The study area

The study was conducted in Delanta and Dawnt districts (locally called woreda) of the Amhara region, Central Highlands of Ethiopia. The study districts are located between 11°20’ to 11°50’ North latitude and 38°40’ to 39°30’ East longitude (Fig. 1). Delanta district is subdivided into one urban (Wogeltena) and 34 rural kebeles (lower administrative units of Ethiopia). Dawnt district is also divided into 1 urban (named as Chet) and 14 rural kebeles. Out of these, 18 and 7 kebeles in Delanta and Dawnt districts, respectively, belong to Dega agro-climatic (cool, humid highlands) zone.

Fig. 1: Map of the study districts.

Map of the study districts showing location of Amhara region (Light Grey) in Ethiopia (White), Delanta and Dawnt districts (Lime Dust) in Amhara region (Light Grey) map and Sample kebeles (light yellow) in Dawnt and Delanta districts (Source of the data/shape file: Ethiopian Central Statistical Authority, 2007).

As shown in Fig. 2, the study districts have a complex topography with wide relief differences ranging from 1348 to 3658 m above mean sea level (m amsl). The major landforms are plateau, escarpments, gorges and hills. The presence of such diverse landforms has caused the districts to have diversified agro-ecology. Based on traditional classification system (Hurni, 1998), the study area consists of kolla, weyna dega, dega agro-ecological zones, covering 3.0, 40.8, 41.2 and 15% of the landmass, respectively (Fig. 2). The topography and agro-ecological diversity have yielded different soil types in the study districts. The steeper slope areas are dominantly covered by shallow soils mainly Leptosols and Regosols while the plateaus are covered by clayey soils, which could be described as Vertisols and Vertic Cambisols (FAO, 1984).

Fig. 2: Agro-ecology and elevation of study districts.

The map is generated by Geographical Information System (GIS) based 30 m resolution digital elevation data analysis using Digital Terrain Model (DTM).

The Beshilo River and its biggest tributary river called Jita drain the study area that shortly joins the Abay (the Blue Nile) River. The plateau parts of these districts receive high rainfall (Fig. 3) during the main rainy season, with poor soil drainage causing water logging and inundation. Cereals and pulses are the major crop types grown in the study sites. Owing to soil properties (clayey and poor drainage characteristics of the plateau lands) and climate (cool and moist), grass pea is among the widely cultivated crop in the plateaus of study sites. Projection of CSA (2008) data revealed that the population of Dawnt and Delanta was estimated to be 73,894 and 137,970, respectively, in 2018.

Fig. 3: Wogeltena station mean monthly rainfall (Source: Wogeltena Meteorology Station, mean monthly rainfall of 1981 to 1988).

Due to missing data, it was not possible to cover period after 1988 in the analysis.


The research applied mixed method that involved both quantitative and qualitative research design to collect pertinent data. The mixed method approach helps to triangulate and understand the contradiction between the findings from both methods (i.e., quantitative and qualitative) and yields more robust results. The quantitative method mainly relied on approaches used to collect and analyse the determinants of land use for grass pea cultivation such as socioeconomic and demographic characteristics. The qualitative research approaches aimed to validate and support the quantitative results.

Data collection

Quantitative data were generated through semi-structured questionnaires survey. Moreover, qualitative data was generated using focus group discussions (FGDs) and key informants interviews (KIIs). The FGDs were conducted with group of people in each sampled kebele to supplement data collected through household survey. In total, six FGDs were conducted, i.e., one FGD in each sample kebele. Each FGD consisted of 10 to 12 participants organised from people of different age, sex and social categories (children, youth, adult and elderly, men and women) and persons with lathyrism disability. FGD participant’s selection was done in joint consultation with kebele administration, health extension workers and agriculture office extension agents based on pre-set criteria of the research team. KIIs were conducted with health extension workers, and development agents at kebele level, district offices of health, agriculture and lathyrism victims association’s offices. In total, 13 KIIs were conducted, where KII participant’s selection was based on involvement in community-based agricultural and health extension services and vulnerability to lathyrism.


Accordingly, using statistical formula (Cochran, 1963), we selected a sample of 384 farm household from the two districts of Delanta and Dawnt. These districts were purposefully selected owing to the relatively large proportion of land allocation for grass pea production despite the continued occurrence of lathyrism in the study sites (Damene, 2014). In total, the survey covered 384 households from the two districts, 128 HHs (33% of surveyed HHs) from Dawnt and 256 HHs from Delanta district (67% of surveyed HHs) (Table 1).

Table 1 Sample HHs by districts and kebeles.

Multi-stage sampling techniques were used to select sample kebeles. Number of sample kebeles was determined based on population of the two districts. Hence, out of the total 48 rural kebeles in the two districts, we decided to sample six kebeles. The kebeles of dega agro-ecological zone were targeted in sampling as this zone is the major production area of grass pea. Thus, the major criteria included in selecting sample kebeles from selected kebeles in the dega agro-ecology were: prevalence of lathyrism, extent of grass pea production, agro-ecological zone and the landform (i.e., plateau sites).

The agro-ecological zone and the landform of the area were generated from 30 m resolution digital elevation data using Digital Terrain Model (DTM) through Geographical Information System (GIS) technique. This map was used to identify potential study kebeles for the sampling. Accordingly, out of 25 kebeles in the dega agro-ecological zone of the districts, 12 kebeles (four from Dawnt and eight from Delanta) with larger proportion of flat and depression landform were shortlisted at desk review stage before the actual field work commenced. Then, in the field, out of shortlisted 12 kebeles, six kebeles were purposefully sampled based on the above selection criteria using information obtained from the respective districts’ agriculture and health offices. Accordingly, Atsefit-Yekendat and Debir-Agonat kebeles from Dawnt and Chew-Katir, Kachin-Ara, Baba and Silana-Tikrena from Delanta were sampled for the survey.

After the selection of sample kebeles, sampling of survey households was carried out from the list of households residing in the sample kebeles (list obtained from the respective kebele administration offices). Household sampling intervals were determined in such a way that number of households (HHs) of the sample kebele was divided by sample size allocated to the kebele. Sample size of each kebele was determined based on household number of the kebele. Then, HHs randomisation was performed indiscriminately among grass pea producers and non-producers by lottery method from the kebele HH list. The first HH was randomly selected from the list by lottery method and then sampling continued by adding sampling interval until required sample size was attained. In the case of unavailability of the sampled HH head or spouse in the study area due to unforeseen circumstances (travel, death or other reasons), the samples were replaced from the list just by picking the name before or after the selected household. Survey HH interval calculation and sampling considered 5% contingency to replace HHs that would be temporarily unavailable for the interview due to unforeseen circumstances.


Then, enumerators were recruited and trained for the household survey from the two districts in collaboration with the two districts administration council and agriculture offices. Their training focused on survey procedures and methods, questionnaire and ethical consideration. In order to avoid communication gap among enumerators, respondents and researchers, the survey questionnaire was translated into Amharic (local language) at the preparation stage. Each question of the household interview questionnaire used for the survey was written both in Amaric and English that ran together on continuous line so as to reduce any language induced communication gap. With all these, the research team together with the data collectors first conducted a questionnaire pre-test in Arka Chinka (non sample) kebele of Delanta district, and then it was improved before administration. Then, primary quantitative and qualitative data were collected through household survey, focus group discussion (FGDs) and key informant interviews (KIIs). Researchers intensively supervised the survey. In this study, crop and livestock production survey covered only 1 year, i.e., the period between June 2015 and May 2016.

Ethical consideration

As the study involved human participants, the research proposal and assessment tools were submitted and approved by Addis Ababa University for compliance with the rules and regulation of the university code of ethics in line with national and international ethical standards. Accordingly, before starting the survey/interview, enumerators informed each respondent about the purpose of the study, data management strategy and confidentiality statement and asked for the willingness of each interviewee telling that they can quit at any time whenever they feel uncomfortable. Thus, our enumerators interviewed 384 respondents who gave their full consent for the interview and allowed those who refrained (7 individuals), thanking them for coming to the interview. Moreover, codes were used during analysis and reporting instead of respondents’ name or other features that could lead to the identification of the individual. As agreed with respondents, the researchers used all records only for research purpose and kept them strictly confidential at all time, as per national and international ethical guidelines.

Data analysis

Final data cleaning was done before encoding the data in SPSS pre-made data entry template. This was followed by quality assessment of the entered data by the research team using randomly selected (10%) raw data to guide the entry and make corrections. The qualitative and quantitative data collected through household survey, FGDs and KIIs were analysed and interpreted using different methods. A thematic analysis of the qualitative data was used to substantiate and explain quantitative data results obtained from statistical analysis. The quantitative data generated through household survey were analysed using descriptive statistics (SPSS) and Heckman two-stage selection model in STATA version 16.

Estimation of survey households’ wealth status

The wealth status of the survey households was assessed from different asset holdings, which is a locally used criterion to measure wealth, as revealed through community FGDs. Household asset variables that were used to measure wealth included: owning of corrugated iron roof house, a pair or more plough oxen, two or more cows, mobile or landline (wireless) phone, bicycle, horse or mules, solar or biogas lamp and number of food sufficient months. All variables except number of food sufficient months were coded as categorical (Yes = 1 and No = 0), while food sufficient months were given scale value ranging between 0 and 1 by dividing number of food sufficient months by 12. Then the wealth status of the household (WsHH) was estimated by adding the variables and dividing number of considered variables (8) and then multiplying the result by hundred as give in the equation below (Eq. 1).

$${mathrm{WsHH}} = left( {left( {mathop {sum}limits_i^8 {Xi} } right)/8} right) ast 100$$


where: i = asset type

Econometric model specification

The decision to produce and the intensity of land allocated to grass pea production by a household was estimated using the Heckman two-stage models. The Heckman model was selected as a plausible model because of the suspected selectivity bias. According to the model, it is presumed that a household follows a two-stage decision process. The first stage describes farm household’s choice decision on whether to grow grass pea or not. This model is estimated using the probit equation (Eq. 2). In the second stage, farm households make the decision on the intensity of land to produce grass pea if they choose to grow the crop in the first stage. This is the outcome equation specified in Eq. (3). Hence, there is selectivity issue as the intensity of land allocated to grass pea is contingent upon household selection decision of growing (or not growing) the crop. The final equation (Eq. 4) that we estimated takes into account the selectivity problem.

Following Heckman (1979), the model is specified as below:

$$Yi = beta X_i + varepsilon _i$$


where Yi represents the intensity (amount) of land allocated to grass pea,

Xi denotes observed household ith variables affecting amount of land used for grass pea by household, and εi represents the stochastic disturbance term.

It is presumed that Yi is observed only for households who decided to grow grass pea. Therefore, the sample selection in this model relates to the fact that household should be a grower of grass pea to observe the amount of land they allocate to grass pea production.

The selection equation relates to the decision or choice to grow or not to grow grass pea, which is specified as:

$$G_i = gamma Z_i + mu _i$$


where Gi is the choice to grow grass pea (Gi = 1, if household grew grass pea, 0 otherwise), Zi represents independent variable in the selection model, and μi denotes an error term.

The model can be specified as:

$$Eleft( {Y_i/G_i = 1,X_i} right) = Eleft( {Y_i/G_iX_imu _i} right) = beta X_i + Eleft( {varepsilon _i/mu _i > – gamma Z_i} right)$$


Heckman approached this problem by considering E(εi/μi > − γZi) as omitted variable problem and proposed that estimating the omitted variable could solve the problem of sample selection bias (Heckman, 1979). In general, the model can be written as:

$$Eleft( {varepsilon _i/mu _i > – gamma Z_i} right) = rho _{varepsilon u}sigma _varepsilon lambda _ileft( {Z_igamma } right) = beta _lambda lambda _ileft( { – Z_igamma } right)$$


Where, λi(−γZi) is the inverse Mill’s ratio evaluated at the indicated value and βλ is unknown parameter equal to ρεuσε.

Identification of independent variables

The detailed description of variables controlled in the model and hypothesised effect on intensity of land allocation to grass pea is presented in Table 2. Household wealth and income are expected to reduce land use for grass pea but the effect of livestock ownership and landholding is expected to be positive because grass pea straw is used for cattle feeding. Following Hillocks and Maruthi (2012) and Mwaura and Adong (2016), landholding size is a key factor affecting household decision on the choice of crop type grown.

Table 2 Description of determinants and their expected effects on intensity of land allocation to grass pea production.

Higher level of education is hypothesised to reduce land use for grass pea since it improves people’s knowledge and awareness of the repercussion of grass pea use (Getahuna et al., 2002). Positive perception of grass pea also induces increase in land use for its cultivation since the crop is drought resistant (Getahuna et al., 2002). Similarly, farmers closer to market centres are expected to allocate more land to grass pea because the crop is grown for cash. Access to credit is also a critical factor determining farmers land use decision (Nguyen et al., 2017). Farmers who have contact with agricultural agents or extension workers are likely to get more training to diversify their crop and gain more awareness about the adverse effect of grass pea on health and are thus expected to allocate less amount of land to the crop (see Greig, 2009).


Source link

Leave a Reply

Your email address will not be published. Required fields are marked *