University of Amsterdam Amsterdam School of Economics
Master of Science in Economics
Property Rights & Rural Development in Tanzania
Andreas M. Knabe July 2021
supervised by
Prof. Menno Pradhan, PhD
Track: Development Economics
Student ID: 13436120
Statement of Originality
This document is written by Student Andreas Knabe who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
UvA Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
Empirical evidence on the impact of formal property rights on agricultural development is scarce and mixed, particularly in the case of Sub-Saharan Africa. Using longitudinal data from 2008 to 2020 covering entire Tanzania, this study reports more benevolent outcomes: Land cer- tification raises productivity and welfare, primarily through increased tenure security and land transferability, less so through improved credit access from collateralization. Certification further contributes to gender equality by granting wives more ownership rights and is congruous with equity considerations by giving primarily low-income households better credit access. Statisti- cally, Two-Way Fixed Effect estimates produce a robust positive investment response if titling is endogenous, even after accounting for a potential self-selection bias. Dynamic Difference-in- Difference with Staggered Treatment Adoption and Two-Stage Least-Square measures in contrast do not suggest a significant investment effect.
Introduction
Conventional economic theory maintains that secure property rights in land, especially individual property rights, are a prerequisite for economic development and growth (Miceli, Sirmans & Kieyah 2001). This is corroborated by recent cross-country studies on income differentials that highlight the role of institutions, in particular that of property rights (Acemoglu & Johnson 2005; Acemoglu, John- son & Robinson 2004). Generally, microeconomic investigations have confirmed these findings, but less so for Sub-Saharan Africa (see Lawry et al. 2017 for a survey). One reason is that comprehensive data are oftentimes lacking. Another is that land titling usually is endogenous or variation is insuffi- cient due to poor enforcement of de-facto legal land reforms. Many examples of large-scale changes in property rights were also accompanied by major social unrest (Banerjee, Gertler & Ghatak 2002).
This is the first rigorous quantitative study on the impact of a major land reform and its subsequent implementation in Tanzania. Peculiarly, it asks whether individual land titling in the form of granting Certificates of Rights of Occupancy (CRO) has raised agricultural productivity and investments of rural landowners. Using longitudinal data from 2008 to 2020 that cover whole Tanzania, I employ two-way fixed effect regressions, thereby controlling for constant idiosyncratic household characteris- tics and common time trends. Further, I examine the channels through which formal property rights might affect agricultural decision-making.
Land is a cross cutting sector which affects all sectors of the economy. Despite, there is a shortage of reliable evidence on the mechanisms by which property rights to land improve economic performance in general (Field 2007), and for the case of Sub-Saharan Africa in particular (Jacoby & Minten 2007).
Even more so, the existing empirics are mixed and inconclusive, both on the overall effect of having property titles and on which channels are crucial (Do & Iyer 2008). Early papers on South East Asian or South American countries, both in urban and rural settings, mostly find positive invest- ment effects from titling, and suggest that this works primarily through increased tenure security and land transferability. This picture has not emerged for the limited number of cases investigated in Africa. Proposed explanations include that informal tenure arrangements are well established, thus formalizing land rights might not change investment behaviour and a duality of competing formal and informal claims might actually deteriorate tenure security (Lanjouw & Levy 2002; Besley 1995).
Property certification might further be of limited impact if credit or land markets are poorly devel- oped (Ravallion & van der Walle 2006). Commercial lending to rural farmers might generally be too risky and agricultural land as collateral of little value (Atwood 1990). Formalization may also reduce transaction costs only for outside investors, potentially fostering land concentration and marginaliz- ing smallholders (Binswanger, Deininger & Feder 1995). Moreover, low levels of income and wealth in many African countries might constrain households to exploit opportunities that arise from land certification (Lawry et al. 2017). As economic performance remains stagnant and food insecurity a prominent issue, the question is particularly pressing for Sub-Saharan Africa.
Tanzania is an interesting case to study the impact of reforming property rights on rural develop- ment. Only about 5% of its land is registered, the lowest rate in the world (World Bank 2019a). The majority of people live in rural areas and with 58% agriculture continues to be the dominant sector in terms of employment (World Bank 2019b). In 2018, half of the population lived below the 1.90 USD per person per day international poverty line and half of them report to have run out of food in the past months (World Bank 2019b). Most of the Tanzanian poor live in rural areas where they operate farms for subsistence needs (National Bureau of Statistics 2019). As land is these citizen’s main asset, moves to alter the legal status of land rights are matters of immediate and vital importance to the lives and livelihoods of some of the world’s poorest populations (Boone 2019). It is evident that agricultural productivity is critical to reduce poverty.
This paper contributes to the literature in several distinctive ways. First, it adds to the research examining the impact of formalizing property rights on agricultural outcomes. It finds that house- holds with a certificate for their land increase maize yields per acre and plant more permanent trees.
Increasing maize yields are shown to be unconditional on the area used for maize cultivation and thus are productivity gains. Both outcomes suggest a more stabilized and diverse food production and the emergence of income-generating opportunities. Households also increase agricultural invest- ments, primarily into fertilizers and pesticides, although the overall evidence is inconclusive and does not suggest that these short-term investments are the main driver. Further examining the channels through which such changes have been hypothesised to work, this study finds that official ownership documentation strongly raises perceived tenure security. It challenges the argument that Sub-Saharan African land reforms have failed to translate into economic growth due to informal tenure arrange- ments working efficiently. Although baseline tenure security is relatively high, the incremental effect from formalization is sizable. In a similar vein, the study shows that documentation greatly raises perceived land value and the propensity to rent out plots as proxies for the functioning of land mar- kets. Of moderate degree, it also adds to the few studies that have identified a positive link between property rights to increase collateral and access to credit. It finds that households with CROs are more likely to take up a loan and that the amount borrowed increases, although insignificantly so.
But credit uptake is significantly positive for male headed households and for low-income households.
Evidence does not suggest that this works through increased commercial lending though, emphasiz- ing that improved land rights not necessarily facilitate formal credit access if smallholder lending is generally considered prohibitive. The improved efficiency of land markets neither appears to induce commercial lenders to perceive land as collateral as of higher value.
Second, this paper adds to the scarce evaluations of how a major Tanzanian land reform, namely the
1999 Village Land Act, has influenced agricultural households. While a number of papers have ana- lyzed its companion reform for urban settings, primarily in the capital city Dar es Salaam, evidence on the rural yet more populous context remains rare. In addition, the impact of the so far biggest implementation project, largely financed by the World Bank, has not been examined yet (World Bank 2019b). By using all five waves of the Tanzania National Panel Survey from 2008 to 2020, this paper is the first not only to make use of the very recent 2020 wave, but also to thoroughly exploit the longitudinal nature of the dataset to measure the impact of changes in household certification status on rural agricultural outcomes. I am not aware of any study in this context that goes beyond a cross-sectional analysis.
Third, this paper adds to the literature that links changes in property rights to female empowerment in developing countries. It finds that female headed households invest more into their land through certification, but considerably less so than male headed households. Still, women experience larger gains in agricultural productivity and profitable tree plantation, reaching or even surpassing male levels. Turning to channels, female headed households do not seem to gain from improved collater- alization at all, a credit channel is non-existent for them but significantly positive for males. But women benefit somewhat more from increased security and see a stronger increase in land value, suggesting that females have a lower bargaining power to successfully negotiate land transfers in in- formal settings and benefit more from official documentation. Perhaps more importantly in absolute numbers of affected women, the analysis shows that certification leads to a large increase in the share of joint land titling. That is, women are more often registered as plot owners together with their spouse, with potentially beneficial effects on gender equality in intra-household decision-making.
Fourth, endogeneity concerns are addressed both by implementing Dynamic Difference-in-Difference estimators with Staggered Treatment Adoption and by instrumenting individual certification status.
The former novel approach has been tested in only a small number of cases so far and has been shown to potentially produce different magnitudes and even different directions of effects. It enables implementation of an event-study methodology and testing of households’ pre-treatment behaviour.
I am not aware that this method has been used for this application so far. The study further uses the leave-out-mean district level certification share as a source of exogenous variation to instrument household’s individual titling status. Both methods largely confirm the main results, but produce an insignificant investment effect, contrasting a robust positive effect based on Two-Way Fixed Effects estimates. They cast doubt on a causal interpretation of findings that link changes in property rights with changes in investments in the presence of endogenous title uptake.
This paper is structured as follows. The first section reviews the literature and outlines the land re- form in Tanzania. Section 2 describes the dataset used and sample selection, Section 3 the empirical approach. Results are discussed in Section 4, followed by a series of robustness checks in Section 5.
The final section offers some reflecting thoughts and concluding remarks.
1 Research Background
1.1 The Role of Property Rights in Economic Development
The academic literature generally advocates a strong link between individual property rights, eco- nomic development and poverty reduction (Field 2007; Acemoglu, Johnson & Robinson 2004). Several empirical studies find that the codification and protection of ownership rights stimulates investments into physical and human capital, and that these factors are used more efficiently to achieve a greater level of income (Acemoglu, Johnson & Robinson 2004). Thus, among policy makers property titling is increasingly considered to be one of the most effective forms of government intervention for targeting poor landowners and encouraging economic growth (Field 2007).
Theoretically, three major channels have been identified. First, there is an emphasis on the link be- tween property rights and freedom from eviction (Besley 1995; Demsetz 1974). It dates back to Adam Smith who observed that cultivator’s fear of expropriation or loss of land on which investments have been made might deter such investment (Goldstein & Udry 2008). According to this security-induced investment demand effect, individuals do not invest if the returns from their investments are seized by others. Assuming rational behaviour, agents invest as long as expected returns surpass associated costs; a lower risk of losing the land translates into lower discount rates, higher present values and subsequently more investments are profitably undertaken. Farmers might invest into better fertilisers or more productive machinery in light of better protection of agricultural outputs. Improved tenure security might also be visible via long-term maximization behaviour, such as leaving land fallow or preventing soil erosion (Garc´ıa Hombrados, Devisscher & Mart´ınez 2015). This can ultimately trans- late into higher yields, higher net returns and superior agricultural productivity (Carter, Wiebe &
Blarel 1989).
Second, economic development and property rights may be linked through credit markets (Besley 1995; Feder 1988). If better rights increase the stock of land that can be collateralised, funding constraints may be eased. This is the supply side effect of titling: better possession rights can lower transaction costs and information asymmetries, raising lenders willingness to issue loans and amounts that can be borrowed. Similarly, property rights can lower equilibrium interest rates and since ratio- nal agents equate marginal returns from investing on land to marginal costs, investment is increased (Besley 1995). Carter, Wiebe & Blarel (1989) argue that it is important to distinguish between de- mand and supply effects, as they have different welfare and policy implications. Whereas the latter collateral channel might be addressed by other actions, such as the formation of borrowing groups, the call for land tenure reform might be more pronounced in the case of tenure insecurity. But Feder
& Nishio (1998) limit that lower interest rates, larger loan amounts and longer term duration are key advantages of formal loans over informal lending arrangements that might already be in place.
In addition, as Binswanger, Deininger & Feder (1995) outline, rural credit markets are difficult to develop and sustain. Given the high transaction costs, the unit costs of borrowing are a declining function of loan size, leading to the exclusion especially of smallholders. Raising interest rates on small loans does not overcome this problem since it also leads to adverse selection. Without collater- alisable capital, high-ability borrowers cannot be screened from low-ability borrowers and the formal lender adversely selects low-ability borrowers with low risk-aversion (Kemper, Klump & Schumacher
2011). Further, formal land ownership in itself can serve as a sign of creditworthiness (Binswanger, Deininger & Feder 1995) and justify institutional intervention.
Besley (1995) suggests a third channel. Formal property rights potentially translate into improved transfer rights that may lower transaction costs in renting or selling. In turn, liquidity of land and efficiency of land factor markets are increased. Particularly outside investors without access to village information networks may be willing to pay a premium for titled land (Jacoby & Minten 2007). But also renting out land creates new sources of income to plot holders if titling provides the security necessary to be willing to lease when there is danger of expropriation by tenants (Deininger, Ali &
Alemu 2011; Jacoby & Minten 2007). Both can resolve temporary allocative inefficiencies by enabling potentially higher-value users of land to offer prices above those of low-value users, ultimately leading to productivity gains (Feder & Nishio 1998). This is in fact one of the central points famously made by Peruvian economist Hernando de Soto. He argues that the major barrier to prosperity and poverty reduction in developing countries is the inability to convert property into usable assets due to a lack of clear and legally recognized rights (Payne, Durand-Lasserve & Rakodi 2009; Do & Iyer 2008). As the most internationally-prominent advocate of land titling, he argues that customary tenure keeps the poor locked out of the market economy (Boone 2019). For de Soto, the process of transforming
”dead capital” into capital through collateralisation is only possible if the government reduces the costs of formal titling (Kerekes & Williamson 2010). Equipped with readily investable capital, pro- ductive projects can be undertaken to promptly increase labour productivity and income (Galiani
& Schargrodsky 2010). On a broader scale, Binswanger, Deininger & Feder (1995) argue that land reform has often been necessary in order to allocate land to smallholders and that the social costs of failing to undertake such reform may extent to peasant revolt or civil war.
Early empirical evidence on the benefits of land titling on investments is provided by Besley (1995).
The author documents that if titling is not endogenous, cultivators who hold more secure rights are more likely to grow trees. Banerjee, Gertler & Ghatak (2002) find that a tenancy reform in India which provided sharecroppers with more secure tenure translated into higher agricultural produc- tivity. More recently, Do & Iyer (2008) exploit different speed of implementing a land reform in Vietnam in 1993 to employ a Difference-in-Difference approach. They find that in provinces with greater progress in land titling the proportion of cultivated area devoted to multiyear plots was in- creased. The authors subsume that the main driving force underlying these changes is increased tenure security. Markussen (2008) provides supportive evidence for Cambodia. The author finds that rural plots held with a document of ownership have higher output in crop agriculture and higher land values. He instruments for the potential endogeneity of property titles with the mode of acquisition of the plot and argues that the main channel of causality is perceived tenure security. Similarly, benefi- cial effects on housing investments in urban contexts have been documented by Field (2007) for Peru or Galiani & Schargrodsky (2010) for Argentina. Both exploit natural experiments in formalizing urban squatting. The former study suggests another channel through which better property rights might raise welfare: improved intra-household labour allocations (Field 2007; de Janvry et al. 2015).
Field (2007) shows that by strengthening formal ownership rights, the burden of maintaining tenure security was shifted from individuals to local institutions, reducing opportunity cost of employment and enabling households to make unconstrained labour decisions. Do & Iyer (2008) corroborate this
for rural Vietnam by finding that official ownership documents significantly increase the time that households spend on non-farm activities.
The majority of these studies argue that enhanced tenure security or better functioning land markets are the most likely channels at work, only few find a credit effect. One of the first is Chalamwong &
Feder (1988) who use cross-sectional data for Thailand. They find that land prices significantly in- crease through legal ownership and attribute this mainly to improvements in credit access rather than perceived security. Kemper, Klump & Schumacher (2011) find that the Vietnam land certification program increased borrowing, but only from formal banks with a collateral-based lending policy. In a review of twenty quantitative and nine qualitative studies, Lawry et al. (2017) conclude that land tenure recognition can provide substantial productivity and income gains, but that the main effect may operate through gains in security. In fact, most studies fail to detect a measurable collateral channel (Panman & Lozano-Gracia 2021; Galiani & Schargrodsky 2010; Kerekes & Williamson 2010;
Do & Iyer 2008; Jacoby & Minten 2007; Torero & Field 2005; Place & Migot-Adholla 1998).
However, theoretical arguments of enhanced investments and welfare from tenure recognition are not unequivocal and a number of papers have only found mixed or even negating results, especially in the case of Sub-Saharan Africa. In his seminal paper, Besley (1995) emphasizes the role of informal arrangements. If individuals care equally about all members of the community and if consumption is shared, individual property rights should not change investment behaviour (Besley 1995). Lanjouw
& Levy (2002) note that a duality of formal and customary systems operating can actually increase ownership uncertainty. Well entrenched informal tenure arrangements have frequently been observed in many African countries (Lawry et al. 2017; Smith 2004; Lanjouw & Levy 2002; Place & Migot- Adholla 1998). If institutional processes replace traditional tenure arrangements, some secondary claims might be extinguished or some groups even prevented from accessing land they previously shared as a community (Atwood 1990). In the worst case, land reform might foster corruption and elite capture, letting powerful accumulate land at the expense of poor farmers. Such economic and institutional distortions can lead to a reduction in terms of efficiency and equity (Feder & Nishio 1998; Binswanger, Deininger & Feder 1995; Carter, Wiebe & Blarel 1989).
With respect to land transactions, Atwood (1990) believes that formal rights are not necessary to enable them. He claims that transfers are widespread in many African tenure systems and that land registration perhaps reduces the risks only faced by an outsider. Further, if potential purchasers tend to see land as an investment with a high potential for appreciation or as a hedge against inflation, rather than as a factor of production, reducing transaction costs might actually lead to less efficient land use (Atwood 1990). This may hold even if land is used for agricultural production. An inverse relationship between average farm size and agricultural productivity has occasionally been observed, particularly in developing countries (Garc´ıa Hombrados, Devisscher & Marti´ınez 2015; Bandiera 2007), as has been that formal land titles can raise land concentration (Binswanger, Deininger &
Feder 1995). Connecting both findings leads to potentially less affirmative efficiency expectations.
Another argument is that land factor markets are likely to be poorly developed in rural areas of agrarian economies, again limiting efficiency gains from freeing up land transactions (Ravallion & van de Walle 2006).
Atwood (1990) also challenges the view that improved property rights translate into better credit ac- cess. In the absence of functioning financial markets, a higher collateral stock might be of little value if agricultural lending is generally prohibitive due to high risks and transaction costs. Income from agricultural production is inherently unstable and increased investment opportunities from property rights reforms might still be unprofitable to a risk-adjusting lender. Similarly, banks might be reluc- tant to accept remote farming plots as a pledge if foreclosure in the case of default appears to be difficult and costly. This is all the more true if property titling does not create a liquid land market or if informal tenure arrangements continue to be predominant. These obstacles are likely to be more difficult to overcome for small-scale farmers (Binswanger, Deininger & Feder 1995).
Early evidence from Kenya, considered the African test case for tenure reform, shows little if any economic impact of land registration (Place & Migot-Adholla 1998; Carter, Wiebe & Blarel 1989). In Madagascar, Jacoby & Minten (2007) compare land-specific investments in titled and untitled plots cultivated by the same household to find that having a title has no significant effect on plot-specific investment and correspondingly little effect on land productivity and land values. This extends to a number of other researched African countries (Holden, Deininger & Ghebru 2009). It has been sug- gested that in Africa 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 main- taining land titles (Deininger, Ali & Alemu 2011; Jacoby & Minten 2007). As a consequence, various efforts to formalize land tenure have been abandoned before ever being implemented (Ali et al. 2014).
Apart from apparently efficient customary tenure arrangements, a potential explanation is that low levels of wealth of African farming families imply further constraints to translate tenure recognition into commercial activity (Lawry et al. 2017). This has changed policymakers’ early perception of land titling as a cost-effective means to boost investments: A World Bank discussion paper notes that in some cases title deeds were not worth the paper they are written on (van den Brink 2003; Smith 2004).
Overall, microeconomic evidence on land tenure recognition indicates substantial productivity and income gains, but the results largely differ by region (Lawry et al. 2017). In their literature review, Lawry et al. (2017) summarize that reported effects are strong in South America and Asia, but that evidence on Africa is much weaker. In a review on land tenure and agricultural productivity in Africa, Place (2009) documents that the empirical findings are at best mixed. The author finds that land titling increases only some types of investments and some of those only have very low impacts.
Further, he only finds a weak relationship between tenure security and productivity gains. The au- thor reviews a number of studies that do not find any positive impact from changes in land rights (Place 2009). Deininger, Ali & Alemu (2011) subsume that the history of land titling in Africa is one of failure rather than success. In the case of Sub-Saharan Africa, the question of whether property rights can stimulate economic growth appears to be open but is particularly pressing, as economic performance remains stagnant and food insecurity a prominent issue.
1.2 Land Reform in Tanzania
In Tanzania, in 1999 the Village Land Act was passed, reforming regulation of land in rural areas.
The goal of the government was not to redistribute land but to increase agricultural productivity
and economic growth by securing tenure, providing greater credit access and improving land mar- kets (Pedersen 2010). It was mainly a response to rapid population growth and rising investment in commercial agriculture that increased land scarcity and conflicts over land (Schreiber 2017; Pedersen 2010). Low registration of rights made these disputes difficult to solve. The law requires community lands to be surveyed and titled with a Certificate of Village Land. Before issuance of individual title deeds can take place, village registries and district land registers are further requirements. Since the 2007 Land Use Planning Act it is also necessary that villages have land use plans in place. Following, village councils are enabled to provide individual rights by issuing Certificates of Rights of Occupancy (CRO). The latter has been seen as a centerpiece of the 1999 reform (Biddulph 2018). Although tech- nically all land belongs to the President of Tanzania, CROs are granted in perpetuity. The certificate allows the holder to transfer and mortgage it. Formally, there are two types of certificates: granted CROs and customary CROs. The former is issued to residents occupying general land, whereas the latter to individuals holding customary rights in village land and to others who have used specific village lands for at least 12 years. According to the Village Land Act, both types are in every respect of equal status and effect (Boone 2019).
Essential for the empirical analysis is that individuals need to apply to obtain a certificate. Although information on time and costs to obtain a title are inconclusive and heterogeneous across regions and time, the procedure to secure a CRO is generally considered to have been costly and time consuming (Schreiber 2017; Collin, Zandefur & Zeitlin 2015; Ali et al. 2014). Fairley (2013) outlines that an application is submitted to the village council which must decide on granting a title within 90 days.
This aligns with the 77 days the World Bank reports as the duration to register a CRO in 2005 (World Bank 2013). The applicant then has another 90 days to accept and to pay for the title before it is reg- istered and issued (Fairley 2013). To give an indication regarding costs, Fairley (2013) refers to about 10 USD per CRO based on interviews conducted in Tanzania in 2009 and 2010. Schreiber (2017) estimates that titling projects between 2008 and 2017 ranged from 9 USD to around 47 USD in costs per title, whereas Ali et al. (2014) report an average price for a formal land title of approximately 64 USD in urban Dar es Salaam in 2012. For reference, at that time households’ mean monthly income was about 183 USD (292,500 TSZ, National Bureau of Statistics 2013a).1 Conducting an experiment, Ali et al. (2014) find that the willingness to pay for land titles is very high with on average between 40 USD and 50 USD. This lets Deininger, Hilhorst & Songwe (2014) claim that in Tanzania even poor households are willing to pay for formal rights.
In comparison to most other reforms in Sub-Saharan Africa, the case of Tanzania is considered to be quite radial in its far-reaching protection of land rights and decentralized allocation of power, thereby building on existing institutions (Peterman 2011). The act came into effect in 2001 but implemen- tation was initially slow and fragmented. Pedersen (2012) describes that the decentralized approach of the reform made implementation too complex for villages to handle on their own but at the same time was out of reach of the responsible ministry. Insufficient funding to survey and register land nationwide was another part of the story (Pedersen 2012). Implementation in the following years was thus largely project-driven, often financed by international donors and NGOs. The first significant moves came in 2003 after Hernando de Soto addressed Tanzania’s cabinet on the importance of land
1Income based on Household Budget Survey 2011/2012. Average exchange rate in 2011 was 1:1600.
and property ownership (Schreiber 2017). De Soto was invited by the Tanzanian president to help establish a program that aimed at imparting formal land titles and business registration to the poor (Ali et al. 2014). The first title deeds in villages were not issued until during a pilot project in Mbozi District in 2004 (Pedersen 2010). By far the largest implementation project since the passing of the land acts was carried out by the Ministry for Lands and Human Settlements Development (Pedersen 2010). It started in 2006, but activities in rural areas did not begin until a couple of years later.
It was funded jointly by a credit from the World Bank and by the Government of the Netherlands, amongst others. Implementation of the land reform was among its main activities.
The World Bank project primarily aimed at creating sustainable conditions for enterprise creation and growth as part of the government’s BEST program (Business Environment Strengthening for Tanzania). A major component was to deepen the financial sector and the land reform played a key role in this. At that time, the Tanzanian government issued the target to become a middle-income country by 2025 and to transform the agriculture-based economy into a competitive semi-industrial economy (World Bank 2005). In its project appraisal the bank notes that in 2005, Tanzania ranked 140 of 155 countries in terms of ease of doing business and that the de Soto diagnostic study reported that valuation, planning, surveying and title procedures can take up to eight years (World Bank 2005). In 2013, the project was restructured and received additional funding (World Bank 2013).
Focus on implementing the land reform increased and the number of days to complete a CRO regis- tration was now a key indicator. It was targeted to be lowered to 40 by the end of the project. By then a ten times less expensive systematic approach relative to the existing demand approach was successfully tested, and some 100,000 CROs were issued. In addition to the theoretical expectations outlined above, it was motivated by the view that better land policies would create jobs and food security in the agricultural sector and give industrial sectors better credit access. Individuals would become more mobile to exploit job and other opportunities and most vulnerable groups would be protected, maintaining social stability in Tanzania (World Bank 2013). The project was extended several times but terminated in 2018. In its final report, the World Bank (2019a) summarizes that an integrated land management information system was implemented, geodetic infrastructure was over- hauled, six land related laws were prepared and enacted, awareness was raised, 22 district housing and land tribunals as well as six zonal land offices were established, covering the whole country. The boundaries of 11,000 of around 12,000 villages in Tanzania were surveyed, of which more than 7,000 were registered and 1,000 villages were covered with land use plans. The time to issue a title deed was reduced to 37 in 2017. In a beneficiary survey, the bank documents rather positive outcomes (World Bank 2019a). Over 70% of respondents find it easier to acquire land in 2018 than ten years ago, primarily due to a decrease in the price and time for getting a title, increased awareness, as well as less corruption. Yet, an impact assessment is still to be carried out (World Bank 2019a). Apart from this, findings from interview-based evidence is inconclusive. Fairley (2013) summarizes that ”CROs are certainly useless as collateral in the large majority of cases”. In contrast, Schreiber (2017) cites:
”The banks now saw villagers as potential customers, and pension funds wanted farmers with titles to join their funds”. Ultimately, the matter needs to be settled quantitatively.
Empirical evidence for the case of Tanzania is scarce, especially in rural areas. A number of papers
have recently examined land titling in urban contexts. Panman & Lozano-Gracia (2021) investigate the impact of formal titles on real estate premia in urban Dar es Salaam. They use a hedonic price model to measure housing quality and do not observe a market premium for holding formal tenure documentation, subsuming that local institutional frameworks for land and housing markets work sufficiently well. Ali et al. (2014) are the first to provide experimental evidence on land titling in Dar es Salaam and women’s access to property. They report that formal land title costs were 64 USD at baseline, relative to a monthly median income of sample households of 200 USD. Using price discounts, estimated price elasticities show that costs are the major obstacle to a broader inclusion in the land registry. Demand significantly increases through price vouchers. Further, conditionality to register women as joint landowners does not depress demand. With small price incentives the authors show that almost gender parity in certification can be achieved (Ali et al. 2014). Collin, Sandefur
& Zeitlin (2015) exploit missing satellite photos as a natural experiment in unplanned settlements in Dar es Salaam. Through a regression discontinuity design the authors compare titled with untitled residents, all of which were initially eligible for formalizing leasehold titles. They detect significant positive effects on housing investments, indicative evidence for increased tenure security and reduc- tions and land sales, but no evidence on improved access to credit markets.
Covering rural Tanzania, Peterman (2011) uses the Kagera Health and Development Survey to draw longitudinal inferences about women’s access to property rights. She finds that changes in inheri- tance rights significantly contribute to a female seeking formal employment or self-employment and to raising average wages. She also notes that strong customary laws in rural areas make the case for female empowerment more difficult, as knowledge about land and property rights is lacking. Garc´ıa Hombrados, Devisscher & Mart´ınez (2015) use the first wave of the Tanzania National Panel Survey (NPS) from 2008 to examine the impact of land titles on agricultural outcomes. They do not find any significant impact on production, investments or long-term behaviour and neither regarding credit or perceived tenure security. But their analysis is cross-sectional and their definition of titled households broader than what was affected by the land reform, namely CROs only. Stein et al. (2016) conducted 1,500 household interviews that they econometrically analyze. They use simple OLS to find an in- significant effect of CRO ownership on the probability to have a loan. The authors claim exogeneity of this variable by arguing that people typically do not self-select into the group of CRO holders, since virtually all titling is initiated exogenously by NGOs and government departments (Stein et al.
2016). The authors’ insights alleviate endogeneity concerns as interviews were conducted in a number of different regions in Tanzania. Still, I assume a more conservative stance on the distribution of title ownership. Kassa (2018) uses the second wave of the NPS in 2010 to identify investment behaviour of rural households based on titling status. The author uses inheritance status of a plot as an instrument to find that a title raises permanent tree plantation only if household heads are in favour of local village governance. Aikaeli & Markussen (2017) make use of the third round of the NPS to investigate the effects of private property rights on agricultural investment, land valuation and access to credit.
The authors find a considerable increase in the value of land but not for agricultural investments.
Yet again, their approach is cross-sectional. Besides, the authors consider any kind of ownership document and distinguish four categories of which none explicitly covers CROs. I am not aware of any study that examines progress of the land reform by making use of the longitudinal nature of the
Tanzania National Panel Survey, particularly not of more recent waves. The fifth wave was published during the time of writing.
2 Data
Table 1: Overview of Tanzania National Panel Survey
Wave 1 2 3 4 5
Year 2008 2010 2012 2014 2020
Households 3,265 3,924 5,010 989 1,184
Rural 63.2% 67.0% 64.3% 57.5% 57.6%
Rural Landowners 59.5% 56.4% 54.1% 48.1% 55.4%
Households with CRO 2.4% 3.0% 3.2% 7.0% 15.0%
Attrition 3.0% 3.5% 9.2%
Note: Table summarizes the nationally representative dataset used. In 2014, the sample was refreshed and only a random subsample was reinterviewed. The in- creasing number of households is only due to individuals splitting off. Split-offs are matched with their original household. By reassinging split-offs, the more than 5,000 households in the third wave are observed for three waves, roughly one thousand of which are also observed for the fourth and fifth wave, yielding more than 17,000 observations. The sample of analysis constitutes of rural landowners, in total 9,300 observations.
The dataset used for the analysis is the Tanzania National Panel Survey (NPS). It is a series of nationally representative household panel surveys that collect information on a wide range of topics in- cluding consumption expenditures, income generating activities and a wealth of other socio-economic characteristics (National Bureau of Statistics 2016). It is part of the Living Standards Measurement Survey with Integrated Survey on Agriculture, which supports governments of African countries in generating nationally representative household panel data with a strong focus on agriculture and rural development (National Bureau of Statistics 2009b). Specifically, it includes a separate questionnaire on agricultural production that collects detailed information on plot level characteristics, production choices and crop sales. The individual waves are being made available by the World Bank. So far, five rounds of data have been collected, as depicted in Table 1. For the first wave in 2008, a sample of 3,265 households has been surveyed, of which almost 60% are rural landowners. All of the households have been re-interviewed for the second and third wave in 2010 and 2012, respectively. In 2014 the sample was refreshed and only a nationally representative randomized sub-sample was selected to continue as part of an extended panel (National Bureau of Statistics 2016). As the previous waves, it covers all major regions of Tanzania. These extended households have been re-interviewed for the fifth wave in 2019/2020.
The increasing sample size is purely due to original households splitting-off, building at least two separate observations. A question is how to treat those split-off households. They technically only exist from the period of emergence but as they can be traced back to an original household, adding such data points provides more information. I decided to follow the latter approach which is also the one undertaken in a uniform panel set generated by the World Bank. However, since this panel does not cover the most recent wave the dataset used here is an adapted version.2 By reassigning split-offs,
2The fifth wave was merged with the uniform panel to obtain household identifiers. These were then merged with all relevant data sections of the respective waves. Data sections of the agricultural questionnaire distinguish between
the more than 5,000 households in the third wave are observed for three waves, roughly one thousand of which are also observed for the fourth and fifth round, yielding more than 17,000 observations. My analysis focuses on rural farmers, thus I only consider the fraction of the sample of rural plot owners.
I further abstract from families moving to urban areas. Families moving to another rural area are still included. However, it might be that an initially rural area transforms into an urban one over time, in its strongest form a general equilibrium effect. I consider this possibility by checking whether households are identified as being located in their original dwelling. Only a minor number of families stays in the same location but is classified as residing in an urban location in later survey rounds.
The results are not sensitive to in- or excluding these households. To avoid that large-scale farmers are driving the results, I checked for outliers both by plotting histograms and by capping observations roughly at the 99.50% interval based on farm size, total crop yield and total land value. Yet, none of these observations are driving results. Another question is whether to include households from Zanzibar in the analysis. Zanzibar is an island in the Indian Ocean in proximity to Dar es Salaam, the capital of Tanzania. I compared demographic and agricultural characteristics of households living on Zanzibar or mainland Tanzania and did not find significant differences. Thus, rural farmers on Zanzibar are included. Restricting the sample to mainland Tanzania produces identical results, as is shown in Section 5.
Total household attrition up to the third round is 4.84%, minimizing the potential for attrition bias within the data waves (National Bureau of Statistics 2014). In absolute numbers, 97 households drop out after the first round, 71 after the second, 81 after the fourth. There is no information on the response rate of the random subsample selected in 2014. As long as attrition is random it is of no concern, but selective attrition might bias estimates. I test for systematic attrition by regressing a probit model of baseline characteristics on the propensity to drop out. Table A1 indicates that smaller households might be more likely to not be reinterviewed. I use inverse probability weighting to reweight observations. However, results are insensitive to the inclusion of correction for attrition bias. The predictive power of baseline characteristics on attrition is small and the correlation between CRO share and attrition is zero.
Figure 1 depicts the share of households owning at least one CRO throughout the survey waves. As can be seen, the share of households with a title amounts to just 2.4% in 2008. Almost a decade after the 1999 Land Act, it shows that a change in de-jure rights not necessarily translates into changing behaviour without appropriate enforcement. Although the World Bank project was already under way, the share increased only marginally until 2012. It was not before 2014 that uptake considerably increased to 7.0%, largely keeping pace to 15.0% in 2020. This implies sufficient variation in the share of certified cultivators that is likely to extrapolate into the future. Although the World Bank project terminated in 2018, shortly after assuming office, President John Magufuli announced ambitious plans to register land parcels, develop land-use plans, and to issue many more property titles (Schreiber 2017).
short- and long-rainy season. Sections were combined to get a comprehensive overview of agricultural production. All datasets were downloaded from microdata.worldbank.org on 30/04/2021. Specifically, they include all five waves of the NPS, as well as the uniform panel dataset 2008-2015. Detailed information can be found under National Bureau of Statistics in the references.
Figure 1: Households with CRO, 2008 to 2020
Note: Graph plots share of sample households with a Certificate of Rights of Occupancy (CRO) for at least one of their owned plots.
3 Empirical Approach
The availability of panel data enables analyzing changes in plot titling status and agricultural decision- making over time, thereby controlling for constant idiosyncratic household characteristics and com- mon time trends. To optimally make use of the repeated household- and plot-level data I have, I employ a two-way fixed effects linear regression (TWFE) as follows:
Yi,t= β1CROi,t+X
k
βk+1Xk,i,t+ αi+ λt+ εi,t (1)
The outcome variables Y for household i in period t comprise of three agricultural outcomes, as well as the channels through which potential effects are likely to occur. The coefficient of interest is β1, it captures the effect of a household i in period t having at least one certificate of Rights of Occupancy (CRO). The agricultural questionnaire contains a question whether a household possesses either a Granted or a Customary Right of Occupancy. As noted above, both can be considered to be equal, as they are treated in the questionnaire. Households who do not answer this question are unlikely to have a title and thus treated as not having one. Since individual plots are titled, I define CROi,t to be the share of the total area a household owns that is titled, it thus ranges from 0 to 1. The vast majority of households, however, is either fully or not titled at all. Household fixed effects αi control for any time-invariant household or plot specific omitted variables. Similarly, λt controls for any characteristics that change over time but that are common to all households or plots. Vector Xk,i,t
further controls for time-varying household characteristics. Standard errors are on the household level and robust standard errors are employed throughout.
Specifically, I measure maize yield per acre, the number of permanent trees planted in the past 12 months per acre and the logged amount of investments per acre as agricultural outcomes. The yield is measured as harvested quantity relative to farm size. Maize is the most productive food crop in Tan- zania, and one of the top export products to African countries (World Bank 2019b, World Bank 2005).
Trees are both profitable in their own right and enhance nutrient recycling, conserve soil moisture, maintain fertility, and reduce soil erosion (Bellemare 2013; Bandiera 2007). Tree production reflects a form of long-term productive investments and can be a critical step in the transition from shifting to stabilized cultivation and food consumption security (Lawry et al. 2017; Holden, Deininger & Ghebru 2009; Besley 1995). With 0.39 in 2018, there is a clear positive correlation between GDP per capita and the share of tree coverage of agricultural land in Sub-Saharan Africa.3 The investments measure is a composite of the value of used fertilizer, pesticides or herbicides, and seed purchases in the past agricultural season. To control for potentially changing land under business over time, all outcomes are measured relative to the total size of the farm. I am also interested in the channels through which having a CRO might impact agricultural business decisions. As discussed above, having formal doc- uments should improve tenure security which might translate into enhanced investments. I proxy for this by using a question that asks respondents whether they would feel comfortable in leaving a plot uncultivated for several months without being worried of losing it. This measure is plot-specific and thus related to the number of plots a household owns, consequently ranging from 0 to 1. I measure access to credit by indicating whether a household has taken up a loan in the past twelve months.
Finally, official documents might facilitate the emergence of a formal land transaction market. I do not have information on land transactions specifically, but proxy for this by examining how having a CRO changes plot owners’ perceived value of their plot. It has been suggested as an interesting measure in itself (Lanjouw & Levy 2002; Besley 1995). Landowner’s likelihood to rent out their land is also an adequate indication of the functioning of land markets and tested as an alternative outcome in the Appendix. This has been used by, amongst others, Deininger, Ali & Alemu (2011) and Lanjouw
& Levy (2002).
Control variables consist of time-varying household-specific characteristics, namely a household’s size and its assets, and plot-specific controls. Assets specifically do not contain any farming related im- plements and count the number of certain durable goods a household possesses; information on their value is not available. Plot related variables comprise of changes in soil quality, slope and distance from dwelling. While soil quality can change both on an individual plot and as a composite, the latter two mostly account for the fact that landholdings of a household might change over time. Soil quality is a measure between 1 and 3, indicating bad, average or good quality; slope ranges from 1 to 4, indicating flat bottom, flat top, slightly sloped or very steep; and distance is the distance of an individual plot to the family’s home measured in kilometres. Bivariate correlations (Table A2) indicate a potentially harmful multicollinearity between changes in composite slope and soil quality of a household’s landholdings. I reran regressions leaving out either one without changing conclusions.
Several statistical issues might arise in estimating Equation 1. First, there may be omitted vari- able bias. By exploiting the panel nature of the data, fixed effects absorb unobserved heterogeneity that is common to all households or constant over time. For instance, more entrepreneurial house- holds might be more likely to title their plots but also have higher ability and skills in agricultural production, creating a correlation between the CRO share and the error term that is eliminated by including household fixed effects. But control variables are added in alternative specifications to allow
3Adapted from Bandiera (2007) who finds a correlation of even 0.94 for Central America in 1998. Land use data are from FAOSTAT, GDP information from World Bank Development Indicators.
for the possibility that some household characteristics are time-varying. More problematic would be time-varying unobserved heterogeneity idiosyncratic to individual plots as the unit of analysis is the household. But there are several reasons why the latter is preferred over a plot-level analysis. First, the mean share of titled land of CRO owners is 78%. Thus, I am fairly confident to say that certifica- tion is a household rather than plot specific decision. In other words, variation from comparing titled to untitled plots within the same household is of limited insight for the given data, as the number of such observations is small. Moreover, vital part of this analysis is to measure access to credit which affects households as a whole but not individual plots. There is no reason to find a link between land rights on a particular field and investment on that plot as farmers can collateralize any of their land holdings to invest in another (Besley 1995). As such, Carter & Olinto (2003) designate a household being titled is at least some of its land is held with mortgageable property rights. To be sure about plot-specific time-varying characteristics, plot-level controls are added as described above. Adding such controls further facilitates the fact that some households change their land holdings over time.
Households which purchase already titled land might be problematic as increasing land holdings may induce disproportionally high investments that are not fully related to the act of certification. I test for this in Section 5 by excluding households that have acquired land and find that the results are robust. Yet another issue can be that households that secure tenure might be different to untitled households along dimensions that are unobserved and not accounted for by fixed effects. One way to deal with this is to restrict the sample to households that have a positive share of titled land any- time throughout the panel and to only exploit variation in timing or magnitude of household-specific titling. This is also done in the robustness checks. The results are again unchanged.
Another potential problem might be measurement error which would attenuate the effect of having a title. But given the considerable effort and costs associated with obtaining a CRO, it is unlikely that households misclassify having an official title or not. In addition, limited adoption of CROs in the years initially after the reform jointly with several information campaigns suggest low awareness of the benefits of titles and does not give any reason why households would either hide or claim having a CRO if the opposite was true.
Most importantly, as outlined above, applying for obtaining a CRO was an endogenous decision, raising the concern of reversed causality. This appears to be be most logical in terms of investment behaviour and plot value. It might be that households obtain a title because they invested more into their land, rather than investing more due to having a title. Most plausibly, title uptake is then pos- itively determined by investments, creating an upward bias in the estimates. Similarly, idiosyncratic increases in plot values could induce households to certify their land. TWFE associates certifica- tion and investment behaviour within a household and reduces the reversed causality problem, but does not fully solve it (Verbeek 2004). For illustration, Jacoby & Minten (2007) aim to deal with endogenous take-up of land titles by comparing titled to untitled plots within the same household, thereby employing a household fixed effect. This procedure takes account of a selection effect but the concern of reversed causality is not resolved. Analogously, Banerjee, Gertler & Ghatak (2002) outline that registration to benefit from a tenancy reform in Indian districts was ultimately a choice of the tenant. They argue that as long as individual characteristics are constant over time, they should not be a problem if district fixed effects are allowed for. Although the potential reversed causality
bias in the absence of exogenous treatment in this type of study is well known, few studies have controlled for the endogeneity of land rights (Holden, Deininger & Ghebru 2011; Brasselle, Gaspart
& Platteau 2002; Besley 1995). The key identifying question here is to what extent interventions are as good as random, conditional on time and group fixed effects (Bretrand, Duflo & Mullainathan 2004). Alternative econometric methods address this question in Section 5.
4 Results
Table 2: Summary Statistics
CRO Share > 0 CRO Share = 0 Difference Mean St. Dev. Mean St. Dev. p-value Household Head
Female 0.23 0.42 0.23 0.42 0.01
Age 49.81 15.85 48.88 16.02 -0.92
Literacy Rate 0.73 0.44 0.66 0.47 -0.07∗∗∗
Attends Village Meeting 0.69 0.46 0.68 0.47 -0.01 Indigenous to Village 0.62 0.49 0.63 0.48 0.01 Household
Household Size 6.42 4.64 6.07 3.63 -0.35∗∗
Assets 0.71 0.58 0.71 2.79 -0.00
Consumption (log) 14.62 0.67 14.51 0.69 -0.11∗∗∗
Part of SACCO 0.06 0.23 0.05 0.22 -0.01
Land Characteristics
Total Area 7.17 10.00 6.71 13.97 -0.47
Number of Plots 2.54 1.40 2.40 1.34 -0.14∗∗∗
Plot Size 3.13 4.69 3.09 6.65 -0.03
Distance from Dwelling 1.15 5.32 1.58 9.80 0.43∗
Soil Quality 1.11 5.57 1.10 5.17 -0.01
Slope 1.06 6.07 1.08 3.83 0.02
Yield per Acre (log) 4.52 2.00 4.58 1.90 0.06
Year possessed 1991 21.56 1988 34.02 -3.01∗∗∗
Observations 1,316 7,984 9,300
* p < 0.10, ** p < 0.05, *** p < 0.01. Note: Table compares characteristics between households that obtain a Certificate of Rights of Occupancy for at least one of their plots and households that do not obtain a certificate. Since CROs are granted for individual plots, CRO share is a continuous variable between 0 and 1. Variables for household heads include their gender (Female; 1 = yes, 0 = no), their age (in years), whether they can read and write (Literacy Rate; 1 = yes, 0 = no), attended village meetings in the past 12 months (1
= yes, 0 = no) and whether they are indigenous to the village they live in (1 = yes, 0 = no).
Household variables include household size, the number of non-farming durable assets, total consumption in the past 12 months in log terms, and whether any household member is part of an informal savings group (SACCO; 1 = yes, 0 = no). Plot variables contain the total area of land holdings (in acres), number of owned plots, their average size (in acres), average distance from household dwelling (in kilometres), average soil quality (1 = bad, 2 = average, 3 = good), average slope (1 = flat bottom, 2 = flat top, 3 = slightly sloped, 4 = very steep), average crop yield per acre in log terms and the average year plots were possessed.
Summary statistics can be found in Table 2. Variables are compared between households that obtain at least one CRO and households that never possess a CRO. The share of female headed households is not different between both groups, 23% of households have a female head. Heads of titled house- holds are slightly older and have a significantly higher literacy rate than heads of untitled households.
Titled households are only marginally more likely to attend village meetings. As outlined above, the decentralized approach of the land reform suggests that farmers who regularly attend village meetings might be more likely to obtain certification. But the question whether the household head attended all or some of the village meetings in the past year only exists for the first two waves and might thus be less indicative. Household heads with certification are also not more or less likely to be born in the village they reside in. But titled households have on average more members. This does not suggest that bigger households with a greater capacity to secure tenure are less inclined to obtain formal documentation as has been hypothesized in urban contexts (Field 2007). Titled and untitled households do not differ in their assets, although this measure has shortcomings as noted above. A better proxy for a family’s wealth and income might be the consumption composite. It consists of household food consumption, health expenditures and utilities, amongst others, but only exists for the first three waves. Although I could replicate the index for the fourth and fifth wave in nomi- nal terms, transformation in real terms requires spatially and temporally disaggregated price indices which are not available. Households with a title consume significantly more, at least until 2012. It is a reasonable assumption that this difference can be extrapolated and points to titled households being wealthier and having a higher income than untitled households. Titled households are with 6%
only slightly more likely to be part of an informal savings or credit group (SACCO). Being part of such a group could suggest that certified households have more experience in borrowing or a higher propensity to expand credit amounts by registering land. Similarly, these households might want to substitute informal with formal lending, in an attempt to increase loan size and lower interest rates. Turning to plot characteristics, titled households possess insignificantly more agricultural land than untitled households. Their total land holdings amount to on average 7.17 acres, relative to 6.71 acres for untitled households. As can be seen, this is not due to larger plots which are being titled but due to more plots in possession. Certified plots are located closer to the family’s dwelling, but not different with reference to soil quality or slope. Interestingly, households with a CRO do not seem to be more productive in terms of aggregate agricultural output per acre. But households with certification occupied their land holdings later than untitled ones, suggesting lower tenure security.
Households who seek certification on average received plots in 1991, considerably later than farmers without any CRO. Most of the differences in variables are taken care of by employing fixed effects, the remainder is part of time-varying controls.
The main results are shown in Table 3. Certification strongly raises productivity in terms of maize yield per acre. Specifically, having fully titled land relative to no CRO increases the yield per acre by around 63%. This effect is robust to adding control variables. One might argue that this is not driven by productivity gains but by merely devoting more land to maize production, with potentially detrimental effects on cash crop cultivation. Table A3 in the Appendix shows that this is not the case. Column (2) reveals that the change in land area from certification that is devoted to maize production is almost zero. Dividing maize yields by the area used for maize production as done in Column (3) produces identical results to Table 3. In addition, the number of planted permanent trees per acre following certification increases by around 55. This is significant on the 10% level, with controls on the 5% level. Slightly more trees are associated with a plot’s distance from a household’s
Table 3: Main Results
Maize Yield per Acre (log)
Tree Seedlings per Acre
Investments per Acre (log)
(1) (2) (3) (4) (5) (6)
CRO Share 0.637∗∗∗ 0.633∗∗∗ 56.38∗ 54.99∗∗ 0.826∗∗∗ 0.825∗∗∗
(0.245) (0.245) (29.52) (27.62) (0.257) (0.258)
Household Size -0.0112 -1.036 0.0333
(0.0184) (2.213) (0.0206)
Assets 0.0028 -0.674 0.0238∗
(0.0131) (1.649) (0.0124)
Quality -0.0006 1.021 0.0371∗∗
(0.0145) (8.174) (0.0188)
Slope 0.0064 7.573 0.0226
(0.0126) (8.972) (0.0191)
Distance -0.0016 3.770 0.0137∗∗
(0.0049) (3.091) (0.0060)
Constant 2.016 2.079 107.5 99.35 4.738 4.431
(0.0088) (0.1160) (1.056) (17.28) (0.0092) (0.1290)
Household FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Adjusted R2 0.00 0.00 0.00 0.01 0.00 0.01
Observations 9,300 9,300 9,300 9,300 9,300 9,300
* p < 0.10, ** p < 0.05, *** p < 0.01; robust standard errors in parentheses. Note: Table shows two-way fixed effects regression estimates based on Equation 1. CRO share is certified land area of a household relative to the total area owned. Maize yield per acre is the harvested quantity of maize relative to the total area owned in log terms (in kg). Tree seedlings per acre refers to number of permanent trees planted in the past 12 months relative to the total area owned.
Investments per acre is a composite of purchased fertilizer, pesticides or herbicides and seeds relative to the total area owned in log terms (in TSZ). Disaggregated results for investments are shown in Table A5. Even columns add time-varying control variables; plot characteristics may change on average due to changes in land holdings.
dwelling and with a plot’s slope. Steeper slopes make processing conventional crops more difficult.
Yet, no time-varying characteristic is a major determinant. Given that households plant on average more than 100 new trees per acre and year, this amounts to an increase of more than 50%. Not only does higher tree plantation point to increased food security and diversity, looking at the kind of trees also suggests higher profitability. As the first column in Table A4 in the Appendix shows, the majority of newly planted trees are permanent cash crops, such as coffee or sugar cane, which sell for higher prices. Changes in household size have a slight negative yet insignificant effect, perhaps due to the lower labour intensity of permanent crops. The coefficient for the alternative outcome is higher since the dependent variable refers to the total number of trees per acre, not recently planted ones.
The findings of increased maize production and tree plantation are not contradictory but mutually reinforcing: Trees prevent soil erosion and annual crops can be planted underneath, increasing both productivity and profitability (Bandiera 2007). Obtaining a CRO also significantly raises the amount households invest into their land. Being fully titled increases investments per acre by more than 82%.
Bigger households tend to invest more into their land. On the one hand, more household members increase the number of dependents and might negatively affect disposable income for investments.
On the other hand, bigger households generally have more manpower and thus benefit from more labour-augmenting investments into physical capital. Assets are positively correlated with agricul- tural investments, as is soil quality. It can be expected that households invest more into more arable land. Surprisingly, households also invest more the farther their plots are from the home dwelling.
Deininger & Jin (2006) argue that if investments can enhance tenure security, then tenure insecurity might induce higher levels of investment. It appears logical that tenure insecurity is positively related to a plot’s distance from the farmer’s dwelling. But none of the components of the investment aggre- gate reflect any sort of long-term or visible investment which might signal ownership. Table A5 in the Appendix disaggregates the investment measure into its individual components. Cultivators pre- dominantly invest more into pesticides and herbicides when having a CRO. This is highly significant and suggests a potential explanation for increased maize yields. Households also spend significantly more on fertilizers. Seeds purchases also increase, yet insignificantly so. Changes in fertilization might imply pronounced scale returns: The UN (2014) notes that the use of technological inputs in Tanzania remains remarkably low compared to other countries: while Tanzania uses an average of 9kg of fertilizer per hectare, Malawi uses 27kg, and China even 279kg.
Looking at the channels through which the increasing agricultural output and investments can be understood, Table 4 shows intermediate outcomes. At baseline, more than 80% of respondents do not feel uncomfortable in leaving a plot fallow for several month without the fear of losing it. This points to informal arrangements working well, as land ownership even without formal certification appears to be quite secure. Despite, being fully titled increases tenure security by more than 6%, which is highly significant and robust to adding controls. As can be expected, bigger households feel more secure. This is in line with the argument that simply more individuals are available to protect land holdings (Field 2007). Also in line with what can be hypothesized, distance of a plot from a household’s home decreases tenure security. Increased tenure security contributes to explaining both the observed rise in investments, as well as the higher number of trees. Security likely is the driving explanation for the latter, as tree plantation requires a long-term planning horizon.
One of the primary goals of the World Bank project was to facilitate credit access by increasing the formality of land ownership. I find that this was partly the case: Being titled raises credit access by more than 3%, which is close to but yet not meeting standard levels of significance. It is still remarkable considering that at baseline only around 8% of households report having taken up a loan in the past year, corresponding to an increase of more than one third. Interesting is to see what borrowers use credits for. As shown in Figure A1, the majority of debtors, namely more than 50% use it to fulfill subsistence needs. Although this is not represented in agricultural outcomes, it can have important welfare implications by enabling households to smooth consumption. Further 12% use it to cover medical costs or pay school fees, respectively. Only around two percent use a credit to purchase land or agricultural machinery, respectively. But 23% use borrowed money to finance other agricultural inputs, providing suggestive evidence of a connection between increased farming investment and improved credit access from certification. This is confirmed econometrically
