Large‐scale land investments, household displacement, and the effect on land degradation in semiarid agro‐pastoral areas of Ethiopia

Agro‐pastoral areas in Ethiopia have been targeted by large‐scale land investments, particularly for the establishment of sugar plantations, since the 1970s. This has led to the displacement of local communities. We investigate the impact of this displacement due to large‐scale land investment on land degradation in semiarid agro‐pastoral areas in Ethiopia. We conducted a survey of 866 households in two agro‐pastoral sites in Ethiopia in 2019, where extensive large‐scale land investment was implemented. We use an endogenous (switching) treatment model to assess the effect of the displacement of households on land degradation. The result shows that 75% of the surveyed households experienced moderate–severe land degradation. Forestlands and grasslands are ranked as the most degraded areas. About 43.7% of the households face a reduction in herd size and 55.8% lost land due to large‐scale land investment, while 86% of the households show a substantial decline in crops and livestock productivity due to land degradation. The results also reveal that the displacement of households leads to a significant increase in land degradation. Household exposure to drought and conflict, the number of livestock, overgrazing, and sharecropping are other drivers of land degradation. Market access, extension services, household income, and mobility, on the other hand, limit the occurrence of land degradation. We conclude that the shifts in property rights from common land used by pastoralists to private land in large‐scale plantations aggravate land degradation in semiarid drylands.

low pastoral value (Sombroek & Sene, 1993). Land degradation is the most serious problem in drylands and a major threat to the world's ability to achieve zero hunger (Nigussie et al., 2017;Waswa, 2012). It is severe in developing countries and particularly in Africa, where the economy is driven by land-based activities such as agriculture and pastoralism (Tilahun et al., 2015).
Land degradation in drylands is, to a large extent, a natural process. However, land-use changes by humans, and especially agricultural activities, aggravate land degradation (IPPC, 2019). Land-use changes by industrial sugar plantations, which we refer to as largescale land investments (LSLI) in this study, can modify natural habitats and land conditions by intensive use of water, agrochemicals discharge, runoff of polluted effluent, and air pollution (World Wildlife Foundation, 2005). Moreover, in dryland pastoral areas, LSLIs have been primarily located on fertile and water-abundant lands leading to pasture scarcity and aggravation of land degradation.
The Ethiopian government initiated LSLIs in the 1970s. With the goal of development, the state captured large tracts of land, often with minimal consultation and compensation to the pastoral communities that resided on the land (Rettberg, 2010). This has led to the displacement of several pastoral communities. For instance, in the Afar pastoral region, over 400,000 ha of land were taken in the last five decades for LSLIs, parks, and wildlife reserves (Mousseau & Martin-Prével, 2016). Since 2010, the Karrayyu and Afar agro-pastoralists in Fentale and Dubti have lost over 80,000 ha of pasture land due to sugar plantations (Rettberg, 2010). The LSLIs control fertile lands and large rivers that pass through the dryland regions. As a result, pastoralists have lost access to highly productive commons (pastures, water, and forests) that they have been using for centuries. This further increases the pressure on land resources. Some reports claim that the introduction of LSLIs in the agropastoral areas of Ethiopia has harmed pastoral welfare and livestock productivity (Ibrahim, 2016;Mekuyie, Jordaan, & Melka, 2018;Mousseau & Martin-Prével, 2016). However, these reports are mainly qualitative and largely ignore the potential land degradation effects of LSLIs. The majority of the existing studies in Ethiopia focus on landuse and environmental changes (Berihun et al., 2019;Meaza et al., 2019;Nyssen et al., 2014;Tsegaye, Moe, Vedeld, & Aynekulu, 2010), land management (Chesterman et al., 2019;Shiferaw & Holden, 1998), and land tenure (Nega, Adinew, & Gebresillase, 2003;Taddese, 2001). Furthermore, these studies have mainly been conducted in highland, non-pastoral areas. No studies have quantitatively investigated the impact of LSLI-induced displacement on land degradation in the agro-pastoral context. Moreover, the use of local pastoral knowledge in assessing land degradation has seen little application by scientists. Therefore, this paper investigates the effect of LSLI-induced displacement on land degradation from a community and household perspective. The study addresses the following specific research questions: (a) What is the impact of household displacement due to large-scale land investments on land degradation? We applied a two-stage stratified random sampling method in which we first selected the two woredas purposively for the presence of LSLI, and then we identified four kebeles, 2 two adjacent kebeles, and two distant (out of five-kilometre radius to the LSLI). The sample was taken from the 2018 population of Fentale (113,902) and Dubti (102,936) (Central Statistical Agency, 2013), by Yamane's formula (Yamane, 1973) at a 95% confidence level, ±5% precision, and 8% contingency. A sample of 440 households from Fentale and 430 households from Dubti were interviewed. After data screening, 866 households were included in the analysis. We adopted the methodology for the local-level assessment of land degradation in drylands developed by the Food and Agriculture Organization of the United Nations (FAO) (Nachtergaele & Licona-Manzur, 2008). Following this methodology, we first conducted community focus group discussions (FGDs) before the formal survey. A total of 43 male and 16 female elders participated in eight FGDs. We conducted the FGDs to identify the major land-use types, livelihood activities, and indicators, causes and trends of land degradation over the last 30 years. We follow Nachtergaele and Licona-Manzur (2008) to rank indicators of the severity of land degradation as none, light, moderate, and severe. Based on Willemen et al. (2018), we identified 15 indicators of land degradation common to all kebeles of the study areas. Insights from the FGDs were used to develop the household survey. Apart from the indicators of land degradation, the questionnaire consists of household characteristics, socioeconomic, institutional, and environmental factors.

| A conceptual model of the study
Large-scale land investments have led to the displacement of pastoral communities, limiting their access to common property resources (grazing, forests, and water) from which the community previously derived their livelihood (Rettberg, 2010). They deprive the customary land rights of the pastoral communities who used the areas as dry season grazing (Ibrahim, 2016). Displaced communities may end up landless or lose entitlements to dry season grazing.
Hence the loss of access to common property resources by displaced people increases pressure on forest lands, rivers, and grasslands, negatively affecting their ecological resilience (Terminski, 2013).
While pastoral production systems require mobility for access to grazing and water, LSLIs restrict access to dry season grazing by blocking paths to water and pasture (Mousseau & Martin-Prével, 2016). As a result, livestock overgrazes sparse grazing land leading to more land degradation (Mousseau & Martin-Prével, 2016). The denial of grazing land forces the displaced people to destroy natural resources for survival (charcoal and firewood selling) and over-exploitation of the remaining grazing areas (Fernandes, 2001;Terminski, 2013). Thus, displacement leads to natural resources degradation. Figure 2 illustrates how land degradation is caused by displacement 3 via loss of access to common resources and overgrazing. Other factors, such as household and farm characteristics, socioeconomic, institutional, and environmental factors also influence land degradation (Berry, 2003;Kertész, 2009;Young, 1994).

| Econometric model
In observational studies where there is no baseline information available, quasi-experimental designs such as matching techniques, F I G U R E 1 Map of the study woredas [Colour figure can be viewed at wileyonlinelibrary.com] instrumental variables, endogenous (switching) treatment, and inverse probability weighting methods can be applied (Baker, 2000). The appropriate method depends on the nature of the selection process.
In our case, the displacement of households is non-random, and not all agro-pastoralists are exposed to the treatment of displacement leading to selection bias. Those households living in areas with good soil and water are likely to be more vulnerable to displacement because LSLIs have been located on good quality land (Lay, Nolte, & Sipangule, 2018). As a result, unobserved factors that drive LSLIs and hence displacement may also affect land degradation (simultaneity) and, therefore, displacement is considered to be endogenous. To address this endogeneity problem, we adopt an endogenous (switching) treatment regression models (Maddala, 1983). The endogenous treatment regression model (ETRM) uses a linear model for the outcome and a constrained normal distribution to model the deviation from the conditional independence assumption imposed. The endogenous switching regression model (ESRM) accounts for observed and unobserved bias by estimating a simultaneous equation model (Lokshin & Sajaia, 2004). The ETRM is nested in ESRM and is widely applied in the analysis of the welfare impact of policies and technology adoption (Adebayo, Bolarin, Oyewale, & Kehinde, 2018;Adego, Simane, & Woldie, 2019;Adela & Aurbacher, 2018;Heckman, Tobias, & Vytlacil, 2003;Mekonnen, 2017). Other relevant studies have applied ESRM to analyse the effect of land-use changes on water quality (Abildtrup, Garcia, & Kere, 2015), the effect of a large dam on agricultural production (Chen, Hsu, & Wang, 2018), the effect of climate exposure on afforestation (Oyekale & Oyekale, 2019), and the effect of forced displacement on income (Do Yun & Waldorf, 2016).
The specification 4 of the endogenous switching regression model is as follows (Lokshin & Sajaia, 2004): T i is the treatment of the ith household, 1, if displaced and 0 otherwise.
Y ji is the land degradation index for household i in treatment j.
Z i is a vector of factors that influence the probability of the treatment. 5 X i is a vector of factors that influence the level of land degradation.
β 1 , β 2 , and γ are vectors of parameters, and ε 1, ε 2 , and u i are the error terms (a trivariate normal distribution with mean vector zero and non-singular covariance matrix).
The average treatment effect of the treated (ATT) and the average treatment effect on controls (ATC) in a counterfactual framework can be defined as follows:

| Dependent variable
Land degradation is complex, and there is no widely accepted measurement of land degradation (Dubovyk, 2017;IPPC, 2019). Generally, the use of multiple indicators for land degradation assessment is advised (Walpole, 1992). Following Walpole (1992)  (erosion), gullies, soil pollution, water pollution, salinity, loss of wildlife, forest loss, depletion of soil nutrients, landslides, dried up water bodies, lost wetlands, weed invasions (Prosopis 6 ), pollution of the land, and loss of soil dry matter. The majority of the indicators of land degradation that were identified by the community in our study were also reported in research in the Ethiopian Rift Valley areas (Meaza et al., 2017;Nyssen et al., 2014). In the survey, households rated each indicator as not visible, light, moderate, and severe on a 1-4 scale (see Appendix 1). We then constructed a land degradation index (LDI) for each household, which is the average value of the 15 indicators, and used this as the dependent variable in the model.
I is the value of each land degradation indicator, i is a household, j is the type of indicator, and n is the total number of indicators.

| Independent variables
The main independent variable of interest is the displacement caused by LSLI. This variable measures whether a household was displaced from their land due to LSLI in the last 30 years or not. About 24.5% of the sample households were displaced because of LSLI in the last three decades. Large-scale land investment locations are determined by land quality, water availability, and infrastructure (Lay et al., 2018).
Thus, distance from LSLI (LSLI_km) and the number of family members employed in the formal sector (N_employed), which drive displacement, were used as exclusion criteria. 7 Table 1 displays all the independent variables with their descriptive statistics. The drivers of land degradation are derived from the literature (Berry, 2003;Kertész, 2009;Nyssen et al., 2014;Tsegaye et al., 2010;Vu, Le, Frossard, & Vlek, 2014;Waswa, 2012;Young, 1994 increases farmers' ability to conserve land (Mango, Makate, Tamene, Mponela, & Ndengu, 2017). Age of the household head and household size has been found to improve the adoption of land conservation practices and limit the occurrence of land degradation (Kirui & Mirzabaev, 2015). Population density leads to more land degradation in highland Ethiopia (Nyssen et al., 2014). Farm characteristics also influence land degradation (Kirui & Mirzabaev, 2015;Kosmas et al., 2016). The number of livestock may increase land degradation through the effect of overgrazing (Jolejole-Foreman et al., 2012).
Cooperative membership helps households to share knowledge, labour, and skills and helps to acquire inputs to combat land degradation . Proximity to the market decreases the adoption of sustainable land management (Kirui & Mirzabaev, 2015).
A higher income may help to invest in the sustainable use of land and the conservation of land. Thus, credit access can also contribute to reducing land degradation (Kirui & Mirzabaev, 2015). Mobility is a strategy for the efficient use of scarce pastures (Davies et al., 2016).
Extension services can include training on the sustainable use of natural resources (Mango et al., 2017). Farmland tenure helps to combat land degradation (Kirui & Mirzabaev, 2015) because the security of land ownership provides incentives for sustainable land use .
Finally, climate factors, such as rainfall and temperature, may also affect the extent of land degradation (Meaza et al., 2017;Meaza et al., 2018).
To capture the effect of climate change, we include households' drought exposure. Drought indicates long dry seasons with the absence of rainfall and very high temperatures. Drought shocks are expected to lead to more land degradation (Ariti, van Vliet, & Verburg, 2018;Davies et al., 2016;Demeke, Guta, & Ferede, 2006). Pastorals live in conflictprone areas because of the nature of mobility, and conflict exposure is expected to increase land degradation. Only less than 1% of the employees in the LSLI are from pastoral communities.

| Level and causes of land degradation
We asked respondents to rate the extent of land degradation for each  (Table 3).
Forestlands and grasslands were rated as the most severely degraded land use, followed by grazing land (areas grazed by livestock including grasslands and shrublands) and water resources. A higher proportion of the displaced households was affected by moderate/severe land degradation compared to the control households.  Next to droughts, deforestation is seen as a major cause of land degradation. In this respect, the respondents indicate LSLI (83%), charcoal (57%), and firewood collection (54%) as the main drivers of deforestation. Charcoal production was introduced to the area by the highland labour migrants that were attracted by the LSLIs. Furthermore, LSLIs cleared natural vegetation and forests that had existed on the land before their establishment (Ibrahim, 2016 About 28.64% of households point to high-intensity overgrazing, contributing to land degradation. Less than 25% of households report the causes of severe land degradation to be wind and water erosion, over-cultivation, settlement, and poor irrigation practices. On average, displaced households give a higher weight to the different causes of land degradation than control households, except in the case of poor irrigation practices. Figure 4 illustrates the perceived effects of land degradation on people's livelihoods by control, treated, and total household categories. More than 86% of the respondents associate land degradation with lower crop and livestock productivity. Moreover, 71.6% of households associate land degradation with increased death of livestock and 48.9% with crop failure. FGD participants explain this by the loss of access to common resources and the poor quality of the soil since LSLI establishment. For instance, maize yield declined from 1.5 metric tt ha -1 to 0.7 metric t ha -1 in Dubti after the establishment of the plantation (Planel & Labzaé, 2016). In the survey year, a household, on average, reported the death of 10 goats, six sheep, three cows, five oxen, four camels, two poultry, and two donkeys. This is in line with studies that show increasing livestock mortality in the region (Ariti et al., 2018;Ibrahim, 2016;Ioras et al., 2014). Similarly, milk production has been declining in the past 30 years for camels from 15 L to less than 2 L per day, for cows from 10 L to 1 L per day, and for goats from 5 L to less than 1 L per day. As a result, there is no or only a little surplus of milk to be marketed. 97.8% of the respondents claim that desertification has been rising.

| Empirical model results
Tables 4 and 5 show the ETRM model results and the average treatment effects of being displaced, and Appendix 4 reports the ESRM regression results. The full information maximum likelihood jointly estimates the selection and treatment equations efficiently (Lokshin & Sajaia, 2004). For both ETRM and ESRM models, the Wald tests show that the regression models fit the data well. The likelihood-ratio test for independence of the treatment and outcome equations also suggests the rejection of the null hypothesis of no correlation between the treatment and outcome errors, indicating an endogeneity problem that should be solved. In ETRM, the correlation coefficients between the error terms of the displacement and land degradation equations are negative and significant. The significance and negative correlation between error terms, respectively, show the presence of a selection bias and unobservables that raise LDI while lowering displacement. In

| DISCUSSION
Overall, we find evidence that displacement due to LSLI increases the intensity of land degradation, causing the deterioration of livelihoods.
The finding is in line with the observation that displacement is causing environmental degradation in Africa (Mohamed, 2016). Over 75% of the respondents have encountered a moderate-severe level of land degradation. Displaced households, on average, have 0.56-0.91 units higher levels of land degradation compared to non-displaced households. Drought, deforestation, and LSLI were seen by community members as the key drivers of land degradation. LSLI-induced displacement was also identified as a significant driver of land degradation in the econometric analysis.
There have been two general debates in the academic literature regarding the causes of land degradation, Hardin's tragedy of the commons (Hardin, 1968) and Ostrom's counter-argument to the tragedy of commons (Ostrom, Burger, Field, Norgaard, & Policansky, 1999).
The "tragedy of the commons" hypothesis argues that the communal ownership of resources leads to their degradation and recommends privatizing property rights (Hardin 1968 However, neither of these two debates acknowledges the tragedy that may result when powerful external groups take control of resources to gain personal advantage without consultation or compensation of local communities. 10 The latter is exactly what was found in the current research: LSLIs restrict pastoralists' access to grasslands and water, leading to increased scarcity of dry season grazing and pressure on pasture lands. Moreover, as LSLI capture the most productive land that has been used for dry season grazing, the overall productivity of the grazing land declines (Abbink et al., 2014) and, for instance, in Afar, the appropriation of land for LSLI has increased the incidence of overgrazing (Sonneveld et al., 2010). In contrast to the "tragedy of the commons" concept, overgrazing in the study area is not the result of the accumulation of livestock and the free-rider problem (Cox, 1985;Hardin, 1968). Instead, it is due to the denial of access to grazing land that disrupted the mobility pattern of pastoralists and their livestock (Beza & Assen, 2017) (Beza & Assen, 2017;Cox, 1985;Said, 1994). [FDRE], 2019). Second, in Ethiopia, pasture land governance has welldefined user rights, access conditions, set rules and norms (Beyene, 2016), and exclusion criteria to prevent outsiders from exploiting the resource (Beyene, 2006).
As the state owns all the land in Ethiopia, pastoralists can easily be removed from their ancestral land and the result for the effect of farmland tenure on land degradation should be interpreted with caution. On the one hand, we do find that land tenure security reduces land degradation and ensures sustainability (Ariti et al., 2018;Kirui & Mirzabaev, 2015). However, the notion of land certification poorly fits with pastoral livelihoods. Considering the transhumance nature of pastoralist systems, what matters for pastoralists is access to pastures rather than a specific piece of land (Dell'Angelo et al., 2017). Loss of access to the commons undermines pastoralist livelihoods unless there is compensation with a land of equivalent or superior quality (Vanclay, 2017). Therefore, the recognition of collective tenure rights to the commons and mobility is a cornerstone of sustainable development and optimizing scarce pastures (Butt, 2010;Davies et al., 2016).
According to Dwivedi (2002), there are two arguments regarding development-induced displacement (such as LSLIs

| CONCLUSIONS AND POLICY RECOMMENDATIONS
This study provides evidence on the effect of LSLI-induced displacement on land degradation in agro-pastoral areas of Ethiopia. The results reveal that LSLI areas expand by displacing households and restricting access to pastures and other resources in the study area.
This aggravates the scarcity of pasture lands and hence leads to land degradation.
LSLI aggravated land degradation directly by destroying common resources (clearing of vegetation and grass) in favour of plantation production and by dispossessing grazing land and exacerbating overgrazing. Significant proportions of the households in our study perceive that poverty and conflicts have been increasing while herd size has shown a declining trend as a result.
LSLI-induced displacement significantly worsens land degradation. There is strong evidence that the land of the displaced households has suffered significantly more degradation than that of the control households. Loss of access to productive dry season pasture and dispossession of former pasture is a major driver of overgrazing.
This is also related to the absence of property rights for the commons.
We conclude that displacement increased the severity of land degradation. However, also non-displaced households may face negative externalities from LSLI, such as the discharge of polluted water and a decline in ecosystem services (loss of native vegetation and forest products Finally, we thank the section editors, the review editor, Katrin Prager, and a number of anonymous reviewers.

CONFLICT OF INTEREST
The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
2 Kebele is the fifth administrative level in Ethiopia (Federal-Regional-Zonal-Woreda-Kebele). 3 The concept of displacement is adopted from Bartolome, De Wet, Mander, and Nagraj (2000) and refers to the alienation of the individual and community customary rights and permanent dislocation of the social and economic organization. The displacement is induced by policy (in our case, LSLIs). The dislocation of people from their homeland territory without social support in the new place of residence is a violation of the most fundamental human rights (Terminski, 2013). 4 Without loss of generality, we specify the models based on treatment effect literature, and we adopt a single equation for the ETRM and ESRM models (Heckman et al., 2003;Lokshin & Sajaia, 2004 ) such as the treatment equation, the outcome equations on treated and the untreated with trivariate endogeneity error terms correlations. 5 We include in the vector Z i some variables that do not belong to the vector X i to make the estimation more robust and improve identification. We recognize that some of the important determinants of the treatment are exogenous factors (policy decisions) and are difficult to obtain. Hence we use as exclusion restrictions, LSLI_km, ROAD_km, N_employed that affect the selection variable but not the outcome variables.
6 Prosopis was introduced into Ethiopia in the 1970s as a soil conservation measure, with high drought tolerance. In Afar region, the plant is now covering over 1.2 million hectares (FDRE, 2017). 7 Exclusion criteria for treatment in ESRM are not strictly required for identification as the non-linearity assumption of the error term. We include them for a more robust estimation of the regression. 8 During the survey year 2019 for January on average 1 USD equals 28.11 Ethiopian Birr. 9 We use etregress for ETRM and movestay for ESRM. Both models capture the treatment with high predictive power. The ATT in ETRM is lower than ESRM. This may be because the treatment effects from ETRM are constrained in the absence of interaction between treatment variable and covariates of the outcome variables (except for MARKET_km).
10 In Dubti-Afar study sites, the FGD participants reported that the government negotiated with clan leaders about taking the land for sugar plantation. The actual people affected by the Tendaho plantation did not receive compensations; however, a few traditional leaders and elites received money and forced the community to relocate as the government's development plan for sugar plantation is compulsory.