1 Introduction

Many efforts have been made to reduce global poverty and hunger over the past few decades, but more than 795 million people are undernourished at a global level [1]. Among these, the majority are found in developing countries, mainly in Africa [2]. In Africa, land quality degradation is the core threat to agricultural production and has severe implications for acute poverty and food insecurity [1,2,3]. Land degradation, in conjunction with climate change, has had an impact on all aspects of food security, including availability, accessibility, utilization, and food stability [4, 5]. In response, emphasis has shifted to the development of means and strategies for improving agricultural productivity through the empowerment of smallholder farmers in order to reduce land degradation and increase food security [6]. This is because most of the rural population in Africa relies on subsistence agriculture characterized by low use of agricultural and land management technologies [7].

In Ethiopia, the landscape potential for agricultural development has been continually affected by severe land degradation [6,7,8]. The problem, in particular, is aggravated by soil degradation, drought, erratic rainfall, population pressure, the inappropriate use of improved sustainable land management (SLM) technologies, and poor agricultural practices [8,9,10,11,12]. Low agrarian productivity contributes to chronic food insecurity and extreme poverty in the country [8]. Accordingly, the country has continued to be dependent on foreign food aid, and outmigration has been a common phenomenon in the last few decades. Ethiopia is one of the most economic deprived countries in the world [13]. About 33% of the rural population is found below the national poverty line, and nearly 14% of the non-poor are vulnerable to extreme poverty [14]. Food insecurity is a communal issue in semiarid lowlands and mountainous highlands areas where the majority of the population relies on volatile rain-fed smallholder agriculture activities [15]. There is no exception in Gojjam; the study area is a part of it, which is well-known for the surplus agricultural production in the past decades [16]. In this region, acute soil erosion, climate variability and change, fragile ecosystem services, and inappropriate land management have been the critical factors in agricultural yield reduction in the last few decades. Because of these, the community has changed from a food surplus to a food deficit [17, 18].

Ethiopia has tracked an agricultural production intensification method to boost agricultural productivity with small farmland sizes through the application of modern agricultural inputs by promoting Integrated Sustainable Land management (ISLM) technologies [19, 20]. This aimed to feed the rapidly growing population [19,20,21]. To achieve this aim, the government has incorporated several SLM programs and strategies in the last three decades. For example, the country’s 5 year Growth and Transformation Plans (GTP I and II) give special emphasis on implementing natural resources management through community mass mobilization at a watershed level [4, 5]. The government promotes ISLM technologies, including soil fertility management, structural soil and water conservation (SWC), and improved seed technologies [4]. Similarly, in the study area, to reduce acute land degradation and increase land productivity, development agencies and the government have been implementing various ISLM technologies for the last few decades [7, 19, 20].

Despite some progress, the impact of ISLM investments on household food security in Ethiopia and the north Gojjam sub-basin in particular is not well documented. As a result, our central question is, do ISLM technologies adoptions affect the food security of rural households who adopt? Regarding this issue, most previous studies focused on the welfare effect of single land management technology, and most of the results are inconsistent [8, 22,23,24,25]. To address this gap, this study examines the impact of ISLM technologies adoption on household food security in high agricultural potential areas of Ethiopia’s highlands, where various land resources conservation programs have been implemented. Therefore, the main objective of this study is to analyze the effects of ISLM technologies adoption on rural household food security in the north Gojjam sub-basin of the upper Blue Nile, Ethiopia.

As a result, this research will contribute to several aspects of the existing knowledge on the effect of ISLM technology adoption. First, it is very important to know the best of our facts about the effects of ISLM on adopter households’ food security. The additional benefit of this study is for advising policymakers and other development agency partners. Lastly, it may be useful to future researchers as a source of literature to conduct similar but not identical studies.

1.1 The concept of integrated sustainable land management (ISLM)

The concept of ISLM has developed as an important global environmental and economic development issue attributable to accelerating land degradation, including Ethiopia. SLM refers to continues use of suitable technologies that farmers implement on their farmland to increase agricultural productivity through improving cultivated land fertility [4, 26]. SLM technologies can be grouped into structural, agronomic, vegetative management to reduce soil loss, control nutrient depletion, and improve soil productivity respectively [26]. On the other hand, ISLM refers to the use of a combination of land management technologies aimed at improving soil fertility to maintain land productivity while integrating socio-economic principles with environmental concerns in order to maintain or enhance production, protect the potential of natural resources, reduce the level of production risk, prevent soil and water degradation, be economically viable, and be socially acceptable land management technologies [11, 12, 27,28,29,30,31]. This study focuses on the integration of different (i.e. structural, agronomic, and vegetative management) SLM technologies that are implemented in in the north Gojjam sub-basin in particular to achieve rural households’ food security. As presented in Table 2 about 53.5% of farmers adopt ISLM technologies on their farmland during the survey season in the north Gojjam sub-basin. Consequently, in this study, we evaluate the effect of ISLM technology adoption on household food security in the north Gojjam sub-basin.

1.2 Conceptualizing and measuring methods of food security

The concept of food security has been widely used at the household level as a measure of households’ welfare status [32]. A household is considered food secure if all members have physical and economic access to safe, sufficient, and nutritious food that meets their dietary needs and food choices for an active and healthy life at all times [33]. However, due to the complexities of measuring food security at the household level, no agreed-upon methods exist [34]. Different studies have used diverse methods ranging from caloric intake, dietary quality, and anthropometric estimates to capture the four pillars of food security. Most of these measures, however, are costly and time-consuming [35]. In this study, we used time-saving and cost-effective standard food security measures, which include Household Food Consumption Expenditure (HFCE), Household Dietary Diversity Score (HDDS), household food gap, and Household Food Insecurity Access Scale (HFIAS).

The adoption of SLM technology is expected to affect household food consumption and expenditure because most inhabitants in the study area are agro-based subsistence farmers [36]. Therefore, we used HFCE as one indicator to evaluate the effect of ISLM on rural households’ welfare. A 7 day recall period was used to capture household food consumption expenditure, and the result was a report in terms of the 2017/18 national average price, which is the Ethiopian Birr per adult equivalent (AE). Likewise, the use of ISLM technologies is expected to improve household food diversity by increasing agricultural productivity. HDDS can be described as the arithmetic sum of 12 food groups consumed by household members during the past 24 h at home [37]. Following FAO-HDDS guidelines, we used the following 12 food groups to measure food diet diversity, namely: cereals, roots and tubers, pulses/legumes/nuts, vegetables, fruits, fish and seafood, eggs, meat and poultry, milk and milk products, oils and fats, sweets/sugar/honey, and miscellaneous food items such as spices. As suggested by [38], we assume that there were no special occasions, such as funerals and fasting, within the sampled households that might influence their food consumption pattern during the 24 h survey period.

The food gap is another subjective measure of food security and refers to the number of months in the past 12 months that households have had difficulty satisfying their food needs [32]. We used food gap measures to evaluate the effect of ISLM technology adoption on household food security in the north Gojjam sub-basin. The impact of ISLM was evaluated by another perception-based measure of food insecurity, HFIAS. The HFIAS consists of nine questions and three frequencies. When administered in a population-based household survey, it allows for estimating the percentage of households affected by three different severities of hunger. The response to each question was coded: 0 = Never; 1 = Rarely (once or twice); 2 = Sometimes (three to ten); 3 = Often (more than ten) times in the past 30 days. The sum of these responses gives the HFIAS score, which ranges from 0 (no hunger) to 27 (severe hunger). Households were interviewed in March and June 2018, which are free from the fasting period of the lean season, and hence, it was an appropriate period to use the HFAIS in the study area.

2 Material and methods

2.1 Study area

The study was conducted in the Ethiopian highlands mainly in the North Gojjam sub-basin, located between 38.2 and 39.6° E longitude and 10.8–11.9° N latitude [19, 20, 36]. Relatively, the sub-basin is found in the Amhara regional state, particularly on the border of West Gojjam, East Gojjam, and South Gonder Zones (Fig. 1). The sub-basin is one of the major headstreams of the Blue Nile/Abbay River and covers 1,431,360 hectares [19, 20, 36]. The total population of the sub-basin and the surrounding villages was 3,565,892 [19, 20, 30, 39]. The majority of the population of the sub-basin was settled in foot hillside areas. The average population density ranged from 260 to 270 people per km2 [19, 20].

Fig. 1
figure 1

Topography and location of the study area [19, 20, 36]

Full size image

The elevation of the north Gojjam sub-basin ranges from 1044 to 4048 masl. The dominant agroecological zone is characterized by tepid to cool moist middle highlands, and cold to very cold moist sub-afro-alpine to afro-alpine highlands. But the eastern and southeastern parts of the sub-basin are hot to warm moist lowlands. The dominant climate condition of the sub-basin is the tropical highland monsoon [19, 20, 36, 40]. The mean minimum and maximum temperature of the study area varies from 24.6 to 28.1 °C and 11.0°–14.51 °C respectively. The mean annual temperature is 19.41 °C [36, 40]. The rainfall pattern is closely correlated with the annual migration of the Inter-Tropical Convergence Zone (ITCZ), and is most of the rainfall occurs in summer, from June to September with high in August [19, 20, 36, 40].

The spatial distribution of rainfall across the sub-basin is not uniform; the highland region tends to be wetter than the lowland region. The meteorological data record from the stations within and around the study area (1986–2017 years) showed that the mean annual rainfall is 1334.48 mm with a minimum of 810 mm and a maximum of 1815 mm [41]. In general, the dominant climate condition of the study area is the tropical highland monsoon rainfall [41]. According to these authors, the climate in the area is mainly influenced by altitude and global weather systems.

The area coverage of vegetation is very small in north Gojjam sub-basin [19, 20]. Natural forest coverage is very low and is found on some river banks, hillsides, and church compounds. Eucalyptus globules forest is dominant among the introduced trees, particularly in the highlands, where its cover has increased. Unreliable rain-fed agriculture is the primary source of livelihood for the majority of the population in the sub-basin. The dominant soil types are leptosols, vertisols, luvisols, and alisols. The geology of the sub-basin is mainly dominated by basalt, but the lowland is dominated by sandstone [19, 20, 40]. It is characterized by a smallholder mixed crop and livestock production system. Different types of crops have been produced to a great extent. The dominant crop types are cereals, pulses, and oilseeds, but vegetation and fruits have been produced to some extent in the area. The common livestock types are cattle, sheep, goats, horses, donkeys, mules, and poultry. Soil erosion, climate variability/change, and water shortage explain to a large extent the prevailing food insecurity and poverty in the sub-basin [19, 20, 38].

2.2 Data and sampling procedures

In order to select the required data for this study, a multistage sampling technique was applied. Based on the information obtained from a panel discussion with stakeholders and the authors’ prior knowledge about the study area, in the first stage, the north Gojjam sub-basin was selected from the upper Blue Nile sub-basins. The sub-basin is characterized by different agroecological and administrative zones. Accordingly, in the second stage, based on the administration and agro-ecological location criteria, three districts were selected. In the fourth, nine rural villages (three villages from each district) were carefully chosen from the lower, middle, and upper parts of the districts. Finally, 414 household heads were selected using systematic random sampling techniques from the lists of households obtained from local agricultural offices. The sample size in the sample districts and villages was decided based on the proportional stratifying sampling procedure (Table 1).

Table 1 Distribution of respondents by agro-ecology and administrations (Source: [30])
Full size table

To get more quantitative data for triangulating the quantitative data analysis results, 127 Focus Group Discussion (FGDs) participants were selected from within different social groups, including leaders of religious and community organizations, elderly persons, youths, and male and female household heads that lived in the selected villages. For a similar purpose, 54 key informants were purposely selected from farmers, Development Agents (DAs), and districts’ agricultural and natural resources experts.

2.3 Data collection procedures

For this study, objective and subjective data types were collected from randomly selected households through open and closed-ended questionnaires from May to June 2018. The questionnaire comprises various demographic, infrastructure, institutional, topographic, plot characteristics, and land users’ perceptions on the effect of ISLM technologies on food security. Besides, household food security indicators such as household food deficit, Household Dietary Diversity Score (HDDS), Household Food Consumption Expenditure (HFCE), and Household Food Insecurity Access (HFIAS) were collected. In order to evaluate the reliability and validity of the questionnaire, several tasks were made. At its inception, it was drafted based on a review of the previous empirical studies. Then, the draft was contextualized by development and extension agents working in the study area. After these tasks, the survey was pre-tested via 20 randomly selected farm household heads in a similar, but non-sample district. Based on the advice obtained from the inputs, certain questions were amended, rearranged, and omitted, and then translated into Amharic, the local language. The survey was administered by trained enumerators under the close supervision of the first author and the field assistants. Moreover, a series of FGDs and key informant interviews were held to complement and triangulate the quantitative results. FGDs were held in the development agents’ offices, churches, and other suitable fields to collect detailed qualitative information. FGD participants were limited to 6–8 members in all 18 series of FGDs. All FGDs were facilitated by the first author guided by the checklists. In each study village, two series of FGDs were held. Similarly, in-depth interviews were conducted with key informants in each village.

2.4 Data analysis methods

In this study, descriptive statistics were used to summarize the demographic, socio-economic institutional, plot, and asset characteristics of the respondents. The Endogenous Switching Regression (ESR) model was used to show the effect of ISLM technologies adoption on households’ food security status using the Full Information Maximum Likelihood (FIML) estimation technique.

2.4.1 Econometric model specification and estimation procedure

Evaluating the effect of agricultural technology adoption on household food security through non-experimental observation is not an easy task due to the challenge of quantifying the counterfactual result [4, 42,43,44]. Moreover, farmers are not randomly assigned to the two groups (adopters and non-adopters), but rather may make the adoption choice based on their experience and the predictable profit of the technology they will be adopting [24, 43, 44]. As a result, farmers who use ISLM technologies may differ from those who do not. Farmers’ unobservable characteristics may affect both ISLM adoption decisions and the welfare outcome of the household. This leads to a false conclusion on the estimation of the adoption effect on household food security unless sample selection bias is considered [7].

Widely used estimation methods for addressing such sample selection bias include Propensity Score Matching (PSM), Instrumental Variable (IV), and Heckman’s two-stage selection [44,45,46]. The methods, however, have their own limitations. For instance, both Heckman selection and IV methods tend to impose a functional form assumption by assuming that farmers’ adoption has only an intercept shift and not a slope shift in the outcome variables [32, 46]. In this regard, PSM assumes that selection is based on observable variables, and it can address these problems by avoiding the functional form of assumption. However, unobserved heterogeneity is likely because farmers' skills and motivation are likely to influence their adoption behavior [46].To address these problems, we applied the ESR model. The selection process in the parametric ESR model can also be addressed using the inverse Mills ratio derived from the probit criterion equation [47]. This approach is widely used in estimating the impact of farm technology adoption on households’ welfare. The premise of this study is that farmers adopt ISLM technologies when the utility of adopting them is greater than the utility of not adopting them in the study area. Considering this, the latent variables were specified following the method of [7, 47].

$$S_{h}^{*} = \gamma Z_{(h)} + \varepsilon_{h} ,\,\,\,S_{h} = (1\,if\,S_{h}^{*} > 0\,and\,0\,otherwis)$$
(1)

where, \({S}_{h}^{*}\) is the unobservable variable and \({S}_{h}\) is observable, i.e., the adoption of ISLM technology is equal to 1 if the farm household is an adopter (improved seed in any crop, compost and/or manure and structural SWC simultaneously) and 0 otherwise. h is a subscript that represents the household head. \({Z}_{h}\) denotes the vector of exogenous variables that affect the decision of adoption technology (Table 2). \(\gamma\) is the regression coefficient and \({\varepsilon }_{h}\) is random error terms. As previously stated, the ISLM technology adoption decision switching regression model was used to estimate the impact of ISLM adoption on household food security while accounting for sample selection bias. As a result, two regime regression models (adopter and non-adopter) were specified in this study as follows:

$$Y_{h1\,} \, = \,\beta_{1} X_{h1} \, + \,u_{h1} \,\,\,\,\,\,{\text{if}}\quad S_{h} \, = \,1$$
(2)
$$Y_{h0} \, = \,\beta_{0} X_{h0} \, + \,u_{h0} \,\,\,\,\,\,{\text{if}}\quad S_{h} \, = \,0$$
(3)

where \({Y}_{h1}\) and \({Y}_{h0}\) food security indicators for adopters and non-adopters, \({X}_{h1}\) and \({X}_{h0}\) are all the independent variables for adopters and non –adopters respectively, while \({\beta }_{1}\) and \({\beta }_{0}\) are parameters to be estimated for the adopter and non-adopter regimes respectively. And \({u}_{h1}\) and \({u}_{h0}\) are error terms for Eqs. 2 and 3 respectively. In this case, the three equations error terms \(( {u}_{h1}\), \({u}_{h0}\) and \({\varepsilon }_{h}\)) are assumed to have a joint-normal distribution a mean vector of 0, and a covariance matrix specified following the equation used by [47].

$${\text{cover}}\left( {{\text{~}}\varepsilon _{h} ,~~u_{{h1}} ,{\text{~~}}u_{{h0}} } \right) = \user2{~}\left[ {\begin{array}{*{20}c} {\sigma _{{\varepsilon _{h} }}^{2} } & {\sigma _{{\varepsilon _{h} u_{{h1}} }} } & {\sigma _{{\varepsilon _{h} u_{{h0}} }} } \\ {\sigma _{{\varepsilon _{h} u_{{h1}} }} } & {\sigma _{{h1}}^{2} } & {.~} \\ {\sigma _{{\varepsilon _{h} u_{{h0}} }} } & . & {\sigma _{{h0}}^{2} } \\ \end{array} } \right]$$
(4)

where \({\sigma }_{{\varepsilon }_{h}}^{2}\)= var(\({\varepsilon }_{h}\)), which is supposed to be 1 as \(\gamma\) is only admirable up to a scale factor [48] \({\sigma }_{h1}^{2}\)= var(\({u}_{h1}\)), \({\sigma }_{h0}^{2}\)=var(\({u}_{h0}\)), covar (\({\varepsilon }_{h}\) and \({u}_{h1}\)) = \({\sigma }_{h1\varepsilon h}\), covar (\({\varepsilon }_{h}\) and \({\sigma }_{h0\varepsilon }\)) = \({\sigma }_{{\varepsilon }_{h}{u}_{h0}}\). Since \({Y}_{h1}\) and \({Y}_{h0}\) are not observed at the same time, the covariance between \({u}_{h1}\) and \({u}_{h0}\) are not defined [47]. The expected values of the error terms \({u}_{h1}\) and \({u}_{h0}\) can be specified following the work of [23, 32].

$${\text{E}}\,\left( {u_{h1} /S_{h} \, = \,1} \right)\, = \,\sigma_{{\varepsilon_{h} u_{h1} }} \left( {\frac{{\phi \left( {\gamma Z_{h } } \right)}}{{\phi \left( {\gamma Z_{h } } \right)}}} \right) = \sigma_{{\varepsilon_{h} u_{h1} }} \lambda_{h1}$$
(5)
$${\text{E}}\,\left( {u_{h0} /S_{h} = \,0} \right)\, = \,\sigma_{{\varepsilon_{h} u_{h0} }} \left( {\frac{{ - \phi \left( {\gamma Z_{h } } \right)}}{{1 - \phi \left( {\gamma Z_{h } } \right)}}} \right) = \sigma_{{\varepsilon_{h} u_{h0} }} \lambda_{h0}$$
(6)

where \(\phi\) (.)is the standard normal probability density function and \(\phi\) (.) the standard normal cumulative density function respectively, and \(\lambda_{h1} = \frac{{\phi \left( {\gamma Z_{h } } \right)}}{{\phi \left( {\gamma Z_{h } } \right)}}\,{\text{and}}\,\lambda_{h0} = \frac{{ - \phi \left( {\gamma Z_{h } } \right)}}{{1 - \phi \left( {\gamma Z_{h } } \right)}}\), where, \({\lambda }_{h1}\) and \({\lambda }_{h0}\) are the inverse mills ratio calculated from the selection equation. If the estimated covariance \({\sigma }_{{\varepsilon }_{h}{u}_{h1}}\) and \({\sigma }_{{\varepsilon }_{h}{u}_{h0}}\) are statistically significant, the decision to practice ISLM technologies and welfare outcomes are correlated and this is the evidence for endogenous switching that indicates the existence of sample selection bias. The outcome model is also known as a “switching regression model with endogenous switching” [48]. A widely used method to estimate self-selection is the two-stage instrumental variable procedure [48]. However, this technique is incorrect and highly criticized since it requires some adjustment to derive consistent standard errors and shows poor performance when there is high multi-collinearity between the covariates of the selection equation and the covariates of the outcome equation [48].

Table 2 Definitions and description of selected variables used in the regression
Full size table

The appropriate and efficient method to estimate the ESR model is the FIML method. This method is preferable to other approaches in many instances. The FIML method estimates both selection and outcome equations simultaneously [47] using the movestay command in STATA [47]. Besides, it allows for restrictions to be applied and permits the construction of likelihood ratio tests on the restriction variable [47] when similar variables affect the adoption decision (Z) and the subsequent outcome equations (X); the absence of identification of the model will be a problem. For identification, we used exclusion restrictions, i.e., we needed at least one variable that affected farmers‘ adoption decisions, but it does not directly affect any of the households‘ food security indicators [47, 49]. By considering the agricultural technology adoption literature on the significance of information access on farmers‘ agricultural technology adoption decisions [24, 47, 49], we used information access related to land management as the instrumental variable. We proposed that farm households that do not face constraints in accessing information on land management practices are more likely to understand the benefits of new technology and adapt it to their farmland. Accordingly, following the method used by [24, 48, 49], the information access variable for the validation instrument was established by carrying out a falsification check. i.e., if a variable is a fitting choice instrument, it will affect the ISLM adoption choice but not the food security indicators of non-adopter farm households.

2.4.2 Method of estimating the average treatment effects

In this section, we describe how the endogenous switching treatment regression model is used to compute the average treatment effect and counterfactual case [24, 49, 50]. ESR is critical when comparing the expected food security of rural farm households that adopted ISLM (7) versus farm households that did not adopt (8). It is also used to investigate the expected farm households’ welfare in the counterfactual case, which is when farm households that had adopted did not adopt (9) and when farm households that had not adopted did adopt (10). The conditional expectation for household welfare in the four cases can be expressed mathematically as follows:

$${\text{E}}\left( {Y_{h1} /S_{h} = 1} \right)\, = \,\beta_{1} X_{h1 } + \sigma_{{\varepsilon_{h} u_{h1} }} \lambda_{h1} \,\,\,\,\,\,{\text{if}}\,S_{h} = 1$$
(7)
$${\text{E}}\left( {Y_{h0} /S_{h} = 0} \right)\, = \, \beta_{0} X_{h0 } + \sigma_{{\varepsilon_{h} u_{h0} }} \lambda_{h0} \,\,\,\,\,\,\,\,\,{\text{if}}\,S_{h} = 0$$
(8)
$${\text{E}}\,\left( {Y_{h0} /S_{h} = 1} \right)\, = \,\beta_{0} X_{h1 } + \sigma_{{\varepsilon_{h} u_{h0} }} \lambda_{hp1} \,\,\,\,\,\,\,\,{\text{if}}\,S_{h} = 1$$
(9)
$${\text{E}}\left( {Y_{h1} /S_{h} = 0} \right)\, = \, \beta_{1} X_{h0 } + \sigma_{{\varepsilon_{h} u_{h1} }} \lambda_{h0} \,\,\,\,\,\,\,\,{\text{if}}\,S_{h} = 0$$
(10)

Based on Heckman [51], the average effect of treatment ‘to adopt’ on the treated (ATT) was calculated as the difference between the expected value of the food security indicator outcome and the counterfactual cases (7 and 9).

$${\text{E}}\,\left( {Y_{h1} /S_{h} \, = \,1} \right) - \left( {Y_{h0} /S_{h} \, = \,1} \right)\, = \,X_{h1} \left( {\beta_{1} - \beta_{0} } \right)\, + \,\lambda_{h1} \left( {\sigma_{{\varepsilon_{h} u_{h1} }} - \sigma_{{\varepsilon_{h} u_{h0} }} } \right)\, = \,{\text{ATT}}$$
(11)

ATT represents the average effect of integrated SLM technologies on the food security of the farm households that actually adopt the SLM technologies. Similarly, the average treatment’s impact on the untreated (ATU) was estimated as the difference between what non-adopters would have benefited from had they adopted SLM technologies and the actual observed food security (8 and 10).

$${\text{E}}\,\left( {Y_{h1} /S_{h} \, = \,0} \right) - \left( {Y_{h0} /S_{h} = 0} \right)\, = \,X_{h0} \left( {\beta_{1} - \beta_{0} } \right)\, + \,\lambda_{h0} \left( {\sigma_{{\varepsilon_{h} u_{h1} }} - \sigma_{{\varepsilon_{h} u_{h0} }} } \right)\, = \,{\text{ATU}}$$
(12)

3 Results and discussion

3.1 Socio-economic characteristics of sample households

Table 2 illustrates the description and summary statistics of the variables employed in the model. According to the table, the majorities of respondents in the study area were male-headed (83%), while the remainders (17%) were female-headed and were mostly widows and divorced. Also, the average age of the study household heads was 50 years, with a minimum of 22 years and a maximum of 85 years during the survey season. On the other hand, the survey found that 46.5% of framer respondents were illiterate, while 53.5% could read and write. The majority graduated from adult education. During the survey season, the respondents’ family sizes ranged from 2 to 11 and had an average of 5.4 family members. Moreover, the study population's average dependency ratio was 0.8.

The average landholding size of the surveyed households was 1.03 ha, ranging from 0.15 to 4 ha, and their plots were frequently not spatially adjacent, so they cultivated three plots on average. The plots in the study area are a 58 min one-way round trip from the farmer’s house. Most plots were used for annual cereal-legume crop production. While evaluating the plots, on average, about 21% of respondents perceived that their farmland was steep, 43% moderate, and 36% had a plain slope. About 35% of respondents perceived their cultivated land as fertile. The majority of farmlands were perceived by the land user as having a moderately to severely eroded status.

The respondents were also requested to answer about their satisfaction with possible advice from development agents; here, about 62% of them said that they have received adequate advice. Similarly, about 39% of respondents said they attend formal SLM-related training. Besides, about 35% of respondents had their own radio or mobile during the survey season, and about 87% of the sampled households were involved in the farmers’ cooperative group. Farmers in the study area walk about two hours on average to arrive at the nearest agricultural input–output market from their residence.

As Table 2 ’s result shows, the surveyed farm households on average possessed about 4.13 tropical livestock units (TLU). Furthermore, about 18% of respondents were involved in off-farm activities, while about 45% of sample households had received credit from formal institutions and earned ETBFootnote 1 2421.6 per year. Besides, the descriptive statistics of the outcome variables, as noted in Table 2, show that the average food consumption expenditure was nearly ETB 149.69 per AE during the survey season. Moreover, the results also show that respondents have consumed an average of 6.03 food items. The average HFIAS was estimated to be a 4.57 index in the survey season. Similarly, rural households in the north Gojjam sub-basin faced a food deficit of about 1 month and 45 days.

3.2 Effect of integrated sustainable land management (ISLM) technologies adoption on households’ food security

The primary focus of the study is to analyze the effect of ISLM technology adoption on rural household food security using four measurement indicators (HFCE, HDDS, household food gap, and HFIAS). To evaluate the effect of ISLM technology adoption on a household’s food security ESR model was applied, and the results are presented in Tables 3 and 4. The ESR model contains the selection equation and separate outcome equations for adopters and non-adopters that are evaluated concurrently [24, 32, 51]. The selection equation refers to the determinants of ISLM adoption (Additional file 1: Appendix 1), which we did not include in this study because they were not the study's objective and were previously presented in a paper [31]. In all food security measure equations, the correlation coefficient of error terms for the ISLM technologies adopter (ρ1) and non-adopter (ρ0) results are statistically significant, except for the food gap. This shows that there is self-selection bias as a result of unobservable factors that occurred in the ISLM technologies adoption decision. Therefore, the use of the ESR model is appropriate for this study, which accounts for both observable and unobservable factors [47]. The coefficients of the factors are not similar in all food security measure results, which further validate the presence of different samples and effects. Likewise, all variables that illuminate the food security of adopters do not affect the non-adopters, and vice versa.

Table 3 ESR results for household food consumption and dietary diversity
Full size table
Table 4 ESR results for food gap and household food insecurity access
Full size table

In all ESR specifications, the variable representing the information constraint of the farmers on whether it is difficult to implement the ISLM technology is used as identifying instrument. The result revealed positive and statistically significant in the adoption equations, but not in all outcomes (food security measures) equations. This satisfies the instrument relevance of all conditions in this model. The positive coefficient confirms that those households that have access to information are more likely to be ISLM adopters. As well, the significance of the likelihood ratio tests for the independence of the equations also indicated that there is a joint dependence between the equations.

3.2.1 The determinant of food security on integrated sustainable land management (ISLM) adopter and non-adopter households

To look at factors that affect food security in ISLM adopter and non-adopter groups, we examined the different socio-economic and plot characteristics related to four food security indicators. As shown in Tables 3 and 4, plot size and livestock number increase the probability of HDDS and HFCE and similarly reduce the HFIAS and household food gap of both adopters and non-adopters. Similarly, FGD participants and key informant interviewees confirmed that households with large farmlands can produce agricultural products, eat a lot of food, and diversify their diet. Furthermore, households with larger farm sizes are not only more encouraged to grow food crops, but they also earn more money from increased crop and livestock yield. This finding is consistent with the latest finding of Teklewold et al. [4], who conclude that large farm sizes increase the food security status of farm households in Eastern Ethiopia. Similarly, FGD participants and key informant interviewees explained that livestock has benefited food security through the quantity and diversity of food, either by selling directly or through livestock products. This shows that livestock is a main source of income to access food and produce food items such as eggs, meat, milk, and milk products. Thus, livestock assets influence food security positively without disaggregating the ISLM adopter and non-adopter groups. Moreover, the ESR model results confirmed that a gentle slope land with treated SLM technology increases HFCE and HDDS while decreasing household food gaps and HFIAS through increased productivity (Tables 3 and 4). Likewise, FGD participants and key informant interviewees indicated that plain slope land is less vulnerable to erosion and suitable to produce household food if it is treated with SLM technology. This finding underlies the justification given by Biru et al. [5] and Shiferaw et al. [25], who testified that land management with better land quality results in better productivity and increases the likelihood of household food security.

The ESR results in Tables 3 and 4 shows that households headed by a male significantly reduce the likelihood of a household food gap and HFIAS while increasing HFCE and HDDS in both adopters and non-adopters. Similarly, participants in the FGDs argued that female-headed households are more likely to have extra months of food inadequacy and their household members are more likely to experience food inaccessibility than male-headed households, whether they are adopters or not. According to the findings of the FGD, male-headed households are more likely to have a diverse food diet and consume more than female-headed households because they lack the labor force and information to access their family's food. Furthermore, in the study area, women have limited access to land and other livelihood resources needed to feed their families. The finding of this research is in line with studies conducted by Kassie et al. [52], who found that female-headed households are more likely to be food insecure than male-headed households in Ethiopia. Distance from the house to the farm input–output market significantly reduces the HDDS and HFCE of both adopters and non-adopters; it increases the household food gap and HFIAS. Similarly, the FGD results show that farmers who are far from the market may not receive as much information from the market to exchange farm inputs and outputs as their counterparts. Likewise, households headed by elders have decreased the food diversity of non-adopters compared to adopters. Furthermore, marital status, credit access, local cooperation, and off-farm activities all improve positively and significantly the HFCE of the adopters. Similarly, only adopters’ family size, training, education, and land fragmentation increase HDDS positively and significantly.

Moreover, training and education significantly reduce the adopters' HFIAS and household food gap. In FGDs and key informant interviews, participants reported that large family size and a high labor force increased food production and consumption expenditures, as well as diversified their food items by managing their cultivated land and generating more income from off-farm activities. This has shown that households with large family sizes are more likely to be food secure than their counterparts by investing in their farmland.

On the other hand, land fragmentation increases households’ food diet diversity; farmers with a large number of plots may produce a variety of food items on different plots in the study area. The positive and significant effect of education and training on food diet diversity demonstrated an increase in farmers’ knowledge about accessing a variety of food items from farmland as well as markets to maintain their health status, but not in HFCE. These showed that households’ consumption expenditures may not always be parallel with food diet diversity. The result of this study is similar to the finding recently conducted by Tambo and Wünscher [32], who reported that education and training increase the food security of farm innovation technology adopters in northern Ghana.

As illustrated in Table 4, there are noticeable alterations across the two household food insecurity (i.e., food gap and HFIAS) indicators. For instance, farmland fragmentation, credit access, education level, marital status, dependent ratio, and the age of the household head do not have a similar effect on the two food security indicator functions. The result further shows that households that have a higher age are subjected to a longer period of food shortage than non-adopters. Access to credit, on the other hand, is more likely to reduce the adopters' food deficit. Those farmers who have farmland with a gentle slope are more likely to have the potential to decrease HFIAS than non-adopter groups. In other words, education, gentle slope farmland, training, and households participating in a local cooperative have the probability of reducing HFIAS in the ISLM adopter group only. The result showed that access to education and training reduces food insecurity among ISLM technology adopters. Similarly, age and marital status have a probability of reducing HFIAS in the non-adopter farm groups.

3.2.2 Average integrated sustainable land management (ISLM) technologies adoption treatment effects on food security

Table 5 presents the average treatment effect of the ISLM technology adoption on household food security in the north Gojjam sub-basin. This table shows that household heads who use ISLM technologies have a higher HDDS and an ETB 38 (27%) higher household food consumption expenditure (HFCE) per adult equivalent (AE) than those who do not use ISLM technology. Likewise, HFCE would have increased by ETB 25 (17.8%) and HDDS by 1.5 (33%) if non-adopters had adopted the ISLM technologies. This is the average variance in food diet diversity and food consumption expenditure of similar pairs of households that belong to different ISLM technologies. The result demonstrates that the adoption of ISLM technologies significantly increases HFCE and HDDS in the north Gojjam sub-basin. The findings are similar to those of the study conducted by Biru et al. [5], who found that SLM adoption improved the HDDS of the adopter farmers in Eastern Ethiopia. On the other hand, as Table 5 shows, ISLM adoption has also had an interesting effect on reducing the depth of food insecurity in the north Gojjam sub-basin. According to the findings, using ISLM technologies reduces the likelihood of household food deficit and HFIAS by one and a half months and 46%, respectively. Likewise, the household food gap would have been reduced by one and a half months and HFIAS by 35.5% if non-adopters had adopted the ISLM technologies on their plot of land.

Table 5 Average treatment effects: endogenous switching regression model result
Full size table

Overall, the results showed that the adoption of ISLM technologies increases the rural household's food security status. The finding confirmed that the adoption of ISLM increased household food security significantly (i.e., increased household food consumption expenditure, diversified food diets, and reduced food-deficit periods and food insecurity access) in the north Gojjam sub-basin. The present study is similar to the recent empirical evidence conducted by Biru et al. [5] and Kassie et al. [53] in Ethiopia and elsewhere Russell et al. [33], which revealed that the adoption of land management technology provided higher welfare returns for rural households.

4 Conclusions and policy implications

Improving national food security through land management is the main policy priority of the Ethiopian government to feed a rapidly growing population. The main objective of this study is to evaluate the effect of ISLM technologies adoption on household food security. The collected data from 414 household heads were analyzed using the Full Information Maximum Likelihood Endogenous Switching Regression (FIML-ESR) model, which was used to estimate the impact of ISLM technology adoption on households’ food security level.

The result showed that gender, land size, market and road distance from the house, and the number of livestock are the main determinants of food security without disaggregating the adopter and non-adopter groups in the north Gojjam sub-basin. Therefore, the study suggested that female-headed households need special treatment to improve their family’s food security. Furthermore, improving rural infrastructure (i.e., markets and roads) accessibility is critical to increasing rural people's food security in the north Gojjam sub-basin.

Adoption of ISLM technologies would increase household dietary diversity by approximately 14.5% and increase food consumption expenditure by ETB 38 per adult equivalent as compared to non-adoption of ISLM technologies. Moreover, it reduces the likelihood of household food deficit and HFIAS by about 1 month and 45 days and 46%, respectively. Thus, the adoption of ISLM provides a twofold purpose for both household well-being improvement and ecosystem service health. This implies that adoption of ISLM reduces land degradation and, thus, increases food security by increasing households’ revenue and food production through crop and livestock farm productivity. Accordingly, ISLM technologies should be scaled up in the north Gojjam sub-basin in particular and elsewhere in general. An important policy implication is that the current agricultural extension programs focus on the promotion and support of ISLM technology adoption to rescue rural farmers from food insecurity and poverty problems. It is also important to stress that our finding does not suggest neglecting a single SLM technology adoption strategy but rather strengthens the arguments for supporting farmers' use of various supplementary SLM technologies simultaneously to increase land productivity and ecosystem health.