Abstract
While early attempts at land titling in Africa were often unsuccessful, factors such as new legislation, low-cost methods, and increasing demand for land have generated renewed interest. A four-period panel allows use of a pipeline and difference-indifferences approach to assess impacts of land registration in Ethiopia. We find that the program increased tenure security, land-related investment, and rental market participation and yielded benefits significantly above the cost of implementation. (JEL O13, Q15)
I. Introduction
A number of factors have recently led to renewed interest in the formalization of property rights to land in Africa, where the majority of land continues to be held under customary tenures. Since the 1990s, many African countries have passed legislation to remedy perceived shortcomings of existing systems, particularly by strengthening customary land rights, recognizing occupancy short of full title, improving female land ownership, and decentralizing land administration. Advances in information technology and remote sensing offer new tools to reduce the cost of land administration. From the demand side, a number of factors are intensifying existing pressures on land and creating a risk of dispossession for traditional landholders. These include increased prices for food, fuel, and fiber, together with new demands for land by outside investors and alternative streams of income from selling environmental services. It is thus widely felt that defining property rights, at the individual or group level, and establishing a well-governed system of land administration will be critical for Africa to make use of these opportunities in a way that contributes to overall growth while at the same time avoiding socially undesirable outcomes and conflict.
However, while there is undeniable potential, the history of land titling in Africa is one of failure rather than success, primarily due to three reasons. First, lack of understanding of the reality of local rights and attempts to replace them in a top-down way with a “modern” paradigm have generally ended in failure (Easterly 2008). Second, unless accompanied by broad-based information campaigns, efforts to formalize land tenure can easily cause a race for rights that will favor the powerful. By dispossessing traditional right holders, this process can end up reducing rather than increasing social welfare, especially for those with secondary rights (e.g., to grazing). Finally, the use of high-cost approaches that are implemented in other parts of the world often proved unsustainable because the benefits were much below the cost of establishing and maintaining land titles (Jacoby and Minten 2007)
This paper quantifies the early impact of recent land certification in Ethiopia, a program that is of interest for three reasons. First, although implemented by the government with minimum outside assistance, it is arguably the largest land administration program carried out over the last decade in Africa, and possibly the world. Second, it departs from the approach of traditional land titling interventions in a number of ways, namely, by (1) issuing nonalienable use right certificates rather than full titles; (2) promoting gender equity with joint land ownership by spouses and inclusion of their pictures on certificates; (3) using a participatory, decentralized process of field adjudication; and (4) replacing sophisticated mapping with community identification of boundaries. Third, costs are significantly lower than those achieved by alternative models and in line with participants’ capacity to pay, implying that this could provide a model for responding to new challenges.
To assess program impact, we investigate early effects of issuing certificates at household or individual level on perceived tenure security, land-related investment, and land market participation, using a four-period household panel survey from the Amhara region. We use a difference-in-differences strategy combined with a pipeline approach that is justified by the program’s gradual roll-out and the fact that outcome variables moved in parallel in treatment and control areas before the program. Results are robust to whether the household or the village is used as the unit of intervention and, as most of the control had already received information about the program or undergone the registration process, constitute a lower bound of true impact. Three main findings stand out. First, the program had significant impact on the variables of interest, namely, it helped to increase tenure security, investment, and renting out by landlords. Second, a rough estimate of program-induced benefits (based on the net increase in productivity due to program-induced investment)1 suggests a favorable cost-benefit ratio, especially if current investments are maintained. Finally although certification has a positive effect, the fact that considerable tenure insecurity remains, largely from the threat of expropriation, implies that certification alone cannot eliminate or compensate for the effects of the policy environment. To reap the full potential of certification, there is need to not only honor existing certificates but also ensure that policy does not undermine their value.
II. BACKGROUND AND APPROACH
The literature suggests that, in principle, measures to strengthen land rights or improve their enforcement can affect owners’ incentives to make land-related investments, transfer land to more efficient uses through markets, and use land as collateral for credit. Given Ethiopia’s policy environment, impacts are expected to be limited to investment and rental market participation, and key aspects of the way in which the program is implemented are used to formulate hypotheses that can be subjected to statistical tests.
Evidence from the Literature and Implications for Africa
The literature identifies three channels through which more-secure and better-enforced property rights may affect economic outcomes. First, well-defined property rights to land and the ability to draw on public enforcement lower the risk of eviction, increase incentives for land-related investment (Besley 1995), and reduce the need for land owners to expend resources to stake out or defend their claims. Thus, groups, such as women, who were traditionally discriminated against (Joireman 2008), could benefit even more, and security of rights against the state can have broader political ramifications (Boone 2009).
The positive impacts of more-secure land tenure on investment and land values in rural areas have been demonstrated in China (Jacoby, Li, and Rozelle 2002), Thailand (Feder et al. 1988), Latin America (Bandiera 2007; Deininger and Chamorro 2004; Field, Field, and Torero 2006; Fort 2007), Eastern Europe (Rozelle and Swinnen 2004), and Africa (Deininger and Jin 2006; Goldstein and Udry 2008; Holden, Deininger, and Ghebru 2009). In urban areas, efforts to enhance tenure security have increased self-assessed land values (Lanjouw and Levy 2002), investment in housing (Galiani and Schargrodsky 2005), and female empowerment (Field 2005). Receipt of title allowed former squatters, especially women, to join formal labor markets instead of staying at home to guard their land, increasing income and reducing child labor (Field 2007). Joint titles helped reduce fertility (Field 2003), increased investment in children’s human capital (Galiani and Schargrodsky 2004), and improved education (Galiani and Schargrodsky 2005). How property rights can be exercised affects governance (Lobo and Balakrishnan 2002) and performance of local institutions (Deininger and Jin 2009).
The size of tenure security and investment benefits depends on the availability of investment opportunities and the reduction in enforcement efforts afforded by formal recognition. Moreover, it will also be critical to design a process that does not generate losses by threatening secondary rights, weakening existing institutions, or setting off speculative clamors for land that increase conflict. Benefits will be larger in settings with potential for land-related investment with high payoff, where tenure has been insecure or the level of conflict high, and where the certificates generated will be respected and affect behavior by third parties. To assess whether intervention is warranted, these benefits need to be compared to the cost, both of first-time registration and of sustaining the required institutions over time. A combination of limited benefits, inappropriate processes, and failure to account for the cost of maintenance has limited the impact of past efforts to register land in many African contexts (Bruce and Migot-Adholla 1994; Jacoby and Minten 2007; Pinckney and Kimuyu 1994).
A second potential benefit from land registration is that low-cost access to reliable information about individuals’ land ownership via a public registry will reduce the cost of exchanging land in rental or sales markets. Rental allows land owners to tap new sources of income while retaining their land as a means of insurance or old-age protection. At the same time, those who remain in farming can consolidate and cultivate larger farm areas. Formal documentation of land rights can allay fears that rented-out land will be lost, either to the government through redistribution, or to tenants who do not vacate it at the end of the lease period. This can be useful in contexts where migration requires land owners to be temporarily absent or if the number of transactions increases beyond the capacity of informal, local mechanisms to handle them transparently. In China, rental activity contributed to occupational diversification and was estimated to have increased productivity by about 60% (Deininger and Jin 2009). In Vietnam, awarding certificates is estimated to have prompted not only investment in perennials (by 7.5% compared to no certificates) but also an 11- to 12-week expansion of the time households spent in nonagricultural activity, an effect that was particularly pronounced for the poor (Do and Iyer 2008).
Finally, in many settings, a key benefit of formal land titles is the ability to sell land to strangers and the associated ability to use land as collateral for credit (de Soto 2000).2 The reason is that, if a reliable land registry provides a formal and low-cost way to identify land ownership without the need of physical inspection or inquiry with neighbors, land is ideal as collateral. However, for credit effects from formal registration to materialize, households will need to have otherwise bankable projects, be credit-worthy, and be willing to take the associated risk (Boucher, Carter, and Guirkinger 2008). Moreover, land markets need to be sufficiently liquid to make quick sales feasible. While credit effects of land titling are reported in the literature (Feder et al. 1988), positive impacts were often limited to larger owners (Carter and Olinto 2003; Mushinski 1999) or may have failed to materialize even in settings where they were expected (Field, Field, and Torero 2006; Fort 2007). Even if profitable projects exist, restrictions on land sales (Sundet 2004), limited commercial value of the land under question (Galiani and Schargrodsky 2005; Payne, Durand-Lasserve, and Rakodi 2008), or social/political considerations limiting foreclosure (Field and Torero 2006) may undermine realization of credit effects, which might thus be less readily achieved than originally hoped for.
Hypotheses on Program Impact and Outcome Variables
In one of the largest land registration programs in the world,3 three of Ethiopia’s four main regions have, over the last 5 years, registered more than 20 million parcels of rural land to some 6 million households.4 Certification is initiated by a team of experts from the woreda (district) that guides the process from a village meeting to the election of an independent village land use and administration committee (LAC).5 The LAC then assumes responsibility for systematic fieldbased adjudication of rights through a public process with the presence of neighbors and help from elders to resolve conflicts. The adjudication process produces preliminary registration certificates that identify size and neighbors for each of a holder’s plots.6 Results are then displayed in public and, after a period for raising complaints, entered into registry books, copies of which are to be kept at kebele (village) and woreda levels. Thereafter, certificates with pictures of the land holders (husband and wife in case of joint ownership) are issued by the woreda. Certificates also include space for maps and spatial information expected to be added in a second stage.
Evidence suggests that decentralized and participatory implementation with emphasis on the provision of information, issuance of certificates rather than titles, and a focus on gender equality helped avoid some of the problems raised in the literature on land titling in Africa. A nationwide survey (Deininger et al. 2008) highlights evidence that access to information and certificates was biased against neither women nor the poor. It also suggests that the process was generally implemented as planned; in particular (1) public meetings were held before and during the certification process, (2) land use committees (LACs) represented most of the sub-kebeles, and (3) adjudication relied on village elders to resolve disputes and included demarcation in the field with neighbors present. Survey data suggest that the quality of the certification process was high; certificates could be issued in 95% of cases where there were no disputes about ownership, compared to 80% in many titling projects. Case study evidence also points to reductions in conflict when registration involved identification of borders and systematic conflict resolution.7 Also, at less than $1 per parcel, program cost is an order of magnitude lower than the $20 to $60 per parcel for traditional titling reported in the literature. 8 In fact, more than 80% of sample households were willing to pay an amount in line with the cost of service provision to replace certificates if lost or to transfer them, suggesting that the program could be selfsustaining.
To assess possible impacts, it is important to note that, in Ethiopia, land is state property that can neither be sold nor mortgaged, implying that we would expect no credit effects from land certification. Land rental also remains restricted in all regions except Amhara. There is, however, scope for positive impacts by increasing tenure security due to two factors. First, the constitutional guarantee of land access by every adult and the government’s ability to resort to often discretionary land redistribution to implement it has long been a threat to land users.9 In Amhara, where our survey was undertaken, a highly politicized land redistribution in 1997 reduced tenure security and increased conflict on a large scale (Ege 1997).10 The topic acquired new urgency when, starting in 2006, the Tigray region began enforcing a proclamation (law) to take away land from rural residents who had left their village for more than 2 years. Second, urban expansion and government-supported land grants to investors continue apace. In both cases, possession of a certificate can improve negotiating power or at least provide a basis for compensation, again implying that certification could lead to potentially large tenure security effects.
III. DATA, DESCRIPTIVE EVIDENCE, AND ECONOMETRIC APPROACH
Our four-period household panel allows tracing adherence to regulations for implementing the program, so as to illustrate the evolution of outcome variables before and after certification and describe the econometric approach chosen to identify hypothesized program impacts.
Data and Identification Strategy
To assess program impact, we use data from four waves of a panel survey of rural households conducted in September–October 1999, July–August 2002, September–November 2004, and July–August 2007 in the East Gojjam zone of the Amhara region.11 The survey, originally aimed to assess impacts of a sustainable development program to which certification was added at a later stage, was undertaken by the Department of Economics of Addis Ababa University in collaboration with Gothenburg University, Ethiopian Development Research Institute, and the World Bank. Each round includes information on a panel of 900 households that had been randomly selected in the first phase, and more than 4,000 plots cultivated by those households, although the plots cannot be matched over time.12
The panel’s first three rounds covered the period before certification and, in view of the program’s perceived success, the pilot was eventually extended to the entire region. While implementation started simultaneously in all woredas, lack of manpower led to adoption of a gradual roll-out that in general expanded from the initial village. We use the geographic discontinuity thus created to identify program impacts. At the time of the fourth survey round, some (treated) villages or households had been certified for more than 12 months, so that impacts on perceived tenure security, investment decisions, and land market participation will be observed in our survey’s reference period. Attrition, at household or plot levels, was very low in comparison with similar surveys from other countries and thus unlikely to be a concern in interpreting results.13
The village-level certification process follows a sequence of information campaign and LAC formation, field adjudication and distribution of registration receipts, and eventually issuance of final certificates. Selection of villages to be certified was the responsibility of woreda officials, who determined a roll-out plan in campaign style, moving from village to village to maximize targets. However, the fact that field work is possible only during the dry season (January to July/August), when agricultural labor demands are limited, creates a discontinuity that we can utilize for identification. Thus, in most cases the process was initiated after the harvest and registration completed before the start of the next growing season. Woreda officials then used the agricultural growing season to complete the paperwork and distribute certificates as and when they were ready and road conditions permitted.14 Plot-level data given in the bottom panel of Table 1 point toward some differences in program implementation.
Program Characteristics at Household and Plot Levels
Household evidence on process in the top panel of Table 1 reinforces earlier notions: 85% in certified and 78% in control villages attended an average of 3.5 public information meetings, and 85% and 68%, respectively, thought they were well informed about the program. At the time of the survey, 87% of households in treatment villages had received a certificate, which they had held for an average of 17 months, compared to 36% and 8 months in controls, with 77.5% in treatment and 2.3% in control areas having held certificates for longer than 12 months so that it could affect decisions (e.g., on investment) during the 12 month-recall period of our survey. Almost all the plots (92%) in treatment villages were measured (92% with rope) in the field with presence of more than half of neighbors in 60% of the cases and between one-third and half in 20% of the cases. Field measurement was done in less than two-thirds of registered plots in control villages; 35% of these cases involved eye estimation only, and more than half or more than one-third of the neighbors were present only in 35% and 11% of cases, respectively.
Table 2 illustrates the timing of program implementation for the seven villages (kebeles) in the three districts (woredas) of our survey. With the exception of two “pilot” villages (A. Gullit and Telma), one in the treatment and one in the control, the program was introduced in treatment villages in early 2004 and in early 2005 or even 2006 in control villages. 15 In “regular” villages it took an average of 11 months to complete registration and another 5 months for the issuance of certificates. Three control villages had completed registration and two had started issuance of certificates at the time of the survey. In those that had started certification, the process commenced some 15 months later than in treated ones. While in all cases certification is at the household level, the fact that even in villages that we define as certified (as the majority of households received certificates) households may be excluded for exogenous reasons (e.g., lack of sufficient forms) raises the issue of whether to define the intervention at household or village level. As it is not clear a priori which one to prefer, we report estimates based on either the village or the household as the relevant unit of observation throughout.16
Program Characteristics by Village
We use a difference-in-differences (DID) approach comparing difference between pre- and post-program outcomes together with a phased and discontinuous roll-out. This will provide an unbiased estimate of program effects if there are no unobserved differences between treatment and control units that could affect changes in outcome variables over time. While this cannot be tested, an ability to show that unobservable differences between villages did not affect the rate of change in outcome variables before the program had been announced will increase our confidence in this condition being satisfied. We explicitly test the assumption of parallel trends in preintervention years for key variables of interest.
Our strategy is conservative in two ways. First, by defining treatment as receipt of a certificate rather than completion of the registration process that is certain to lead to a certificate, we implicitly assume that registration has no effect and that the expectation of a certificate will not impact behavior. As illustrated in Table 2, at the time of the fourth round of our survey all villages had received information about the program and many households in noncertified villages had undergone the registration process. If this led them to modify their behavior, it would reduce the size of the effect estimated here, implying that our estimate will constitute a lower bound of the true certification impact. Second, to the extent that we define the intervention at the village rather than the household level, our treated category includes households that did not receive a certificate. If there is self-selection so that the benefit of receiving certificates for recipients is above the average, this would exert downward bias on the estimated effects of certification.
Descriptive Evidence
To illustrate the evolution of key dependent variables, Table 3 displays key household characteristics by participation status and year for the 356 and 477 households in treatment and control villages, respectively. There are a number of significant differences between households; for example, treated villages have slightly higher endowments of land per household (but not per capita), higher levels of human capital as proxied by literacy of the household head, and more livestock and other animals. Also, some time-varying factors (e.g., a drought in 2002) affected both types of villages similarly. Attributes for the 3,972 and 4,699 plots in treatment and control villages, respectively, averaged over all periods as reported in Table 4, point to a mean plot size of 0.3 and possession by the current owner for 21 years. With 4% of the plots having access to irrigation, it is rare in both villages. Even though subjective land attributes (quality and incline) differ between treated and controls, there are negligible disparities between villages in the share of flat and gently sloped as well as good- and medium-quality land together.
Household Characteristics by Treatment Category over Time
Plot Characteristics by Treatment Category
Levels and changes for our outcome variables, namely, perceived tenure security, landrelated investment, and rental market participation, are displayed in Table 5. High levels of tenure insecurity prior to program start and quick changes in this variable over time are particularly notable.17 Possibly as a result of the 1997 redistribution, perceived tenure insecurity in 1999 was very high, with 78% and 75% who expected a change of land holdings due to administrative intervention in treatment and control villages, respectively. In the 5 years before the certification program, this decreased to 38% in treatment and control villages.18 With the program, however, the trends start to diverge, dropping to 24% in treatment villages while increasing to 39% in controls. The share expecting an increase dropped from 19% to 4% in treatment and to 11% in control areas, whereas the share expecting a decrease remained unchanged in treatment areas but increased from 19% to 28% in control villages.19
Outcome Variables by Treatment Category
We use a dummy for whether households constructed new soil conservation structures (e.g., terraces, soil or rock bunds, and hedgerows) or repaired existing ones at the plot level, and hours spent on conservation, to measure land-attached investment. Although comparable data is available only for the last two periods, the preprogram share of plots that had investment or repairs and the amount of time spent on such investment were significantly higher in control as compared to treatment villages. The difference in both narrowed significantly and reversed for construction of new structures. For example, a decline in the share of plots where households voluntarily constructed new structures or repaired existing ones and the number of hours spent (from 36% to 24% and 8.2 to 5.5 hours, respectively) in control villages contrasts with an equally large increase (from 12% to 25% and 2.3 to 4.4 hours) in treatment villages. We observe a narrowing or reversal in the share of plots with any conservation structure (from 44% to 34% in control and 22% to 32% in treatment villages) and the share of constructing new structures in the last 12 months (from 10% to 8% and 7% to 10% in control and treatment villages, respectively).
Finally, a dummy for the type of net rental market participation and the amount of land transacted is used to capture potential rental market impacts of land certification. Before certification, the landlord share and mean area rented out were consistently higher in treatment than in control villages, a difference that narrowed from 2004. After certification, we note a clear increase in rental market participation in both areas. While the increase in renting out (7% vs. 5%) is marginally higher in treatment groups as compared to the control group, the opposite is true for renting in, implying that more rigorous evidence will be needed to assess whether certification can be said to have had a significant impact on land market participation or whether, possibly as a result of the rather restrictive policy regime, no such impact materialized.
Econometric Approach
To estimate program impacts on perceived tenure security, we use data from all four rounds to estimate
[1]where yit is a dummy variable that takes a value of one if household i expects an increase (or a decrease) of its landholdings due to administrative intervention in the 5 years following the survey; wit is the policy variable of interest (one for posttreatment period if household i lives in a treated village and zero otherwise; and equivalently for the case where treatment is defined at the household level); xit is a vector of controls at the household level that include the head’s age, gender, education, household assets (oxen, value of other livestock, roof material), and land size;20 ci captures household-specific unobserved effects, λt is a full set of time dummies; and uit is an iid error term. The null hypothesis that certification increases tenure security would imply τ < 0, which is a negative and significant marginal effect of the treatment indicator, wit. A random effects probit model is appropriate if ci is normally distributed with mean 0 and variance 
and independent from all right-hand side variables. As this may be unrealistic, we use Chamberlain’s random effects probit (Chamberlain 1980; Wooldridge 2002), which allows correlation between ci and the means of time-varying covariates at the household level according to
where 
is the vector of the average of time-varying household covariates for household i over all periods, and ai is an error term. All that is required is that xit and ai are independently and normally distributed with mean zero and variance 
. Adding 
as explanatory variables to equation [1] in each time period allows estimation of the parameters λ, τ, γ, ψ, ξ, and in 
a random effects probit model. This methodology has been used successfully by Holden, Deininger, and Ghebru (2009) to assess the impact of a similar program on land-related investment and productivity. While one could estimate a conditional logit model with household fixed effects, doing so would require dropping a large part of the sample and also provide less flexibility to compute marginal effects (Wooldridge 2002). Despite its limitations,21 we use a fixed effects linear probability model to check the sensitivity of the Chamberlain random effects probit, which is our preferred model. Results from doing so are reported in the Appendix.
Although we are able to control for time-invariant fixed effects using the Chamberlain method, the estimated marginal effect of the treatment indicator, wit, will be an unbiased estimate of program impact only if no unob-servable factors affect changes in outcome variables differently between control and treatment. Although this assumption itself is not testable, data from the survey rounds prior to certification can be used to test whether during this period, outcome variables moved in parallel in the two sets of villages. To do so, we estimate different intercept terms for the treatment and control villages for each year and then test for the equality of the differences in the intercept terms between treatment and control villages over all the periods.
In contrast to the household-level analysis in equation [1], impacts of certification on land-related investment are assessed using data at the plot level. Dependent variables for land-related investment take the value of one if the plot owned by the household received soil or water conservation investment during the past 12 months, or the number of hours spent in undertaking such investment during the past 12 months. Recall that a single certificate is issued for all of a household’s plots and that transfer of land to residents from outside the village is prohibited. Thus, although information on land-related investment is available at the plot level, we can identify (average) impacts at the household level only, controlling for right-hand side variables. Using earlier notation, the random effects probit or tobit (depending on choice of the dependent variable) model for land-related investmen on plot j by household i is specified as
[2]where the only difference is the inclusion of pjit, a vector of plot-level characteristics including size, soil quality, slope, and length of possession, and addition of a plot-specific error term ujit. The hypothesis of higher investment incentives through certification translates into τ > 0. As comparable information on investment is available only for the last two rounds (and plots cannot be matched over time), we are unable to include initial stocks of land investment or adopt a more sophisticated model.
Similar random effects probit and tobit specifications for participation on either side of the rental market and the amount of land transferred, respectively, are estimated for our rental market outcomes. As rental market participation may be persistent over time, for example, due to nonconvex transaction costs (Holden, Deininger, and Ghebru 2008), we also estimate a specification that allows for state dependence of rental market participation. The implied need to include the lagged dependent variable on the right-hand side of equation [1] gives rise to a nonlinear dynamic model that may suffer from the initial condition problem, that is, the correlation between the unobserved effect and the initial observation of the dependent variable. To account for this, the distribution of the unobserved effect is modeled conditionally on the initial observation in addition to the time-varying household-level covariates (Wooldridge 2005). The reduced form equation to be estimated is
[3]where yit−1 is the lagged dependent variable and yi0 is the first realization of the dependent variable. The parameters in equation [3] are estimated using standard random effects probit or tobit, depending on the type of the dependent variable. As this procedure requires data from at least four periods, we are forced to drop one of the villages (A. Gulit), which was added to the survey during the third round. The hypothesis of a positive impact of certification on the propensity to rent out land translates into τ > 0 in the probit and tobit equations for renting out or the area rented out. Again, we can test for the parallel trend assumption between treatment and control areas.
IV. ECONOMETRIC RESULTS
Results corresponding to the three main hypotheses suggest that, despite the limited time elapsed since its completion, certification has had a positive economic impact and improved tenure security, investment, and supply of land to the rental market. Even conservative estimates and a rough calculation of monetary benefits suggest that benefits exceed the cost significantly, allowing us to discuss ways in which the sustainability of impacts could be enhanced and other countries could benefit from Ethiopia’s experience.
Perceived Tenure Security
We find that, although certification failed to eliminate tenure insecurity, it has significantly reduced fear of land loss by some 10 percentage points, an estimate that is robust across specifications. Table 6 reports results from estimating equation [1] to identify the impact of certification on perceived risk of land loss or gain through administrative redistribution over the next 5 years. We report random effects probit estimates with treatment defined at village and household level (Columns 1 and 3 and 2 and 4, respectively). In all cases, results suggest that land tenure for treated households is significantly more secure, as evidenced by the fact that those households expect less administrative intervention. Estimated marginal effects from the Chamberlain specification with village-level treatment indicator suggest that certification leads to a decrease of about 14 percentage points in the share of those expecting to gain and of about 9 percentage points in the share of those expecting to lose from land redistribution. The estimated size of impact with treatment defined at the household level is the same for land loss and slightly lower for land gain. As a robustness check, we also estimate a household fixed effects linear probability model, results from which are reported in Appendix Table A1. They are consistent with those obtained from the random effects probit for the case of decreases, though insignificant for increases.
Impact of Certification on Perceived Land Tenure Security: Marginal Effects from Chamberlain Random Effects Probit Model
Tests of the parallel trend assumption (Appendix Table A2) imply that the share of those expecting land loss through administrative intervention, arguably the more relevant indicator, had moved in parallel during all of the three pretreatment periods. The share of households expecting an increase in land holdings moved in parallel between 1999 and 2002 but started to diverge in the 2002–2004 period, consistent with the notion that initial dissemination of the certification program in some villages before September–November 2004 had given rise to some speculation about possible increases in land endowments.22
Signs for coefficients on other variables are largely as expected. Coefficients on the time trend are highly significant and of large magnitude for gains but less significant for land losses, in line with descriptive evidence that points toward a reduction over time in the share of households that expect their holdings to increase rather than those that expect to lose land. Older household heads are more likely to fear land loss, consistent with the notion that administrative measures aim to redistribute productive assets among generations. A higher per capita land endowment relative to the village median increases fear of land loss and reduces perceived likelihood of gains, as expected in a system that aims to distribute a limited amount of overall land equitably among rural residents. The opposite is true for higher shares of good quality land, which could suggest that officials are either not good at assessing land quality or do not take it into account in making their decisions. Nonland physical assets, oxen, education, and possession of an iron roof all have little impact on the perceived threat of land loss or gain. Interactions between the treatment dummy and assets, land, the head’s gender all are insignificant throughout, providing little support to the notion that certification-induced tenure security effects are differentiated by wealth or gender.
Land-Related Investment
From an economic point of view, higher tenure security should manifest itself in landrelated investment. Table 7 reports estimated marginal effects from probit and tobit models for new investment in or repairs of conservation structures over the last 12 months. Results from the Chamberlain specifications with village- and household-level treatment indicators (Columns 1 and 2 for probit and 3 and 4 for tobit) point toward significant and economically meaningful impacts. This is consistent with the fixed effects linear probability model in Column 3 of Appendix Table A1. According to our estimates, the propensity to invest in soil and water conservation measures increases by between 20 and 30 percentage points. The number of hours spent on such activities increases significantly with program participation, although the log transformation of the dependent variable precludes determination of the average partial effect.
Impact of Certification on Propensity and Magnitude of Investment in Soil and Water Conservation: Marginal Effects from Chamberlain Random Effects Model
Coefficients on other variables suggest that the propensity to make land-related investment increases with plot size but decreases in total holding size. This is consistent with the notion that the presence of some fixed cost element increases payoffs from investment in conservation for larger fields, but that, on larger holdings, there is increased competition among plots for investment. The propensity to undertake investment is significantly lower on flat plots, consistent with the fact that such plots are less prone to erosion and land degradation than hilly plots, implying less need to guard against these through adoption of soil conservation measures. As the investments considered do not involve any cash outlays, there is little reason to expect impacts to be differentiated by wealth, as is suggested by the lack of significance of the certification dummy’s interaction with the various measures of wealth (not reported) throughout.
The potentially large magnitude of increases in the propensity to invest makes it of interest to assess the economic impact of certification. In light of the recent nature of certification, such investment will not yet have affected production in our survey. To obtain a measure of the size of the investment impact, we estimate a household fixed effects production function with a dummy for the presence of a functioning conservation structure.23 Results, as reported in Appendix Table A3, suggest that such a structure increases output by about 9 percentage points, implying that, with mean annual output of ETB 3,300 (with 9.6 ETB to $1), investment-induced certification impacts are between ETB 56 and 87/ha (0.29 or 0.19*0.09*3,330).24 Even assuming that some of the investment involves repairs of existing structures or labor is valued at the prevailing wage, our estimate implies that the increment in output from certification-induced investment in the first year alone could be sufficient to cover program costs ($1 per plot or $3.2/ha). Although a few issues, such as the addition of a cadastral index map, inclusion of common property resources, and a mechanism to keep records up to date, will need to be added to ensure sustainability of Ethiopia’s program, these are unlikely to increase the cost beyond what would be warranted in light of the benefits obtained, which could be further increased by changes in policy.
Rental Market Participation
Rental markets are likely to become of increased importance as a catalyst of the local nonfarm sector, and having a certificate can, in principle, be an important incentive for farmers to rent out. Tables 8 and 9 present results from probit and tobit estimates of equation [3] that allow us to test whether, as expected, certification affected the propensity to rent out but left demand for land rental unaffected. In both cases, results from the Chamberlain specifications of the rent-out regressions in Columns 1–3 (with treatment defined at village or household level) strongly support our hypothesis. Estimated marginal effects suggest that certification consistently increases the amount of land rented out by about one-tenth of a hectare at the mean, and the propensity to rent out by 13 or 9 percentage points for treatment defined at village or household level, respectively.25 Estimated impacts for renting in are insignificant as expected if treatment is defined at the village level and marginally negative if treatment is defined at the household level (Table 8, Column 6). Our dynamic analysis (Columns 2–3 and 5–6 in both tables) suggests that participation decisions and the amount of land transacted on both sides of the rental market are strongly and positively state dependent. Thus, policy interventions affecting market participation at any given point in time will affect households’ long-term trajectories.
Certification Impact on Rental Market Participation: Marginal Effects from Chamberlain Random Effects Probit Models
Certification Impact on Size of Land Rented: Marginal Effects from Chamberlain Random Effects Tobit Models
Our results also point toward a significant impact of land endowments on renting out (positive) and renting in (negative), as would be expected if rental markets contribute to equalization of factor input ratios. Total owned area has a positive and significant effect in the leasing-out regressions as compared to a negative and significant effect on the leasing-in regressions. However, the absolute value of the marginal effect of total owned land on the amount of land rented out or in (Table 9) is less than one, indicating that rental market participation allows only partial adjustment toward desired area of cultivated land (Bliss and Stern 1982). Contrary to what is found in studies from other countries, but consistent with evidence from Ethiopia (Deininger, Ali, and Alemu 2008), rental markets transfer land from relatively resource-poor households (mainly in terms of oxen power) that are often female headed, to comparatively resource-rich households. Appendix Table A2 demonstrates that the parallel trend assumption holds throughout the preintervention period for renting out (Columns 3 and 5), the variable of primary interest in our analysis, implying that unobserved factors did not lead to differential evolution of this variable over time in treatment and control villages.
Rental markets in Ethiopia also have strong gender implications, as sociocultural norms and factor market imperfections make selfcultivation of land by female-headed households extremely rare, implying that they either rent out their land or, often due to insecure tenure, leave it fallow (Adal 2005). This is borne out by our results where gender of the household head and the number of oxen have significant impacts on the nature and magnitude of rental market transactions in terms of encouraging renting in and discouraging renting out. Older households are more likely to rent out, and literate households are more likely to rent out larger areas of land. The significant coefficient on possession of an iron roof in the rent-in equation may point in the same direction by highlighting imperfections in financial markets that make renting in easier for those with greater wealth. To the extent that they allow productive use of plots that had been left uncultivated, or greater freedom in the choice of transaction partner to transfer land to those with higher levels of ability, certification-induced rental market effects could enhance productivity of land use. Such impacts could come about if (female) landlords were able to enter into longer-term contracts or to select more-productive tenants beyond their immediate social network, due to the increased security provided by certificates. As virtually all land is rented under sharecropping contracts, any productivity effects would translate directly into improved welfare for (female) landlords. Although beyond the scope of this paper, further study of impacts on women would be of interest.
V. CONCLUSION AND POLICY IMPLICATIONS
This paper was motivated by the fact that, despite a combination of supply and demand factors that has led to renewed recent interest in land registration, evidence on the impact of specific interventions is lacking, so that it is not clear whether land tenure should be of greater concern to policy makers. To explore this, we use a four-period household panel to assess short-term impacts of Amhara’s certification program on perceived tenure security, investment, and land market participation. Double differences and a pipeline comparison with household fixed effects provide estimates of program effects, noting that trends in both treated and control areas moved in parallel before the program had been announced.
We find certification to have resulted in a significant reduction of tenure insecurity and an increase in land-related investment and, to a lesser extent, supply of land to the land rental market. Although tenure insecurity decreased markedly due to certification, it remains large, pointing to the need for complementary action on the policy front if the full potential of this intervention is to be realized. Estimated investment effects are similarly large, and our results suggest that, if the voluntary investments made following certification will be maintained or if additional investment will be forthcoming in the future, benefits exceed program cost. Furthermore, contrary to what was experienced in most other cases, the cost of maintaining the land administration system will not be a major constraint to its sustainability in the long term. Implementing a decentralized, transparent, and cost-effective process of land registration under African conditions is not only possible but, by reducing (but not eliminating) tenure insecurity, can have impacts such that the magnitude exceeds the cost of implementation, even in Ethiopia, where policy restrictions rule out any credit effects a priori. This suggests that, in the many situations where population growth, urban expansion, or land sales to outsiders pose a threat to tenure security, a community-based process to certify and register rights may be economically and socially beneficial by securing existing rights and allowing right holders to individually or collectively make decisions on how to use such rights.
Given the short time elapsed since implementation of the program, our study is able to provide evidence of short-term impacts only. Longer-term effects could be larger or smaller than the ones ascertained here, due to a number of factors related to physical maintenance and expansion of the system, future policy initiatives, and the extent to which certificates provide a reliable basis for predicting behavior by third parties, especially government officials. Follow-up research to explore the extent to which benefits from land certification are affected by the policy environment, as well as their distribution and longer-term trajectory, could provide important inputs into the policy dialogue in Ethiopia and beyond.
Acknowledgments
The authors would like to thank AAU staff for expert data collection; M. Carter, A. de Janvry, S. Holden, G. Kohlin, Y. Liu, E. Sadoulet, for valuable comments and suggestions; and two anonymous reviewers. Financial support from the Global Land Tools Network and the Norwegian ESSD Trust Fund (Environment Window) is gratefully acknowledged The views expressed in this paper are those of the authors and do not necessarily reflect those of their respective institutions.
APPENDIX
Impact of Certification: Household Fixed Effects Linear Probability Model
Test of Parallel Trend Assumption Using Pretreatment Data for Perceived Tenure Security and Rental Market Participation: Marginal Effects from Chamberlain Specification
Determinants of Value of Crop Output: Household Fixed Effect Estimates
Survey Districts (Woredas) in East Gojjam Zone
Footnotes
The authors are, respectively, lead economist, World Bank; economist, World Bank; and assistant professor, Addis Ababa University, Addis Ababa, Ethiopia.
↵1 Given the policy-induced limits to the functioning of land, labor, and credit markets in Ethiopia, further productivity impacts of land certification through these channels will be small at best. To obtain a conservative estimate, we focus on investment-induced productivity effects only.
↵2 The large differences in the ratio of credit to GDP across countries is used as a key argument to justify interventions to formalize land rights that could then allow greater use of land as a collateral to access credit (Besley and Ghatak 2008; de Soto 2000).
↵3 The program is similar in size to the 11 million certificates awarded in Vietnam from 1993 to 2000 and the issuance of 8.7 million titles in Thailand during 1980–2005. Its accomplishments compare favorably to what was achieved by other land administration programs, for example, the 2.7 million titles (1.2 million urban and 1.5 million rural) issued in Peru from 1992 to 2005 and the 1.8 million titles issued in Indonesia since 1996.
↵4 The fourth largest region in Ethiopia, Tigray, had implemented a similar program in 1998 (Holden, Deininger, and Ghebru 2009). Although a number of modificationswere undertaken, the relative success of this program wasone of the reasons for other regions to initiate certificationprograms.
↵5 The fact that the LAC is directly elected in a democraticfashion rather than being part of the (often politicized)administrative structure was mentioned repeatedly as an importantmerit in interviews with groups as well as individualvillagers.
↵6 Although LAC members repeatedly emphasized thedemanding nature of this task, it is critical to ensure transparency,especially in the identification of communal areas.This reduces the scope for error that could arise from theuse of office records that may not be up to date.
↵7 In one site, the volume of court cases is reported tohave reduced from 20 to at most 2 per week (Adal 2008).This is important as, according to local government statistics,land conflict accounts for some 80% of rural crime. Italso has a relevant gender aspect, as in some cases widowswere able to win court cases to hold on to their land ratherthan, as dictated by local tradition, have it revert back to thehusband’s lineage at the point of his death. In polygamoussettings, the requirement to have separate certificates for anyspouses beyond the first one is linked to a reduction in (reported)polygamy. Even male farmers acknowledge thatjoint titling increased their wives’ willingness to work andinvest as official co-owners. Households in areas where urbanexpansion is imminent are reported to be particularlyeager to get certificates that could help them substantiatetheir claims for compensation if their land is taken for urbanexpansion. In fact, observers link the ability to use certificatesfor demanding compensation to the emergence of innovative,in-kind compensation arrangements in a numberof peri-urban areas.
↵8 For a detailed computation of program cost see Deininger et al. (2008).
↵9 The proclamation (law) in Amhara allows land redistributionif properly researched and decided upon at thecommunity level. Tigray has recently started redistributingthe land of anybody absent from the village for more than2 years with a minimum income ($100 per month).
↵10 Data from the study villages confirm the widespreadnature of the government-sponsored land redistribution inAmhara region in 1997. The first-round data that were collectedin 2000 show that about 46% and 48% of the sampledhouseholds experienced a decrease and an increase in theirholdings, respectively, since the formation of their household.Of these amounts more than 80% of the changesoccurred due to village-level land redistribution and reallocation,and about 61% of the decreases and 33% of the increasesin landholdings had happened in 1997. The fact thatthe highest percentage of parcels (22%, while the secondhighest was just 10% in 1975, i.e., right after the radical landreform) was acquired in 1997 is an additional evidence forchanges in landownership in the region in the specified year.
↵11 The East Gojjam zone was selected purposefully torepresent surplus producing areas of the region. The districtsand the villages in each district were also selected based onsimilar criteria, while the households in each village wereselected randomly. For agricultural production, the referenceperiod is the main season (meher, i.e., June to February) inthe 1998/99, 2000/01, 2003/04, and 2006/07 crop years. AppendixFigure A1 depicts the location of the three sampledistricts (woredas) in the East Gojjam zone of the Amhararegion.
↵12 Information from one of the treatment villages (A.Gulit), where the second round of the land certification surveyhad been undertaken, is available only for the last tworounds, as it was added to the sample during the third round.
↵13 As restrictions on migration or labor-market participation(e.g., a residency requirement to avoid loss of landuse rights) are prevalent in rural Ethiopia, it is not too surprisingthat the rate of attrition was, with less than 1% peryear, lower than in any of the seven longitudinal surveys forwhich attrition rates are reported by Alderman et al. (2001),and 93% of households from the first round were still presentin round four. At the plot level, only 10% of surveyed householdsreported any increase or decrease of their landholdingover the period covered by our survey. As most of thesewere transfers within the extended family that should becorrelated with observables, this is unlikely to systematicallyaffect our results.
↵14 A number of factors that can range from delays indelivery of printed certificates to the woreda, to the sequencingof batches to sign, or to the lack or loss of owners’pictures can lead to delays in issuance of final certificates toindividuals even in villages where the registration processhas been completed and a majority received certificates.
↵15 Both A. Gullit (treatment) and Telma (control) wereused to pilot different methods of certification. In the former,high-precision surveys were conducted with advanced technologyincluding GPS and total stations, leading to a delayof 28 months between the start of the program and issuanceof the first certificate. In the latter, some of the participatoryprocesses were initially tested, and the fact that the woredawould not be able to count it toward its achievement led toits temporary abandonment. Dropping either does not affectthe substantive results reported here; results from doing soare available upon request.
↵16 As discussed earlier, defining treatment at the villagelevel is likely to result in an underestimate of impact owingto inclusion of noncertified households, whereas defining itat the household level would overestimate impacts.
↵17 We use the response to the question of whether thehousehold expects a change (increase or decrease) of landholdings through administrative action over the next 5 yearsand note that, as the question format was identical in all foursurvey rounds, concerns about potential halo effects areunfounded.
↵18 As the generalized expectation of an increase in holdingsize could exert considerable pressure on policy makers,both of these outcomes may be relevant for tenure security.
↵19 While less robust, a plot-level variable asking ownerswhether they were concerned about land conflict, which wasintroduced only in the last round, points to significantlyhigher levels of tenure insecurity in the control villages(20%) as compared to the treatment villages (14%) in a simplecross section.
↵20 To allow for relative rather than absolute land size toaffect the risk of expropriation, we use the amount of ownedland per adult equivalent relative to the median of this variablein the village, although results are similar if absoluteland size is used. The rationale for including the other household-level characteristics, x, in all the specifications and theplot-level characteristics, p, in equation [2] is that they willindependently affect the likelihood of redistribution or landmarket participation and cost of and return from investment.
↵21 Despite its attractiveness in terms of being able tohandle unobserved heterogeneity if explanatory variablesare discrete in nature (Wooldridge 2002), shortcomings ofthe linear probability model in situations such as ours wherethis is not the case include that error terms are unlikely tosatisfy the implied normality assumption and inferior efficiencyof the resulting estimates as compared to nonlinearprocedures (Maddala 1983).
↵22 As an additional check, we follow the suggestion bya reviewer to conduct a “placebo test” that randomly assignsvillages or households to treatment and or to control whilemaintaining their relative shares in the total. Given the smallnumber of villages, we focus on households. Results from1,000 repetitions for the tenure insecurity regression implythat the null hypothesis (having no positive or negative impact)is not rejected in more than 90% of the cases, boostingconfidence on the plausibility of our methodology. The analysisis repeated for the fixed effects linear probability model,and the results are found to be similar.
↵23 With constant returns to scale as assumed in the aboveanalysis, estimates using the value of crop output per hectare will not be different from those obtained using total outputvalue, as is indeed confirmed by the data.
↵24 All investments in soil and water conservation structuresin the survey villages are labor intensive with few materialinputs. It should be noted that the benefits from theprogram would definitely be reduced if the shadow value offamily labor was considered in the analysis.
↵25 The generalized method of moments estimation of thelinear dynamic probability model suggests that this effect isless robust than earlier ones, making us more cautious in itsinterpretation. Nevertheless, the estimates from the dynamicfixed effects model may not be very reliable, as the Sargantest rejects the validity of the overidentifying restrictions.Moreover, the random effects probit model allows us to effectivelydistinguish the effects of state dependence fromthose of unobserved heterogeneity in the form of randomeffects (Alessie, Hochguertel, and van Soest 2004).
References
In this issue
Jump to section
Related Articles
Cited By...
- Centennial Anniversary of Land Economics: A Bibliometric Retrospective
- Financial Assistance and Environmental Compliance: Evidence from the Clean Water Act and the Clean Water State Revolving Fund
- Prospect Theory and Tenure Reform: Impacts on Forest Management
- Migration Pressure, Tenure Security, and Agricultural Intensification: Evidence from Indonesia
- Partial Land Rights and Agricultural Outcomes: Evidence from Thailand