1.4 Objetivos de la Investigación: 7
2.2.3. Riesgos Laborales 19
The sample design for the papers slightly varies. For instance, the focus of analysis for the papers on the impact of finance on productivity of smallholder agricultural farmers and the impact of finance on the welfare of smallholder farm households in Ghana are on farmer level whereas the analysis for the papers on market participation of smallholder farmers in Northern Ghana as well as the integrated soil fertility management (ISFM) and productivity of smallholder farmers are focused on farm level.
The Impact of Finance on Productivity of Smallholder Agricultural Farmers:
This study evaluates the impact of access to finance on the productivity of smallholder farmers. For the data, we applied a combination of convenient, stratified and proportional sampling techniques. The
16 record consisted of 27,856 farmers across the Northern Region of Ghana who participated in the ACVF project. These farmers are into farming of selected staple crops, namely maize, rice, soyabean and groundnut. Following a four-stage approach, the population was classified into different subgroups or strata, then the final subjects were proportionately selected at random from the different population groups or strata. First, we selected seven communities from each of the 22 districts representing the Northern Region of Ghana. The choice of these seven communities was influenced by the number of beneficiary farmers within a community. This brought the total number of communities to 154.
The second stage was to randomly select a sample of 1,700 farmers from the 154 communities for data collection. After data editing and cleaning of outliers and various inconsistencies, we had 1,564 farmers. To achieve the objective for this study, we focused on maize farmers bringing the data to 1,152 farmers. The maize farmers were chosen because finance (production credit) was allocated to only them. At the third stage, we categorised the data into two separate groups, namely maize farmers with access to finance (treatment group) and maize farmers who are financially constrained (control group). The data indicated a total number of 154 farmers with access to finance and 998 farmers who are financially constrained. At the fourth stage, 398 maize farmers were sampled from the 998 farmers who were financially constrained for analysis.
To ensure robustness in the checking of results, the study used a second control group. The second group of farmers are the non-beneficiary group of AVCF. Data for the non-beneficiary group was also collected on farmers in selected communities within the Northern and Brong Ahafo (BA) regions, with a total number of 295 farmers and 200 farmers respectively. The selected areas for this group are within the same agro-ecological zone as the beneficiary group. They share similar agricultural practices as well as community and socioeconomic characteristics. The selection of the areas or communities of these farmers was influenced by the fact that they are remote from communities where Government provides agricultural extension services to farmers and by the fact that non-governmental organizations (NGOs) are not there to provide agricultural services. After data cleaning and editing, the total number of smallholder farmers who did not benefit from the AVCF project intervention was recorded at 466. Of this number of smallholder farmers, the data revealed 366 of them were maize farmers.
17 The Impact of Finance on the Welfare of Smallholder Farm Households in Ghana:
This study evaluates the impact of access to finance on the welfare of smallholder farmers’ household. For the data, we applied a combination of convenient, stratified and proportional sampling techniques. The record consisted of 27,856 farmers across the Northern Region of Ghana who participated in the ACVF project. These farmers are into the farming of selected staple crops, namely maize, rice, soyabean and groundnut. Following a four-stage approach, the population was classified into different subgroups or strata, then the final subjects were proportionately selected at random from the different population groups or strata. First, we selected seven communities from each of the 22 districts representing the Northern Region of Ghana. The choice of these seven communities was influenced by the number of beneficiary farmers within a community. This brought the total number of communities to 154.
The second stage was to randomly select a sample of 1,700 farmers from the 154 communities for data collection. After data editing and cleaning of outliers and various inconsistencies, we had a number of 1,564 farmers who either owned a maize farm or groundnut farm or soyabean farm or rice farm only, or a farmer owning a combination of farms of different crops. At the third stage, we categorised the data into two separate groups, namely farmers with access to finance (treatment group) and farmers who are financially constrained (control group). The data indicated a total number of 176 farmers with access to finance and 1,388 farmers who are financially constrained. At the fourth stage, 208 farmers were sampled from the 1,388 farmers for purposes of matching and estimation.
To ensure robustness in the checking of results, the study used a second control group. The second group of farmers are the non-beneficiary group of AVCF. Data for the non-beneficiary group was also collected on farmers in selected communities within the Northern and Brong Ahafo (BA) regions, with a total number of 295 farmers and 200 farmers respectively. The selected areas for this group are within the same agro-ecological zone as the beneficiary group. They share similar agricultural practices as well as community and socioeconomic characteristics. The selection of the areas or communities of these farmers was influenced by the fact that they are remote from communities where Government provides agricultural extension services to farmers and by the fact that non-governmental organizations (NGOs) are not there to provide agricultural services. After data cleaning and editing, the total number of smallholder farmers who did not benefit from the AVCF project intervention was recorded at 466. Of this number of smallholder farmers, 233 farmers were sampled for estimation.
18 Market Participation of Smallholder Farmers in Northern Ghana:
This study examines the determinants of market access and market participation of smallholder farmers. The focus of analysis for this paper is on the farm level. The data for this study consisted of two groups of farmers (the AVCF group and the non-AVCF group). Both groups consisted of farmers who are farming in either one or a combination of the following crops: maize, rice, soyabean and groundnut. The total number of the beneficiaries of AVCF consists of 27,856 farmers across the Northern Region of Ghana.
To obtain the sampled data, we applied a combination of convenient, stratified and proportional sampling techniques. This was made possible following a two-stage approach: we first selected seven communities from each of the 22 districts representing 154 communities from the Northern Region of Ghana. In the second stage, we randomly selected 1,700 farmers from the 154 communities. After data cleaning and editing we had data on 1,608 farmers. The total number of plot farms owned by the 1,608 smallholder farmers was recorded at 2,724 plot farms covering all the four crops. Of the total number of 2,724 plot farms, 1,163 were for maize plot farms, 698 were for groundnut plot farm, 645 soyabean plot farms and 218 rice plot farms.
The data for the non-AVCF group was collected on farmers in selected communities of the Northern and Brong Ahafo (BA) regions with a total number of 295 farmers and 200 farmers respectively. The selected communities for this survey have in common the same agro-ecological zone and areas where agricultural practices and socioeconomic characteristics of the farmers are similar to the beneficiary group. After data cleaning and editing, the total number of smallholder farmers was recorded at 484. The total number of plot farms owned by the 484 smallholder farmers stood at 701 plot farms covering all the four crops. The data reveals 369 maize plot farms only, 261 groundnut plot farms only, 44 soyabean plot farms and 27 rice plot rice.
Integrated Soil Fertility Management (ISFM) and Productivity of Smallholder Farmers:
Two groups of farmers were sampled for identifying the impact of ISFM on farm-level productivity. The beneficiary group is the group that participated in the AVCF project and received the ISFM training and the non-beneficiary group is the group of farmers who did not participate in the AVCF project and so did not receive the ISFM training. Both groups consisted of farmers who were farming in either one or a
19 combination of the following crops: maize, rice, soyabean and groundnut. This paper focuses on farm level analysis.
A three-stage sampling approach was used in the case of the beneficiary group. A combination of convenient, stratified and proportional sampling techniques was used. The rationale was to segment the entire population into different subgroups or strata, then randomly select the farmers proportionately from the different population groups or strata. In the first stage, we selected seven communities from each of the 22 districts used for this study in the Northern Region of Ghana. The selection of a community was influenced by the size of farmers who received ISFM training. That is, farmers were selected from communities with a large number of farmers who received ISFM training. The second stage was to randomly select a sample size of 1,700 farmers from a total number of 154 communities. After data editing and cleaning of outliers and various inconsistencies, we had 1,608 of farmers who either owned a maize farm or groundnut farm or soyabean farm or rice farm only, or farmers who owned a combination of farms of different crops. The data shows a total number of 2,724 farms covering all the four crops. Of the total number of 2,724 farms, 1,163 were for maize farm plots only and 698 were for groundnut farm plots. In the third stage, we sampled 292 maize farms and 209 groundnut farms from the 1,163 maize farms and 698 groundnut farms respectively. The choice of maize and groundnut for analysis is due to the fact that we did not have enough observations for soyabeans and rice for matching.
Data for the non-beneficiary group were also collected on farmers in selected communities within the Northern and Brong Ahafo (BA) regions with a total number of 295 farmers and 200 farmers respectively. The selected areas for this group are within the same agro-ecological zone as the beneficiary group. They share similar agricultural practices as well as community and socioeconomic characteristics. The selection of the areas or communities of these farmers is influenced by the fact that they are remote from communities where Government provides agricultural extension services to farmers and by the fact that NGOs are not there to provide agricultural services. After data cleaning and editing, the total number of smallholder farmers who did not benefit from the AVCF project intervention was recorded at 484. The total number of plot farms owned by the 484 smallholder farmers stands at 701 plot farms covering all the four crops. From the 701 plots farms, there are 369 maize plot farms and 261 groundnut plot farms only. The data was limited to maize plot farms and groundnut plot farms due to the limited observations for soyabeans and rice.
20 1.6.3. Data Analysis
For the study, various estimation methods were used in line with the above stated objectives and research questions. Research question 1 was estimated using the instrumental variable (IV), question 3 was estimated using the double hurdle model (DHM), while research questions 2 and 4 were estimated following propensity-score matching (PSM) techniques.
We also estimated the impact of access to finance on agricultural productivity of smallholder farmers. Access to finance (production credit) to the smallholder farmers was not carried out at random; thus, a problem of possible selection bias could arise. According to Heckman (1979), non-randomisation fuels the problem of selection bias caused by either an individual’s self-selection or selection methods used by the project implementing agencies. The problem of selection bias may also emerge because of unobservable or missing characteristics. Under this prevailing circumstance, an ordinary least squares (OLS) estimator will not produce a result which is consistent due to bias in the selection of farmers with access to finance and also endogeneity of access to finance (Baum, 2006). To control for such biases and to produce an estimation that is consistent, we adopted the IV estimations technique. The IV with exclusion restrictions is preferred as it establishes causality and addresses selection bias (Cuddeback, Wilson, Orme & Combs-Orme, 2004; Bushway, Johnson & Slocum, 2007).
This study also adopted the DHM proposed by Cragg (1971) to access the determinants of smallholder farmers’ market participation and intensity. The DHM is considered much more flexible than the Tobit model, which assumes the factors influencing market participation and intensity of participation are jointly made. Meanwhile, the application of the DHM assumes that the adoption decision and the intensity are separable. The factors influencing the decision to participate on the market are not the same factors influencing the extent of market participation (Mather et al., 2013).
According to Hacking (1988), Burtless (1995) and Loux (2015), randomisation (experimentation) is generally viewed as the most robust evaluation approach as it controls for selection bias. It is described as a highly reliable evaluation technique that helps to easily assign the difference in the average outcome to the treatment. However, the design of AVCF, which was the project of study, was not randomised. Meanwhile, the characteristics of the AVCF project offered the opportunity to use the propensity scores matching the PSM estimation model as the alternative approach for the estimations. The choice of the PSM estimation is to control for selection bias and endogeneity. According to Rosenbaum and Rubin
21 (1983), propensity score is “the conditional probability of assignment to a particular treatment given a vector of observed covariates”. In other words, the propensity score is the probability of participating or receiving a treatment depending upon the pool of the observed characteristics. On that note, data was collected from both the beneficiary and the non-beneficiary and non-equivalent control groups who answered the same questionnaire for the purpose of estimating the average treatment effect. Similar studies by Jalan and Ravallion (2003) used PSM to estimate “the benefit incidence of an antipoverty program in Argentina” and while Wendimu, Henningsen and Gibbon (2016) also adopted the PSM model for the estimation of “the effects of compulsory participation in sugarcane outgrowers schemes in Ethiopia”.
1.7. AN OVERVIEW OF THE DANISH INTERNATIONAL DEVELOPMENT AGENCY’S