Soil Quality in Uganda: Do Transfer Rights Really Matter?

Abstract

This paper uses a unique decade-long panel data with lab-tested soil measures from farmers in Uganda to examine the relationship between transfer rights and soil quality. We find that transfer rights are associated with better soil. This possibly has to do with the ways farmers with transfer rights manage their land compared to farmers without transfer rights. We find strong positive correlation between transfer rights and use of chemical fertilizers. Although we find no strong correlation between transfer rights and agricultural intensification in terms of manure use, the strong positive relationship found between transfer rights and soil quality may suggest that farmers with transfer rights may be investing in soil improvement through other means such as use of compost, crop residues, and fallowing.

Introduction

Economists have long argued that property rights to land that are secure and easily transferable increase investment incentives (Feder and Feeny 1991, Besley 1995, Brasselle et al. 2002, Goldstein and Udry 2008, Abdulai et al. 2011, Deininger et al. 2011, Fenske 2011). In agricultural societies of Sub-Saharan Africa (SSA), tenure security and transferability of land can incentivize farmers to invest in land improvement. This may in turn increase the quality of land and boost agricultural productivity. Moreover, land tenure security and transferability of land induce intensification of farming (Otsuka and Place 2001). Notwithstanding, tenure insecurity¹ and lack of transfer rights remain problems in SSA (Namubiru-Mwaura and Place 2013). This may negatively affect investment in land improvement and agricultural intensification, thus contribute to deterioration of the quality of soil. Soil quality matters in SSA since majority of its people live in rural areas and derive their livelihoods from agriculture, a land based sector.

In many parts of SSA, there are concerns of soil quality exhaustion (Sanchez 2002, Henao and Baanante 2006, Tully et al. 2015, Mugizi and Matsumoto 2020, 2021a). At the same time, the region’s fertilizer usage rate to replenish the soil is very low (FAO 2015). Soil quality deterioration may seriously affect many people who derive their livelihoods from agriculture. Indeed, declining soil fertility is the primary biophysical reason for stagnant per capita food production in SSA (Sanchez et al. 1996, Sanchez 2002, Place et al. 2003, Morris et al. 2007). Conversely, an increase in soil fertility has been associated with increase in yields; Lal (2006) estimates that for every ton/ha increase in soil organic carbon, maize, wheat, and rice yields increase by 30–300, 20–70, and 10–50 kg/ha, respectively, across the developing world.

In this paper, we examine whether transfer rights (ability of farmers to transfer their land without seeking approval) is associated with soil quality. If farmers can transfer their land without seeking for approval, then their right to land is more secure. We provide more details of this variable in sub-section Measures of land rights and agricultural intensification. We also examine the relationship between transfer rights and agricultural intensification by focusing on increased fertilizer use. We use panel data collected from farmers in Uganda in 2003 and 2012. The data contains six lab-tested soil variables–carbon, nitrogen, phosphorus, potassium, calcium, and soil pH. Since none of them can in isolation provide an extensive picture of quality of the soil; we use all of them to construct the soil quality index—details are in sub-section Measure of soil quality. We use Uganda as an ideal case study for three reasons. It used to be one of the countries with most fertile soils in the tropics (Chenery 1960); to-date it is one of the countries with the highest nutrient depletion in Africa (Henao and Baanante 2006), more so, in highly populated areas (Mugizi and Matsumoto 2021a), the areas that used to be potential for agriculture. Second, the country’s current fertilizer use intensity is one of the lowest in SSA (FAO 2015). Third, land in Uganda is increasingly becoming scarce due to population growth, hence, concerns of tenure insecurity including land conflicts are high (Mugizi and Matsumoto 2021b).

Although the link between transfer rights and incentives to invest in land improvement has been studied, our distinctive contribution is the use of comprehensive actual soil information with six soil quality measures. Moreover, these measures are decade-long panel data, thus makes it possible to study soil quality dynamics. Besides, extant studies examine whether transfer rights induce investment in soil management and conservation; none explicitly examine whether transfer rights improve soil quality. We go beyond the existing literature and examine whether such investments translate into improved soil quality. Because soil quality is deteriorating in Uganda (Mugizi and Matsumoto 2021a), it is important to understand whether transfer rights influence soil quality. Our decade-long panel data allows us to use household fixed effects model to control for unobservable household or parcel specific time-invariant characteristics that could bias our estimates. We also include region-year specific effects to control for unobserved region-year specific characteristics that would affect soil quality hence helps to clean our estimates of time trends region bias. We find that transfer rights are positively correlated with soil quality. We also find a strong positive correlation between transfer rights and use of chemical fertilizers. We, however, find no strong correlation between transfer rights and manure use.

Theoretically land rights increase incentives to invest in soil improvement through three main channels (Besley 1995, Brasselle et al. 2002, Goldstein and Udry 2008, Abdulai et al. 2011, Fenske 2011). First, better land rights reduce farmers’ uncertainties about ownership of land. This assurance increase famers’ incentive to invest in land especially long-term investment such as use of manure, crop rotation, fallowing, and mulching. Second, with better land rights or greater land transferability, farmers may relax their credit constraints by using land as collateral to access credit from lenders and use the borrowed funds to invest in land improvements. Third, better land rights reduce transaction costs and increase the incentive to invest on land. The right to transfer land to others, encourages investment on land, as it makes it easier to convert land into liquid assets through selling or renting. Thus, the benefits of the investment made can be realized at any time. By incentivizing farmers to undertake short-term investment in soil management or improvement including use of chemical fertilizer, and long-term investment such as use of manure, mulching, crop residues, compost, crop rotation, fallowing, or other practices that maintain soil fertility, higher levels of transfer rights should improve the soil quality.²

Dozens of studies have examined the link between transfer rights or tenure security and investment in land improvement (Besley 1995, Gavian and Fafchamps 1996, Deininger and Jin 2006, Kabubo-Mariara 2007, Deininger and Ali 2008, Deininger et al. 2008, Goldstein and Udry 2008, Abdulai et al. 2011, Deininger et al. 2011, Fenske 2011, Bambio and Bouayad 2018). Some studies have examined the short-run investment like use of chemical fertilizers and others have looked at the long-run investment such as application of manure.³

With regard to transfer rights and long-term investment, Deininger et al. (2008) found that in Uganda perceived transfer rights had marginal significant impact on the propensity to undertake soil conservation measures. Abdulai et al. (2011) revealed that owner operated with transfer rights are more likely to invest in organic fertilizers in Ghana. Gavian and Fafchamps (1996) unveiled no relationship between manuring and whether local customs allow land sales, but farmers who cultivated both borrowed and owned fields diverted manure toward owned fields in Niger. Similarly, in Ghana farmers with local political power fallow their land over long duration than their counterparts (Goldstein and Udry 2008). In West Africa, Fenske (2011) found strong link between tenure and fallow, but weak link for manure. Jacoby et al. (2002) showed that organic fertilizer use is significantly higher on private plots in China.⁴ Regarding short-term investment, such as use of chemical fertilizers,⁵ Abdulai et al. (2011) found that owner operated with transfer rights were more likely to use mineral fertilizers than those without transfer rights in Ghana. In contrast, Fenske (2011) found a weak link between tenure and chemical fertilizer uses in West Africa. Similarly, in China Jacoby et al. (2002) showed that chemical fertilizer use is not significantly higher on private plots.

Notwithstanding some few exceptions, the documented positive association between transfer rights and investment in land improvement suggests that transfer rights may affect soil quality. Surprisingly, however, none of the existing studies examines whether transfer rights affect the quality of soil. This remains an empirical question of policy relevance in SSA where soil fertility exhaustion is one of the major reasons for low agricultural productivity. We fill the existing void in the literature by examining whether transfer rights are associated with soil quality. We hypothesize that parcels perceived to have transfer rights have better soil than those perceived to have no transfer rights. We also examine how transfer rights affect agricultural intensification. As discussed earlier, one of the pathways through which transfer rights may increase the soil quality is by incentivizing owners to invest in land improvement. We, therefore, expect to find a positive correlation between transfer rights and agricultural intensification.

Against this backdrop, the remainder of this paper is organized as follows. Section Property rights in Uganda describes property rights in Uganda, highlighting the country’s tenure systems in relation to land rights. Section Data and descriptive statistics describes the data. Section Estimation strategies lays out the estimation strategy. Section Estimation results presents and discusses the results. Section Conclusion and policy implications concludes and draws out implications for policy.

Property Rights in Uganda

Land rights in Ugandan context can be understood by looking at the four land tenure systems—mailo, leasehold, freehold, and customary that exist in the country. These have different levels of land rights and tenure security. Mailo land tenure system is dominant in central Uganda. It was introduced by the 1900 Buganda Agreement—an agreement between Great Britain and the Kingdom of Buganda that was signed during colonial era whereby about 19,600sq. miles of land were given to some individuals to own in perpetuity. The 19,600sq. miles were divided into mile blocks (mailo) and awarded to Buganda king, chiefs and other officials in Buganda kingdom (West 1965). Local peasants who were previously on the land did not get any share. Consequently, they became tenants on land and had to pay rent to the landlords (Mwesigye and Barungi 2021).

Under mailo system, landlords own certificates of title and have full rights while tenants have usufruct rights. This tenure is also characterized by overlapping rights. This is because landlords own titles but because tenants have been on the land for long time, they consider it theirs (Mwesigye and Barungi 2021). Consequently, this landlord-tenant relationship has been one of the sources of land conflicts (Deininger and Castagnini 2006). Thanks to the 1998 Land Act—it contains some steps to increase tenure security of tenants and grants some protection to tenants. As per the Act, tenants that have been on land for 12 years cannot be removed without compensation. Also, if the landlords want to sell land, priority should be given to current tenants. Moreover, with the consent of the registered owner, mailo tenants can apply for certificate of occupancy that includes rights to give, lease, mortgage or inherit land (Deininger et al. 2008).

Customary land tenure is a system where ownership and disposition of land is in accordance with customary rules. In Uganda, there are three different types of customary tenure. In some communities, land is held communally; in others, it belongs to a particular clan; in other communities, individuals hold land (Amone and Lakwo 2014).⁶ Therefore, in communities where land rights are purely communal, transfer rights are restricted. In communities where individual land rights are weak, individuals are required to seek approval of clan heads or extended family members before transferring land. In communities where the land rights are more privatized, individuals have rights to sell, give or bequeath land without seeking consent from the extended family or clan members (Mwesigye and Barungi 2021).

Freehold tenure system is mainly in western part of the Uganda. Under this system, land is held in perpetuity and the owner is given a certificate of title. This system allows ownership of land indefinitely giving full powers over its use and disposition. Individual farmers enjoy full rights, such as transfer rights, rights to bequeath and give, and the right to use land as collateral; however, only a small amount of land in Uganda falls under this system. Because owners have certificate of titles, tenure security is likely to be high.

Lastly, leasehold refers to the use of land for particular period—owning land on contract. According to Ugandan laws, a person can receive a lease for a period not more than 99 and 49 years, for citizens and non-citizens, respectively. Like freehold, under leasehold system, land titles are granted to owners, hence tenure security is likely to be high. Individual farmers have full transfer rights such as rights to bequeath and give. However, because of the high costs of surveying and demarcation involved in obtaining a land title, this regime also account the smallest share of land in Uganda.

Data and Descriptive Statistics

Data

The main source of data is Research on Poverty, Environment and Agricultural Technologies (RePEAT) project. We complement the RePEAT data with rainfall, temperature, and wind data sourced from the National Aeronautics and Space Administration. The National Graduate Institute for Policy Studies, Japan conducted the RePEAT surveys in collaboration with Makerere University of Uganda. The sample for the RePEAT project largely builds upon and complements a completed research project on policies for improved land management in Uganda which was conducted by International Food Policy Research Institute (IFPRI) and Makerere University between 1999 and 2001 (Pender et al. 2004). The IFPRI research project involved a survey of 107 communities (Local Council 1 s—LC1’s—the lowest administrative unit—usually a single village) selected from two-thirds of the regions in Uganda, including more densely populated and more secure areas in the southwest, central, eastern and parts of northern Uganda and representing seven of the nine major farming systems of the country. Within the study region, communities were selected using a stratified random sample. The stratification based on development domains defined by the different agro-ecological and market access zones, and differences in population density. The RePEAT project was conducted in 29 out of 32 districts and in 94 out of 107 LC1s studied by IFPRI.

The RePEAT questionnaires have detailed information including farm input use and land issues. The RePEAT data consist of five survey rounds: 2003, 2005, 2009, 2012, and 2015. In this paper, we use 2003 and 2012 rounds—the only years with soil data. Except for the 2015, the remaining rounds covered three regions—East, Central, and West. The Northern region was not surveyed due to insecurity concerns. During the first round, ten households were randomly selected from each of 94 communities, making a total of 940 households (Yamano et al. 2004). The fourth survey round conducted in 2012 successfully traced 779 households of the original households. This gives us an attrition of 17%. To examine whether this attrition is random, we estimate a Probit model of 2003–2012 attrition on a number of 2003 household characteristics. In column 1 of Table 8, we present the results of this attrition model. We return to this issue in Section Estimation strategies, under estimation of Eq. (3).

Along with the 2003 and 2012 surveys, soil samples were collected from largest maize plot and in very few cases from non-maize cereal plot if the household did not cultivate maize. No sample was collected if the household did not cultivate any of these crops—only 16% and 9% did not cultivate the mentioned crops in 2003 and 2012, respectively. In both rounds, the soil samples were: taken from the same plots,⁷ collected at the same time of agricultural season, which is March through July, and collected before application of fertilizer. To account for intra-field variability, the samples were collected at a deepness of 0–20 cm from five different positions within each plot and properly mixed (Yamano and Kijima 2010). Then, they were analyzed in the soil laboratory at the World Agroforestry Centre in Nairobi. In testing the samples near-infrared reflectance spectroscopy, a method appropriate for large sample sizes was used (Shepherd and Walsh 2002). Prior to testing, the soils were air-dried and ground to pass through a 2 mm sieve, after which they were stored in paper bags at a reasonable room temperature. Then, for a selection of soil samples (20% of the total soil samples), soil properties were measured and calibrated to soil reflectance. The resultant calibrations and soil reflectance were used to predict the soil properties for the entire soil samples. After calibrations, an evaluation of prediction performance on predictive and actual observations was done using the coefficient of R² and root mean square error. Both tests revealed that the method had high level of prediction accuracy. However, detailed information on the mentioned procedures is only available in wave 1 and not in wave 2. In wave 2, what we know is that soil samples were analyzed by calibrations, but not the detailed procedures. Therefore, it is possible that soil analysis methods slightly differed across rounds. However, even if this may be the case, as long as the style of analysis changed wave 1 to wave 2 in the same way for everyone, not correlated with land rights, it should not bias our results (Mugizi and Matsumoto 2021a).

Because the soil samples were only collected from maize or non-maize cereal crops together with the fact that some got spoiled before they were analyzed (Matsumoto and Yamano 2009), we have only 559 households with soil samples in the 2003 data. In 2012, we have samples from 741 households.⁸ After the cleaning process, we remain with a balanced panel data of 409 households.⁹ However, there is some missing information on parcel identification for some households in 2003,¹⁰ making it difficult to know for those households the exact parcel from which soil samples were collected. Therefore, we restrict the analysis to a subsample of households with the same soil parcel in both rounds. This gives us a balanced panel data of 313 households out of the original of 940 households—in Section Estimation strategies, we explain how we deal with this attrition when estimating Eq. (2). We use this balanced soil sample data in the main analysis—regressions that include soil quality. For regressions that do not include soil variables, we use the full sample.

We supplement the RePEAT data with rainfall (mm), temperature (degrees C) and wind speed 10 m (m/s) data sourced from the National Aeronautics and Space Administration-Prediction of Worldwide Energy Resources (NASA-POWER). NASA reports daily averaged air temperature (degrees C) at 2 m above the earth surface, daily wind speed at 10 m (m/s) above the earth surface, and average precipitation (mm/day). By specifying the Geographical Positioning System (GPS) coordinates, we downloaded the data. The RePEAT surveys have community-level GPS coordinates. We use the GPS coordinates to merge the two sets of data. Because NASA records daily data, before merging, we generate annual variables for five consecutive years (the survey year and 4 years preceding the survey). Then, we calculate the 5-year average for each of the variables. Because soil is also affected by biophysical factors such as climate, temperature, rainfall, and wind (Jenny 1995, Nkonya et al. 2005) which influence soil nutrient balances and soil quality in various ways,¹¹ we control for these agroclimatology data in most of the regressions.

Measure of Soil Quality

To measure soil quality, we use six soil variables contained in our data—carbon, nitrogen, phosphorus, potassium, calcium, and soil pH to create a soil index.¹² We use principal component analysis (PCA) technique. The PCA extracts a linear combination of all the variables, which best describes and transforms them into one index (Gray 2011). It also determines weights intrinsically and assigns them to each indicator by its relative importance. The first principal component, which captures the greatest variation among the set of variables is used as the index. Except soil pH (which is normally distributed), other variables had some outliers; therefore, they were transformed to logarithms before creating the index. Because we are using the same soil data used by Mugizi and Matsumoto 2021a, we follow exactly their approach to construct the soil quality index as follows (Eq. 1):

$$SQI_{it} = \mathop {\sum}\nolimits_{n = 1}^6 {V_n} \left[ {\frac{{z_{itn} - z_n}}{{sd_n}}} \right],$$

(1)

where, SQI_it is the soil quality index of farmland of household i in year t, V_n is the weight of each of the soil variables in the PCA model, and z_itn is the soil variable of the farmland of household i in year t. z_n and sd_n are the sample mean and standard deviation of variable n, respectively. Soil pH has ranges that are problematic—soil pH below 6.6 is acidity and above 7.3 is alkalinity (both not suitable for some crops, although it is very common to find cultivated fields outside of this range). Thus, an index in which soil pH enters as a continuous variable could be problematic.¹³ As a robustness check, we construct two alternative indices by using: five soil variables except soil pH; and, all the six soil variables but soil pH enters as a dummy, i.e., 1 if neutral (soil pH ≥ 6.6 & soil pH ≤ 7.3), and 0 otherwise. In addition, phosphorus and calcium reveal a dramatic increase (Table 1). The observed odd changes over time may have come from a change in the analysis protocol.¹⁴ To further check the robustness of our index, we construct other three alternative indices—the first excludes phosphorous, the second excludes calcium and the third excludes phosphorous, calcium, and soil pH.

Although the soil index is likely to be nonlinear, it is difficult to know the optimal level. Moreover, optimal level is likely to be site and crop specific (Srivastava et al. 2006, Musinguzi et al. 2013). Loveland and Webb (2003) and Musinguzi et al. (2013) document that the desired level of soil carbon content is 2%, regardless of the crop type. Regarding some specific crops, the optimal level of soil carbon content for maize is 1.9–2.2% in, that of soil pH for maize is 6.0–7.0 and for beans it is 6.0–6.5 (Musinguzi et al. 2016). The optimal level of phosphorous amount for corn is 13 mg/kg (Mallarino and Blackmer 1992). Although it may be difficult to establish the optimal level, anecdote evidence suggests that given the rate of soil degradation and low fertilizer usage in SSA, soil fertility is far less than the optimal. In Fig. 1, we show how the soil quality index is positively related to carbon content—the proxy for soil fertility (Marenya and Barrett 2009, Yamano and Kijima 2010). This is suggestive evidence that our index is a good measure of soil quality.

Fig. 1

Relationship between soil fertility index and carbon content in the soil

As a second suggestive evidence that our index captures soil fertility, in Fig. 2 we plot the relationship between soil fertility index or carbon content and the share of land allocated to various crops. We see that land allocated for cultivation of banana is high when soil fertility index or carbon is high; land allocated for other crops (cassava, sorghum, and potatoes) is high when soil quality index or carbon is low. The latter are drought resistant crops and grow relatively well on poor soils. We, however, acknowledge that creating an index for soil fertility, more so, in panel data setting is difficult (Michelson et al. 2013), and any one index is surely imperfect. That is why for robustness check, we re-estimate the regressions by using several soil indices and by separately using the six soil properties.

Fig. 2

Relationship between soil fertility index or carbon content and share of land allocated to various crops

Measures of Land Rights and Agricultural Intensification

Another key variable worth discussing is measure of land rights. In this study, we focus on transfer rights. The household has full transfer rights over a parcel if it can sell the parcel without having to seek approval or consent from extended family members, clan or local authority; else, it has partial or no transfer rights (Bellemare 2013, Besley 1995). This variable is created from the following survey question: Do you have a right to sell this parcel?¹⁵ Although our transfer rights variable is perception of the household, and these rights may not necessarily correspond exactly to the rights that farmers care about in planning their investment, examining these self-reported transfer rights may shed some light. Notwithstanding, there are three concerns regarding our perceived transfer rights variable. First, beliefs on transfer rights take a long time to form and are unlikely to change much once formed. However, in a decade-long period, it is very likely that these beliefs may have changed. Indeed, it is this change of the perceptions that may have implications on household’s investment decisions on land improvement. Second, is whether this perception on transfer rights is short-run or long-run. For example, if the perception is short-run and in the long-run households do not think they will still have the right; a household may appear in the dataset perceiving full transfer rights but not making investment that will be good for the long-run. Unfortunately, in our data, it is difficult to know whether the perception on transfer rights is short-run or long-run. Third, is whether the same person answered the question in both 2003 and 2012 surveys. If the same question is asked of a different individual, their perception may be different, leading to measurement error. Indeed, as Table 1 shows, for some few households the person who answered the question may not necessarily be the same in two rounds. We see that household characteristics such as headship, changed in two rounds (Table 1). However, even if the person who answered the question in 2003 and 2012 may have changed for some few households, this should not be a problem since it is this change of the perceptions that may have implications on household’s investment decisions. Moreover, the person who answered the questions in both rounds is the household head (whether he/she changed or not), who is also a decision maker on investment in land management.

With regard to agricultural intensification, we focus on maize farming. Maize is one of the crops grown by many farmers in Uganda—76% and 85% of the surveyed households produced maize in 2003 and in 2012, respectively.¹⁶ Moreover, the soil samples were collected mainly from the maize parcels. We measure intensification by using two intensive farming practices-related variables: the amount of chemical fertilizer applied per hectare of the land cultivated, and the amount of manure applied per hectare of the land cultivated.

Descriptive Statistics

Table 1 reports the summary statistics. The proportion of households with perceived transfer rights decreased from 64% in 2003 to 57% in 2012.¹⁷ With regard to soil quality, during the decade period, the overall soil quality depleted although some soil nutrients improved—phosphorus, potassium and calcium increased. On the other hand, nitrogen and soil pH decreased from 0.22 to 0.17 and 6.62 to 6.19, respectively between 2003 and 2012. Carbon content almost remained stable. As stated earlier, we see an odd change for some soil variables. This may have come from a change in the analysis protocol. That is, although the method for the calibration should be the same, their reference samples (the samples with wet chemistry) used for the calibration are different and, hence, the regression coefficients estimated for the calibration. These differences affect the level and distribution of predicted contents. Therefore, the over-time comparison of the values of soil-contents may be problematic. That is why the trends of some soil variables in Table 1 are contrary to our expectations. Notwithstanding, overall the descriptive statistics suggest that both, land with perceived transfer rights and soil quality have decreased during the period this study covers.

Table 1 Summary statistics and t-test for equality of means of key variables

Estimation Strategies

Transfer Rights and Soil Quality

To examine whether soil quality differs by transfer rights, in Eq. (2) we restrict the analysis to a subsample of households with the same soil parcel in both survey rounds. We use household fixed effects model with region-year specific effects. The former helps to control for unobserved time-invariant household or parcel¹⁸ characteristics that could be correlated with both soil quality and land rights. The latter affords control of unobserved region-year specific characteristics that would affect soil quality. The identifying assumption is that unobservables such as soil type, farm management ability, managerial skills, farmer’s risk preferences and entrepreneurial drive that might simultaneously affect the soil quality as well as the perceived transfer rights are time-invariant, hence will be cancelled out by fixed effects. Moreover, the inclusion of region by time trends should mitigate the omitted variable bias that may arise from region-year specific unobservable characteristics that would affect the outcome variable and the explanatory variables of interest.

Although we use fixed effects and include many relevant controls, we cannot rule out the possibility that transfer rights may be endogenous to soil quality in such cases where farmers are more likely to acquire transfer rights of land with good soil quality. For example, it is possible that households with fertile lands have more incentives to demand and acquire private lands (this would increase their perception on transfer rights) to protect their fertile lands. To mitigate this concern, in alternative specification we use a time-invariant perceived transfer rights dummy. We exclude households who changed their perceptions from perceiving to have full transfer rights to perceiving having partial or no transfer rights or vice versa.¹⁹ To estimate this time-invariant variable, we interact it with year dummies (Wooldridge 2010 p.170). If its coefficient turns out to be positive, it will be suggestive evidence that full transfer rights are associated with better soil quality. It may also suggest that the soil quality of households with full transfer rights has increased over time (Wooldridge 2010 p.170). However, the omission of the households whose perceptions on transfer rights changed may not fully address the endogeneity issue because the resulting sample is obtained based on the endogenous variable. Therefore, the estimates cannot be assumed causal, but given the identification strategies explained above, it seems quite likely that the relationships estimated are at least partially causal.

Another concern is that the final sample—balanced soil sample may be biased and non-random. This is because out of the original of 940 households, our balanced soil sample has only 313 households. We follow Fitzgerald et al. (1998) and estimate a probit model of 2003–2012 attrition on a range of 2003 household characteristics. All, except two household characteristics (household head’s age and land holdings) are not significant (Table 8, column 1). We find that households headed by older people and households with more lands were less likely to have remained. Another determinant of the soil sample attrition is the region dummies, which represent the soil sample spoilages and losses. Therefore, we do not think that the soil sample attrition is systematically correlated with household characteristics to create attrition biases. However, in Eq. (2) we control for any possible attrition bias with the inclusion of attrition weights²⁰ estimated based on the methods of inverse probability weights (Fitzgerald et al., 1998, Wooldridge, 2010). Moreover, because we use the household fixed effects models, if any attrition biases exist, they would be minimal.

$$SQI_{ijrt} = \pi _0 + \pi _1Pr_{it} + \pi _2H_{it} + \pi _3V_{jt} + \gamma _{rt} + \alpha _i + \varepsilon _{ijrt}$$

(2)

where, i, j, r, t denotes household, community, region, and year of survey, respectively. SQI_ijr is the outcome variable—the soil quality index or each of the soil variables. Our parameter of interest is π₁. Pr_it is a time-variant perceived transfer rights dummy equal to 1 if the household has complete transfer rights over a parcel and 0 otherwise. H_it is a vector of household specific controls. As explained earlier, the decision-maker on land uses, investments and agricultural production activities matters in association with that of transfer rights. In our case, the actual manager of land is the head of household. We, therefore, control for a number of household head’s characteristics including whether household head is female, household head’s years of schooling and household head’s age. In H_it, we also control for household size, number of male adults, number of female adults, average years of schooling of male adults, average years of schooling of female adults, asset value, and livestock value. V_jt is a set of observable community characteristics including population density, rainfall, temperature and wind. γ_rt is expected to capture region-year specific unobservable characteristics which would affect soil quality as well as land rights. α_i is included to remove the effects of time-invariant household or parcel characteristics such as soil type, parental material, elevation and soil management ability that may bias our estimates. ε_ijrt is the error term that may be heteroskedastic and correlated within the community. We account for intra-community correlation by using robust standard errors clustered at the community level. We also include attrition weights to account for possible attrition that may result from non-randomness of households with soil samples.

Why Transfer Rights Improve Soil Quality?

Earlier, we argued that transfer rights might improve soil quality by incentivizing farmers to invest in land improvement. To understand why transfer rights improve soil quality, it is necessary to examine the effect of transfer rights on agricultural intensification as well. We do so by estimating Eq. (3).

$$AI_{pijrt} = \beta _0 + \beta _1Pr_{pit} + \beta _2FS_{pit} + \beta _3H_{it} + \beta _4V_{jt} + \gamma _{rt} + \varphi _i + \mu _{pijrt}$$

(3)

AI_pijrt takes two variants: quantity of manure applied (kg/ha) and amount of chemical fertilizer used (kg/ha). The variable of interest is Pr_pit. FS_pit denotes farm size in hectare, H_it is a set of household characteristics. V_jt is a vector of community-level variables,²¹γ_rt, is the region-year fixed effects, and φ_i is the household fixed effects or parcel fixed effects. μ_pijrt is the error term that may be correlated within a community. We, therefore, use robust standard errors clustered at the community. To control for possible attrition bias that may result from non-random of households (779 out of 940), in Eq. (3) we include attrition weights estimated from this attrition level.

Estimation Results

Transfer Rights and Soil Quality

Table 2 presents the results of the relationship between transfer rights and soil quality (Eq. (2)). The outcome variables are the six different measures of soil quality index. The explanatory variable of interest is a time-variant transfer rights dummy. The results show that parcels with complete transfer rights are statistically significant associated with better soil than those with partial or no transfer rights. The results suggest that households who perceive to have complete transfer rights over their lands are associated with around 0.4 points higher soil quality index compared to their counterparts who perceive to have partial or no transfer rights over their lands. To check the robustness of the results, in Table 9 we re-estimate the regressions with year-district fixed effects instead of year-region fixed effects. The significance levels remain the same and the correlations are statistically significant and in the same direction. This provides further evidence that endogenous spatial variation is well soaked by our estimation strategy.

Table 2 Transfer rights and soil quality

To shed light on how each of our soil variables is correlated with transfer rights, in Table 3 we report the regression results of the correlation between transfer rights and each of the six soil variables. In columns 1–5, we see that perceived complete transfer rights is positively correlated with carbon, nitrogen, phosphorous, potassium and calcium contents, respectively. However, except for potassium and calcium, other soil variables are statistically insignificant when considered individually. In column 4, those households who perceive their lands to have complete transfer rights their soils have 15 percentage points of potassium content higher than those with partial or no transfer rights. Similarly, households who perceive their lands to have complete transfer rights their soils have 10 percentage points of calcium content higher than those with partial or no transfer rights (column 5).

Table 3 Transfer rights and soil nutrients

As a further check of the robustness of the main results, in Table 4, we use a time-invariant transfer rights dummy interacted with year dummy. As explained in Section Data and descriptive statistics, we exclude households who changed their perceptions from perceiving to have complete transfer rights to perceiving to have partial or no transfer rights or vice versa.²² The coefficient of the interaction term between the time-invariant complete transfer rights dummy and year dummy is positive and significant. This provides further suggestive evidence that households who perceive to have complete transfer rights of their lands are associated with better soil quality than those who perceive to have partial or no transfer rights. The positive and significant coefficient of the interaction term may also suggest that the soil quality of households with complete transfer rights of their parcels has increased over time. We also re-estimate the regressions with year-district fixed effects instead of year-region fixed effects (Table 10 in the appendix). The results are consistent in general. The significance levels remain more or less the same and the correlations are statistically significant and in the same direction. We also show how each of the soil variables correlates with complete transfer rights (Table 11); our estimates bear the same signs and significance levels like those of Table 3.

Table 4 Transfer rights and soil quality: Interacting transfer rights and year dummy

To examine more closely those farmers who changed their perception on transfer rights, we isolate them. We do so in order to exploit the variation to see how soil quality relates to farmers who change their perceptions on transfer rights (Tables 12, 13). The results largely corroborate our main findings in Table 2 and Table 3, respectively—the size of the coefficients of interest and significance levels remain largely the same. Out of those who changed their perceptions on transfer rights, we further examine whether the move from not perceiving to have complete transfer rights to perceiving to have complete transfer rights is associated with an improvement in soil quality. We find that moving from not perceiving to have complete transfer rights to perceiving to have complete transfer rights is indeed strongly associated with an improvement in soil quality (Table 5).

Table 5 A subsample of households who moved from not perceiving to have complete transfer rights to perceiving to have complete transfer rights

Similarly, in Table 6 we report the regression results of the correlation between transfer rights and each of the six soil variables of this subsample of households that moved from not perceiving to have complete transfer rights to perceiving to have complete transfer rights. The results strongly support our hypothesis. Except soil pH, other soil variables are statistically positively correlated with transfer rights. In column 1, we find that moving from not perceiving to have complete transfer rights to perceiving to have complete transfer rights increases the soil carbon content by 43 percentage points. Similarly, this move increases phosphorous, potassium, and calcium contents by: 118 percentage points, 128 percentage points, and 88 percentage points, respectively. In addition, the move increases nitrogen content by 11 percentage points; however, this is not statistically significant. Conversely, as expected moving from perceiving to have complete transfer rights to not perceiving to have complete transfer rights is not associated with improvement in soil quality (Table 14 and Table 15). All these provide further supportive evidence that transfer rights indeed are associated with better soil quality.

Table 6 A subsample of households who moved from not perceiving to have complete transfer rights to perceiving to have complete transfer rights

Transfer Rights and Agricultural Intensification

In the previous analysis, we showed that households who perceive to have complete transfer rights of their lands are associated with better soil quality. To understand why transfer rights improve the soil quality, we estimate Eq. (3) and present the results in Table 7²³. We find a positive correlation between complete transfer rights and application of manure (column 1). This relationship, however, is not statistically significant. This suggests no significant difference in manure use between farmers who perceive to have complete transfer rights and those who do not. One possible explanation for this result is that in our sample few people used manure−only 3% and 5%, in 2003 and 2012, respectively. Thus, regardless of the nature of transfer rights, the use of manure is very minimal in Uganda.

Table 7 Transfer rights and agricultural intensification

Interestingly, however, we find that parcels perceived to have complete transfer rights are associated with more use of chemical fertilizers than their counterparts (column 2). Although the use of chemical fertilizer is statistically positively associated with transfer rights, the effect of chemical fertilizers on soil quality can be complex. Studies have shown that the use of chemical fertilizers increase the fertility of soil especially in short-term because chemical fertilizers have more nutrient contents and are consumed quickly by crops (see, Chen et al. 2010). Conversely, other studies (Dong et al. 2012, Guo et al. 2010) caution that excessive, long-term and inappropriate use of mineral fertilizers may negatively affect the quality of soil. Because of these complexities, agricultural experts should educate farmers to use chemical fertilizers sustainably. Notwithstanding, chemical fertilizer usage in Uganda is still very low.

Although the correlation between transfer rights and agricultural intensification is statistically significant on chemical fertilizers but not on manure use, the strong positive correlation between transfer rights and soil quality found suggests that the latter is affecting the former through other soil conservation means. These may include use of compost, crop residues, mulching or other practices that improve or maintain soil fertility. In fact, Yamano et al. (2004) show that application of crop residues is the mostly widely practiced in Uganda. However, in this paper, we are unable to analyze these pathways due to data limitations. Nonetheless, existing studies on Uganda shed some light. Deininger and Ali (2008) found that full ownership rights provide significantly higher incentives to invest in soil conservation and manure application than occupancy rights. Similarly, greater transfer rights are correlated with significantly higher levels of long-term investment in Uganda (Deininger et al. 2008).

Conclusion and Policy Implications

Theoretically, transfer rights incentivize individuals to invest in land improvement. Empirical studies have documented a positive relationship between soil conservation-related measures and transfer rights. This suggests that transfer rights can play a vital role in land management and improvement. This is especially important in most SSA countries where soil quality deterioration is a public concern. Yet, gaps on whether transfer rights may help to improve the quality of the soil remain. This is partly due to lack of objective panel soil measures in many usual surveys. We use a decade-long panel data with lab-tested soil measures from farmers in Uganda to examine the relationship between perceived transfer rights and soil quality.

We find that households who perceive to have complete transfer rights of their parcels have better soil than those with partial or no transfer rights. This suggests that households who perceive to have complete transfer rights have more incentives to invest in soil conservation than those who do not. The findings support the theoretical argument that secure land rights or transfer rights incentivize farmers to invest in soil management and improvement. Through which pathways does the positive relationship between transferability of land rights and soil quality we find operate? There are potential three main pathways. First, better land rights reduce farmers’ uncertainties and increase their assurance about ownership of land—this in turn increases farmers’ incentive to invest in land improvement. Second, with greater transferability of land rights, farmers may use land as collateral to borrow from lenders and use the borrowed funds to invest in land improvement. Third, the right to transfer land to others makes it easier to convert land into liquid assets through selling or renting, hence encourages investment on land. Although in this paper we are unable to empirically test each of these pathways explicitly (due to data limitations), the strong positive relationship between transferability of land rights and soil quality revealed suggests that at least some of these pathways may be behind the obtained results. Certainly, it has to do with the way farmers perceiving to have complete transfer rights are managing their land compared to farmers perceiving not to have. Moreover, although we do not find strong positive correlation between transfer rights and use of manure, the fact that strong positive correlation between transfer rights and soil quality is found suggests that farmers perceiving to have complete transfer rights may be investing in soil improvement and conservation through other means such as use of compost, crop residue and fallowing. However, due to data limitations we could not test these plausible pathways.

Although our findings may not be generalized to every part of SAA, they have important policy implications. The first policy implication is that the findings underscore the need to promote transfer rights. The second is to promote agricultural intensification. One way to achieve the latter is through policies that stimulate investment in land improvement. Such policies may include encouraging farmers to use external inputs such as chemical fertilizers and locally available inputs such as manure, compost, and crop residues. However, because the effect of chemical fertilizer on soil quality is complex and still an empirical issue, agricultural experts should assist farmers to use it sustainably.

Appendix

Table 8

Table 9

Table 10

Table 11

Table 12

Table 13

Table 14

Table 15

Table 8 Determinants of attrition in the household survey and soil samples

Table 9 Transfer rights and soil quality

Table 10 Transfer rights and soil quality

Table 11 Transfer rights and soil nutrients

Table 12 Transfer rights and soil quality: A subsample of households who changed their perceptions on transfer rights

Table 13 Transfer rights and soil quality: a subsample of households who changed their perceptions on transfer rights

Table 14 A subsample of households who moved from perceiving to have complete transfer rights to not perceiving to have complete transfer rights

Table 15 A subsample of households who moved from perceiving to have complete transfer rights to not perceiving to have complete transfer rights