, ‘branch’ and ‘employees’. For females, the unbalanced variables for individual characteristics are ‘age’, ‘born in Dar es Salem’, ‘branch’, ‘work hours’ and ‘sales’. In terms of the group and group members characteristics, for the male sample, the unbalanced variables are ‘number of treated females in group’, ‘number of treated group members in commerce’ and ‘own loan versus treated group members’. For females, there is just one unbalanced variable for group and group member characteristics – the ‘number of group members’.

9. The exchange rate at the time of the survey was 0.0008 USD/TZS. Daily income per capita in the household is around 6 800 TZS (5.3 USD) for females and 5 800 TZS (4.5 USD) for males, indicating that most of the households are above the standard extreme poverty line defined by the United Nations.

10. Typical investments include for example chairs, cooking equipment and refrigerators for restaurants, a new premise for commerce businesses, sewing machines for tailors and hair-cutters for hairdressers and barbers. Note that investment does not include purchases of stocks/goods for resale, which is an important cost component in the service sector.

11. When log-transforming continuous outcome variables (sales, profit, investment, loan size), which may take zero-values, we add the smallest positive value observed for all observations and log-transform this number. In general, in order to make estimates as comparable as possible to the estimates of the direct treatment, we estimate the impacts similarly to Berge et al. (2015a).

12. Given that the treatment dummies are uncorrelated with unobserved explanatory factors, there is no need to include a covariate matrix to obtain unbiased estimates. However, including control variables should make the estimations more precise (see Angrist & Pischke, 2009).

13. Net profits also deduct salaries and investments, as defined in 5.1. If we estimate a profit measure where we also deduct estimated monthly interest-costs, based on outstanding debt, to get operating profits – salaries – investments – interests; results remain very similar (that net profits have increased for the combined treated males-coeff: 0,476, p = 0,056).

14. We also observe a negative impact of grants on marketing for females. We have no particular explanation for this finding.

15. We note that among the directly treated, as reported in Berge et al. (2015a), we see similar effects, with those receiving both training and the grant being more involved in commerce and less in services and manufacturing, as well as investing more in their businesses and having higher loans. Note also that manufacturing is a male dominated sector. This can also explain, at least partially, why males are more likely to move out of this sector than women entrepreneurs, since there are few women in manufacturing initially.

16. Although not significant (and not shown in the table), male entrepreneurs with combined treated peers increase borrowings from relatives and business partners by 25%.

17. In Appendix 2 (see supplementary material), we also show the results for different absolute loan sizes. These are fairly similar to the results in Table 7.

18. Unfortunately, our sample size does not allow to split up the sample further into either having only females in the group, only males, and both male and females in the group.

Additional information

Funding

This work was supported by the Prof. Vilhelm Keilhaus Minnefond [Travel grant]; Bank of Norway [Small Research grant]; Sparebanken Vest, Bergen, Norway [Grant funding field experiment].