Land Use Policy 72 (2018) 270–279 Contents lists available at ScienceDirect Land Use Policy journal homepage: www.elsevier.com/locate/landusepol Land ownership and technology adoption revisited: Improved maize T varieties in Ethiopia ⁎ Di Zenga, , Jeffrey Alwangb, George Nortonc, Moti Jaletad, Bekele Shiferawe, Chilot Yirgaf a Centre for Global Food and Resources, 6.24 Nexus 10, University of Adelaide, Adelaide, SA 5005, Australia b Department of Agricultural and Applied Economics, 215I Hutcheson Hall, Virginia Tech, Blacksburg, VA 24061, United States c Department of Agricultural and Applied Economics, 205B Hutcheson Hall, Virginia Tech, Blacksburg, VA 24061, United States d International Maize and Wheat Improvement Center (CIMMYT), P.O. Box 5689, Addis Ababa, Ethiopia e World Bank Group, 1818 H Street NW, Washington DC 20433, United States f Ethiopian Institute of Agricultural Research (EIAR), P.O. Box 2003, Addis Ababa, Ethiopia A R T I C L E I N F O A B S T R A C T Keywords: The lack of land ownership can discourage agricultural technology adoption, yet there is scarce evidence of the Land ownership impact of land rental contracts on the adoption of improved crop varieties in developing countries. The current Technology adoption study investigates such impact using a nationally representative survey of Ethiopian maize farmers. In contrast to Maize many previous studies, we show in a simple model that cash-renters are as likely to adopt improved maize Improved varieties varieties as owner-operators, while sharecroppers are more likely to adopt given that such varieties are prof- Land contract Ethiopia itable. Empirical analysis reveals a significant impact of sharecropping on improved maize variety adoption, and no significant impact from cash-rental, lending support to the above hypotheses. These results imply that im- provements in land rental markets can potentially enhance household welfare through crop variety adoption in agrarian economies where land sales markets are incomplete or missing. 1. Introduction Empirical findings of this literature, however, are mixed the hy- pothesized impacts can sometimes bear opposite signs and their mag- Land ownership, or land tenure, has been increasingly investigated nitudes are usually small (see Brasselle et al., 2002; Place, 2009; as a factor affecting modern agricultural technology adoption in Sub- Fenske, 2011 for literature syntheses from different perspectives). Such Sahara Africa (SSA). From both theoretical and empirical perspectives, inconclusiveness is partly due to the failures to differentiate the varying Gavian and Fafchamps (1996) find secure tenure encourages invest- characteristics of agricultural technologies. For example, Deininger and ments in soil conservation technologies in northern Ethiopia. Abdulai Jin (2006) show that the lack of land ownership, or tenure insecurity, et al. (2011) conclude that land ownership tends to facilitate invest- can either discourage agricultural technology investment (if ownership ment in soil-improving and natural resource management practices in security is exogenous) or encourage investment (if ownership security is Ghana. Oostendorp and Zaal (2012) also suggest that transfer rights, a endogenous). The latter observation accords with earlier literature that measure of land ownership, stimulate the adoption of soil and water the threat of non-renewal may cause tenants to work harder and pro- conservation technologies in Kenya. It is generally hypothesized that duce more (Cheung, 1969). Place (2009) further shows that the di- land ownership encourages agricultural technology adoption, while the vergent impacts of land ownership on the adoption of different tech- lack of land ownership discourages it. The underlying argument is that nologies in a comprehensive literature review. Hence, characteristics of the lack of landownership, as usually reflected in land rentals, may agricultural technologies need careful differentiation to help disen- preclude tenants from future technology-induced benefits due to the tangle any confounding impacts in search of policy implications. risk of eviction. Land ownership, on the contrary, can safeguard cash While most studies in this literature focus on resource-conserving flows over time and facilitate asset liquidation given transferrable land technologies, modern agricultural technologies also include pro- rights and can also enhance access to resources such as credit (Feder ductivity-enhancing ones such as improved crop varieties and fertilizer and Nishio, 1998). All these factors can incentivize the adoption of (Ersado et al., 2004), and possible impacts of land ownership on the technologies that require investments and that potentially increase the latter need to be better understood. This literature bias could be partly value of land. driven by the belief that land ownership affects only long-term ⁎ Corresponding author. E-mail address: di.zeng@adelaide.edu.au (D. Zeng). https://doi.org/10.1016/j.landusepol.2017.12.047 Received 11 March 2016; Received in revised form 18 October 2017; Accepted 15 December 2017 0264-8377/ © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/). D. Zeng et al. Land Use Policy 72 (2018) 270–279 investments related to natural resource management but not short-term hiring were legalized. However, as permanent land transfer continues input use decisions. However, this is not generally true because even to be prohibited by enforced policies, the land sales market is still seasonal crop variety choices may have deferred impacts on pro- nonexistent in rural Ethiopia. Land inheritance is allowed and creates ductivity or risk-mitigation that could affect investments, which in turn incentives for land rentals (Crewett and Korf, 2008). As a result, the depends on land ownership. Although a few recent studies have ana- short-term land rental market is expanding, and plot rentals are lyzed the impact of land ownership on the adoption of fertilizer (e.g. common due to land fragmentation (Benin et al., 2005). The scenario of Abdulai et al., 2011; Ali et al., 2012), these studies fail to consider crop Ethiopia therefore provides a unique context of study as the land rental variety choices which potentially affect fertilizer application decisions market plays an active role to meet the expanding land demand without (Heisey and Norton, 2007). Despite land market imperfections, im- land sales market, and possible policy implications of the current study proved crop varieties have been a major driver of agricultural pro- may also apply to other agrarian economies in SSA where land sales ductivity growth in SSA (see Evenson and Gollin, 2003 for a compre- markets are underdeveloped and land rental widely exists. hensive cross-country analysis), which further results in welfare Maize is one of the most important food and cash crops in SSA. In improvements in terms of poverty reduction (Kassie et al., 2011; Zeng Ethiopia, maize accounts for the largest share of production by volume et al., 2015), and food security (Shiferaw et al., 2014). Understanding and is produced by more farmers than any other crop (Chamberlin and how land ownership affects improved crop variety adoption is therefore Schmidt, 2012). During the 2009–2010 production year, Ethiopia highly relevant in assisting ongoing market-oriented land reforms in produced 3.89 million tons of maize on 1.77 million hectares of land SSA. (Central Statistical Agency, 2010). The average productivity of 2.20 Empirical identification of the hypothesized impact is difficult due tons per hectare was the highest among all cereal crops in the country. to confounding effects. For instance, resource-conserving practices such In the last four decades, more than 40 improved maize varieties as tree planting can be adopted to demonstrate and strengthen claims to have been developed through joint efforts of the Ethiopian Institute of land rights (Place and Otsuka, 2002), while productivity-enhancing Agricultural Research and the International Maize and Wheat practices such as organic fertilizer that improves soil capital can also be Improvement Center (CIMMYT). Improved maize seeds have been dif- adopted by tenants to increase the chance to continue land operation in fused mainly through the Ethiopian Seed Enterprise, the major seed the future (Abdulai et al., 2011). In both cases, causality can be reverse, producer and distributor, while regional seed enterprises such as but such potential endogeneity is not commonly recognized (Brasselle Oromia Seed Enterprise, Amhara Seed Enterprise, and Southern Seed et al., 2002; Fenske, 2011). Moreover, although the lack of land own- Enterprise also produce and sell maize seeds. Improved varieties are a ership is mainly manifested through land rental contracts,1 few studies major contributor of maize productivity growth. Recent literature as- differentiate contract types such as cash-rental or sharecropping due to sociates the adoption of improved maize varieties in Ethiopia to a data limitations. These complexities need to be clearly understood by 47.6%–63.3% yield increase and consequently a 0.8–1.3 percentage policy makers who hope to improve rural welfare from this perspective. reduction of poverty headcount ratio (Zeng et al., 2015). The current study assesses the impacts of land rental, as associated Improved maize varieties can be categorized as either hybrid or with two most important land rental contracts (cash rental and share- open-pollinated improved varieties (OPVs). Hybrids have the highest cropping), on improved crop variety adoption using a nationally re- yield, but are more costly to adopt as the restoration of hybrid vigor presentative survey of maize farmers in Ethiopia. We show in a simple requires purchasing new seeds in each cropping season. The yields of model that, unlike the case of resource-conserving technologies, land OPVs are generally lower than those of hybrids (though still much rental does not discourage the adoption of improved crop varieties for higher than those of local varieties), but OPV seeds cost less than those cash-renters, but encourages adoption for sharecroppers if such vari- of hybrids and may be recycled for up to three cropping seasons without eties are profitable. Empirical evidence is robust in support of these significant yield loss. Many OPVs have specific traits which make their hypotheses, suggesting that improvements in land rental markets can yields robust against challenging conditions such droughts and pests. potentially enhance household welfare in agrarian economies where Seed recyclability also makes them especially attractive for areas with land sales markets are incomplete or missing. underdeveloped seed markets (Jaleta et al., 2013). Adoption of improved maize varieties varies across agro-ecological regions throughout Ethiopia (Jaleta et al., 2013). Our data suggest that 2. Land ownership and maize production in Ethiopia adoption rates as measured by area are higher in places of higher maize potential. No single variety dominates the whole adoption scenario, but Land tenancy in Ethiopia has a long history, which stems from the hybrids are more popular than OPVs in general. feudal system that existed before the Derg government took power in 1974. Land distribution was skewed and a large share of land was op- 3. Theoretical framework erated by tenants. Early literature shows that the share of rented land was over 40 percent, and operating tenants represented a similar pro- To illustrate the potential relationship between land rental contracts portion of the total population (Rahmato, 1984). Sharecropping was the and the adoption of improved crop varieties, we build a simple theo- dominant type of land rental (Holden et al., 2008). retical model below. Our model is comparable to the mainstream lit- Land rental has been present in Ethiopia throughout history. erature that links land ownership and agricultural technology adoption Arbitrary eviction of tenants was a major feature of the land rental (e.g. Deininger and Jin, 2006), but extends it to differentiate cash-rental system in the feudal society (Deininger and Jin, 2006). The land reform and sharecropping contracts and to capture the specific characteristics in 1975 confiscated all land as state property, and cultivators were left of productivity-enhancing technologies. In this model, each farmer i is with only user rights but prohibited from land rentals and labor hiring categorized as an owner-operator (O ), a cash-renter (H ), or a share- (Holden et al., 2008). Further, land redistribution through govern- cropper (S ). Regardless of land ownership, farmer i maximizes the mental power was common during the Derg regime under the stated present value (PVi ) of current cropping returns (Ri ) , net of total costs objectives of overcoming inequality and landlessness (Fenske, 2011). (Ci ), plus the expected future net returns, Vi , consisting of all future Since the current government took power in 1991, land redistributions revenues assumed to be realized in the second period and possibly were largely reduced (with the exception of land redistribution in downscaled by a tenure risk indicator, θi (0 ≤ θi≤1), due to the risk of Amhara region in 1997–1998) and short-term land renting and labor losing land use rights (with r denoting the discount factor): Vi 1 Land rental contracts can be either written or oral, and it is often the latter in rural maxPVi = Ri−Ci + θi 1+r (1) Ethiopia (Holden et al., 2008). 271 D. Zeng et al. Land Use Policy 72 (2018) 270–279 The total cost, Ci , captures both production/technology inputs, Ii , This result is intuitive because crop revenues are received after each and possible land cost, Li , i.e. Ci = Ii + Li . Ri and Ci can be functions of cropping season, and therefore the cash-renter adopts improved crop household characteristics, plot characteristics, and crop variety adop- varieties as long as they are profitable in the current period. The risk of tion decision. For illustrative purposes, we consider the adoption de- eviction in the future simply does not play a role as it is not changeable cision, Ai , as a continuous argument of both Ri and Ci . Ai also reason- by current adoption, though it can increase current and thus the total ably captures the effort level of farmer i in face of extra investment discounted profit over time, PVH . This result differs from that for other reflected in Ci . In reality, it can be either the acreage of improved crop agricultural technologies such as conservation strategies which produce varieties (measured in hectares) or the area share of improved crop future benefits. In the latter cases, land rental can preclude tenants from varieties among the whole cultivated area (measured as a ratio between future technology-induced benefits through eviction and may dis- zero and one). Assume that both Ri and Ci are fully differentiable with courage adoption (Soule et al., 2000). respect to Ai and satisfy standard properties that suggest the existence The case of the sharecropper is more complicated than that of the of a unique optimum.2 On the other hand, unlike other agricultural cash-renter for two reasons. First, unlike either owner-operators or technologies that affect long-term productivity, Vi is not likely to be cash-renters who receive all revenues and bear all costs, a sharecropper affected by season-specific crop variety choices.3 For this reason, Vi is assumes only a share of each and the shares can differ. Instead of paying assumed not to vary among farmer types, and so VO = VH = VS = V . Eq. a separate land rent, the sharecropper forgoes a proportion of the (1) can therefore be written as: cropping revenue in exchange of land use rights. Second, although fu- ture revenues (V ) are not likely altered by crop varieties adopted in the V maxPVi = Ri (Ai )−Ii (Ai )−Li (Ai ) + θi current period, the chance of continuing cultivation on the plot, again 1+r (2) captured by the tenure risk indicator, can indeed be affected as a profit- Eq. (2) describes the problem for all types of land ownership con- driven landlord may choose to continue contracting with a share- sidered in the analysis: owner-operator (O ), cash-renter (H ), or share- cropper who generates higher returns for him/her through the adoption cropper (S ). For the owner-operator, LO = 0 as he/she does not pay for of improved crop varieties. Given these differences, we can write the the owned land, and θi = 1 as there is no risk of eviction from the op- sharecropper’s problem as: * erated land. The optimal adoption level ( AO ) is derived according to the V following first-order condition: maxPVS = αRS (AS )−βIS (AS ) + θS [(1−α ) RS (AS )] ∙ 1+r (5) AO (∂RO / ∂AO −∂IO / ∂AO ) = 0 (3) where α is the revenue share (0 < α<1); β is the cost share (0 < β<1); θS Eq. (3) satisfies complementary slackness: either AO > 0 and is the tenure risk indicator (0 ≤ θS≤1 due to the risk of eviction); and ∂RO / ∂AO −∂CO/ ∂AO = 0 , or AO = 0 and ∂RO / ∂AO −∂CO/ ∂AO < 0 .4 α ≠ β in general. The third term is the tenure-risk weighted, discounted The case of the cash-renter is different as the land cost is no longer expected future net returns received by the sharecropper. In this case, zero, i.e. LH > 0 . However, since the land rent is fixed, it does not vary the land cost of the sharecropper is computed as the difference between with the tenant’s adoption behavior, i.e. ∂LH / ∂AH = 0 . The tenure risk the forgone revenue and the compensated input cost (shared by the indicator, θH , can take any value between zero and one, yet it is not landlord), i.e. (1−α ) RS (AS )−(1−β ) IS , and therefore Eq. (5) describes a likely affected by varietal adoption either since the rent is pre- specific scenario derived from Eq. (2). Unlike the cash-renter case, θS is determined free of crop variety choices.5 Moreover, the expected future now affected by the landlord’s share of cropping revenue, (1−α ) RS , net returns are again not affected by adoption for reasons above. where the total revenue (RS ) is further affected by the sharecropper’s Therefore, the first-order condition for the cash-renter can be expressed adoption decision, AS . The first-order condition can then be written as: using similar notation as: ∂RS ∂I ∂θ ∂R V α −β S + (1−α ) S ∙ S ∙ =0 ∂RH / ∂AH −∂IH / ∂AH = 0 (4) ∂AS ∂AS ∂RS ∂AS 1 + r (6) Therefore, the first-order condition for the cash-renter as well as the Eq. (5) intuitively suggests the equimarginal principle: at the op- optimal level of adoption is exactly the same as that for the owner- timum, the marginal benefit of improved crop variety adoption, in- operator. This leads to our first testable hypothesis: cluding current marginal revenue plus future marginal revenue due to Hypothesis 1. Land rental does not affect the adoption decisions of cash- increased chance to cultivate the same land in the future, is equal to the renters, who are as likely to adopt improved crop varieties as owner-operators. marginal cost of adoption the sharecropper bears in the current period. To derive the comparative statics of interest, i.e. the impact of sharecropping contract on improved crop variety adoption, ∂AS / ∂θS , 2 For the rest of the analysis, it is only necessary to explicitly assume that ∂Ri / ∂Ai > 0 , we formally assume ∂θS / ∂RS ≥ 0 . It is intuitive that higher revenue in i.e. adoption of improved crop varieties increases revenue with higher yields. It is trivially most cases may financially incentivize a landlord to continue the con- satisfied for most improved crop varieties. 3 This argument could be challenged if improved maize varieties are adopted with tract with a sharecropper. In the first (and rare) case where the landlord other technologies, particularly fertilizer as the residual of fertilizer might go to the next is profit-neutral and an interior solution exists, i.e. ∂θS / ∂RS = 0 , Eq. (5) season. In our survey, fertilizer application is very common (utilized by more than 90% of reduces to: surveyed farmers), and so its association with improved maize variety adoption is not strong. Including fertilizer use is therefore less informative given its limited variation and α (∂RS / ∂AS )−β (∂IS / ∂AS ) = 0 (7) is problematic due to endogeneity and associated bias. Therefore, we choose to exclude fertilizer use in our empirical modeling. Under the equal share rule that α = β , it is straightforward to see 4 For the rest of the analysis, we shall focus on interior solutions to derive comparative that the first-order condition and optimal level of adoption for the statics. Obviously, some operators would not adopt improved crop varieties for a number sharecropper is exactly the same as that for the owner-operator, as of reasons regardless of the tenure type (owner-operator, cash-renter and sharecropper), shown in Eq. (2). In the real world, however, departures from the equal yet identification of those constraints are beyond the scope of the current study. Therefore, it is of our primary interest to see how tenure arrangements would affect the share rule are also observed, where issues such as output uncertainty level of improved variety adoption which, for illustrative purposes, could be better cap- and input monitoring difficulty may result in contracts where α ≠ β tured by comparative statics assuming an interior solution. (Braverman and Stiglitz, 1986). In these cases, a sharecropper can be 5 It could be argued that the profitability and increasing adoption of improved maize either more or less likely to adopt improved crop varieties, but all these varieties might increase cash rents over time. If that is the case, a positive correlation cases would be trivial in the real world as most landlords are profit- would be expected between cash-rental and maize variety adoption. Empirical in- vestigation of this, however, would require multiple rounds of observations and is not driven (∂θS / ∂RS > 0 , the conventional rationality assumption). supported by our data. Nevertheless, the current approach does not lose generality in the As it is assumed in footnote3 that ∂RS / ∂AS > 0 , ∂θS / ∂RS > 0 further short-term investigation of the hypothesized impacts. implies that the last term of Eq. (5) is positive while the sum of first two 272 D. Zeng et al. Land Use Policy 72 (2018) 270–279 terms is negative regardless of the values α and β take. With implicit excluded and included households and the plots they operated are similar, function results, we finally derive Eq. (7) and our second testable hy- and find that they are. pothesis: Maize variety information was also reported at the plot level. Exact variety names recorded in the survey were classified as hybrid, OPV or ∂AS (1−α ) V = <0 local after detailed communication with breeding scientists in Ethiopia. ∂θS (1 + r )[α (∂RS / ∂AS )−β (∂IS / ∂AS )] (8) Although both hybrids and OPVs are widely adopted, few plants are genetically pure in maize cropping practices as inbred lines are usually Hypothesis 2. Sharecroppers are more likely to adopt improved crop crossed through open pollination. Also, yields of hybrids (3,543 kilo- varieties than owner-operators given that these varieties are profitable and grams per hectare) and OPVs (3,068 kilograms per hectare) are similar the landlord is profit-driven. in our data (with no statistical significance at 5% level through pairwise This result is again intuitive: as long as improved crop varieties are t-test). Therefore, varieties are only categorized as either improved profitable, a sharecropper will try to secure future profit flows through (including both hybrids and OPVs) or local in our analysis. As suggested extension of the land rental contract by adopting improved crop vari- by local breeding scientists, any hybrid variety ever recycled or an OPV eties. As most landlords are profit-driven, and given the fact that recycled for more than three seasons are categorized as local due to sharecropping has been the predominant land rental contract type in substantial productivity loss. Finally, the maize variety was unique for Ethiopia (Holden et al., 2008), we further expect a positive overall any plot in our data, and each plot is finally identified as having either impact of land rental on the adoption of improved crop varieties. Such an improved or a local maize variety. overall impact is also evaluated in our empirical analysis after the as- Based on the binary categorization of maize varieties, our survey sessment of contract-specific impacts of cash-rental and sharecropping. presents 541 non-adopters (who did not adopt any improved maize variety), 486 full adopters (who purely grew improved maize varieties) 4. Data description and 273 partial adopters (who grew both improved and local varieties on different plots). Household characteristics are reported in Table 1. The current analysis uses data from a household survey of Ethiopian Larger, wealthier land holders and those with more family members are maize farmers conducted during 2009–2010. Four regions were cov- more likely to adopt improved maize varieties. Adopting households, ered: Tigray, Amhara, Oromia, and Southern Nations, Nationalities, and including both full and partial adopters, differ systematically from non- People's Region (SNNPR), which together accounted for more than 93% adopting households in that the latter are poorer and are more likely of maize production in Ethiopia (Schneider and Anderson, 2010). The female-headed ones. Non-adopters also have smaller land holdings, live data were collected using a stratified random sampling strategy that farther from main markets, and observe lower maize yields. Moreover, appropriately accounted for the representativeness of areas with non-adopters are less likely to participate in land rental markets, varying maize potential, and therefore are nationally representative for through either cash-rental or sharecropping. the maize growing areas. Partial adoption of improved crop varieties widely exists in the The survey covered 1396 farm households from 124 villages (ke- developing world (e.g. Smale et al., 1994; Radhu et al., 2015; Kathage beles) in 30 districts (woredas) across the four regions, of whom 1359 et al., 2016). In our data, partial adopters have the largest total land- households grew maize on 2496 plots during the surveyed period. Basic holding, wealth and total maize area with market access better than demographics, including characteristics of the household head, were non-adopters but worse than full adopters (Table 1). The pattern that reported at the household level. Total land and asset holdings, and larger farms tend to be partial adopters is also observed in Radhu et al. infrastructure conditions such as distances (measured in walking min- (2015). While identification of the determinants of partial adoption is utes) to the nearest agricultural extension office (where most Ethiopian beyond the scope of the current study and is difficult with our cross- farmers buy improved crop seeds and meet extension personnel) and sectional observational data, literature generally suggests risk-averse main market, were also reported at this level. Detailed maize cropping farmers would partially adopt to reduce perceived downside production practices and physical conditions such as fertility, soil depth and slope risk of improved varieties (Smale et al., 1994; Krishna et al., 2016), were assessed by farmers at the plot level. especially in the absence of formal insurance markets (Baumgärtner Land use rights (ownership), crop choice and technology adoption and Quaas, 2010). Therefore, it is primarily the intermediate small- were reported at the plot level. Maize plots are classified as either “owned” holders that are fully incentivized to intensify crop production through or “rented in” (through either cash rental or sharecropping contracts). full adoption of improved maize varieties (Radhu et al., 2015). Further classification of the two types of land rental contracts, either cash- Among the 2359 maize plots operated by these 1300 households, rental or sharecropping, is facilitated using the information of in-kind 1965 were operated by owners, 89 by cash-renters and 305 by share- payments to land recorded in the crop utilization section of the survey. croppers. Plot characteristics are reported in Table 2. Rented-in plots After intensive discussions with local experts, sharecropped plots are are farther from operators’ homes and sharecropped plots are generally identified as those of positive in-kind payments to land of maize and cash- larger and more fertile. Adoption rates are higher on rented plots. rented plots are those of zero in-kind payments to land. In our survey, crop Sharecropped plots have slightly higher maize yields than owner-op- utilization is recorded as crop- and season-specific rather than plot-spe- erated ones, yet such difference is only marginally significant and could cific, and differentiation of cash-rental and sharecropping would have largely result from the higher adoption rate of improved maize varieties been impossible if a household operated more than one rented plot. with sharecropping. Finally, although “reverse tenancy” is common in Fortunately, 46.4% of households operated only a single maize plot in the Ethiopia (resource-poor landlords rent out land to resource-rich tenants surveyed period and 85.53% of households who cultivated multiple maize in change of fixed income to hedge against production risks, ses Ghebru plots owned all these plots. To help identify sharecropped plots, the and Holden, 2008; Holden and Bezabih, 2008), our data show no evi- magnitudes of in-kind payments are further checked with total plot-level dence for either land holding or wealth differentials by land ownership. maize outputs, and a few observations where in-kind payments exceed total output are identified as misreported and dropped. Finally, house- 5. Empirical strategy holds with either unidentifiable contract types or missing values are dropped, and 2359 maize plots operated by 1300 households remain in Our empirical analysis is implemented in three steps. First, we our empirical analysis with all necessary information. To minimize con- consider the contract-specific impacts of land rental on improved maize cerns about sample selectivity, we further check if the characteristics of the variety adoption and estimate separate impacts of cash-rental and 273 D. Zeng et al. Land Use Policy 72 (2018) 270–279 Table 1 Household Summary Statistics by Adoption Type.a Non-adoptersb (n = 541) Full adoptersb (n = 486) Partial adoptersb (n = 273) Household size 6.26 (2.22) 6.53 (2.42) 6.91 (2.40)**, † Total assets (ETB)c 13,474 (30,511) 18,982 (35,721)** 22,819 (61,171)** Head age (years) 43.66 (12.61) 41.86 (13.00)* 43.30 (11.40) Head gender (male = 1; female = 0) 0.92 (0.27) 0.95 (0.21)* 0.98 (0.15)**, † Head marital status (married = 1; other = 0) 0.91 (0.29) 0.95 (0.22)* 0.96 (0.19)** Head education (years) 2.54 (3.03) 2.94 (3.36)* 2.97 (3.13) Main market distance (walking minutes) 114.5 (59.97) 83.60 (54.11)** 97.20 (60.52)**, †† Extension office distance (walking minutes) 29.96 (30.50) 28.25 (27.78) 34.17 (35.02)† Unmet credit need (yes = 1; no = 0) 0.03 (0.15) 0.03 (0.16) 0.02 (0.17) Years living in village 37.21 (13.99) 36.55 (14.51) 37.78 (13.24) Cooperation membership (yes = 1; no = 0) 22.55 (0.401) 23.67 (0.412) 23.34 (0.404) Total land holding (ha) 2.10 (0.97) 2.13 (1.17) 2.37 (1.46)**, † Total maize area (ha) 0.55 (0.54) 0.69 (0.64)** 1.09 (0.99)**, †† Avg. maize plot area (ha) 0.37 (0.38) 0.45 (0.41)** 0.37 (0.28)†† Maize yield (kg/ha) 2170 (1,483) 3479 (2,176)** 2753 (1,380)**, †† Cash-renter proportion (%) 4.62 5.76 7.69 Sharecropper proportion (%) 12.75 20.78 23.81 a Standard deviations are in parentheses.* and ** indicate the variable mean differs from that of non-adopters at 5% and 1% levels, respectively. † and †† indicate the variable mean differs from that of adopters at 5% and 1% levels, respectively. b Non-adopters are those who did not grow any improved maize variety in the survey period; full adopters are those who purely grew improved maize varieties; and partial adopters are those who simultaneously grew both improved and local varieties on different plots. c Computed as the sum of reported current values of all itemized assets in Ethiopian Birrs (ETB). Daily average exchange rate in 2010 is 1 USD = 14.38 ETB. Table 2 Plot Summary Statistics by Tenure Type.a Owned (n = 1965) Cash-rented (n = 89) Sharecropped (n = 305) Plot size (ha) 0.39 (0.39) 0.41 (0.32) 0.49 (0.41)** Soil fertility (fertile = 1; otherwise = 0) 0.08 (0.27) 0.09 (0.29) 0.20 (0.40)**, † Soil slope (flat = 1; otherwise = 0) 0.67 (0.47) 0.63 (0.49) 0.67 (0.47) Soil depth (deep = 1; otherwise = 0) 0.41 (0.49) 0.42 (0.50) 0.41 (0.49) Distance from home (minute) 9.38 (16.96) 23.38 (31.29)** 24.28 (46.07)** Cropping season (long = 1; short = 0) 0.93 (0.26) 0.97 (0.18) 0.91 (0.29) Total household land holding (ha) 2.28 (1.42) 2.19 (1.16) 2.25 (1.40) Total household asset (ETB)b 19,580 (43,564) 25,846 (79,201) 19,882 (67,475) Maize yield (kg/ha) 2751 (2,024) 2614 (1,694) 2960 (2,029) Adopting area proportion (%)c 38.34 41.79 55.93 a Standard deviations are in parentheses. * and ** indicate the variable mean differs from that of owned plots at 5% and 1% levels, respectively. † and †† indicate the variable mean differs from that of cash-rented plots at 5% and 1% levels, respectively. b Computed as the sum of reported current values of all itemized assets in Ethiopian Birrs (ETB). Daily average exchange rate in 2010 is 1 USD = 14.38 ETB. c Sampling weights are accounted for. sharecropping (“contract model” hereafter). To see possible overall household i had unmet credit need in crop production during the sur- impact regardless of contract types, we further apply a similar proce- veyed year.6 The vector Pij includes plot size, soil fertility, soil slope, dure using a binary land rental indicator without differentiation of soil depth, distance from home and the cropping season. Although Tij is contract type (“tenure model” hereafter). Multiple robustness exercises also a plot feature, we further set it apart from other plot characteristics are finally performed to build confidence in the findings. as it is of our main interest. In the contract model, Tij is a vector of two Empirical modeling is implemented at the plot level, as a farmer binary indicators (of cash-rental and sharecropping) which take the could operate both owned and rented-in plots at the same time. As value of one for rented plots and zero for owned plots. While in the discussed in our theoretical model above, for plot j of farmer i , both tenure model, Tij is one binary indicator of land rental. Tij is treated as crop revenue (Rij ) and cost (Cij ) are assumed to be functions of house- endogenous given its choice nature as reflected by possible correlations hold characteristics, Hi , plot characteristics, Pij , adoption decision, Aij , with unobservables (εij ). and unobservables, εij . Use Tij to denote land tenure. Our empirical Econometric approaches to binary choice models with endogenous model with adoption as the dependent variable can be conceptually regressors include linear probability model with instrumental variable specified by solving the optimization problem for plot j of farmer i : estimation, multivariate probit model with maximum likelihood esti- mation and control function probit model with two-stage estimation. Aij = Aij (Hi , Pij , Tij, εij ) (9) Estimation of a linear probability model by two-stage least squares where εij is the random disturbance that comes into play in the re- (2SLS) is suitable in our case as the causal impact of land rental on gression model. Specifically, the vector Hi includes household size, total household wealth, total land holding the age, gender, marital 6 This measure is constructed using two survey questions: 1) whether a household had status and education of the household head, the walking distances to credit need in crop production in the surveyed year, and 2) whether such need was the nearest main market and agricultural extension office, two social eventually met. Some farmers might not need credit, while some others in need of credit network indicators (number of years living in the village and farmers’ might have successfully secured it. Hence, this measure suggests whether access to credit cooperative membership indicator) and a binary indicator suggesting if was a “real” constraint in maize production during the surveyed period. 274 D. Zeng et al. Land Use Policy 72 (2018) 270–279 adoption is our main interest (Angrist, 2001). Although a nonlinear may affect improved maize variety adoption, which we view as highly model may fit the conditional expectation function better for limited unlikely.8 Therefore, the exclusion restriction should again be satisfied. dependent variable models, the difference in terms of marginal effects is usually indistinguishable (Angrist and Pischke, 2009). A multivariate probit model with maximum likelihood estimation is an alternative, 6. Results where the determinants of multiple correlated binary outcomes are jointly estimated. However, the multivariate probit model is more re- The contract model is first estimated. Results are presented in strictive than the linear probability model as it relies on the assumption Table 3. Of our main interest is the linear probability model estimated of joint normality of error terms that is usually violated. Moreover, it is via 2SLS, which appears to be appropriately identified according to test vulnerable to incorrect first-stage specifications, but the 2SLS estima- statistics. Moreover, the significant error correlation in the bivariate tion of a linear probability model is robust against such misspecification probit model suggests the simultaneity of improved maize variety (Angrist, 2001; Lewbel et al., 2012). Unlike either of these procedures, adoption and land rental decisions. Therefore, it makes sense to draw the control function probit model with two-stage estimation is designed inference based on the 2SLS estimates while using the bivariate probit for continuous endogenous variables (Rivers and Vuong, 1988), and our results for robustness check purposes. We also estimate simple linear binary endogenous regressors would violate the distributional as- probability model and probit model without instrumentation in search sumption of this approach. Therefore, we choose the linear probability of the sign of selection. Finally, standard errors in linear probability model as our main estimation strategy, while we also estimate multi- models are clustered at the district level, the primary sampling unit. variate probit models, namely a trivariate probit contract model with Sharecropping contracts significantly increases the probability of three binary choices (adoption, cash-rental decision, sharecropping improved maize variety adoption by 0.142 among sharecroppers. Such decision) and a bivariate probit tenure model with two binary choices change is very close to, though slightly smaller than, the trivariate- (adoption, land rental decision), for comparison purposes. probit marginal effect of 0.159. On the other hand, the impact among Successful identification of our empirical models requires the cash-renters are much smaller and statistically insignificant, which is availability of appropriate instruments, which should be correlated intuitive and accords with Hypothesis 1 above. As a comparison, the with land rental decision but not maize variety choice other than impact magnitudes estimated by both OLS and simple probit procedures through land rental decision. Two instruments are included in our are much larger and even significant for cash-renters, suggesting the analysis: 1) the change of total land holding in the last five years, and 2) existence of positive selection. In other words, farmers may in- for each household, the proportion of smaller holders in the village who tentionally rent in land to cultivate improved maize varieties. For rented-in land during the study period. Both instruments are worth sharecroppers, this is intuitive as maize plots under sharecropping careful discussion regarding their validity. contracts are more fertile and larger than owned ones (see Table 2), and The first instrument is computed as the acreage difference between it can be reasonably speculated that sharecroppers may realize both household-level land holdings of the current period compared to that of productivity and the economy of scale at higher levels. five years ago. As land sales market is nonexistent in Ethiopia (Holden Among the covariates, larger plots, deeper soil and better education et al., 2008), change of land holdings mostly occurs through govern- of the household head are positively associated with adoption, while mental land redistribution that is exogenous to the household older age and longer distance from the nearest main market are nega- (Deininger and Jin, 2006), except for land transfers through inheritance tively associated with adoption. Farmers tend to adopt improved maize which should be rare in the short run. We posit that recent loss of land varieties in the long rainy season (meher) rather than the short season encourages land rental for a tenant, while recent gain of land dis- (belg), which is a common practice as the time span of the long rainy courages it. However, such change is unlikely to affect maize variety season (from mid-June to mid-September) better meets the water needs choices directly. Moreover, sufficient variation exists as 17.4% house- of most maize varieties, and higher maize yields are usually reached in holds (226 out of 1300) observed changes in land holdings. the former. Most of these effects, however, are relatively small. The second instrument reflects the thickness of village land rental Given the estimates above, the overall impact of land rental on market. Based on within-village ranking of household land holdings, improved maize variety adoption can be positive if it is sufficiently the proportion of households who own a smaller acreage of land as driven by the impact of sharecropped plots that consist of 77.41% of all compared to each household is constructed.7 The renter proportion rented plots. Therefore, we further estimate such overall impact in the over smaller holders rather than of the whole village is employed given tenure model with a single binary land rental indicator rather than the notion that the smallness of land holding could stimulate land rental separate dummies for cash-rental and sharecropping. Results are pre- market participation to meet household food demand, and the instru- sented in Table 4. Land rental increases the probability of improved mental variable construction in this way also allows within-village variety adoption on that plot by 0.135, which is highly significant. The variation of the instrument among households. It is speculated that the bivariate probit model suggests a similar impact of 0.151. Moreover, as village land rental markets are thicker and associated transaction costs in the contract model, the impact magnitudes using OLS and simple are lower if more farmers rent in land. Hence, this instrument should probit procedures are much larger than those with instrumentation, directly affect land rental decision but not improved maize variety again suggesting positive selection. Covariate coefficients in the tenure adoption. In fact, it is recently found that poorer and smaller holders model bear similar patterns to those in the contract model. Likewise, may as likely adopt improved maize varieties as richer and larger few such effects compare in magnitudes to that of the land rental in- holders in Ethiopia (Zeng et al., 2015). This leaves only one potential dicator. source of bias: the correlation between the position of a household in These results provide consistent and fairly strong evidence for our the within-village land holding ranking that may systematically affect main hypotheses in Hypotheses 1 and 2: land rental does not affect crop the variation of the instrument and unobservable characteristics that variety adoption decisions of cash-renters but encourages it for 8 The validity of this instrument would be threatened, for example, if more capable farmers (with higher ability, more experience, etc., which could affect improved maize variety adoption) would be able to expand his/her land holding over time (which therefore would change his/her within-village land holding ranking). However, as land 7 To further illustrate our instrumental variable construction, consider a village where sales market is absent in Ethiopia where the only sources of acreage change are the rare 12 households were randomly surveyed. For a household whose total land holding ranked events of governmental land redistribution and land inheritance, such concern should be 4th over these 12 households and who observed 3 renters (either cash-renter or share- minimized. Therefore, the validity of both instruments rely critically on the plausible cropper) among the 8 smaller holders, the instrument takes the value of 3/8=0.375. exogeneity of land holding change, a special feature of the Ethiopian land market. 275 D. Zeng et al. Land Use Policy 72 (2018) 270–279 Table 3 Impact Estimates of Contract Model (n = 2359).a Linear probability models Maximum likelihood proceduresb OLS 2SLS Probit Trivariate probit Cash-rental 0.115 (0.054) * 0.069 (0.055) 0.107 (0.048) * 0.084 (0.055) Sharecropping 0.304 (0.037) ** 0.142 (0.031) ** 0.344 (0.046) ** 0.159 (0.033) ** Plot size 0.039 (0.008) ** 0.051 (0.015) ** 0.041 (0.011) ** 0.038 (0.007) ** Soil fertility −0.041 (0.028) −0.028 (0.023) −0.032 (0.027) −0.019 (0.034) Soil slope 0.032 (0.027) 0.032 (0.022) 0.027 (0.019) 0.040 (0.026) Soil depth 0.069 (0.022) ** 0.074 (0.020) ** 0.063 (0.022) ** 0.066 (0.021) ** Dist. from home 0.001 (0.000) 0.001 (0.000) ** 0.001 (0.001) 0.001 (0.001) * Cropping season 0.127 (0.035) ** 0.145 (0.038) ** 0.131 (0.042) ** 0.119 (0.042) ** Household size 0.007 (0.005) 0.006 (0.006) 0.008 (0.005) 0.006 (0.005) Total asset −0.000 (0.000) 0.000 (0.000) −0.000 (0.000) −0.000 (0.000) Head age −0.002 (0.001) −0.002 (0.001) * −0.002 (0.001) −0.003 (0.001) ** Head gender 0.051 (0.056) 0.065 (0.048) 0.052 (0.073) 0.074 (0.074) Head marital status 0.036 (0.064) 0.044 (0.062) 0.066 (0.068) 0.051 (0.067) Head education 0.001 (0.005) 0.003 (0.001) * 0.001 (0.004) 0.004 (0.002) * Main market dist. −0.002 (0.000) ** −0.002 (0.000) ** −0.002 (0.000) ** −0.002 (0.000) ** Extension office dist. 0.001 (0.001) −0.001 (0.000) −0.001 (0.000) −0.001 (0.000) Total land holding −0.001 (0.000) −0.001 (0.002) −0.001 (0.001) −0.001 (0.001) Unmet credit need −0.009 (0.004) * −0.014 (0.004) ** −0.007 (0.004) * −0.009 (0.003) ** Years living in village 0.007 (0.011) 0.004 (0.011) 0.014 (0.015) 0.004 (0.014) Cooperation membership 0.022 (0.012) 0.017 (0.043) 0.023 (0.016) 0.012 (0.020) Constant 0.192 (0.167) 0.112 (0.173) F statistic 16.11 ** 22.79 ** F test of 1st stage IV 47.41 ** Underidentificationc 36.88 ** Weak identificationd 27.26 ** LR / Wald chi-square 293.05 361.44 ** ρadoption and cash-rental 0.198 ** ρadoption and sharecropping 0.345 ** ρ cash-rental and sharecropping 0.127 * a Dependent variable is the binary improved maize variety adoption indicator. Standard deviations are in parentheses. * and ** indicate statistical significance at 5% and 1% levels, respectively. b Marginal impacts are reported. c Kleibergen-Paap rank Lagrange Multiplier statistic is reported. d Kleibergen-Paap rank Wald F statistic is reported. sharecroppers. Moreover, a positive overall impact is found as the Meteorology Agency of Ethiopia. Specifically, we consider the first two majority of rented plots were operated under sharecropped contracts. moments of yield, the mean and variance, both computed at the district The impact magnitudes of land rental are larger than most covariate level using plot-level measures. Impact estimates with these additional coefficients, suggesting the substantial explanatory power of this factor covariates are reported in the second panel of Table 5, which tend to be in crop variety adoption decision. Therefore, it is towards the robust- slightly smaller yet in no cases do they lose statistical significance. Since ness of these impact estimates that we now turn. our inference is mainly qualitative, this again supports our main results. We implement several robustness check procedures in addition to We further implement two subsample analysis to test our main re- the multivariate probit model estimation. The first two procedures sults. One concern may arise from cropping season. If land rental address model specification issues. First, it is arguable that village-level market conditions vary systematically with cropping season, partici- unobservable heterogeneity may possibly exist, which could arise from pation in this market would be affected and our impact estimates could local land rental market conditions, distinctive agroecological en- be inconsistent. This, however, is unlikely to be the case as Ethiopian vironments, different agricultural production practices, or varying so- farmers mainly cultivate maize in the long rainy season for reasons cial norms across localities. To check if this possibility threatens our discussed above. As a result, only 172 of 2359 plots were cultivated in main results, we control for village fixed effects and re-estimate all the short rainy season in our data, and related variation is rather lim- models. As seen from the first panel of Table 5, these new estimates are ited. Nevertheless, we re-estimate all specifications with a homogenized very close to our main estimates in all specifications. subsample that only includes 2187 plots cultivated in the long rainy The second procedure regards production risks that could affect season to check if our main impact estimates are robust. As reported in improved maize variety adoption. One limitation of our cross-sectional the third panel of Table 5, the new estimates are slightly larger than our data is the lack of ex ante yield statistics (existent before the cropping main results, suggesting that the latter are rather conservative. There- season) that could factor into crop variety choice made at the beginning fore, this concern should be minimized. of that season. Still, there is a need to detect any potential effect from As a final robustness check, we make use of a unique subsample of production uncertainty, which could confound the impacts of land our survey data: 743 maize plots cultivated by 273 partial-adopters who rental that is of our primary interest. This is more likely the case for grew both improved and local maize varieties on different plots in the sharecroppers who, unlike cash-renters, do not individually bare pro- surveyed period. Specifically, we consider a first-difference type model duction risks. To test this, we have to refer to the ex post yield statistics where the plot-level maize variety adoption decision is first demeaned at the district level and assume that the yield distribution realized after within the household, and then regressed against both demeaned plot the surveyed season is representative of the real yield variations. This is characteristics and demeaned land rental indicator(s). Household a reasonable assumption as 2009–2010 was a fairly good cropping year characteristics would drop out as there is no within-household varia- according to climate statistics we obtained from the National tion. The model as linearized from Eq. (8) can be expressed as: 276 D. Zeng et al. Land Use Policy 72 (2018) 270–279 Table 4 Coefficient Estimates of Tenure Model (n = 2359).a Linear probability models Maximum likelihood proceduresb OLS 2SLS Probit Bivariate probit Land rental 0.365 (0.142) ** 0.135 (0.033) ** 0.277 (0.063) ** 0.151 (0.038) ** Plot size 0.033 (0.007) ** 0.040 (0.011) ** 0.044 (0.014) ** 0.044 (0.009) ** Soil fertility −0.037 (0.039) −0.016 (0.023) −0.012 (0.030) −0.021 (0.041) Soil slope 0.039 (0.020) * 0.033 (0.021) 0.030 (0.025) 0.015 (0.025) Soil depth 0.067 (0.019) ** 0.060 (0.017) ** 0.079 (0.020) ** 0.073 (0.022) ** Dist. from home 0.000 (0.001) 0.001 (0.000) 0.000 (0.000) 0.001 (0.001) Cropping season 0.122 (0.036) ** 0.141 (0.042) ** 0.154 (0.051) ** 0.136 (0.039) ** Household size 0.012 (0.008) 0.006 (0.005) 0.002 (0.005) 0.006 (0.004) Total asset −0.000 (0.000) 0.000 (0.000) −0.000 (0.000) 0.000 (0.000) Head age −0.001 (0.001) −0.002 (0.001) −0.002 (0.001) ** −0.002 (0.001) Head gender 0.062 (0.055) 0.033 (0.057) 0.075 (0.066) 0.096 (0.062) Head marital status 0.041 (0.061) 0.050 (0.072) 0.070 (0.041) 0.050 (0.034) Head education 0.002 (0.003) 0.003 (0.002) * 0.004 (0.003) 0.003 (0.002) * Main market dist. −0.002 (0.000) ** −0.002 (0.000) ** −0.002 (0.000) ** −0.002 (0.000) ** Extension office dist. 0.000 (0.000) −0.001 (0.000) −0.001 (0.000) −0.000 (0.000) Total land holding −0.001 (0.001) −0.001 (0.001) −0.001 (0.001) −0.001 (0.001) Unmet credit need −0.019 (0.003) ** −0.014 (0.003) ** −0.008 (0.003) * −0.009 (0.001) ** Years living in village 0.008 (0.005) 0.005 (0.014) 0.013 (0.020) 0.004 (0.014) Cooperation membership 0.012 (0.015) 0.012 (0.042) 0.021 (0.018) 0.003 (0.022) Constant 0.127 (0.148) 0.189 (0.095) ** F statistic 16.81 ** 16.33 ** F test of 1st stage IV 79.19 ** Underidentificationc 64.95 ** Weak identificationd 54.11 ** LR / Wald chi-square 286.26 536.60 ** ρadoption and land rental 0.213 ** a Dependent variable is the binary improved maize variety adoption indicator. Standard deviations are in parentheses. * and ** indicate statistical significance at 5% and 1% levels, respectively. b Marginal effects are reported. c Kleibergen-Paap rank LM statistic is reported. d Kleibergen-Paap rank Wald F statistic is reported. Table 5 Impact Estimates via Robustness Check Procedures.a Linear probability models Multivariate probitb Contract model Tenure model Contract model Tenure model Instrumental variable estimation with village fixed effects (n = 2359) Cash-rental 0.068 (0.045) 0.062 (0.033) Sharecropping 0.139 (0.047) ** 0.160 (0.054) ** Land ownership 0.121 (0.028) ** 0.137 (0.059) * Instrumental variable estimation with yield risk measures (n = 2359) Cash-rental 0.071 (0.052) 0.069 (0.054) Sharecropping 0.097 (0.035) ** 0.134 (0.042) ** Land ownership 0.115 (0.039) ** 0.122 (0.036) ** Instrumental variable estimation using long-season subsample (n = 2187) Cash-rental 0.081 (0.057) 0.065 (0.046) Sharecropping 0.164 (0.030) ** 0.185 (0.033) ** ** ** Land ownership 0.142 (0.035) 0.162 (0.041) Ordinary least square estimation using partial-adopter subsample (n = 743) Cash-rental 0.207 (0.141) Sharecropping 0.228 (0.065) ** Land ownership 0.205 (0.064) ** a Dependent variable in the upper and central panel models is the binary improved maize variety adoption indicator, while in the lower panel it is the demeaned adoption indicator. Standard deviations are in par- entheses. * and ** indicate statistical significance at 5% and 1% levels, respectively. b Marginal effects are reported. − − − Aij = β1 Pij + β2 Tij + uij (10) cash-rented and sharecropped plots appear to be much larger than the full-sample 2SLS results. This provides strong evidence for the observed Eq. (9) cancels out both observed and unobserved household-level positive selection that farmers intentionally rent in plots to cultivate heterogeneity. With further control for plot-level heterogeneity, the improved maize varieties. All these results suggest the robustness of our impact of land rental is captured by the estimate of β2 , which can be main impact estimates. unbiasedly and consistently recovered through OLS estimation. As shown in the bottom panel of Table 5, the impact estimates for both 277 D. Zeng et al. Land Use Policy 72 (2018) 270–279 7. Concluding remarks References We assess the impact of land rental as manifested in two major Abdulai, A., Owusu, V., Goetz, R., 2011. Land tenure differences and investment in land contracts on improved crop variety adoption. While many previous improvement measures: theoretical and empirical analyses. J. Dev. Econ. 96 (1), 66–78. studies hypothesize that the lack of land ownership discourages agri- Ali, A., Abdulai, A., Goetz, R., 2012. Impacts of tenancy arrangements on investment and cultural technology adoption (Gavian and Fafchamps, 1996; Abdulai efficiency: evidence from Pakistan. Agric. Econ. 43 (6), 1–13. et al., 2011; Oostendorp and Zaal, 2012), this is not the case for im- Angrist, J.D., 2001. Estimation of limited dependent variable models with dummy en- dogenous regressors: simple strategies for empirical practice. J. Bus. Econ. Stat. 19 proved crop varieties. The theoretical model shows that land rental (1), 2–16. does not affect crop variety adoption of cash-renters, but encourages Angrist, J.D., Pischke, J.-S., 2009. Mostly Harmless Econometrics: An Empiricists adoption by sharecroppers. Empirical analysis of Ethiopian maize Companion. Princeton University Press, Princeton NJ. Baumgärtner, S., Quaas, M.F., 2010. Managing increasing environmental risks through farmers provides robust evidence in support of these hypotheses. As the agrobiodiversity and agri-environmental policies. Agric. Econ. 41 (5), 483–496. benefits of improved crop varieties are widely documented in literature Benin, S., Ahmed, M., Pender, J., Ehui, S., 2005. Development of land rental markets and and their adoption is widely proposed as a potential means of welfare agricultural productivity growth: the case of Northern Ethiopia. J. Afr. Econ. 14 (1), 21–54. improvement in SSA, it is necessary to reevaluate the role of land rental Brasselle, A.S., Gaspart, F., Platteau, J.P., 2002. Land tenure security and investment markets as a means of stimulating improved crop variety adoption, incentives: puzzling evidence from Burkina Faso. J. Dev. Econ. 67 (2), 373–418. which makes the current study policy-relevant. Braverman, A., Stiglitz, J.E., 1986. Cost-sharing arrangements under sharecropping: Although our findings contradict many previous results concerning moral hazard, incentive flexibility, and risk. Am. J. Agr. Econ. 68 (3), 642–652. Central Statistical Agency, 2010. 2010 National Statistics (Abstract). Addis Ababa, land tenure and agricultural technology adoption, they are not incon- Ethiopia. . sistent with arguments for tenure security in SSA, which has been Chamberlin, J., Schmidt, E., 2012. Ethiopian agriculture: a dynamic geographic per- shown to be associated with multidimensional welfare improvements spective. In: Dorosh, P., Rashid, S. (Eds.), Food and Agriculture in Ethiopia: Progress and Policy Challenges. University of Pennsylvania Press, Philadelphia PA. Kassie et al., 2011; Shiferaw et al., 2014; Zeng et al., 2015). In fact, Cheung, S.N.S., 1969. The Theory of Share Tenancy. University of Chicago Press, policy implications would directly arise from the theory of second best, Chicago IL. which justifies governmental intervention in related markets to offset Crewett, W., Korf, B., 2008. Ethiopia: reforming land tenure. Rev. Afr. Polit. Econ. 116, 203–220. the efficiency loss in a primary market that observes irremovable dis- Deininger, K., Jin, S., 2006. Tenure security and land-related investment: evidence from tortion (Lipsey and Lancaster, 1956). In our context, as land sales Ethiopia. Eur. Econ. Rev. 50 (5), 1245–1277. market is not legal in Ethiopia (and incomplete in most SSA countries, Ersado, L., Amacher, G., Alwang, J., 2004. Productivity and land enhancing technologies in Northern Ethiopia: health, public investments, and sequential adoption. Am. J. see Holden et al., 2008), improvements in the land rental market in- Agr. Econ. 86 (2), 321–331. stead could substantially reduce the gap between land demand and Evenson, R.E., Gollin, D., 2003. Crop Variety Improvement and Its Effect on Productivity: supply and improve household welfare through improved crop variety The Impact of International Agricultural Research. CABI Publishing, Wallingford UK. Feder, G., Nishio, A., 1998. The benefits of land registration and titling: economic and adoption. It is recently found that land rental markets in Kenya promote social perspectives. Land Use Policy 15 (1), 25–43. farm productivity and significantly raise the incomes of land-con- Fenske, J., 2011. Land tenure and investment incentives: evidence from West Africa. J. strained farm households (Jin and Jayne, 2013). Therefore, potential Dev. Econ. 95 (2), 137–156. policies that ease transactions in local land rental markets may com- Gavian, S., Fafchamps, M., 1996. Land tenure and allocative efficiency in Niger. Am. J. Agr. Econ. 78 (2), 460–471. plement existing strategies to stimulate improved crop variety adop- Ghebru, H.H., Holden, S.T., 2008. Factor market imperfections and rural land rental tion. For example, agricultural extension services that promote in- markets in northern Ethiopian highlands. In: Holden, S.T., Otsuka, K., Place, F.M. formation exchange in land rental markets, such as a village (Eds.), The Emergence of Land Markets in Africa: Impacts on Poverty, Equity, and Efficiency. Resource for the Future, Washington DC. information board, may immediately reduce the search costs of both Heisey, P.W., Norton, G.W., 2007. Fertilizers and other farm chemicals. In: In: Evenson, tenants and landlords. On the other hand, government actions to reduce R., Pingali, P. (Eds.), Handbook of Agricultural Economics Vol. 3 Elsevier, policy uncertainty in land rental markets may potentially help for- Amsterdam Netherlands. Holden, S.T., Otsuka, K., Place, F.M., 2008. Land markets and development in Africa. In: malize these markets to allow longer term contractual arrangements Holden, S.T., Otsuka, K., Place, F.M. (Eds.), The Emergence of Land Markets in Africa: and incentivize farmers to participate. Although our results do not Impacts on Poverty, Equity, and Efficiency. Resource for the Future, Washington DC. speak directly to the merits of these strategies, they are still worth Holden, S.T., Bezabih, M., 2008. Gender and land productivity on rented land in Ethiopia. In: Holden, S.T., Otsuka, K., Place, F.M. (Eds.), The Emergence of Land Markets in policy consideration as a possible means to improve farm household Africa: Impacts on Poverty, Equity, and Efficiency. Resource for the Future, welfare in SSA. Washington DC. A first assessment of the impact land rental on improved crop Jaleta, M., Yirga, C., Kassie, M., de Groote, H., Shiferaw, B., 2013. Knowledge, adoption and use intensity of improved maize technologies in Ethiopia. Invited Paper variety adoption, our study has several limitations. The cross-sectional Presented at the 4th International Conference of the African Association of data do not support the investigation of the dynamic impacts associated Agricultural Economists. September 22-25, 2013, Hammamet, Tunisia. with contractual length and/or renewal frequency. Also, limited data Jin, S., Jayne, T.S., 2013. Land rental markets in Kenya: implications for efficiency, variation regarding other productivity-enhancing technologies, espe- equity, household income, and poverty. Land Econ. 89 (2), 246–271. Kassie, M., Shiferaw, B., Muricho, G., 2011. Agricultural technology, crop income, and cially fertilizers, does not allow the modeling of joint adoption deci- poverty alleviation in Uganda. World Dev. 39 (10), 1784–1795. sions. These open questions are worth further investigation with Krishna, V., Qaim, M., Zilberman, D., 2016. Transgenic crops, production risk and available data. At the minimum, our partial-equilibrium analysis does agrobiodiversity. Eur. Rev. Agric. Econ. 43 (1), 137–164. Kathage, J., Kassie, M., Shiferaw, B., Qaim, M., 2016. Big constraints or small returns? confirm the overall positive impact of land rental on crop variety Explaining nonadoption of hybrid maize in Tazania. Appl. Econ. Perspect. Policy 38 adoption. Empirical results imply that improvements of land rental (1), 113–131. markets could potentially improve household welfare through partici- Lewbel, A., Dong, Y., Yang, T.T., 2012. Comparing features of convenient estimators for binary choice models with endogenous regressors. Canadian J. Econ. 45 (3), pation in this market and associated profit-maximizing crop choices 809–829. given the prominent role of agriculture in small household income Lipsey, R.G., Lancaster, K., 1956. The general theory of second best. Rev. Econ. Stud. 24 generation in Ethiopia and other SSA economies. (1), 11–32. Oostendorp, R.H., Zaal, F., 2012. Land acquisition and the adoption of soil and water conservation techniques: a duration analysis for Kenya and The Philippines. World Dev. 40 (6), 1240–1254. Acknowledgements Place, F., 2009. Land tenure and agricultural productivity in Africa: a comparative ana- lysis of the economics literature and recent policy strategies and reforms. World Dev. 37 (8), 1326–1336. Data collection was funded by the Bill & Melinda Gates Foundation Place, F., Otsuka, K., 2002. Land tenure systems and their impacts on agricultural in- and led by CGIAR Standing Panel on Impact Assessment (SPIA). The vestments and productivity in Uganda. J. Dev. Stud. 38 (6), 105–128. views are solely from the authors. Radhu, P.T., Erenstein, O., Böber, C., Krishna, V.V., 2015. Adoption and outcomes of hybrid maize in the marginal areas of India. Q. J. Int. Agric. 54 (2), 189–214. 278 D. Zeng et al. Land Use Policy 72 (2018) 270–279 Rahmato, D., 1984. Agrarian Reform in Ethiopia. Scandinavian Institute of African and impacts on household food security in Ethiopia. Food Policy 44, 272–284. Studies, Uppsala Sweden. Smale, M., Just, R.E., Leathers, H.D., 1994. Land allocation in HYV adoption models - an Rivers, D., Vuong, Q.H., 1988. Limited information estimators and exogeneity tests for investigation of alternative explanations. Am. J. Agr. Econ. 76 (3), 535–546. simultaneous probit models. J. Econometrics 39 (3), 347–366. Soule, M.J., Tegene, A., Wiebe, K.D., 2000. Land tenure and the adoption of conservation Schneider, K., Anderson, L., 2010. Yield Gap and Productivity Potential in Ethiopian practices. Am. J. Agr. Econ. 82 (4), 993–1005. Agriculture: Staple Grains & Pulses. EPAR Brief No. 98. Evans School Policy Analysis Zeng, D., Alwang, J., Norton, G.W., Shiferaw, B., Jaleta, M., Yirga, C., 2015. Ex-post and Research at University of Washington, Seattle WA. impacts of improved maize varieties on poverty in rural Ethiopia. Agric. Econ. 46 (4), Shiferaw, B., Kassie, M., Jaleta, M., Yirga, C., 2014. Adoption of improved wheat varieties 515–526. 279