Abstract
In several African contexts, households are unable to enhance agricultural production due to land constraints. Few governments have explored the use of resettlement to alleviate land scarcity and facilitate rural-to-rural migration. We examine whether a resettlement project in southern Malawi improved food security in the long term. Our findings indicate resettled households achieved greater long-term food security, owing to additional land coupled with a more diversified crop portfolio. We also find the formalization of property rights improved land security for male and female household heads, but resettlement jeopardized the land security of women in maleheaded households. (JEL Q15, Q18)
I. Introduction
In several African contexts, land scarcity combined with exposure to micro- and macroeconomic shocks jeopardizes food security. Households are unable to enhance production through expansion of cultivated area since land is constrained. Vulnerability to climate disasters may induce risk-averse households to underinvest in agriculture (Rosenzweig and Binswanger 1993). Input price shocks further reduce the available options for those facing severe land constraints to enhance productivity. Under these circumstances, migration can serve as a risk management strategy by providing access to auxiliary income sources (or land) and freeing up resources for households that remain behind (Taylor and Martin 2001).
Community-based approaches to land reform have been used to remedy extreme inequalities in access to land, initially in Latin America (World Bank 2003, 2009a), by mobilizing resources for communities to relocate to areas, involving communities in the selection and negotiation of land purchases, and then providing grants and initial resources for farmers during the transition. These programs most importantly offer individual farmers access to land to produce and sustain livelihoods—albeit in more remote areas, where land is abundant. They also offer features to reduce the costs of relocation and improve the expected benefits of migration.
Due to the politically sensitive nature of these programs (e.g., extent of voluntary participation, contention between resettled households and native households) and the challenge in identifying impacts ex post, many of the assessments of resettlement programs have largely been qualitative and the impacts are very context-specific (Cernea 1997). For example, initial income declines of 40% to 50% caused by the displacement of households by two dam projects in Indonesia may have been offset by the generation of fishing enterprises (through additional financial support from the government, World Bank, and others), which later increased employment opportunities and, hence, living standards of the displaced (Van Wicklin 1999). Another example is from the ongoing resettlement component under the Food Security Program in Ethiopia, which led some households to greater food insecurity, while others prospered (Pankhurst et al. 2013). During the initial stages of resettlement, there was quite a bit of uncertainty as to the expectations of harvests on newly cultivated land, and sometimes delays in the delivery of food rations (Pankhurst et al. 2013). Those households that were able to reduce the uncertainty of food security through taking advantage of social capital built at their origin or through investing in livestock as a form of savings were much more likely to withstand the transition (Pankhurst et al. 2013). The figures on regional return migration, which range from 11% to 46%, suggest some households (as well as destinations) proved more successful than others (Pankhurst et al. 2013). Both examples suggest great lessons can be learned from quantitative evaluations of ongoing resettlement projects in terms of the roles of land reform and migration in improving food security.
We take advantage of a project implemented by the government of Malawi, the Community Based Rural Land Development Project (CBRLDP), to evaluate whether resettlement can improve the agricultural production and food security of households in the long term. A four-year data collection effort (2006-2009) to monitor and evaluate the CBRLDP was commissioned by various stakeholders. The International Food Policy Research Institute collected a fifth round in 2011. We use the baseline and final rounds to estimate the long-term implications of resettlement.
We face two challenges in measuring the impact of resettlement using the CBRLDP panel. First, attrition between rounds can potentially bias impact estimates. Without accounting for attrition, one might be concerned that favorable resettlement impacts are driven by the more successful farmers continuing to reside in the resettlements, rather than actual improvements attributable to the program. Second, participation in the program is voluntary and targeted, implying that simple comparisons of the outcomes of beneficiary households and households in selected control districts are subject to the standard selection problem. We address the two empirical issues by using a nearest neighbor matching framework (Abadie et al. 2004; Abadie and Imbens 2006) with inverse probability weights (Fitzgerald, Gottschalk, and Moffitt 1998). The former technique attempts to control for selection bias by matching each treated household with similar control households before comparing differences in outcomes. The adjusted weights attempt to correct for attrition bias. As an additional measure of robustness, we also employ a propensity score weighted regression, which uses a function of the propensity score and attrition weights as weights (Hirano, Imbens, and Ridder 2003; Busso, DiNardo, and McCrary, forthcoming).
II. Community Based Rural Land Development Project
In 2004, the government of the Republic of Malawi, with financial assistance from the World Bank, began implementing the CBRLDP. The intent of the project was to address poverty and emerging social conflicts in the southern region of Malawi. As the flagship operation of the Malawi Land Reform Program Implementation Strategy (2003-2007), the CBRLDP was devised to address land redistribution issues arising from stark inequality in land distribution, land degradation, rural tensions, high population density, and land market failures. The project aimed to reallocate underutilized land from failing tobacco and tea estates to land-constrained households. Four districts were selected for the program: Mulanje, Thyolo, Machinga, and Mangochi. By 2012, there were 15,142 beneficiary households over the course of the program in four of the original districts plus two districts added in 2008 (Ntcheu and Balaka) (World Bank 2012).
The beneficiaries of the project were chosen on a voluntary basis, but eligibility was conditioned on the following criteria: Malawi citizenship, landlessness or being land poor, food insecurity, low-income status, chronic dependence on external assistance, and vulnerability. Participants were required to form a beneficiary group (10-35 households), be willing to relocate and farm, adhere to transparency and accountability principles, and follow recommended practices. Beneficiary groups attended informational trainings on the identification of suitable land allotments for resettlement (e.g., decent soil quality and water access). Each beneficiary group also formed a project management committee that identified such land, negotiated land acquisition, prepared the farm development plan, and oversaw activities of the group, such as signing all documents on behalf of the group and initiating bank accounts to withdraw funds for the group. A list of estates on sale in the pilot districts were distributed to the group after being vetted by the CBRLDP staff and the district assembly officials in the receiving districts. The group then would send representatives, namely the project management committee, to view prospective estates and negotiate the land price.1
The resettlement terms were as follows: Households in a beneficiary group were each given a cash grant of $1,050, where up to 30% of the grant was allocated to buy land, 8% for resettlement allowance, and 62% for farm development. Each household was demarcated about 2 ha of land from the total land bought, had access to 3 to 5 ha of communal land, and was granted a group-level title deed that was signed by all of their members. The cash grant was given to the beneficiary group as a whole in three tranches: 60%, 20%, 20% (World Bank 2009b). For beneficiaries moving more than 50 km, cost of transportation was also provided. In addition to the cash grant, the project also provided beneficiaries with training in negotiation, financial management, and extension services to inform decisions on input expenditures, farm development such as land resource management, preservation, ridge and plant spacing, and fertilizer and pesticide usage.
Quantitative impact assessments of the CBRLDP have been limited to just two short- and medium-term evaluations. Datar, Del Carpio, and Hoffman (2009) find beneficiaries significantly increased the area of land cultivated and agricultural production (maize and total crop value) over the course of two years. However, the authors did not find maize yields to be positively affected, leading to the inference that the increase in agricultural production was likely driven primarily by the expansion of cultivated land area. Simtowe, Mangisoni, and Mendola (2011) also performed a short-term (comparing effects between 2006 and 2007) as well as a medium-term (comparing effects between 2006 and 2009) evaluation of the resettlement project. They find that both short- and medium-term benefits were realized in the form of increased total landholdings, maize production, total crop production, and maize yields. However, the authors also find that the benefits of the project across maize production and total crop production were markedly less positive in the medium term. We build upon these previous evaluations and examine the impacts of the program five years after its inception, by incorporating a later round of data. Of particular interest is how such programs make participants more resilient or more vulnerable to external shocks such as the global food crisis and the local economic (especially fuel) crisis at the time of the survey.
Evolution of Land Rights in the Midst of Resettlement
One of the key insights missing from prior analysis of the project is the impact of resettlement on land security. Beneficiary groups were given the choice to register their land allotments under customary, freehold, or leasehold status (Simtowe, Mangisoni, and Mendola 2011).2 Relatively less is known about how these arrangements were formally described and presented as options to individuals within the beneficiary group. While it was possible for beneficiary groups to register subdivisions of their allotment, the conditions required for this formality, such as surveying, valuation, and registration fees for the titles to be registered, were likely cost-prohibitive (World Bank 2009b). Unfortunately, attempts to ascertain these details and their effects using the CBLRDP panel have been limited, as only one question related to landownership was asked on the questionnaire prior to 2011.
While it remains to be established how resettlement affects landownership, of equal importance is learning how traditional forms of land inheritance might have been affected. In particular, southern Malawi follows matrilineal customs, and, traditionally, women own the land and their spouses use the land of their wives' families. Daughters of these marriages inherit the land. In their interview with a Lomwe ethnolinguistic beneficiary group from the Thyolo district, Holden, Kaarhus, and Lunduka (2006) find that these traditions resonated. However, in their interview with a Yao ethnolinguistic beneficiary group from the same district, the authors discover that although land rights were in principle inherited from their parent's generation following matrilineal custom whereby couples would normally settle uxorilocally—with the wife's village—men and women had contrasting opinions about land rights. Women claimed that land was either theirs or jointly owned with their spouses, while the men indicated that land allocated through the resettlement program actually belonged to them and that land titles should be issued under their names. Qualitative evidence at least suggests that the resettlement program might compromise the traditional practices of matrilineal land inheritance.
The present study builds on these existing evaluations in three ways. First, we measure long-term food security impacts of the CBRLDP using the most recent data collection effort. Second, we make adjustments for attrition in our estimates using the technique adopted by Fitzgerald, Gottschalk, and Moffitt (1998). Third, we examine how the resettlement project affected land security. We exploit additional questions added to the survey in 2011, to further disentangle how resettlement might have gender-differentiated impacts on landownership.
III. DATA
The International Food Policy Research Institute collected an additional round in August 2011 of an existing panel (2006 through 2009) to evaluate the CBRLDP.3 The survey instrument covers an array of household behavior in order to document improvements in agricultural production, food security, assets, and other income-generating activities attributable to the resettlement program. Figure 1 provides a district map of the beneficiary and nonbeneficiary sites in the panel dataset. The sampling strategy for the long-term evaluation of the resettlement program encompasses participants of the program who resettled in Machinga and Mangochi and nonparticipants living in adjacent districts, Balaka and Chiradzulu. Some participants originated from the resettlement districts, while others were from Mulanje or Thyolo.4 Our impact evaluation is based on comparisons between 220 treated and 333 control households in adjacent (not the same) districts that were present during the 2006 (baseline) and 2011 surveys.5 Such comparisons are reasonable using a matching approach, assuming that, conditional on the baseline covariates we use, the control households have the same mean outcomes as the treated households would have had they not resettled elsewhere (conditional mean independence).
District Map of Beneficiary and Nonbeneficiary Sites
The attrition rate between the baseline and final rounds is 15%, approximately 3% a year. We account for the impact of the attrition in our analysis using inverse probability weights (Fitzgerald, Gottschalk, and Moffitt 1998). We estimate an unrestricted probit regression to explain who was present in the 2006 round and remains in the sample by the 2011 round using variables from the baseline survey. A second restricted probit regression is estimated that omits the instruments (ethnic dummy variables and the village attrition rate6). The ratio of the predicted values from the restricted and unrestricted regression is then used to compute the inverse probability weights applied in the calculation of descriptive statistics and the matching approach. The coefficients and standard error estimates produced from the unrestricted probit regression are presented in Table 1. The statistically significant determinants of remaining in the sample by the final round are ethnicity and age. Those with a Chewa ethnic background are more likely to remain in the sample than those with a Yao ethnic background. Older individuals are more likely to remain in the sample at a decreasing rate. These factors suggest that impact estimates that ignore attrition may bias the impact of resettlement estimates, validating the use of attrition-adjusted weights in the analysis.
Probability of a Baseline Household Remaining in the 2011 Sample
We present descriptive statistics of the household—its food security, agricultural inputs, vulnerability to shocks, and market access indicators by round and program participation—in Table 2. Our measure of production is computed using the total quantities produced by the farmer multiplied by the median output prices constructed from household responses within a district. Total output is divided by the total landholdings possessed by the household to construct a measure of yield. Overall, beneficiary households experienced larger increases in total production and smaller declines in yields than control households.
Descriptive Statistics by Participation in the Resettlement Program
We further disaggregate the production measures by crop in Table 3. By 2011, beneficiary households substantially diversified the portfolio of crops they produced, from mainly concentrating on maize to growing, for example, cassava, sweet potato, sorghum, pigeon pea, cowpea, and groundnut. Control households had lower diversification than beneficiary households but increased production in cash crops common to the area, such as tobacco and cotton. One potential concern is that, with increased diversification, declines in maize production may jeopardize the food security of an average household since its caloric intake predominantly comes from the consumption of maize (54%) (Minot 2010). At a descriptive level, food security concerns are assuaged, since beneficiary households are switching to alternative staple crops and to crops with more nutritional value. Consumption patterns are aligned with improvements in total production and crop diversification, as beneficiary households report greater increases in the number of meals per day consumed in both harvest and lean periods, with a higher magnitude of improvement in the latter.
Crop Choice Portfolio by Participation in the Resettlement Program: Value of Agricultural Production
Another important source of income is through livestock ownership, which both groups of households increased (Table 2). Interestingly, control households experienced a growth rate of 223% in livestock assets (compared to the beneficiary growth rate of 93%). Yet, beneficiary households still generated greater income improvements, even when accounting for increased livestock values.
The expansion in total production is mostly attributable to the additional land provided by the resettlement program. Beneficiary households on average had 0.52 ha (compared to control households who had 0.78 ha) prior to the program's inception. By 2011, the total landholdings of beneficiary households on average increased to 1.38 ha (compared to 0.91 ha on average among control households).7 On average, there are minimal differences in other changes in inputs used across participant and nonparticipant groups over time. However, resettlement households are burdened by remoteness, which affects their profit potential, access to markets, and access to agricultural extension agents. Resettled households also experienced a higher exposure to floods in the last 12 months prior to the 2011 survey. These factors may potentially explain the tendency for resettled households to reduce cash crop production and diversify staple and other crops.
Descriptive figures also suggest resettlement might have improved land security. Thirty-six percent of resettlement household heads claim to have a title for their landholdings, compared to 18%. One of our additions to the survey instrument in 2011 was a module to identify individual ownership of each household plot. Thus, the new cross-sectional data provide a unique opportunity to enhance the traditional differentiations of impacts by evaluating how changes in land acquisition might have differed for individuals within the household according to their gender.8 Using the plot-level data, we will later illustrate in Section V that, despite the appearance of improvements in land security, the resettlement program disadvantaged the majority of women in the matrilineal south.
IV. Methodology
One of the main limitations of estimating the long-term impacts of any voluntary and targeted program is that eligibility and participation are not random. Comparing the outcomes of participants to nonparticipants yields biased results, as unobserved factors that affect participation are likely correlated with the outcomes of interest. To address this selection problem, we use the nearest neighbor matching approach, which compares the outcomes of program participants with similar nonparticipants using a suite of observable characteristics (Abadie et al. 2004; Abadie and Imbens 2006).
There are several advantages to using this matching estimator. First, an analytical expression has been derived for the variance, which avoids the problem of inconsistency among propensity score indicators (Abadie and Imbens 2008). Second, the matching procedure involves estimating an indicator to match each treated individual with one or two control individuals (in our study), in which the indicator is generated using a distance metric based on the norm of the observables (Imbens 2004). Propensity score indicators are based on parametric models with asymptotic properties that may not hold for smaller sample sizes. Lastly, the procedure can easily accommodate more complicated variance structures, such as the inverse probability weights we will use to deal with attrition.
Our measure of the impact of resettlement on those that participate in the program is the average impact of the treatment on the treated (ATT):
[1]
Our resettlement program participation indicator R takes on the value one for beneficiary households that have been resettled in 2006 and zero otherwise. Beneficiary and control household outcomes are represented by Y1 and Y0, respectively. We use a vector of baseline variables X to match similar treated with control households that characterizes the differences between program participants and nonparticipants and explains variation in the outcome of interest, food security. We condition participation on variables that are likely to influence eligibility for the program and one's expected return from participating (ethnicity, sex, and age of the household head, land holdings, the number of adults in the household by sex, durable assets, and livestock assets) and remain correlated with agricultural production and consumption. Since the outcomes of interest are vulnerable to unpredictable changes in the environment, we also include variables that reflect the shock exposure in the last 12 months before the 2011 survey, S (the percent of households within the village that were exposed to droughts and floods).9 A second source of bias comes from the inability to control for all time-invariant characteristics that impact the outcomes. To reduce this source of bias, we apply the difference-in-difference matching estimator,10 which compares the change in 2011 and 2006 outcome levels by resettlement participation:
[2]
Figure 2 illustrates the kernel densities of the propensity score values by program participation, which were predicted based on a logit regression. Logit estimates are included in Table 4. While the range of propensity score values is consistent for each group, a greater number of lower propensity score values is present in the group of control households. We address the issue of overlap and matching quality by providing additional robustness checks. First, we provide nearest neighbor matching estimates using a trimmed sample, that is, using only households with propensity score values that lie between the 5th and 95th percentiles. Second, we apply bias corrections to all of the matching estimates to reduce the bias caused from a combination of the use of numerous covariates and poor matching performance (Abadie and Imbens 2002). Third, we report estimates from regressions that weight observations by a function of the propensity scores and the attrition weights (Hirano, Imbens, and Ridder 2003). Busso, DiNardo, and McCrary (forthcoming) show for finite samples when overlap is poor, the nearest neighbor matching with bias correction is, however, more effective.
Propensity Scores by Program Participation
Probability of Program Participation
V. Results
Food Security
We provide the average treatment on the treated effects for six food security outcomes in Table 5. The most conservative estimate suggests long-term impacts on total production of 59%. The largest estimate indicates a total improvement of 82%. Note the estimates produced by the matching approach are lower in magnitude than those extrapolated from taking the differences in the variable means across treatment and control groups, underscoring that resettled households were positively selected in observed attributes. We did not find robust differences in total yields. Estimates in the maize production improvements attributable to the resettlement program were not robust to all specifications, but weak evidence suggests a range between 35% and 40%.
Average Treatment Effect for the Treated on Food Security Outcomes
To understand how these production differences manifest in consumption, we compare the number of meals per day consumed after harvest and during the lean period, as well as per capita food expenditures (Table 5). We find a statistically significant and robust difference in the number of meals consumed after harvest and during the lean period. Differences in log per capita food expenditure, however, are not commensurate with the measured differences in meals. While beneficiary households appear to experience a slight reduction in per capita food expenditure, the estimates are not statistically significant, which indicates that we cannot reject that there is no difference between the per capita food expenditure values of resettlement participants and nonparticipants.
The timing of the final survey likely explains the lack of effect on food expenditure. While all households in Malawi were exposed to a global food price crisis and a local economic crisis in 2011, beneficiary households were likely more vulnerable. The average distances beneficiary households traveled to the nearest market shop, market to buy farm inputs, and market to sell outputs far exceeded those traveled by control households (see Table 6).11 Similarly, the overall demand for nonstaple food items likely declined due to low purchasing power.
Average Treatment Effect for the Treated on Inputs and Remoteness
Interestingly, the remoteness issue might have influenced the decision of beneficiary households to produce more staple crops, since with the exception of landholdings, there appear to be no statistically significant differences in the changes of other key agricultural inputs (Table 6). Although every single control and treatment household in our sample produced some maize over both time periods, maize production rose substantially among beneficiary households. We further inspect whether changes in decisions to grow other crops vary by resettlement status, by taking the difference in indicators representing whether the household produces a given crop at the baseline and 2011 (Table 5). A greater proportion of resettlement households produce pigeon pea, perhaps owing to the desire to increase the production of another staple crop and the knowledge of the soil quality and yield benefits in intercropping pigeon pea with maize. Moreover, a greater proportion of beneficiary households are producing tobacco than were at baseline. For these settlers, tobacco may be a relatively safe cash crop to diversify their production portfolio because their land was previously part of tobacco estates, meaning that the land may be particularly suitable to tobacco production. Furthermore, because their land originates from large tobacco plantations, access to tobacco markets likely predates the resettlement program.
Therefore, the diversification of alternative crops for profit purposes may be riskier if these areas are remote and the locations of markets to sell these goods have not been identified.12
Property Rights and Gender-Differentiated Impacts
In addition to the production benefits, there is evidence that the resettlement program increased security in the ownership of land. Significantly more heads of households in the resettlement program indicate having a title for their landholdings (Table 7); female household heads even more so than male household heads. Due to the wording of the question, we are not able to distinguish between individual versus beneficiary group ownership of the title. However, when we further ask the house-hold head who decides how the land would be used if he and his family moved away, a greater percentage of resettlement household heads report the decision being made by a head/spouse, farmer group leader, then the traditional authority (TA)/chief. By way of program design, resettlers have fewer family members residing in the resettlement areas, and the farmer group leaders institutionalize property ownership for the beneficiary group. Thus, there may be limited scope and flexibility for maintaining landownership and diversifying household income off-site.
Average Treatment Effect for the Treated on Gender-Differentiated Land Holdings and Rights
Table 7 also presents comparisons between the allocation of land and property rights in 2011 by the sex of the plot owner. We first observe a substantial increase in the total landholdings owned by men in resettlement households relative to households in control sites, and no difference in the overall landholdings owned by women. The landscape of property rights also differs for beneficiary households. Male and female plot owners in beneficiary households have a lower proportion of land that was inherited, which is not surprising, considering the context of migration to the region. However, we observe that land rights of men in beneficiary households are disproportionately stronger than those of women. In particular, male plot owners in beneficiary households tend to have a greater proportion of land purchased with title (10% to 11% compared to 6% to 7%) and leased (46% to 47% compared to 20%) than their female counterparts. This suggests that the implementation of land titling might have disempowered women (most of whom are not female heads) in beneficiary households through the formalization of property ownership on behalf of the men within those households. This is a trend that has been observed in former attempts to formalize property rights in Malawi (Peters 2010), and more broadly in sub-Saharan Africa (Lastarria- Cornhiel 1997).
VI. Conclusion
This paper focuses on the long-term impact of the resettlement program on agricultural production and food security. We find beneficiary households realized a 59% improvement in overall production by 2011. Yields remained the same, which suggests land expansion was the main factor leading to production growth in this region. Considering that Malawi is a land-constrained country, food security and agricultural productivity in the longer term will need to be addressed through the use of sustainable intensification technologies. Land expansion through resettlement may afford only an option in the meantime.
Earlier in the CBRLDP, careful attention was placed to transition the beneficiary households in the short-term. The degree of remoteness, however, may create other disadvantages, such as poor access to markets and increased costs of production, especially for those farmers thinking of diversifying beyond subsistence crops. Increasing access to cheaper inputs and weather-resistant crop varieties can foster yield growth. Linking households to markets will enable programs like voluntary resettlement a stepping stone to improving agricultural growth.
How resettlement affected land security is also relevant given established relationships between investments and production (Besley 1995; Deininger and Jin 2006; Dercon and Ayalew 2007; Holden, Deininger, and Ghebru 2009). We find that significantly more heads of beneficiary households report having a title to their land; a greater proportion of female than male heads. Using plot-level data collected in 2011, we observe that men in beneficiary households tend to have more landholdings than men in control households. Men in beneficiary households also have stronger property rights than women within those households. Customary practices in which women acquire land through inheritance and transfer land to their families have been compromised in resettlement communities as women experience a greater reduction in land that is inherited. The protection of women's rights to land ownership and access in resettlement areas—particularly those women who live in male-dominated households—may be possible by securing their inheritance rights and including them in the land registration and titling process (Lastarria-Cornhiel 1997).
Acknowledgments
We thank Alan de Brauw, Pete Fleming, Joel Phiri, Humphreys Kabota, and the Invest in Knowledge Initiative, who were largely responsible for the implementation of the IFPRI 2011 survey. The paper benefitted from the feedback of two anonymous referees.
Footnotes
The authors are, respectively, research fellow; senior research fellow; senior research assistant, International Food Policy Research Institute, Washington, D.C.; and independent consultant, Johannesburg, South Africa.
↵1 The details regarding land acquisition are extrapolated from a qualitative report (World Bank 2009b) and Machira (2009). The process of land identification and acquisition differed for the beneficiary groups that moved out of their original district since there were no available estates to be purchased in Mulanje or Thyolo. These groups identified land by sending representatives from the Project Management Committee, which were elected by the beneficiary group in a participatory manner, for trustworthiness and gender balance, to view prospective estates and report back. Unlike those who moved within districts, these beneficiary groups were at a disadvantage because they were not familiar with the local estates, and therefore it was more challenging to identify suitable land and negotiate a good price.
↵2 Most of the land in Malawi follows customary tenure, which is allocated by the traditional authority and remains with the family’s lineage (Chirwa 2008). Households in possession of freehold and leasehold land have the use right of the land’ however, the terms are fixed when under leasehold. Freehold land can be further divided or leased out without government authorization (Chirwa 2008).
↵3 The government of Malawi and the World Bank commissioned data collection efforts over the duration of the CBRLDP. With the assistance of a local firm, Invest in Knowledge Initiative (IKI), these data sources were compiled into a four-year panel spanning 2006 to 2009. This panel has been used in a medium-term comprehensive evaluation of the project (Simtowe, Magisoni, and Mendola 2011).
↵4 Two additional groups of households were covered in the panel through 2009: a short-term control group that focused on surrounding areas in Machinga and Mangochi, and a short-term control group that focused on vacated areas in all four districts of origin (Machinga, Mangochi, Mulanje, and Thyolo). The International Food Policy Research Institute did not survey these households in 2011, but additional details regarding these sampling groups are provided by Simtowe, Magisoni, and Mendola (2011).
↵5 The household panel includes more households due to the expansion of the survey in future rounds. As part of the panel expansion, the households who were not part of the initial baseline survey were given retrospective questions to solicit behavior at the time of the baseline survey. For some households, these retrospective questions might have been asked years later (IKI 2009). The recall households tended to be more representative of the Yao ethnic group, had fewer female household heads, had households that consisted of fewer adult females, and tended to use more fertilizer based on t-tests comparing variable means by recall and real-time responses to baseline questions. Based on statistical differences and the potential measurement error associated with recall responses, we focus on the households that were present at the time of the baseline and 2011 surveys.
↵6 The village attrition rate is exogenous to the household by construction, as it is computed using the sample of household responses in the village, omitting in each case the observation for which it is being constructed. This strategy follows previous approaches (e.g., Alderman and Garcia 1993; Sahn, Van Frausum, and Shively 1994).
↵7 The small increase in the total landholdings of households in the control districts is likely attributable to changes in local demographics (e.g., deaths in the household, marriages in the household) or measurement error.
↵8 Since we did not collect production measures specific to each plot, we are unable to further disaggregate production effects attributable to the resettlement program by the gender of a household member (for example, as with Peterman et al. 2011).
↵9 We added a shock module to the questionnaire in our recent round. Thus, there is no information on shock exposure prior to this round, nor are we able to use weather information from secondary data sources due to the few districts (or administrative units) present in our sample.
↵10 This was first developed by Heckman, Ichimura, and Todd (1997) and Smith and Todd (2005) using the propensity score matching approach.
↵11 A number of observers have noted changes in land use and differential rates of fertilizer use in response to Malawi’s Farm Input Subsidy Program (see, e.g., Chibwana, Fisher, and Shively 2012). The fact that we observe no statistically different rates of fertilizer use despite the greater production of maize among beneficiary households suggests remoteness and temporary residency may be a barrier to full participation in the Farm Input Subsidy Program. The latter, in fact, may be at play, since permanent residency in a village was intended to be a criterion for program participation.
↵12 In a qualitative report, there were stated instances when beneficiary groups sold to less profitable but more convenient local vendors due to the lack of alternatives (World Bank 2009b). Moreover, there were cases of beneficiary households, particularly those that relocated to another district, who reported declines in consumption because their contract required the sales of bulk food items at a low price around the same time when households were exposed to erratic weather patterns.
