Results and Discussions

A. Propensity Score Estimations

The first stage in the propensity score matching is to model the probability of being a KB borrower. With that purpose, we include variables that influence the likelihood of borrowing from Khushhali Bank. The rationale behind this is that, if a variable influences participation but not the outcome, there is no need to control for differences with respect to this variable in the treatment versus the control groups. Likewise, if the variable influences the outcome but not the treatment likelihood, there is no need to control for that variable since the outcome will not significantly differ in the treatment versus the control groups. Variables that affect neither treatment nor the outcome are also clearly unimportant. Therefore, only those variables that influence both the treatment and the outcome are needed for the matching and are included in the probit model from which we derive the propensity score.

Table 10 [ PDF 45.9KB | 1 page ] shows the propensity score estimations by probit regression method. In general, the model is well specified with high Likelihood Ratio Chi-squared and Pseudo R-squared coefficients.

The probit estimations show the relatively good fit of the model, expressed by Chi-squared and Pseudo R-squared statistics. Among the covariates, holding some office in the neighborhood, having a literate adult in the household, owning land, and living in remote villages positively affect the probability of borrowing from Khushhali Bank. More importantly, being female and having an experience of borrowing from sources other than Khushhali

Bank greatly increases the odds of borrowing from Khushhali Bank. In contrast, other variables, such as age of household head, dummy for being rural households, adult numeracy, and dummy for being poor do not strongly explain the participation in the Khushhali Bank lending program.

Remote villages have the greater probability of having a KB program, indicating that Khushhali Bank is doing well in reaching the remote areas. On the other hand, dummy for being poor is insignificant, indicating that Khushhali Bank's lending toward the poor is neutral.

After deriving the propensity score, we need to ensure that there is enough common support. This is done by discarding treated individuals with a propensity score lying outside the range of propensity scores for individuals in the control group.

Table 11: Description of the Estimated Propensity [ PDF 45.9KB | 1 page ]

The final number of blocks is 11. This number of blocks ensures that the mean propensity score is not different for treated and controls in each block. The balancing property is satisfied. The table below shows the inferior-bound, the number of treated, and the number of controls for each block.

From Table 12 [ PDF 42.7KB | 1 page ] one can see that the distribution of KB borrowers and nonborrowers along the propensity score is not similar. To the extent that there are substantial differences between treatment group and comparison group, there should be little overlap. There is some overlap between KB borrowers and nonborrowers when the propensity score is between 0.09 and 0.18, implying that the two groups share the same characteristics in these brackets, but there is little overlap over the higher propensity score brackets. As mentioned earlier, a higher propensity score basically means a higher probability of borrowing from Khushhali Bank.

B. Matching and Impact Estimations

Once the common support requirement is fulfilled, we can carry out the matching for all pairwise combinations. Various propensity score matching methods have been proposed in the literature as a means to identify a comparison group. Each of these methods implies a tradeoff between quality and quantity of the matches. The most intuitive matching method is the Nearest-Neighbor (or one-to-one) matching, which matches each treated observation to a control observation with the closest propensity score. In the case of the Nearest-Neighbor method, all treated units find a match. However, it is obvious that some of these matches are fairly poor, because for some treated units the Nearest-Neighbor may have a very different propensity score. The Radius Matching and Kernel Matching methods offer a solution to this problem. With Radius Matching, each treated unit is matched only with the control units whose propensity score falls in a predefined neighborhood of the propensity score of the treated unit. If the dimension of the neighborhood (i.e., the radius) is set to be very small, it is possible that some treated units are not matched because the neighborhood does not contain control units.

Another matching method, stratification, consists of dividing the range of variation of the propensity score in intervals such that, within each interval, treated and control units have on average the same propensity score. For practical purposes, the same blocks identified by the algorithm that estimates the propensity score can be used. Then, within each interval in which both treated and control units are present, the difference between the average outcomes of the treated and the controls is computed. The ATT of interest is finally obtained as an average of the ATT of each block with weights given by the distribution of treated units across blocks.

Once each treated observation is matched to a control group observation, the difference between the outcomes for the treated versus the control observations is computed. This procedure is usually implemented with replacement; that is, each treated individual has one match, but a control group individual may be matched to more than one treated individual. Once each treated unit is matched with a control unit, the difference between the outcome of the treated units and the outcome of the matched control units is computed. The ATT is then obtained by averaging these differences. Dehejia and Wahba (1998) found that matching with replacement improves the performance of the match and is less demanding with regard to the common support requirement.

Table 13 [ PDF 45.7KB | 1 page ] presents the results of the comparison of Khushhali Bank borrowers with nonborrowers, matched by the Nearest-Neighbor Matching method. The first two columns of the table show the number of treated (KB borrowers) and matched nontreated (non-KB borrowers). The ATT is displayed for different outcome variables classified by type of functions. The last two columns provide standard errors and corresponding t-statistics of the ATT estimations. In all, 1,204 households that borrowed from the Khushhali Bank were matched with 663 non-KB borrowers. Those nonborrowers who did not match with corresponding borrowers are eliminated.

The results in Table 13 show that with regard to household production and consumption or income poverty—MDG 1—Khushhali Bank has had positive impact on agricultural production and, in particular, animal-raising activities. On crop production, KB clients use more pesticides and possess more farm equipment. The point estimate of pesticide use shows that KB borrowers on average are 24% likelier to use pesticides than nonborrowers. The implications, however, of the higher use of pesticides should be interpreted with care.

While pesticides enhance agricultural outputs, they are toxic substances that the poor often are not trained to use properly. The inappropriate use of pesticides could lead to negative outcomes on other MDGs, e.g., farmer's health (MDGs 4-6) and the environment (MDG 7), which will be discussed in a later section.

On ownership of farm equipment, KB borrowers possess higher-value farm equipment, valued in Pakistan Rupees (PRs), 12,814² higher on average than nonborrowers. Also, rental income from farm equipment is PRs 882.5 higher.

KB membership has the strongest impact on animal raising. The value of livestock, sales, and profits were all highly positive and statistically significant for animal raising. The value of livestock owned by KB clients is on average PRs 17,705 higher than that of nonborrowers. Also, KB borrowers have PRs 6,494.2 higher profit on livestock than that of nonborrowers. This shows the strong positive effect of KB borrowing on a farmer's poverty situation.

Regarding consumption expenditures, KB members appear to spend less than nonborrowers—particularly on food—although the difference is not statistically significant. This may point to the fact that agriculture loans have led to increased on-farm food production, leading to borrowers spending less on food and more on non-food items than nonborrowers.

Borrowing from Khushhali Bank is not associated with higher durable assets and higher transfers from outside, which is consistent with the fact that the loans are largely for home enterprises, agriculture, and non-agriculture. With regard to non-agricultural enterprise activities, KB clients reported higher values of associated variables (e.g., value of assets, sales, and profit), none of which are significantly higher than nonborrowers. Contrary to most MFIs, since 74% of KB clients in the survey are agricultural households, many in remote areas, this finding simply reflects the low level of nonfarm activities in the communities.

The findings show that clients' households, both rural and urban, invested immediately in animal raising, which requires minimal skills and land. Contrary to lending programs of other MFIs where microenterprise financing is the main use of funds, the impacts of KB's program on microenterprise is yet to be significant, given the limited cycle of the loans and the predominantly agricultural households in the client profiles.

The results show that KB clients do not have significantly longer working hours in crop production and animal raising. The results suggest that there was a shift in labor use from non-agricultural activities to agricultural-related activities.

Since child labor is widespread in Pakistan, we were interested in assessing the impact of the microfinance program on child labor. The evidence in this respect is inconclusive. Similar to the pattern of adult labor use, there is an increase the in working hours of children in animal raising along with a decline in child working hours in non-agricultural activities. The impact of KB borrowing on the education of children (MDG 2) is not significant on any of the education indicators. Impacts on the empowerment of women (MDG 3) also are not significantly visible. Apart from the significantly higher possibility (10%) of KB-client household women using more contraception, other indicators are not significantly different. Women having a say in schooling matters, women having a say in healthcare, and the incidence of domestic violence are, in fact, better in nonborrowing households, although not significantly. The limited cycle of loans may explain these results.

With regard to healthcare (MDGs 4-6), KB membership has a positive impact on the possibility of households seeking medical treatment. The results also showed that the possibility of KB members having funds to pay for medical treatment is significantly higher than that of nonmembers.

With regard to the environment (MDG 7), the significantly higher amount of pesticide use among KB borrowers should be approached with caution. Among poor farmers, particularly illiterate farmers, inappropriate use of pesticides often leads to negative health outcomes and environmental consequences. The public sector, together with the Khushhali Bank may need to explore providing information or training programs such as the Integrated Pest Management Program to ensure that the loan does not lead to worsening health outcomes and negative environmental consequences.

In summary, we found that the microfinance program positively impacted some incomegenerating activities, such as agriculture and animal raising. We mostly failed to confirm the beneficial impact of Khushhali Bank on other outcome variables such as household durables, consumption, savings, education, and healthcare expenditures. These findings can be interpreted in two different ways. On the one hand, they might indicate that the impact of Khushhali Bank lending on households' well-being (i.e., consumption, education, healthcare, and labor) is quite modest. On the other hand, it is possible that most of the KB borrowers are going through an initial phase of capital accumulation, when their increased incomegenerating capacities have not translated into increased consumption, education, and healthcare expenditures.

C. Impact of the Lending Program on Poor Households

To assess the impact of Khushhali Bank on the poor, we carried out the same analysis on a subset of poor households in the sample. (To distinguish the poor from the non-poor, we used the national poverty threshold of 878.6 PRs, deemed necessary to provide 2,350 calories per day). The results are shown in Table 14 [ PDF 46.9KB | 1 page ].

In total, 749 poor households who borrowed from Khushhali Bank were matched with 439 non-poor, non-KB borrowers. Table 14 shows that impacts of the KB lending program on poor households are essentially similar to the impacts on clients in general. The Khushhali Bank membership positively affects animal raising and agricultural activities. While the level of significance is similar for animal raising, the level of significance was less in agricultural production. With regard to non-agricultural enterprise, durable assets, consumption, education, healthcare, and empowerment, the impacts were not significant. Borrowing from KB led, notably, to a significant increase in the time spent on raising animals and a reduction in the time spent on non-agricultural enterprise for both adult and children.

D. Robustness of Results

The Nearest-Neighbor Matching method, which we have used so far, is not the only method of assessing the average treatment effect on the treated. Other methods such as Radius Matching, Kernel Matching, and Stratification Matching have advantages and disadvantages; therefore, their joint consideration offers a way to assess the robustness of the results.

On the other hand, the smaller the neighborhood, the better the quality of the matches. With Kernel Matching, all treated are matched with a weighted average of all controls with weights that are inversely proportional to the distance between the propensity scores of treated and controls.Table A.2 [ PDF 46.3KB | 1 page ] in the Appendix shows the results for the whole dataset obtained by the Kernel Matching method. Standard errors are obtained by bootstrapping, using 50 replications. The 1,204 borrower households are matched with 1,652 nonborrower households.

Compared to Nearest-Neighbor Matching, the results of the Kernel Matching method show the stronger impact of Khushhali Bank on the borrowers. The basic impacts in general remain unchanged; as before, there is strong impact on animal raising and agricultural production. In addition, there is a statistically significant positive impact of the microfinance program on the value of household durable assets, sales, and profits from non-agricultural family business. The other results remain essentially unchanged; thus, the findings are robust.

Results of Stratification Matching, presented in Table A.3 [ PDF 46.1KB | 1 page ] in the Appendix, show that borrowing from Khushhali Bank had a positive impact mainly on income-generating activities—agriculture, animal production, non-agricultural business, and healthcare. We find no significant impact of the lending program on household consumption, assets, savings, education, and female empowerment. The evidence on the other outcome variables such as adult and child labor is mixed.

E. Comparison of Results to an Earlier Study

As our study used the same dataset as, but adopted a different research methodology from, a previous study to assess the impacts of the KB program, in this section we compare the results of the two studies. In the previous study conducted by Montgomery (2005), it was assumed that the selectivity bias was addressed through the survey design and a series of OLS, and Logit Regression was run to assess impacts of different variables. In this study, we employed propensity score matching to correct for self-selection bias and estimate the impact of the Khushhali Bank lending program. The results on selected comparable variables are reported in Table 15 [ PDF 44.3KB | 1 page ].

The comparison of the results presented in Table 15 shows that while there are some common findings on some variables' impacts, the degree of significance differs. At the same time, the findings on certain variables' impacts are directly opposite. The Montgomery study found, for instance, that impacts on agricultural sales to a third party, sales and profit from microenterprise, health expenditure per capita, and amount of pesticide use are positive and significant at the 1% level while the P-Score estimates show the impacts in the same direction but at the lower significant level of 5%. The regression estimate showed positive but not significant impacts on sales and profit of livestock, while the PSM showed a significant impact at the 1% level. On the other hand, OLS estimates showed a significantly positive impact at the 5% level on “women have say in healthcare” and “health expenditure per capita” while PSM showed opposite but not significant results.

Montgomery (2005) reported that both access to and participation in the program had strong positive impacts on all variables tested for income generation. She showed that as the number of loan cycles increased, assets in terms of amount of land cultivated, value of farm equipment, and hours of tractor use increased significantly. There are two reasons, however, which suggest that those results were overestimated. Firstly, the loan size offered by KB is generally limited to 10,000 PRs. Given the small size of loan, even with repeated borrowing, it is questionable if it would have generated large amounts of income to purchase additional land and heavy farm equipment. Secondly, since 70% of the KB clients in the survey were first-time borrowers or borrowed only once, the larger size of cultivated land and the higher value of farm equipment are most likely not the results of the borrowing from KB but reflections of the higher level of wealth among repeat KB clients.

Overall, the results of the PSM estimate showed a lower degree of impacts of KB's lending programs on the households. Given the fact that a selectivity bias does exist as shown in Table 9 in the earlier section, the comparison of the results confirm that running OLS and Logit Regression on the survey sample without correcting for selectivity bias led to overestimating of the impacts of KB's program on the households.

Download this Discussion Paper [ PDF 198.6KB| 31 pages ].

Post a Comment

We welcome your feedback on this publication. Post a comment. ADBI is not obliged to acknowledge or publish comments and may abridge or edit them before web posting.

Comment(s)

There are [0] comment(s) for this entry. Post a comment.