Microcredit and Rural Household Outcomes: Evidence from Habru Woreda’s Smallholder Farmers, Ethiopia

Theoretical literature review

Microcredit has been widely recognized as an essential instrument for enhancing rural household income and reducing poverty by providing financial access to those excluded from formal markets (Ukpe et al., 2016). Credit in general serves as a bridge between income and expenditure, enabling households to invest in productive activities (Beckman, 1962; Reddall & Miller, 1977). Yet, rural credit markets in developing countries are often constrained by collateral requirements, high transaction costs, and information asymmetry, which limit poor households’ ability to borrow from formal institutions (Binswanger & Rosenzweig, 1986; Joan et al., 2022). In Ethiopia, where smallholder farmers account for over 90% of agricultural production, financial constraints remain a critical barrier to adopting modern technologies and improving productivity (Singh et al., 1986). In response, the Ethiopian government formalized microfinance through Proclamation No. 40/1996, enabling institutions such as the Amhara Credit and Saving Institution (ACSI) to provide deposit, savings, and loan services across rural areas (Balcha & Tamare, 2017; Chirkos, 2014; Ramanaiah & Gowri, 2011). While microcredit is theoretically expected to promote income diversification, asset building, and risk management, evidence also suggests that benefits are unevenly distributed, with the “upper poor” often gaining more than the “very poor” (Gashayie & Singh, 2016). Consequently, although microfinance institutions are vital for rural transformation, their effectiveness depends on household characteristics, institutional design, and accessibility (Boltana et al., 2023; Mersland & Strøm, 2010).

Empirical literature review

Empirical studies on the issue of microcredit utilization shows a mixed outcome on rural household’s income and welfare. In Ethiopia, Siyoum et al. (2012) and Lawin (2018) reported that while borrowing improved consumption smoothing and technology adoption, effects on efficiency, profit, and income varied. Geleta et al. (2018) found that beneficiaries of microfinance in Cheliya District had significantly higher incomes than non-users, and Boltana et al. (2023) showed credit programs improved household food security and calorie intake. Similarly, Berhanu et al. (2021) and Dhillon & Moncur (2023) highlighted that microcredit enables smallholders to protect assets, maintain consumption, and improve productivity, thereby enhancing income. Yet, Mandy (2023) observed that while entrepreneurial households benefited from higher income and investment, overall poverty reduction effects were modest. Beyond Ethiopia, Obisesan (2013) found that microcredit positively influenced farm output in Nigeria, while Joan et al. (2022) stressed the ongoing importance of informal lenders for rural households. More recent evidence in the Amhara region indicates that ACSI significantly improved household income (Beyene & Fentaw, 2023; Kassegn & Endris, 2021), though coverage gaps persist.

Description of the study area

Restricted to Ethiopia’s North Wollo Zone, the research was conducted in the Habru District. There are 491 kilometers between it and Addis Ababa. Its total landmass coverage of 1239.79 square kilometers makes it one of the zone’s largest districts. Habru’s economy is based on an unstable rain-fed livelihood system, making it the district in the Amhara Region most vulnerable to drought and food crisis. The district is home to 202,009 people in total, with 99,442 women and 102,567 men, according to the Habru woreda Administration Office (office, 2023). Two urban and thirty-five rural kebeles make up the district. Afar Region borders the district on the east, Guba Lafto borders it on the west, the Alewuha River borders it on the north, and the South Wollo Zone (Ambasel district) borders it on the south as shown by the following Figure 1.

Figure 1. Map of the study area.

Note: 'This figure/table has been reproduced from Kassegn, A., & Endris, E. (2022). Factors affecting loan repayment rate among smallholder farmers got loans from the Amhara Credit and Saving Institution: In the case of Habru District, Amhara Regional State, Ethiopia. International Area Studies Review, 25(1), 73-96. https://doi.org/10.1177/22338659211040993.

The grant is obtained under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).

Type, source and method of data collection

This research used only primary data collected between 20 January and 24 February 2023. Structured questionnaires were used to gather the main data on assets, expenses, farm size, and socioeconomic factors that were thought to limit the use of microcredit.

Sampling method and sample size determination

In the study area, there were 49,880 small holder farm households. To select the representative sample of the target population (small-scale farm household heads), the Cochran (1963) formula of sample size determination for infinite population was used and which is seen as follows.

Where, ${\mathrm{n}}_0$ = sample size, e = the level of precision at 95% of confidence interval.

To obtain the above representative sample, the study was applied three stage probability sampling method. In the first stage, systematic random sampling was used to select seven (6) out of 35 rural kebeles in the woreda. In this stage, the 5^th kebele, was selected through lottery method and the rest was selected systematically via skimming factor. In the second stage, based on the evidence of clients from ACSI in the study area, rural farm households in the selected kebeles was stratified based on the microcredit program as users and non-users and finally, the sample was selected from each stratum in the selected kebelles through proportional allocation.

Model specification

Logistic regression model

Due to the dichotomous nature of the dependent variable-having or not having access to microcredit services-a univariate binary model called the logit model was chosen. The dependent variable was stated as follows:

Where y* is a latent variable expressing the quantity of microcredit loans contracted by small holder farmer i from micro finance institutions, ACSI. This quantity was a function of the small farmers’ household and his farm characteristics (X_i), as expressed in the form of

The cumulative logistic probability function was specified as:

Where Pi is the probability that a farmer was using Microcredit given his household and farm characteristics X_i, and ${\mathrm{\beta }}_0$ and ${\mathrm{\beta }}_{\mathrm{i}}$ are the parameters to be estimated. After a simple mathematical manipulation of the above equation, equation 2, the odds ratio can be written as (Wooldridge, 2004)

By taking the natural logarithms of the odds ratio with the consideration of the expected errors made during the research process, the Logit model becomes

Therefore, the coefficient of the logit model ${\mathrm{\beta }}_{\mathrm{i}}$ , presents the change in the log of the odds associated with a change in the explanatory variables, ${\mathrm{X}}_{\mathrm{i}}.$

Propensity score matching

The program’s features as well as the type and caliber of the data that are currently accessible determine which impact evaluation models should be used. There were no baseline socioeconomic data available in the study area, and it was challenging to locate a non-participant to produce a true counterfactual. Propensity score matching was found to be useful for generating a counterfactual group (comparison) in situations where random assignment is not feasible in order to solve this problem. It was used to calculate how microcredit affected the study area’s small-scale farmers’ revenue and expenses. One of the more useful techniques for estimating impacts with cross-sectional data is the PSM method. Additionally, it provides a useful method for analyzing how credit affects a variety of income indicators, and the outcomes rely less on the operational forms of econometric models. PSM’s key strategy is to maintain as many of the variables as possible constant, ensuring that credit accounts for the difference in income and spending between households that use credit (treated) and those that do not (counterfactual or control group) (Heinrich et al., 2010).

The propensity score is a conditional probability estimator that can be applied to any discrete model, including logit and probit, since they produce comparable outcomes (Caliendo & Kopeinig, 2008). By assuming the logistic distribution of the sample mean and variances, the logit model was utilized in this study to estimate the propensity scores. The model of propensity scores is represented as follows:

Where D = (1, 0), the indicators of improvement in income, it was the binary variable whether a small holder farmers were using credit or not using credit, (1= yes, 0 = otherwise) where ${\mathrm{X}}_{\mathrm{i}}$ is a vector of pre-treatment covariate propensity score, to ensure that matching estimation is done on treatment and control clients that are as similar as possible for $\left(\mathrm{i}\right)$ effective comparison. As a result given a population of units denoted by $\left(\mathrm{i}\right),$ if the propensity score $\mathrm{p}\left({\mathrm{X}}_{\mathrm{i}}\right)$ is known as average treatment effect (ATE), it can be estimated as

Where ATE is the average treatment effect $,{\mathrm{Y}}_{1\mathrm{i}}$ and ${\mathrm{Y}}_{0\mathrm{i}}$ is the potential outcome for the two counterfactual situations of the treatment and control small farm households respectively. $\mathrm{P}\left({\mathrm{X}}_{\mathrm{i}}\right)$ , is the propensity score, $\mathrm{D}$ is the small farm household’s variable, where $\mathrm{D}=1$ if the household participated in microcredit program and 0 otherwise.

Description and expected sign of working variables

The descriptions and expected sign of the working variables were described in the following Table 1.

Description of the respondents

The study was carried out using primary data gathered from 385 smallholder farmers in the north wollo zone’s Habru woreda, of which 126 had taken out microcredit loans from the Amhara Credit and Saving Institution and 259 had not. According to the descriptive statistics, microcredit loan users had greater mean family sizes and access to irrigation schemes, but non-users had higher mean household head ages. The dependency ratio was higher in non-user households than in user households, but a higher proportion of non-users were married. On the other hand, female-headed farm households were more common in the non-user group, whereas the mean number of male farm household heads was higher for users. Furthermore, social organization memberships accounted for 57% of the household heads. More over the full description of the respondent’s characteristic is given by the extended data at Zenodo: https://doi.org/10.5281/zenodo.13829443.

Determinants of microcredit loan utilization

The estimates of the marginal effects from the logit model, as shown by Table 2 below indicates that, in comparison to their male counterparts, female household heads had greater limitations when it came to using microcredit loans. When all other variables are held constant, households headed by men are 24.5% more likely than households headed by women to use microcredit. This result is consistent with the research that Chamboko and Guvuriro (2022) conducted. The likelihood of using microcredit is negatively impacted by the age of the head of the household, and this effect is statistically significant at the 95% confidence level. More specifically, the likelihood of taking out microcredit loans falls by 0.9 percentage points for every year that the head of the household gets older. This outcome agrees with the conclusions of the research that Domanban (2024) conducted.

Family size was another significant factor that positively correlated with the use of microcredit. This implies that the likelihood of obtaining formal financial credit increases by 8 percent for every additional family member added, all other things being equal. This can be explained by the fact that, in comparison to smaller families, larger families are able to engage in more self-directed farming activities and provide more family labor for production, which results in higher income. This result is consistent with the earlier theory and the findings of Kiros and Meshesha (2022). Education also had a beneficial impact because farmers’ access to formal credit institutions can be influenced by their level of literacy. It is believed that farmers with higher levels of literacy would know more about government facilities and the market (Alemayehu, 2020; Ameh & Lee, 2022). The last positive effect was that having access to irrigation schemes increased the likelihood that households would use microcredit by 13.5 percentage points.

At the one percent significance level, the frequency of extension services is another variable that has a significant impact on rural households’ use of microcredit participation. The hypothesis posited that households receiving social work services more frequently would be more knowledgeable about emerging technologies, strategies for boosting output and efficiency, and the advantages and disadvantages of formal credit. This hypothesis is supported by the coefficient of this variable in the current investigation. The results of earlier research, including Alemayehu (2020) and Fonke (2021), which claim that providing better extension services encourages smallholder farmers to apply for microcredit, also corroborate this conclusion.

Participation in formal microcredit programs by rural households is negatively impacted by dependency ratio. Ceteris paribus, for every additional dependent member in the household, the probability of participation decreases by 31.5 percent. This study is also in line with research done by Ding and Abdulai (2020) & Ferede (2012).

The possibility of utilizing microcredit services was found to be positively and significantly correlated with households’ experience of shocks, remittance income, and improved perception of group lending. Precisely speaking, these variables raised the likelihood of obtaining microcredit by 13.9%, 15.8%, and 12.8%, in that order. Additionally, at the ninety percent significance level, the data showed a positive and significant relationship between farm size and loan usage. A one-hectare increase in farm size is associated with a five percent increase in the use of microcredit loans, according to the positive coefficient. Larger farms are probably more capital-intensive and therefore have a higher demand for loan financing because they need more labor and inputs for production. This result is consistent with the study that Ameh and Lee (2022) recently conducted.

The participation of rural households in microfinance institutions (MFIs) was found to be significantly impacted, at the 5 percent significance level, by the distance from the credit source, as previously hypothesized. If all other variables remain unchanged, the marginal effect shows that the likelihood of participating in the microfinance program falls by 0.6 percent for every kilometer added to the distance traveled. The result of the study conducted by Alemayehu (2020) and the outcome of this study is consistent. In comparison to households in closer proximity, the results indicate that those residing further away from credit sources have lower participation rates in microfinance initiatives. To increase rural households’ access to credit services, it is crucial that microfinance institutions (MFIs) have a broad network of branch offices or agents to reach out to them, even in remote areas.

The impact of microcredit use on household asset and expenditure

Propensity score matching (PSM) was used in the study to calculate how microcredit loans affected annual household expenses and assets. A logit model that found multiple significant explanatory variables and had a pseudo R-squared of 0.4011, indicating a good model fit, was used to estimate the propensity score. The propensity score overlap between the treatment (microloan recipients) and control (non-recipients) groups is represented by the common support region, which was found to be between 0.02596857 and 0.98411727. This indicates that six treated households were left out of the matching exercise because they had propensity scores that fell outside of this range.

A kernel density distribution of the propensity scores for the treatment and control groups was plotted in the study to further investigate the shared support. A significant region of overlap between the two groups was found in the kernel density plot, as illustrated in Figure 1, suggesting the absence of a serious common support issue. Propensity score matching was employed by the researchers to compare the outcomes of loan recipients and non-recipients with similar observed characteristics, allowing them to estimate the impact of microcredit loans on household annual expenditure and asset. More accurate program impact estimates are produced with this method, which also helps to mitigate potential selection bias as shown by the following Figure 2.

Figure 2. Kernel density plot of propensity scores.

To make sure that there was no significant difference in the mean of the estimated propensity scores between the treated and control households within each block, the study separated the sample into a maximum of five blocks. This suggests that the matching procedure was effective in forming evenly distributed groups for the impact analysis that followed. Figure 3 below shows the propensity scores’ graphical distribution. The propensity score overlap between the treatment and control groups is displayed in this visualization, which supports the suitability of the matching strategy even more.

Figure 3. Graphical distribution of the propensity scores.

Estimation of the average treatment effect on the treated

The intensity of the matching process is more important in propensity score analysis than the sheer amount of output. By evaluating the balancing property and conducting balance tests, this quality can be evaluated. The estimated propensity scores in this study satisfied the balancing property. All the covariates, or explanatory variables, across all the blocks used in the analysis had the same distributions in households with the same propensity scores. The researchers used a Mantel-Haenszel test statistic to explore the results’ sensitivity in more detail. The absence of hidden bias is demonstrated when the $\mathrm{\Gamma }$ value of this test is 1, indicating that the test statistic’s bounds match the base case. The process of estimating and interpreting the average treatment effect on the treated (ATT) is a crucial output in impact evaluation studies. For the households who actually received the loans, this metric captures the microcredit program’s primary policy-relevant impact. The ATT estimates and their interpretation are shown in the following Table 3. This illustrates how policy makers’ primary concern in impact evaluation studies is the estimation and interpretation of the average treatment effect on the treated.

Strong empirical evidence of a statistically significant impact of the microcredit program on household asset and expenditure status is provided by the kernel matching results. The program has improved participating households’ expenditure levels, as indicated by the positive value of the average treatment effect on the treated (ATT). On the other hand, a negative ATT value indicates that the study area’s microcredit utilization has resulted in the depletion of household assets.

Statistically, the analysis accounted for the variations in demographic and asset distribution traits between participant and non-participant households. The results show that the program has raised the spending of households involved by 1.712% and this outcome aligns with (Choudhury et al., 2017; Phan, 2020; Phan et al., 2023; Yu et al., 2020) while their yearly asset possession has decreased by 8.737%, which is comparable to the earlier research carried out by Ahamad et al. (2021). This result seems to contradict the main goal of microfinance institutions (MFIs), as the small loans have failed to safeguard the wealth of rural households. However, the findings are statistically significant at the 5% level, as demonstrated by the kernel matching estimators (ATT = 1.712, t = 11.561 for annual spending; ATT = -8.737, t = 4.024 for household yearly asset) with bootstrapped standard errors.