Implications for Poverty Reduction in Rural Households in Ghana: Model Specification

Given the fact that there is potential selection bias and issues of endogeneity which have come about because MFIs choose communities in which they want to operate (to lend) deliberately (known as endogenous programme placement), and also because individuals themselves choose to borrow or not borrow (known as endogenous programme participation),
those who are not borrowing are often not a good comparison group for those borrowing. Accordingly, why some individuals choose to borrow is critical to understand, and yet also difficult if not impossible to fully understand. Also, merely observing that some characteristics of borrowers and non-borrowers are similar is not sufficient. Karlan, Harigaya & Nadel again observed that often the unobserved characteristics which may be the propelling factors for MFI programme participation are believed to be the most important. For example individuals with high entrepreneurial spirit are those believed to be beneficiaries of MFI programmes, these unobserved factors more often than not raise the issues of selection bias. When these selection issues are not dealt with properly, the observed difference in outcomes can be attributed to both the programme’s impact and the pre-existing differences between the two groups. The comparison between the two groups will yield the accurate programme impact only if the two groups have no pre-existing differences other than access to the product change being evaluated. Since we are evaluating the impact of access to microfinance and faced with the difficulty of observing the same individual in treated (beneficiary) and untreated (non-beneficiary) states, leads to the use of various population level treatment effects widely used in the biostatistics literature and applied in economics. Also, to deal with the problem of selection bias, then the Treatment Effect Model is employed. This is a version of the Heckman sample selection model, which estimates the effect of an endogenous binary treatment. Again to reduce or eliminate any pre-existing difference between the beneficiaries and non-beneficiaries, the respondents are made up of women who are engaged in agro-processing business.
The objective here was to estimate a model that efficiently estimates the impact of access to microfinance (Q) on household consumption expenditure on basic needs (C); a proxy for poverty which is also the cost of basic needs consumed by the household (food and non-food items). Bank Singapore’s graduates

To do so, let i index individual households and q. denote an access to microfinance indicator, equal to 1 if a participant of a microfinance programme received a microfinance loan, and equal to 0 if otherwise. To describe the treatment effect, two other variables are defined. Let C i0 denote the potential outcome that would occur when person i does not receive a loan ( q. = 0) and c1 the potential outcome when she receives a loan (Q. = 1).
Clearly these are not both observed. One of them will be counterfactual, an outcome that would have occurred if the loan beneficiaries had not received the loan. This therefore calls for the use of an appropriate control group that mimics the treatment or beneficiary group.
We make a strong assumption that the consumption expenditure on basic needs could be used to determine if one is poor or not.
The first is a probit regression predicting the probability of treatment (Q). The second is a linear regression for the outcome of interest (C.) as a function of the “treatment” variable, controlling for observable confounders.
It is assumed that the error terms (s and ц ) are jointly normally distributed and a maximum likelihood methods of estimation was used. Because corr(Q,s) ф 0 then appropriate Instrumental Variable(s) (IV) must be found to solve the problem. In which case IV must be correlated with Q( but not correlated with C..

Tags: , , , ,