Impact of Credit on Technical Efficiency of Maize Producing Households in Northern Ghana Edward Martey1*, Alexander N. Wiredu1,2 and Prince M. Etwire1 1
CSIR-Savanna Agricultural Research Institute (CSIR-SARI), Tamale, Ghana Chair of Rural Development Theory and Policy, Institute of Agricultural Economics and Social Sciences in Tropics and Subtropics, University of Hohenheim, Stuttgart, Germany 2
*Correspondence: P.O. Box TL 52, Tamale, Ghana. Tel: 233-242-344-450. E-mail: [email protected]
; [email protected]
Selected Paper Prepared for Presentation at the Centre for the Study of African Economies (CSAE) Conference 2015, University of Oxford, March 22-24, 2015.
Abstract Production credit is essential in enhancing the technical efficiency and welfare of smallholder farmers in Africa. This paper evaluated the impact of credit on smallholders’ technical efficiency using cross-sectional data from 233 maize-producing households in northern Ghana. Due to the exogenous assignment of credit and assumption of homogeneity in farm technologies, the propensity score matching analysis was used to compare the average difference in technical efficiency between farmers with credit and those without credit. The results show that credit impacted positively on farmers’ technical efficiency. Provision of credit enhanced timely purchase and efficient allocation of factor inputs to produce the maximum output. Income and age also played an important role in reducing smallholders’ technical inefficiency. The major recommendation of this study is that, the credit programme of agricultural interventions should target several resource poor farmers in order to enhance efficiency gains. Financial institutions must collaborate with agricultural and farmer development projects to facilitate credit delivery to smallholder farmers. Keywords Impact, credit, technical efficiency, northern Ghana, propensity score matching
1. Introduction Ghana’s agricultural sector plays an integral role in the generation of employment and income for most households in rural areas. It contributed about 26% to the total gross domestic product in 2011 inspite of the discovery of oil and
gas (MoFA, 2011). The sector however lags behind the service sector in recent times due to a fall in the cocoa, forestry, and logging sub-sectors (ISSER, 2011). Report from the Ministry of food and Agriculture (MoFA) has revealed that about 80% of the population in northern Ghana depends on subsistence agriculture with very low productivity and farm income (MoFA, 2011), largely due to over-reliance on rain-fed agriculture under low farm input conditions. Low soil fertility has further worsened the plight of smallholder farmers who lack the capacity to invest in soil improvement technologies (MoFA, 2010). Depletion of soil nutrient is a threat to agricultural sustainability and poverty reduction. Additionally, inefficiencies in fertilizer distribution networks limit access, and increase the cost of fertilizer in farming communities (FAO, 2005). Developmental efforts in addressing agricultural problems have focused on improvement in land productivity whilst ensuring environmental sustainability. Among the strategies employed in the improvement of productivity and efficiency include the promotion of accelerated adoption of improved crop varieties, increased restoration of soil nutrients through the use of organic fertilizers, inorganic fertilizers, and the adoption of improved soil and water conservation technologies (Oduol et al., 2011). The Alliance for a Green Revolution in Africa Soil Health Project (AGRA SHP) 0051 for instance, is an intervention implemented by Savanna Agricultural Research Institute (SARI) of the Council for Scientific and Industrial Research (CSIR) aimed at contributing to poverty reduction and food security in northern Ghana. The components of the project included technology transfer, capacity building and credit support. The credit component was very important because, productivity and overall welfare of farmers in developing countries is often greatly affected by lack of access to credit (Deb and Suri, 2012). It is intended to assist smallholder farmers’ access production input and subsequently impact positively on their technical efficiency (TE)2. Literatures on the impact of an intervention on smallholder farmers’ technical efficiency have been documented with different estimation methods depending on the underlying assumptions. Whilst some have used the propensity score matching methods, other studies have predicted impact by using the Stochastic Production Frontier (SPF). The traditional two stage estimation approach where efficiency is estimated in stage one and its determinants are identified in stage two by regressing predicted efficiency scores, obtained in stage one, on some explanatory variables have been criticized. Ideally, the efficiencies obtained from first stage are assumed to be normal,
The project was implemented in 2009 and extended to 2014 with financial support from AGRA Technical efficiency reflects the ability of a firm to obtain maximum output based on a given set of inputs (Farrell, 1957) 2
independent and identically distributed. The efficiency scores derived are not identically distributed in the second stage estimation using Ordinary Least Squares (OLS) leading to inconsistent results (Pitt and Lee, 1981). Secondly, Tung (2013) also suggested the use of Data Envelope Analysis (DEA) at the first stage to calculate the technical efficiency scores and subsequently employ the Tobit model in the second stage. Alternatively, Simar and Wilson (2007) suggested using a bootstrap method to estimate bias-corrected efficiency scores during the first step due to serial correlation of the DEA efficiency estimates. In the second stage, Truncation regression is used as supposed to the Tobit model proposed by other authors. Sossou et al. (2014) and Reifschneider et al. (1991) have also used the stochastic frontier truncated-normal with conditional mean model to analyze the determinants of farmers’ technical efficiency. The model allows capturing the double effect of the variable of interest as production input and as determinant of TE of a production system. The model was specifically used by Dinar et al. (2007) to assess the impact of extension access on farms TE.
Asante et al. (2014) estimated the impact of adoption of the New Rice for Africa (NERICA) on the technical efficiency of smallholder farmers in Ghana using cross sectional data. By assuming that impact is heterogeneous across the population, the propensity score matching method was used to estimate the average treatment effect on the treated (ATT). Adoption of NERICA rice varieties was found to have a positive and significant impact on technical efficiency of rice producing households in Ghana. Abate et al. (2013) evaluated the impact of participation in agricultural cooperatives on technical efficiency of smallholder farmers using data from household survey in Ethiopia. The maximum likelihood estimation method was used to estimate the stochastic production frontier and technical inefficiency model. By assuming exogenous cooperative formation, they employed the propensity score matching to compare the differences in average technical efficiency between treatment and control groups. The study concluded that agricultural cooperatives enhance members’ efficiency by easing access to productive inputs and facilitating extension linkages. Hormozi et al. (2012) also adopted the maximum likelihood estimation method to evaluate the impact of mechanization inputs and cultivation systems on the productivity and technical efficiency of rice production in the Khuzestan province in the southwestern part of Iran. They established a great variation in the levels of efficiency, which ranged from 0.15 to 0.99 with a mean of 0.67. They used the strong correlation between the mechanization index and technical efficiency to justify the impact of mechanization on the efficiency of rice producers. Ayaz et al. (2011) employed the maximum likelihood estimation approach to evaluate the impact of
institutional credit on production efficiency of farming sector using Faisalabad district in Pakistan as a case study. Based on the analysis, they found that institutional credit increases production efficiency by 14% and a mean efficiency score of 0.84. In Benin, Sossou et al. (2014) used the stochastic frontier truncated-normal with conditional mean model to study the impact of credit allocation behaviour of farmers. Per the analysis, they concluded that farmers who invest credit in the purchase of pesticides, herbicides, and fertilizers recorded a significant impact of 0.86% in TE. Some studies (Ayaz, 2011; Tan et al., 2010; Mouelhi, 2009; Sidhu et al. 2008; Bashir et al. 2007 and Olagunju, 2007) have equated the coefficient of the treatment variable in the inefficiency model to the average impact. However, the present study extended the frontiers of the argument by using the efficiency scores obtained from both parametric and non-parametric efficiency estimation to compute the ATT by employing the PSM method. The SPF function model was employed by this study to measure the technical efficiency of sampled farm households, as it is effective in estimating the efficiency score of households taking into account factors that are beyond the control of individual producers (Coelli et al., 2005; Kumbhakar and Lovell, 2000). To check the consistency of the SPF estimates, the DEA approach was also used to generate the efficiency scores at the first stage followed by the PSM approach to determine the impact of the intervention on technical efficiency following the approach of Bernard et al. (2008), Francesconi and Heerink (2010), and Godtland et al. (2004).
2. The Agricultural Credit Programme (ACP) in Northern Ghana Northern Ghana has about 7 million hectares of arable land of which 70% is available for agricultural production. The land is generally flat with soils which are predominantly lateritic with less than 0.5% organic matter, thus making the soils inherently poor in fertility. This situation is aggravated by very harsh and unfavourable climatic conditions of short and erratic rainfall patterns (MoFA, 2010). This has led to a consistent decline in agricultural productivity and a widespread poverty over the years, especially in rural communities. As part of efforts to address the problem of low soil fertility, the Integrated Soil Fertility Management (ISFM) technology was advocated by the Alliance for a Green Revolution in Africa (AGRA). AGRA has demonstrated its commitment to improve the health of the soils in Northern Ghana by funding the Soil Health Project 005. Complementary to the promotion of ISFM, was the credit programme managed by the Centre for Agricultural and Rural Development (CARD). The credit
facility managed by CARD was intended to promote household food security, microfinance, and rural development in northern Ghana. The credit facility is in the form of chemical fertilizers, agrochemical, and quality seeds in addition to tractor ploughing. These inputs are procured prior to the start of the cropping season and stored in the warehouse of CARD. The placement of the programme was largely determined by CARD based on the accessibility and the deprived nature of communities engaged in food crop farming system. As part of the selection criteria, the soil fertility was considered to be crucial, followed by the enterprising nature of the target group members, and participation in preseason training organized by the farm management and extension section of CARD. Participation in the credit programme was voluntary but the ultimate decision to grant the loan to beneficiary farmers was based on recommendation to the manager for final decision on loan approval. The credit programme was therefore exogenously assigned to not random farmers. The total amount of credit allocated was usually determined at the beginning of the year and disbursed based on groupings. The process also included rigorous monitoring system conducted primarily by the district coordinators with support from the credit officers. The monitoring system involves on-site visits to farms to provide advisory services to ensure that good agronomic practices are being practiced. Credit repayments are made in-kind with bags of harvested maize. The repayment is done immediately after harvesting at a central location in the beneficiary community and subsequently conveyed to the head office of CARD. At the end of the harvesting season, all outstanding loans are declared delinquent and more intensive measures are used by the credit officers and district coordinators to retrieve the outstanding loans.
3. Methodology 3.1. Measuring Technical Efficiency A production system that achieves a maximum attainable quantity of output from a given inputs is described as technically efficient (Aigner et al., 1977). Parametric and non-parametric methods are the approaches widely applied in the estimation of technical efficiency. In order to check for consistency based on the efficiency scores generated, the two approaches were both employed. The efficiency scores obtained from both approaches were then compared to establish similarity or variance.
The parametric approach applies econometric methods of either deterministic or stochastic modeling. The deterministic model regards all deviations in output as technical inefficiency effects regardless of the fact that, deviations in output could be beyond the control of the producer (Onumah, 2010). The SFP allows for estimation of the household efficiency score by accounting for factors beyond the control of each producer. Additionally, it helps to understand the factors that determine technical inefficiency of farm households. The functional form adopted by the present study is the Cobb-Douglas production function as proposed by Battese and Coelli (1995). The equation is specified as follows:
𝐼𝑛𝑍𝑖 = 𝛽0 + ∑ 𝛽𝑗 𝐼𝑛𝑋𝑗𝑖 + 𝑉𝑖 − 𝑈𝑖 , 𝑖 = 1,2,3 … . . 𝑁
Where 𝑍𝑖 represent the quantity of maize output for the 𝑖𝑡ℎ farmer, 𝑋𝑗𝑖 is the vector of 𝑗𝑡ℎ inputs associated with the 𝑖𝑡ℎ farmer, 𝛽𝑗 is a vector of unknown parameters to be estimated. The 𝑉𝑖 is the random error assumed to be independently and identically distributed as 𝑁(0, 𝜎 2 ). The 𝑈𝑖 is the non-negative (𝑈𝑖 ≥ 0) inefficiency error term. The condition that 𝑈𝑖 is non-negative ensures that all observations lie on or below the stochastic production frontier (Aigner, Lovell and Schmidt, 1977; Coelli et al., 2005; Onumah et al., 2010). Following from Battese and Coelli (1995), the technical inefficiency effect is defined as:
𝑈𝑖 = 𝐺𝑖 𝛿 + 𝑊𝑖
Where 𝐺𝑖 is a vector of explanatory variables associated with the technical inefficiency effect which could include socioeconomic and farm management characteristics. 𝛿 is a vector of unknown parameters to be estimated and 𝑊𝑖 are random variables such that 𝑊𝑖 ≥ 𝐺𝑖 𝛿. Equations (1) and (2) were estimated using the maximum likelihood single-stage estimation procedure to obtain the technical efficiency scores.
The computation of TE in the econometrics literature is based on either input-oriented or output-oriented analysis. According to FAO (2003), the choice between the two depends on the objective of either input minimization or output maximization, without changing any of the other elements. The input-oriented measure was used in the estimation of the TE. The non-parametric data envelope analysis was also used to generate the efficiency scores for
the impact analysis based on the assumption of constant return to scale 3. The DEA creates a non-parametric piecewise surface over the data (Coelli et al. 2005). Assuming there are data on N inputs and M outputs for each of I firms. The column vectors of inputs and outputs are denoted by 𝑥𝑖 and 𝑞𝑖 , respectively for the i-th firm. The Nx1 input matrix, X, and the Mx1 output matrix, Q, represent the data for all I firms. By using the duality in linear programming, an envelopment form of this problem can be derived such that; 𝑚𝑖𝑛𝜃,𝜆 𝜃, 𝑠𝑡 −𝑞𝑖 + 𝑄𝜆 ≥ 0,
𝜃𝑋𝑖 − 𝑋𝜆 ≥ 0, 𝑎𝑛𝑑 𝜆 ≥ 0
Where 𝜃 is the efficiency score of the i-th firm and 𝜆 is a Ix1 vector of constants. The value of 𝜃 must satisfy the restriction: 0 ≤ θ ≤ 1. If θ = 1; it indicates that the firm is on the production frontier and is technically efficient. On the other hand, if θ ≤ 1, then the firm is technically inefficient (Coelli, 2005).
3.2. Measuring Impact Impact evaluation has been estimated using different approaches that ranges from simple to more sophisticated methods. Some studies use the differences in mean outcomes of treatment and control groups. In some cases, simple regression of an outcome equation with a treatment variable as one of the explanatory variables has been conducted. However, such methods have been criticized due to their failure to effectively solve the problem of self-selection and selection on unobservable associated with sampling and data collection (Imbens, and Wooldridge, 2009; Heckman and Vytlacil, 2005; Lee, 2005; Imbens, 2004; Rosembaum, 2002; Heckman and Robb, 1985; Rosembaum and Rubin, 1983; Rubin, 1974). According to Alston and Pardey (2001) and Salter & Martin (2001), establishing a viable counterfactual, attributing the impact to an intervention, and coping with long unpredictable lag times are the three (3) main challenges in conducting impact evaluation. Technically, the counterfactual4 analysis enables evaluators to attribute cause and effect between interventions and outcomes. The key challenge in impact evaluation is that the counterfactual cannot be directly observed and must be approximated with reference to a comparison group. On the other hand, the choice of a specific estimation method depends largely on the assumptions made and the theoretical underpinning.
The CRS assumption states that a given and proportionate increase in all inputs in the long run will result in an increase in output in the same proportion. 4 The ‘counterfactual’ measures what would have happened to beneficiaries in the absence of the intervention, and impact is estimated by comparing counterfactual outcomes to those observed under the intervention.
The treatment effect estimation approach was used by this study to determine the impact of credit on technical efficiency due to its ability to produce consistent estimates of impacted outcomes (Imbens & Wooldridge, 2009). It was assumed that the smallholder farmers had two hypothetical potential technical efficiency outcomes, 𝑍, given access and non-access to credit (T) such that 𝑍 = 𝑍0 𝑖𝑓 𝑇 = 0 𝑎𝑛𝑑 𝑍 = 𝑍1 𝑖𝑓 𝑇 = 1. The average treatment effect for a randomly selected household is expressed as: 𝐴𝑇𝐸 = 𝐸(𝑍1 − 𝑍0 )
Given the treatment status, the average treatment effect on the treated (ATT) which measures the impact of credit on those individuals who had access to the credit (i.e. T=1) is given as: 𝐴𝑇𝑇 = 𝐸((𝑍1 − 𝑍0 )|𝑇 = 1)
However, these parameters are not observable, since they depend on counterfactual outcomes. Given the fact that the average of a difference is the difference of the averages, the ATT can be rewritten as: 𝐴𝑇𝑇 = 𝐸(𝑍1 |𝑇 = 1) − 𝐸(𝑍0 |𝑇 = 1)
The average outcome that the treated individuals would have obtained in the absence of treatment is not observed. In such types of casual inference, the estimation of treatment effects in the absence of information on the counterfactual poses an empirical problem known as the problem of filling in missing data on the counter-factual (Becker and Ichino, 2002; Dehejia and Wahba, 2002; Rosenbaum and Rubin, 1983). The challenge is to find a suitable comparison group with similar covariates and whose outcomes provide a comparable estimate of outcomes in the absence of treatment.
Self-selection among participants is one of the major challenges with ex-post evaluation studies given that participation may depend on the inherent characteristics (preference, expected benefit, skills) especially where there is no ex-ante study to assess the behavior of treated group before being treated (Abate, 2013). Credit was exogenously assigned to non-randomly sampled farmers by the AGRA SHP 005 even though farmers’ willingness to access the credit was one of the considering factors. Assignment of credit and technical efficiency may be interdependent such that technically efficient farmers will be more likely to receive credit whereas farmers with
credit may increase their technical efficiency. This situation results in selection bias since unobservable factors influence both the error terms of the outcome equation, and that of the credit choice equation (Greene, 2003). The use of OLS in estimation of the outcome equation will be biased. The Heckman two-step model has been used to address selection bias especially when the correlation between the error terms is greater than zero. The model however, depends on restrictive assumption of normally distributed errors. The instrumental variable approach has also been used with major limitation of difficulty in the identification of an appropriate instrument in the estimation. Additionally, both OLS and IV procedures tend to impose a linear functional form assumption implying that the coefficients on the control variables are similar for adopters and nonadopters (Ali and Abdulai, 2010). The Propensity Score Matching (PSM) technique was therefore used to control for the selection bias since it accounts for the differences between the outcomes of the treated and control groups (Francesconi and Heerink, 2010; Bernard et al., 2008 and Godtland et al., 2004). The PSM provides unbiased estimate through controlling for observable confounding factors and reducing the dimensionality of the matching problem (Becker and Ichino, 2002; Rosenbaum and Rubin, 1983).
3.2. Propensity Score Matching (PSM) Methods The PSM method was employed in the impact estimation of credit on smallholders’ technical efficiency based on the assumptions of conditional independence and the presence of a reasonable overlap of propensity scores (common support) (Winters, 2010). The method is intuitively attractive as it helps in comparing the observed outcomes of treated with the outcomes of the counterfactual control group (Heckman et al., 1998). It helps to evaluate programs that require longitudinal datasets using single cross-sectional dataset where the former does not exist. The PSM method basically matches observations of farmers with credit and non-credit farmers according to their predicted propensity of accessing credit (Rosebaum and Rubin 1983: Heckman et al., 1998; Smith and Todd, 2005; Wooldridge, 2005). In the first step, the conditional probability of credit (propensity score) was estimated using the Probit model by controlling for observed household characteristics. The second step is the estimation of the ATT using the matching methods. Detailed discussion on the estimation of the ATT through matching procedures can be obtained from Becker and Ichino (2002), Dehejia and Wahba (2002), Heckman et al. (1997), Rosenbaum and Rubin (1983), Smith and Todd (2005), and Todd (2006). The nearest neighbor, Radius, Local linear
and Kernel matching methods were both used to compute the impact estimate and the results compared to check for robustness. Technical efficiency was the outcome variable in the analysis.
The analysis was further subjected to test of unobserved heterogeneity. According to Becker and Caliendo, 2007; Keele, 2010; Rosenbaum, 2002; Rosenbaum and Rubin, 1983, unobserved heterogeneity (hidden bias) occurs when unobserved variables influence both the treated variable and outcome variable simultaneously. In the absence of experimental data, it is not possible to estimate the magnitude of selection bias (Rosenbaum, 2002). Secondly, the Rosenbaum bounds sensitivity analysis was used to evaluate the presence of unobserved heterogeneity when the key assumption is relaxed by a quantifiable increase in uncertainty.
4. Results 4.1. Descriptive Characteristics of Farm Households by Credit Status A summary of the characteristics of the selected farm households by treatment status is presented in Table 1. Based on the sampled households, about 26% were provided credit. The results show significant differences in the means of household size, proportion of educated members, membership of FBO, access to extension service, access to market, and distance to the nearest market among farmers with credit and non-credit farmers. The significant differences in the means show the non-random selection of the households thus the use of mean difference between the outcomes of the control and treated will be erroneous in terms of causal interpretation. Farmers with credit had relatively higher household size and extension access. However, the non-credit farmers had more educated members, higher market access, and travelled longer distance to the nearest market relative to the farmers with credit.
Table 1. Descriptive characteristics by treatment status and test of mean difference Variable Household Size
Educated household heads
Years of farming experience
Proportion of economically active persons
Proportion of educated members
Livestock ownership (Total Livestock Unit)
Member of FBO
Access to extension service
Distance to the nearest market
Access to market
Distance to the nearest agricultural office
Land fragmentation (Simpson index)
Adoption of improved varieties
Quantity of maize harvested Quantity of NPK used
Notes: Standard deviations are in parentheses. The t-statistics was used for two groups mean comparison *** p < 0.01, ** p < 0.05 and * p < 0.10
4.2. Technical Efficiency Score and Estimated Probability of Credit Access by Treatment Status The summary statistics of the impacted outcome variable (technical efficiency scores) derived from both the parametric and non-parametric approaches are presented in Table 2. The results show that about 26 percent of the households were provided with credit. It was observed that on the average, farmers with credit had significantly higher technical efficiency (0.62) compared to the non-credit farmers (0.53) at the 5 percent level of significance. The treated category households can therefore be described as more technically efficient. Consequently, the control group recorded higher minimum and maximum technical efficiency relative to the treated category. However, the standard deviation was almost the same across the sampled households. The mean efficiency scores derived from the DEA shows no significance difference between the treated and control groups. Comparatively, the TE score obtained from the SFM was relatively higher than that of the DEA.
Table 2. Technical efficiency score by credit status Indicators Treated (N=58)
Pooled Sample (N=223)
Technical Efficiency Scores from SFM Meana
Technical Efficiency Scores from DEA
Test of mean difference between treated and control is significant at 5% level (P chi2 Pseudo R2 Log likelihood
Coefficient 0.001 -0.432 -0.867 0.036 1.167 -1.123 -0.028 0.447 1.070 0.227 -0.513
Std. Error 0.007 0.282 0.506 0.026 0.294 0.273 0.017 0.225 0.508 0.235 0.454 223 58.050 0.000 0.227 -98.787
P>|z| 0.929 0.125 0.086 0.168 0.000 0.000 0.109 0.047 0.035 0.334 0.259
Land fragmentation was expected to influence credit both positively and negatively. According to Latruffe et al. (2013), land fragmentation may exacerbate conflicts regarding labor allocation on the farm as it takes time to travel from one plot to another while the labor force could be undertaking more productive tasks. In such cases, financial institutions may be reluctant to provide production credit especially where there is challenge with effective management of the farm plots. The probability of receiving credit among farm households with access to extension services is 117% higher than the household heads that do not have access to extension service. Extension of good agricultural practices plays a major role in stimulating farm level productivity which increases individuals’ chance of receiving credit. It was established per this study that access to market does not necessarily guarantee access to credit. Conditions that guarantee credit goes beyond market access especially in the formal systems of credit supply.
Participation in the AGRA SHP 005 increases the probability of receiving credit since the project provides credit support for smallholder farmers. Participant household heads have 45% chance of receiving credit relative to the non-participant household heads (Table 4). It is assumed that participant household heads have undergone training which guarantees them credit per the activities of the project. Proportion of off-farm income increases the
probability of accessing credit. Most financial institution requires some financial commitment before providing credit to these smallholder farmers. Farm households that earn off-farm income are able to overcome the financial requirement of the financial institutions thus increasing their chances of accessing credit. A balancing test6 was conducted to establish whether farmers with credit and non-credit farmers have the same distribution of propensity scores. The matching was restricted to the two categories with common support in the distribution of the propensity scores to improve the robustness of the estimate. The unmatched sample fails to satisfy the balancing property as there were significant differences between the means of access to extension service, market and proportion of off-farm income before the matching (Table 5).
Table 5. Balancing Test of Matched Sample Variables Experience of household head Educated household head Land fragmentation index Livestock ownership Access to extension service Access to market Distance to nearest market Participation in AGRA SHP Proportion of off-farm income Adoption of improved varieties Labour land ratio
Unmatched Sample Mean Treated Control 30.53 28.81 0.12 0.21 0.76 0.74 4.69 3.09 0.33 0.05 0.10 0.42 6.43 7.31 0.38 0.38 0.37 0.30 0.71 0.63 0.23 0.29
P > |t| 0.464 0.113 0.554 0.175 0.000 0.000 0.275 0.962 0.034 0.284 0.090
Treated 30.81 0.12 0.76 3.66 0.32 0.11 6.44 0.37 0.37 0.72 0.23
Kernel Matching Mean Control 33.26 0.12 0.74 2.76 0.31 0.10 5.59 0.34 0.35 0.72 0.20
P > |t| 0.390 0.942 0.596 0.214 0.945 0.879 0.409 0.736 0.621 0.984 0.475
However, results of the balancing test after matching indicated that there was no statistical difference in observed characteristics between the two groups (Table 5). Based on the result, it was concluded that the comparison was statistically valid. Results of balancing test from other matching methods (nearest neighbor and radius) are reported in Table A1 in the appendix. The balancing test was satisfied by all the matching methods. Table 6 presents the results of the Average Treatment effect on the Treated (ATT) using different matching methods. The standard errors of the impact estimates were calculated by bootstrap using 50 replications for each estimate. The results revealed that access to production credit had a significant positive impact on technical efficiency. For all the matching methods with the exception of local linear, positive impact was realized at the 5
Balancing test compares a simple means of household characteristics within the treatment group to the corresponding comparison groups created by the matching techniques before and after matching as a complement.
percent level of significance (Table 7). Farm households with access to credit reported 8.2% increase in their technical efficiency on average compared to the control farm households. In other to ensure consistency in the result, the Inverse Probability Weighting (IPW) and nearest neighbor method (NN) were used to estimate the ATT using efficiency scores from the DEA. The ATT computed using the efficiency scores from the DEA was similar to the results obtained from the SPF approach. By using the IPW and NN methods the results showed that farmers with credit are 8.2% and 8.8% technically efficient than non-credit farmers respectively.
Table 7. ATT Estimates of impact of credit on Technical Efficiency Computation of impact based on technical efficiency scores obtained from SPF model Matching Algorithm Coefficient (ATT) Bootstrap Std. Error Kernel 0.082 0.036** Radius 0.082 0.037** Nearest Neighbour 0.082 0.038** Local Linear -0.082 0.044* Computation of impact based on technical efficiency scores obtained from DEA Method Average Treatment Effect on Treated Robust Std. Error Nearest Neighbour 0.082 0.039** Inverse Probability Weighting 0.088 0.041** *** p < 0.01, ** p < 0.05 and * p < 0.10
The present study assumed conditional independence implying that credit is based on observable characteristics and that variables simultaneously influencing credit and technical efficiency are observable. The result of sensitivity analysis for both sided significance (positive and negative) for all the matching methods are presented in the appendix (Table 1B). For the purpose of discussion, the result from Kernel matching with the positive significance (sig+) was used. It was observed that the results were insensitive to any hidden bias that may undermine the significant positive impact of credit on technical efficiency centered on the matching methods.
5. Discussion 5.1 Determinants of Production and Technical Efficiency The study has shown variation in the technical efficiency of farmers with credit and those without credit. Farmers with credit were relatively more technically efficient than non-credit farmers. Given the technology currently used, farmers with credit and non-credit farmers were respectively producing maize at about 62 per cent and 53 percent of the potential frontier output. The implication is that in the short run, there is capacity for increasing technical efficiency in maize production by 38 percent and 47 percent respectively without necessarily varying the existing
input levels. Output of maize was determined by area under maize cultivation, seed, fertilizers and herbicides. Farmers in northern Ghana have high land-use intensity which is not complemented with proportional increase in productivity enhancing technologies such as fertilizer use. High cost of fertilizers mostly constrained smallholder farmers in terms of adoption and use which impacted negatively on output. The returns to scale value imply that, on the average, maize farms in northern Ghana are operating under increasing returns to scale. Hence, a percentage increase in all factors of production will result in 1.41percent increase in output, ceteris paribus. The result is consistent with the findings by Amos (2007) and Idiong (2007) who obtained returns to scale values of 1.26 and 1.57 respectively. Interventions that focus on increase in maize output must make long term investment in land productivity and seed. Farm households in the study area must diversify their income source in order to make commitment towards investment in farm technologies thus improving the household food insecurity.
Age, proportion of economically active members, access to production credit, access to market, per capita income and distance to nearest market determined the technical efficiency of farmers in the study area. Older farmers were more technically inefficient relative to the younger farmers. Younger farmers are more dynamic in terms of technology adoption that will impact on their efficiency. It is possible that older farmers rely on the experience which affects their willingness to adopt new technologies. The results confirm the findings of Shaheen, Sial, Sarwar and Munir (2011) and Onumah et al. (2010) who both suggested that younger farmers are more progressive and willing to implement new technologies and are therefore more technically efficient than older farmers. The use of household size may not necessarily reflect the true effect on technical efficiency therefore the proportion of economically active persons was rather used. Economically active members of households were more technically inefficient perhaps due to lack of investment of earned income in agriculture as well as insufficient time allocation to farming. Targeting of farmers with agriculture as principal occupation coupled with training will be more expedient to improve TE gains in northern Ghana. Farmers with reported access to credit have higher TE results. Credit enables farmers to invest in productivity enhancing technologies. As parts of the effort to ensure the adoption and utilization of fertilizers by farmers, Karimov, (2014) and Bozoglu and Ceyhan (2007) found a similar positive impact of credit on TE of farmers in Uzbekistan and Turkey respectively. Contrary to expectations, farmers with access to market were technically inefficient. The results imply that market access is not a sufficient motivation factor for farmers to effectively allocate resource for increase in maize output. The ability to overcome the barriers
to entry is considered to be crucial. Surprisingly, farmers that travelled longer distance to market were more technically efficient. Distance to market imposes a transaction cost that impacted farmers’ attitude towards the purchase and use of inputs. It is also possible that farmers were getting a relatively lower price for the inputs at a distant market. Finally, per capita income was not enough measure of farmers TE. High-earning income farmers may have competing needs which influence the allocation of resources. Allocation of resource towards agriculture is likely to increase efficiency and vice versa.
5.2 Impact of Credit on Technical Efficiency The results of the study showed that credit had a positive impact on technical efficiency of farmers in northern Ghana. Farmers’ technical efficiency increases by 8.2% with an increase in credit supply which is relatively low. Increase in TE can be achieved with increased participation in credit market through reduction in interest charges on credit. Providing credit to smallholder farmers will translate to increase in the use of agricultural technologies such as improved seed, fertilizers and herbicides ceteris paribus. Few skilled laborers exist but are expensive to hire. Most farmers are not able to afford mechanization services provided by private enterprises as a result of lack of funds. The opposite is true for farmers that have credit facilities. The efficiency gains per access to credit support the findings of Karimov (2014). In Ghana, Asante et al., (2014) established a 3.1% increase in technical efficiency for adopters of NERICA rice varieties. Abate (2013) also found that participation in agricultural cooperatives increases technical efficiency by 5.6% in Ethiopia. Chaovanapoonphol, et al., (2009) also obtained similar result where access to agricultural loans was found to significantly increase technical efficiency of rice farmers in the Upper North of Thailand. Deraniyagal (2001) reported a positive impact of technology accumulation on the technical efficiency of the Sri Lankan clothing and agricultural machinery industries.
6. Conclusion This paper assessed the impact of credit on the technical efficiency of smallholder maize farmers based on crosssectional household data from 223 farm households in northern Ghana. By assuming exogeneity in the assignment of the credit programme and non-random selection of farmers, the Propensity Score Matching techniques was employed to compare the average technical efficiency difference between treated and control farm households. Based on the results, it was deduced that in the estimation of ATT, technical efficiency scores generated from both the parametric and non-parametric methods respectively generate the same result.
The study showed that access to credit had a positive and significant impact on the technical efficiency of maizeproducing households in northern Ghana. More specifically, farm households with credit are able to generate maximum possible output from a given set of inputs used by at least 8.0 percent on the average. Credit enables farm households to overcome the difficulty in the purchase of inputs and adoption of technologies that increases productivity. Provision of credit especially in-kind as a policy tool is essential for enhancing the technical efficiency of smallholder farmers in Ghana. The findings of this study also suggests that policy to enhance farmers’ technical efficiency should focus on age, household structure, market access as well as interventions that impacts on the per capita income of household heads.
Acknowledgement The authors of this paper wish to express their profound gratitude to Alliance for a Green Revolution in Africa (AGRA) for the financial support for generation of the data through the AGRA Soil Health Project 005. The administrative support of the host institution, CSIR-SARI is acknowledged. Finally, the reviewers and technicians who worked tirelessly to review the paper and collect the data respectively are also appreciated.
References Abate, G. T., Francesconi, G. N., Getnet, K., 2013. Impact of agricultural cooperatives on smallholders’ technical efficiency: evidence from Ethiopia, Euricse Working Paper n. 50(13), 1-32. Aigner, D. J., Lovell, C. A. K., Schmidt, P., 1977. Formulation and Estimation of Stochastic Frontier Production Function Models. J. Econom. 6, 21-37. Ali, A. Abdulai, A. 2010. The adoption of genetically modified cotton and poverty reduction in Pakistan. Journal of agricultural economics vol.61, No1,2010, 175-192. Alston, J.M., Pardey, P.G., 2001. Attribution and other problems in assessing the returns to agricultural R&D. Agric. Econ. 25, 141–152. Amos, T. T., 2007. An Analysis of Productivity and Technical Efficiency of Smallholder Cocoa Farmers in Nigeria. J. Soc. Sci. 15(2), 127-133. Asante, B. O. Wiredu, A. N. Martey, E. Sarpong, D. B., Mensah-Bonsu, A., 2014. NERICA Adoption and Impacts on Technical Efficiency of Rice Producing Households in Ghana: Implications for Research and Development. Am. J. Exp. Agric. 4(3), 244-262. Ayaz, S., Hussain, Z., 2011. Impact of Institutional Credit on Production Efficiency of Farming Sector: A Case Study of District Faisalabad. Pak. Econ. Soc. Rev. 49(2), 149-162 Bashir, M.K., Zulfiqar, A.G., Sarfraz, H., Sultan, A.A., Khuda, B., 2007. Impact of credit disbursed by commercial banks on the productivity of sugarcane in Faisalabad district. Pak. J. Agric. Sci. 44(2), 361-363.
Battese, G.E., Coelli, T.J., 1995. A Model for Technical Inefficiency Effect in Stochastic Frontier Production for Panel Data. Emp. Econ. 20, 325-345. Becker, S., Caliendo, M., 2007. Sensitivity analysis for average treatment effects. Stata. J., 7, 71–83. Becker, S., Ichino, A., 2002. Estimation of average treatment effects based on propensity scores. Stata. J., 2, 358– 377. Bernard, T., Taffesse, A.S., Gabre-Madhin, E., 2008. Impact of cooperatives on smallholders’ commercialization behavior: evidence from Ethiopia. J. Agric. Econ. 39, 147–161. Bozoglu, M., Ceyhan, V., 2007. Measuring the technical efficiency and exploring the inefficiency determinants of vegetable farms in Samsun province, Turkey. Agric. Sys. 94, 649–656. Chaovanapoonphol, Y., Battese, G.E., Chang, H.S., 2009. The impact of agricultural loans on the technical efficiency of rice farmers in the Upper North of Thailand. Productivity, Efficiency, and Economic Growth in the Asia-Pacific Region. Berlin, Heidelberg: Springer-Verlag., 279-295. Coelli, T. J., Rao, D. S. P., O’Donnell, C. J., Battese, G. E., 2005. An introduction to efficiency and productivity analysis (2nd ed.). New York, USA: Springer Publishers. Coelli, T.J., 1995. Estimators and hypothesis tests for a stochastic frontier function: a monte carlo analysis. Journal of Productivity Analysis 6,247-268. Deb, R., Suri, T., 2012. Endogenous Emergence of Credit Markets: Contracting in Response to a New Technology in Ghana. J. Devt. Econ. forthcoming Dehejia, H.R., Wahba, S., 2002. Propensity score matching methods for non-experimental causal studies. The Rev. of Econ. Stats 84(1), 151-161. Deraniyagala, S., 2001. The Impact of Technology Accumulation on Technical Efficiency: An Analysis of the Sri Lankan Clothing and Agricultural Machinery Industries. Ox. Dev. Stud. 29(1), 101-114 Dinar, A., Karagiannis, G., Tzouvelekas, V., 2007. Evaluating the impact of agricultural extension on farms’ performance in Crete: a nonneutral stochastic frontier approach. Agric. Econ. 36(2), 135–146. Food and Agriculture Oraganization. 2005. Fertilizer Use by Crop in Ghana, Land and Plant Nutrition Management Service, Land and Water Development Division, Food and Agriculture Organization of the United Nations, Rome. Accessed on September 5, 2014. Available at http://ftp.fao.org/agl/agll/docs/fertuseghana.pdf Farrell, M.J., 1957. The measurement of productive efficiency. J. Royal Stat Soc. Series A (General) 120(3), 253– 290 Francesconi, G.N., Heerink, N., 2010. Ethiopian agricultural cooperatives in an era of global commodity exchange: Does organizational form matter? J.Afric. Econ. 20, 1–25. Godtland, E.M., Sadoulet, E., de Janvry, A. Murgai, R., Ortiz, O., 2004. The impact of farmer field-schools on knowledge and productivity: A study of potato farmers in the Peruvian Andes. Econ. Dev. Cult. Change. 53, 63–92. Greene, W.H. 2003. Econometric Analysis. New York: New York University. Heckman, J. Vytlacil, E., 2005. Structural Equations, Treatment Effects, and Econometric Policy Evaluation. Econom. 73 (5), 669-738. Heckman, J., Ichimura, H., Smith, J., Todd, P., 1998. Characterizing selection bias using experimental data. Econom. 66 (5), 1017-1098.
Heckman, J., Richard, R., Jr., 1985. Alternative Methods for Evaluating the Impact of Interventions. in Longitudinal Analysis of Labor Market Data, ed. James J. Heckman and Burton Singer, 156- 245. Cambridge; New York and Sydney: Cambridge University Press. Hormozi, M.A., Asoodar, M.A., Abdeshahi, A., 2012. Impact of mechanization on technical efficiency: A case study of rice farmers in Iran. Proc. Econ. Fin. 1, 176 – 185. Idiong, I. C., 2007. Estimation of Farm Level Technical Efficiency in Small-scale Swamp Rice Production in Cross River State of Nigeria: A Stochastic Frontier Approach. W. J. Agric. Sci. 5, 653-658. Imbens, G. W., Wooldridge, J. M., 2009. Recent Developments in the Econometrics of Program Evaluation. J. Econ. Lit. 47(1), 5-86. Imbens, G., 2004. Nonparametric estimation of average treatment effects under exogeneity: Rev. Econ. Stat. 86, 429. Karimov, A.A., 2014. Factors affecting efficiency of cotton producers in rural Khorezm, Uzbekistan: Re-examining the role of knowledge indicators in technical efficiency improvement. Agric. F. Econ. 2, 1-16. Keele, L., 2010. An overview of rbounds: An R package for Rosenbaum bounds sensitivity analysis with matched data. Accessed September 2014 at, http://www.personal.psu.edu/ljk20/rbounds%20vignette.pdf. Kumbhakar, S.C., Lovell, C.A.K., 2000. Stochastic Frontier Analysis. New York, NY: Cambridge University Press. Latruffe, L., Piet, L., 2013. Does Land Fragmentation affect Farm Performance? A Case Study from Brittany. Factor Markets Working Paper. No. 40. Lee Myoung,_J., 2005. Micro-Econometrics for Policy, Program and Treatment Effects. Advanced Texts in Econometrics. Oxford University Press. Ministry of Food and Agriculture, MoFA (2011). Annual Report”. Policy Planning Monitoring and Evaluation Directorate. Accra, Ghana Ministry of Food and Agriculture, MoFA. (2010). Agriculture in Ghana: Facts and Figures (p. 53), Accra, Ghana. Mouelhi, R.B.A., 2009. Impact of the adoption of information and communication technologies on firm efficiency in the tunisian manufacturing sector. Econ. Mod. 26, 961–967 Oduol, J.B.A., Binam, A.J.N., Olarinde, L, Diagne, A., Adekunle, A., 2011. Impact of adoption of soil and water conservation technologies on technical efficiency: Insight from smallholder farmers in Sub-Saharan Africa. J. Dev. Agric Econ. 3(14), 655-669 Olagunju, F.I., 2007. Impact of credit use on resource productivity of sweet potatoes farmers in Osun-State, Nigeria. J. Soc. Sci. 14(2), 175-178. Onumah, E. E., Brümmer, B., Hörstgen-Schwark, G. H., 2010. Elements Which Delimitate Technical Efficiency of Fish Farms in Ghana. J. W. Aquac. Soc. 4(4), 506-518 Pitt, M.M., Lee, M.F., 1981. The measurement and sources of technical inefficiency in the Indonesian weaving industry. J. Dev. Econ. 9, 43–64. Reifschneider, D., Stevenson, R., 1991. Systematic departure from the frontier: a framework for the analysis of firm inefficiency. Int. Econ. Rev. 32, 715–723. Rosenbaum, P.R., 2002. Observational Studies (2nd ed.). New York, NY: Springer.
Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score in observational studies for casual effects. Biom. 70 (1), 41–55. Rubin, D. B., 1974. Estimating Causal Effects of Treatments in Randomized and Non-Randomized Studies. J. Educ. Psyc.66, 688-701. Salter, A.J., Martin, B.R., 2001. The economic benefits of publicly funded basic research: A critical review. Res. Pol. 30, 509–532. Shaheen, S., Sial, M. H., Sarwar, G., Munir, R., 2011. Nexus between Human Capital and Technical Efficiency of Cauliflower Growers in Soon Valley, Punjab: A Panel Data Analysis. Int. J. Hum. Soc. Sci. 1(14), 129-135. Sidhu, R. S., Vatta, K., Kaur, A., 2008. Dynamics of institutional agricultural credit and growth in Punjab: contribution and demand-supply gap. Agric. Econ. Res Rev. 21, 407-414. Smith, J., Todd, P., 2005. Does matching overcome LaLonde’s critique of non-experimental Estimators? J. Econom. 125(1-2), 305-353. Sossou, C.H., Noma, F.,Jacob A. Yabi, J.A., 2014. Rural Credit and Farms Efficiency: Modelling Farmers Credit Allocation Decisions Evidences from Benin. Econ. Res. Int. 1-8 Tan, S. Heerink, N., Kuyvenhoven, A., Qu, F., 2010. Impact of land fragmentation on rice producers’ technical efficiency in South-East China. NJAS –Wag. J. L. Sci. 57, 117–123 Todd, P. E. (2006). ‘Matching estimators’, http://athena.sas.upenn.edu/~petra/papers/mpalgrave2.pdf.
Winters, P. Salazar, L., Maffioli, A., 2010. Designing Impact Evaluations for Agricultural Projects. ImpactEvaluation Guidelines. Technical Notes No. IDB-TN-198 Wooldridge, J.M., 2005. Instrumental estimation of the average treatment effect in the correlated random coefficient model. Department of Economics, Michigan State University, Michigan.
Appendix Table 1A: Balancing Test Variables Experience of household head Educated household head Land fragmentation index Livestock ownership (TLU) Access to extension service Access to market Distance to nearest market Participation in AGRA SHP Proportion of off-farm income Adoption of improved varieties Labour land ratio
Unmatched Sample Mean Treated Control 30.53 28.81 0.12 0.21 0.76 0.74 4.69 3.09 0.33 0.05 0.10 0.42 6.43 7.31 0.38 0.38 0.37 0.30 0.71 0.63 0.23 0.29
P > |t| 0.46 0.11 0.55 0.17 0.00 0.00 0.28 0.96 0.03 0.28 0.09
Nearest Neighbor Mean P > |t| Treated Control 30.81 33.50 0.55 0.12 0.12 1.00 0.76 0.72 0.41 3.66 2.73 0.18 0.32 0.30 0.84 0.11 0.07 0.56 6.44 5.72 0.50 0.37 0.32 0.56 0.37 0.36 0.74 0.72 0.73 0.90 0.23 0.22 0.81
Table 1B. Rosenbaum Bounds Sensitivity Analysis for Hidden Bias Kernel Matching Radius Matching Critical Value of Hidden Bias (г) Sig+ SigSig+ Sig1.00 0.005 0.005 0.005 0.005 1.05 0.002 0.011 0.002 0.010 1.10 0.001 0.022 0.001 0.022 1.15 0.000 0.040 0.000 0.040 1.20 0.000 0.068 0.000 0.068 1.25 0.000 0.108 0.000 0.108 1.30 0.000 0.159 0.000 0.159 1.35 0.000 0.220 0.000 0.220 1.40 0.000 0.291 0.000 0.291 1.45 0.000 0.368 0.000 0.368 1.50 0.000 0.448 0.000 0.448 1.55 0.000 0.527 0.000 0.527 1.60 0.000 0.603 0.000 0.603 1.65 0.000 0.673 0.000 0.673 1.70 0.000 0.735 0.000 0.735 1.75 0.000 0.790 0.000 0.790 1.80 0.000 0.836 0.000 0.836 1.85 0.000 0.874 0.000 0.874 1.90 0.000 0.904 0.000 0.904 1.95 0.000 0.929 0.000 0.929 2.00 0.000 0.947 0.000 0.947
Radius Mean Treated Control 30.81 33.51 0.12 0.12 0.76 0.74 3.66 2.72 0.32 0.31 0.11 0.10 6.44 5.54 0.37 0.33 0.37 0.36 0.72 0.72 0.23 0.20
Nearest Neighbour Sig+ Sig0.005 0.005 0.002 0.010 0.001 0.022 0.000 0.040 0.000 0.068 0.000 0.108 0.000 0.159 0.000 0.220 0.000 0.291 0.000 0.368 0.000 0.448 0.000 0.527 0.000 0.603 0.000 0.673 0.000 0.735 0.000 0.790 0.000 0.836 0.000 0.874 0.000 0.904 0.000 0.929 0.000 0.947
P > |t| 0.34 0.91 0.65 0.18 0.97 0.95 0.38 0.71 0.72 0.98 0.45
Local Linear Sig+ Sig0.005 0.005 0.002 0.010 0.001 0.022 0.000 0.040 0.000 0.068 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Γ measures the degree of departure from random assignment of treatment or a study free of bias. The sensitivity analysis is for both-sided significance levels