Food Policy 67 (2017) 26–40 Contents lists available at ScienceDirect Food Policy journal homepage: www.elsevier.com/locate/foodpol Agricultural intensification: The status in six African countries Hans P. Binswanger-Mkhize a, Sara Savastano b,⇑ a University of Pretoria, South Africa b World Bank and University of Rome Tor Vergata, Department of Economics and Finance, Italy a r t i c l e i n f o a b s t r a c t Article history: Boserup and Ruthenberg (BR) provided the framework to analyze the impact of population growth and Available online 13 October 2016 market access on the intensification of farming systems. Prior evidence in Africa is consistent with the framework. Over the past two decades, rapid population growth has put farming systems under stress, while rapid urbanization and economic growth have provided new market opportunities. New measures of agro-ecological potential and urban gravity are developed to analyze their impact on population den- sity and market access. The descriptive and regression analyses show that the patterns of intensification across countries are only partially consistent with the BR predictions. Fallow areas have disappeared, but cropping intensities remain very low. The use of organic and chemical fertilizers is too low to maintain soil fertility. Investments in irrigation are inadequate. In light of the promising outcomes suggested by the Boserup-Ruthenberg framework, the process of intensification across these countries appears to have been weak. Ó 2016 The World Bank. Published by Elsevier Ltd. This is an open access article under the CC BY IGO license (http://creativecommons.org/licenses/by/3.0/igo/). 1. Introduction offset the negative impact of population growth on farm sizes, maintaining or increasing per capita food production, and even Since independence in the 1960s, Sub-Saharan African countries increase a farmer’s income, which we call the BR predictions. Pop- (SSA) have undergone exceptionally fast population growth. They ulation growth provides the necessity for intensification, while also have faced rapid urbanization and some economic growth, market access provides the opportunity.2 The increase in output, which would have tended to increase the demand for agricultural however, comes at the cost of an increase in labor and other inputs products. In more densely populated areas, the rising population per hectare cultivated. The positive outcome has been realized in has resulted in farm sizes now close to East and Southeast Asian those tropical areas of the world where technical change has added levels (Headey and Jayne, 2014; Otsuka and Place, 2014).1 This impetus to productivity growth. means that farmers now have to fend for their livelihood on a much However, another outcome observed by Geertz (1963) in Java reduced area, which requires rapid intensification and productivity prior to the Green Revolution, was that the intensification trig- growth. At the same time, the rising demand for agricultural com- gered by population growth and market access was insufficient modities should be beneficial for them in terms of better market to lead to enough productivity growth to make today’s farmers opportunities and higher prices for non-traded commodities. Both better off than their parents, and that instead, they became worse forces are leading to higher farming intensities, and possibly to off. Geertz called this process agricultural involution.3 Since the higher investments and input use. 1960s, biological technical change in SSA has been lagging behind Under the theory of intensification of farming systems of Ester the rest of the world, and so have fertilizer use, mechanization and Boserup (1965) and Hans Ruthenberg (1980a,b), the BR model of investment in irrigation (World Bank, 2008). The question, therefore, intensification, both population growth and market access can lead is whether there has been agricultural involution in Africa, which to a virtuous cycle of intensification of agriculture: These forces was first addressed by Lele and Stone (1989), who found significant lead to a reduction in fallow, higher use of organic manure and fer- signs of involution. Have increases in farm profits per acre been suf- tilizers to offset declining soil fertility, and investments in mecha- ficient to also lead to an increase in agricultural income per person, nization, land and irrigation. All of these have the potential to 2 In addition to intensification, farmers can diversify into cash crops and buy food, ⇑ Corresponding author. or they can migrate. These opportunities are better in an open economy than in a E-mail address: sara.savastano@uniroma2.it (S. Savastano). closed one. 1 3 In less densely populated countries and regions, it is still possible to maintain A study of agricultural intensification in Africa found signs of involution only in 2 farm sizes as emphasized by Headey and Jayne (2014). of 10 locations they studied (Turner et al., 1993). http://dx.doi.org/10.1016/j.foodpol.2016.09.021 0306-9192/Ó 2016 The World Bank. Published by Elsevier Ltd. This is an open access article under the CC BY IGO license (http://creativecommons.org/licenses/by/3.0/igo/). H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 27 more than offsetting the decline in land per person? This is the the use of new seeds and fertilizers. The total impact includes the research question that needs to be evaluated in Africa, and towards impacts via all pathways by which AEP and UG influence intensifi- which we make a modest contribution. cation, including via population density and market access. What The literature on agricultural intensification in Africa developed we are not able to do, is to measure the components of the total significantly in the 1980s and has resumed over the past decade. As impact that operates via population density and market access, shown in the literature review below, it generally finds that in and therefore the regression we present does not yet constitute a most areas studied, intensification has progressed along the lines rigorous test of the BR framework. predicted by Boserup and Ruthenberg, and that agricultural involu- The measure of a single agro-ecological potential (AEP) variable tion is confined to a few areas. These studies typically used case is based on the modeling of attainable crop yields across all agri- studies across locations. However, Headey and Jayne (2014), using cultural areas of the globe, estimated by IAASA and FAO (Tóth cross country data, have shown that rises in population density et al., 2012). As a proxy for urban demand, we develop a measure have been associated with reduced fallow and more intensive of urban gravity (UG) that a particular location experiences with use of fertilizer, but not in mechanization or irrigation. That would respect to all urban centers in the country with a current popula- make involution very likely, as it is hard to see how yields and farm tion of over 500,000 people.4 We use an estimate of the light emit- profits per acre could increase much under these circumstances. ted at night by each city that is derived from exiting light intensity Testing whether involution is occurring or not would require measures across all pixels of the city.5 The light emitted by each city access to micro-panel data that is not yet available in Africa over is assumed to be highly correlated with its overall GDP. We convert a sufficiently long period. the light intensity to an urban gravity variable that is a negative In this paper, we instead take initial steps towards analyzing the exponential function of the distance of the urban area from the enu- status of intensification processes using national representative meration area (EA) in which the farmers live. household data. They are for six African countries that have been More specifically, this paper will collected under the Integrated Surveys on Agriculture (ISA) that have been imbedded in broader Living Standard Measurement 1. Develop internationally comparable measures of the overall Studies (World Bank, 2009) (Ethiopia, Malawi, Niger, Nigeria, agro-ecological crop potential (AEP) and of Urban Gravity Uganda and Tanzania). These national household data contain (UG) in the farmers’ location. the intensification and technology variables, as well as profits 2. Describe the degree of agricultural intensification across the and household incomes. These will generate panels of five or more countries, and across the agro-ecological zones found in these years of data which will have to be analyzed in the future. In this countries. paper, we use the cross section data from the first year of the stud- 3. Estimate the causal impact of agro-ecological potential and UG ies. We are therefore not able to rigorously test the BR predictions. on population density, infrastructure and market access, and on However, rigorous tests of the BR framework micro-data has to a range of agricultural intensification variables. wait until panel data of sufficient length become available in order to enable an analysis of changes in farming systems that may be As discussed, a rigorous test of the BR framework has to await quite slow. Instead, we are focusing on the description of the status panel data analysis. Nevertheless, some of the country data allow of agricultural intensification in the six countries, including popu- for consistency checks to be made of the observed values with lation density, cropping intensity, fallow, irrigation and use of the BR predictions, and these will also be signaled. inputs. We then check whether there is consistency of the predic- The plan of the paper is as follows: Section 2 reviews the theory tions of the BR framework with respect to these variables, and and findings about agricultural intensification. Section 3 presents among them. the analytical framework needed to test the BR framework rigor- In the Boserup-Ruthenberg framework, the main drivers for ously and to estimate the impacts of AEP and UG on population agricultural intensification are population density and market density and market access, as well as their total impact via all access. These in turn are partly determined by the agro- routes they influence. Section 4 describes how the AEP and UG ecological potential of a village, as people would have migrated variables are constructed and defines the variables for all the inten- more to high potential areas, such as tropical highlands, and have sification variables used in the paper. Section 5 presents the been able to support more children; and governments would have descriptive results while section six presents the regression results. preferred to invest in roads and markets to take advantage of the Summary and conclusions follow in Section 7. food production potential and serve the dense population (Binswanger et al., 1993). Investments in roads and markets are 2. Agricultural intensification: Theory and findings likely to also depend on the strength of urban demand for food, and the distances of urban centers from the villages. In this paper, The general model of the evolution of farming systems origi- we also explore the relationship of the two drivers of intensifica- nates in the work of Ester Boserup (1965) and Hans Ruthenberg tion, population density and market access, to the agro-ecological (1980a,b) – henceforth referred to as the BR theory or framework. endowment and the strength of urban demand impacting on the In the 1980s, these ideas were summarized, partially formalized, survey villages. In order to do so, we develop a single variable for and tested for SSA in books by Pingali et al. (1987), Binswanger the agro-ecological potential (AEP) of each enumeration area, and and McIntire (1987) and McIntire et al. (1992). All these authors a second variable for urban gravity (UG) which reflects the eco- nomic size of the city in question and the travel time from the enu- 4 meration area to the city (see below). Clearly, these two variables We leave out the smaller cities, as their income as measured via light emissions could be affected by the agro-ecological potential of the zone in which they sit, are exogenous to the population density and government invest- making them endogenous to the system analyzed. ments for market access, and we therefore can estimate a causal 5 As a proxy of light intensity, we used the sum of nighttime lights recorded in impact of these two variables on the BR drivers of intensification. 2009. Input values ranging from 0 to 63 indicate average intensity of light The finding is that high AEP and UG have had a significant positive observations, regardless of frequency of observation. Ephemeral events such as impact on population density of the enumeration areas and on bet- lightning strikes and fires have been discarded. The satellite source is DMSP F16, inter-calibrated for comparison between years. The range of 0–63 refers to the pixel- ter infrastructure and market access. level value (the source data are gridded at 30 arc seconds). The variable we are using We can also estimate the total impact of AEP and UG on the var- is aggregated at the 5 arc minute block level (resolution of SPAM, GAEZ and other ious intensification variables, such as cropping intensity, fallow or harvest choice variables), which would include many pixels from the lights data. 28 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 consider the evolution of farming systems, the methods of main- 8. Land rights evolve from general rights of the communities taining soil fertility, the technology in use and the labor input which occupy an area to cultivate in their territory to individ- per hectare as endogenous, being influenced by the both agro- ualized property and use rights to specific plots of land. This ecological and the socio-economic characteristics of the environ- process radiates from the homesteads to more distant areas, ments with which the farmers are confronted. The main driving including land under fallows and pastures. Common property forces of the evolution of the farming systems towards higher resources are progressively privatized. intensification and crop–livestock interaction are population pres- 9. Intensification leads to increases in yields, which is faster sures (often measured as population density) and market access, where new technology or irrigation is introduced, and often both of which define the opportunities and constraints of house- to the diversification from basic staples to higher value crops. holds in the areas.6 Market access consists of two factors: The exter- 10. Value of output per acre increases, but, on account of higher nal demand that emanates from the urban sector and export input costs and/or declining farm sizes, profits per acre and markets, and roads which enable farmers to reach these markets. agricultural incomes per households may increase or In low population density areas (other than the arid zone), crop- decrease. ping is characterized by long forest fallow systems in which the re- growth of the forest after cultivation restores soil fertility in terms We will analyze most of these dimensions of intensification. of nutrients and soil structure, and suppresses weeds. Land is Because on account of population growth and/or higher input cleared by fire, with the ashes further increasing the nutrient con- costs, profits per acre and household income may increase or tent of the soil. Seeding takes place between the stumps, using a decline or, as suggested by the involution hypothesis, in panel data digging stick or a hand hoe. The stumps make the use of a plough it is possible to test for it, but not yet in this paper. impossible. Weeding is not necessary as weed seeds have decayed Formally, the involution hypothesis associated with population during the long fallow period. Farmers hold no cattle. The labor growth can be expressed as follows: Net farm income (input costs) requirements for producing crops are very low. After one or several per capita, Yf , is by definition the product of net farm income per N seasons of cultivation, soil nutrients and soil organic matter Yf L hectare, , and land per capita, N : decline, the soil structure deteriorates, and weeds start to take L over. Declining yields and rising labor requirements for weeding Yf Yf L and land preparation lead farmers to abandon the land and open ¼ : ; ð1Þ N L N new forest areas or re-grown forests for cultivation. If population growth reduces the availability of forests and fal- or in percentage change terms: low land, and if new market opportunities emerge, farmers have to intensify agricultural production. They do it in order to maintain or Yf Yf L D ln ¼ D ln þ D ln ð2Þ increase their food supplies and the income from the sale of crops. N L N The BR effects of higher population density and improved market access in the past have led to the following impacts, which are also If population density is rising, then land per capita is falling, predictions for the future: leading to a loss of income, all else being equal. Of course, the Boserup argument is that all else is not equal because households Y 1. The progressive reduction in fallow length until the land is intensify production (increase output per hectare, Lf ). Thus, the permanently cultivated, and from there onwards, to multiple extent to which net income per capita declines or rises depends cropping per year. on whether changes in net income per capita compensate for decli- 2. Soil fertility must be restored via the incorporation of nearby nes in land per capita. However, it is also necessary to account for vegetation into soils, preparation of compost and/or manure, the higher input cost, will make the income increase needed to and/or artificial (inorganic) fertilizers. compensate for declining farm sizes even larger. Nevertheless, a 3. The appearance of grassy weeds makes hand hoe cultivation second cause of ambiguous welfare effects is that welfare is better much more difficult, and, as tree stumps disappear in the represented by net farm income, or gross income less costs. The short fallow stage, the plough is introduced via animal draft intensification process involves an increase in a number of costs, or tractors. including labor, oxen, modern inputs and land preparation (e.g. 4. Cultivation moves from lighter soils on mid-slopes to heavier irrigation). Even with rapid production growth, net farmer income soils in lower slopes and depressions that have higher water may not rise or may actually fall. retention capacity, and to more fragile soils on the upper Increases in household welfare, where they occur, are often slopes. associated with diversification of agricultural production to a 5. Cultivation in these new areas often requires investment in broader range of high value products that are often less land inten- land for the prevention of soil erosion, and/or drainage and sive (e.g. fruits and vegetables) and that can be marketed through irrigation. improving commercialization channels. Where rising population 6. Farmers and herders start to trade crop residues for cattle pressure and market access lead to increased specialization, and dung, the start of crop–livestock interaction. Eventually, where agricultural technology adoption and input use increase, farmers acquire animals and herders sometimes acquire there may be a beneficial diversification into rural nonfarm cropland, which leads to livestock integration. activities. 7. Labor requirements per unit of land increase for restoring of In contrast to positive intensification processes, under very high soil fertility, weeding, land preparation, for investments in and rising population density and poor policy, institutional or agro- land, and for the maintenance of draft animals. ecological environments, intensification could lead to involution and the diminution of economic and social well-being, and envi- ronmental degradation. Geertz (1963) characterized involution as 6 There are some parallels of the BR model with the Induced Innovation model a situation in which increasing demand for food is met by highly (Hayami and Ruttan, 1985) in that under BR, it is population density and market labor-intensive intensification, but at the cost of very small and access that push agricultural intensification, including new technological innovations and institution that have to underpin it, while under the induced innovation model, it decreasing marginal and average returns to inputs. Because there is technologies, land/labor ratios and institutions that adapt. However, each of the still is relatively little landless labor in SSA, the extra work would theories cannot explain what the other theory explains, and vice versa. often fall on family workers, rather than being supplied by landless H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 29 workers, as in Asia. Of the 10 cases of very high population density African countries.” (Headey and Jayne, 2014, p. 31). These results in SSA studied by Turner et al. (1993), there are signs of involution suggest that the full set of intensification processes discussed in in the humid tropical areas of Imo State in Nigeria, and in the this section have hardly occurred in Africa, which means that the Usambara mountains of Tanzania, where special rules inhibit ero- BR model is only partially supported, a conclusion that is also sion control because it can jeopardize access to land for women.7 reached in this paper. Based on macro- rather than micro-data, Lele and Stone (1989) also suggest that a significant share of the intensification observed in SSA 3. Analytical framework was already showing signs of involution by the mid-1980s. This means that the conclusions from aggregate data are more pes- The analytical framework has to be able to measure the causal simistic than from case studies. impact of population density, infrastructure and external demand Headey and Jayne (2014) find that agricultural intensities in (urban or export demand) on the various intensification variables. much of African agriculture have reached the stage of permanent Hj stand for the vector of intensification variables of an enumera- cropping. Most of the literature is consistent with the theory of tion area j (EA, usually a village); let the X ij variables stand for intensification, in Africa, as well as elsewhere. Baltenwick et al. the drivers of intensification in EAj i.e. population density, an indi- (2003), in an analysis of 48 sites in 15 countries of Africa, Latin cator of access to infrastructure such as roads, and let Z j be a vector America, and Asia, find that the forces of population density and of other conditioning variables for EAj . market access transcend national and continental specificities Then the correct equation for testing the BR hypothesis is and applied well across the study sites in all three continents. Their reviews, following McIntire et al. (1992), focus especially on crop– Hj ¼ a þ b1 X 1j þ b2 X 2j þ cZ j þ ej ð3Þ livestock integration and confirm the general trends and more detailed findings of these authors. Eq. (3) relates the intensification variables to the drivers of intensi- The papers in Pender et al. (2006) report studies of strategies for fication as identified by BR. sustainable farming systems in the East African Highlands, focused The critical coefficients for testing the BR framework are the b primarily on low to medium potential areas. The selection of areas coefficients, which should be greater than zero. However, because of lower agro-ecological potential also implies a bias in the results, the unobservable error term ej influences both the X variables and this time against the BR effects, as in lower potential areas the the intensification variables H, the b coefficients would be esti- work and investment incentives are likely to be lower than in mated with unobservable variable bias. Examples are specific higher potential areas. They find similar corroborating evidence potentials to grow high quality coffee or fruit, special locational for the general impacts, again with variations which will be dis- advantages such as proximity to water sources that could be used cussed in subsequent sections of this paper. They emphasize that for irrigation or proximity to ports, or even cultural factors. Many intensification is progressing especially well where vibrant mar- of these factors are unobserved or unobservable and cannot there- kets are nearby. Much earlier, this had been found to be true in a fore be captured as Z variables. Panel data are therefore required to case study of the agricultural history of Machakos district in Kenya, rigorously test the BR framework. where the output demand from Nairobi played an important role However, we only have cross section data for each of the coun- (Tiffen and Mortimore, 1992). Moreover, the opportunities of earn- tries. These descriptive data can be used to check whether the ing income in Nairobi provided resources for investment in Macha- levels of the various intensification variables are consistent with kos district. Clearly, urban centers present both market and trade each other. For example, if cropping intensity has already reached opportunities, which point is important in interpreting the results 100% and there are no longer any fallow periods, soil fertility must in this paper. Finally, Turner and Fischer-Kowalskic (2010), in a tri- be restored via the application of organic manure and/or chemical bute to Boserup’s 100th birthday, find that the Boserup framework fertilizers. If the proportion of farmers using these techniques is has held up well. low, then these variables have not responded as expected under Headey and Jayne (2014) used FAO data covering recent dec- the BR framework. Alternatively, if population density and crop- ades (1977–2007) from FAOs regular reporting and from their peri- ping intensity are high, substantial irrigation investments should odic agricultural censuses to study the process of agricultural have occurred, but if they are very low, one of the BR predictions intensification in countries from Asia and from Africa.8 As dis- is not satisfied. The descriptive section below performs this cussed in the modeling section, their panel data of countries allowed analysis. them to overcome the endogeneity issues associated with the Over their history, areas of high AEP have attracted more migra- response of population density to agro-ecological potential and tion than those with low AEP, they have been able to sustain higher urban gravity by using the fixed effects model. population growth and therefore they are likely to have higher They find that, in line with the BR model, agricultural intensifi- population densities. Recognizing the agro-ecological potential of cation is an important mechanism to offset declining farm sizes in an area, governments and communities would have been more both Asia and Africa. In response to declining farm sizes, in Asia likely to invest in infrastructure that provides access to markets. yields grew rapidly, while this response is absent in Africa. ‘‘In Similarly governments of urban centers with significant agricul- Africa, we observe no response of yields to land constraints over tural demand would also have been induced to invest in infrastruc- the short run, nor any growth of modern inputs such as fertilizers ture. To test whether these dynamics have been in place, later in or irrigation. Instead, we observe increased cropping intensity driv- the paper we develop measures of agro-ecological potential (AEP) ing around half of the growth in total crop output per hectare. This and urban gravity (UG) which can be a proxy for the demand pull would appear to be an unsustainable intensification path given the of cities, and other influences on rural areas. Our AEP and UG are implied mining of nutrients, and the more limited prospects for exogenous to the intensification variables and their impact can low cost irrigation investments, at least in many high density therefore be estimated via Eq. (4) without giving rise to unob- served variable biases. 7 Women had secure access to unimproved land for their subsistence production, Hj ¼ a þ d1 AEPj þ d2 UGj þ cZ j þ ej ð4Þ but once a parcel was improved via erosion control, would lose such access. 8 They also develop a general model of intensification that can accommodate the The d coefficients will then estimate the sum of the direct impact on per capita income of land expansion, intensification, reduced rural fertility impacts of AEP and UG on the intensification variables, as well as rates, and diversification into non-farm activities. the indirect effect via their impact on population density and 30 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 market access. If these are positive, then either the direct or indi- for a prolonged past period should be used, for which we do not rect effects, or both, have been at work, and the regressions there- have cropping pattern information. The current AEP is likely to fore do not reject the BR predictions. If on the other hand they are be highly correlated with past AEPs, so we also use the current negative, it is likely that the BR predictions for the respective vari- AEP instead. able cannot have been realized. That means that zero or negative We use the data on ‘‘Potential yield” that does adjust for fallow coefficients of AEP and UG can be interpreted as an absence of requirements. GAEZ contains potential yields for 28 crops, however the respective BR effect. On the other hand, a positive coefficient we use those 15 for which international prices are available. These could have been either a direct effect of AEP or UG, or an indirect are wheat, rice, maize, barley, millet, sorghum, white potatoes, cas- one via their impact on population density or infrastructure. sava, soybean, coffee, cotton, groundnut, banana, sweet potatoes, The dependent variables are therefore as follows: Population and beans.12 density; distance to the nearest road and the nearest markets; GAEZ presents potentials yields for low, medium and high input cropping intensity, defined as gross cropped area per net cropped levels, of which the current values at intermediate level13 are the area; the proportion of area currently fallowed and fallowed in most appropriate for the proposed analysis: ‘‘In the case of interme- the past; the proportion of net crop area irrigated; and the propor- diate input/improved management assumption, the farming system tion of households using different technologies that enhance yields is partly market oriented. Production for subsistence plus commer- – high yielding varieties, organic manure, fertilizer, or pesticides. cial sale is a management objective. Production is based on improved Equations are estimated for each of the dependent variables, and varieties, on manual labor with hand tools and/or animal traction in double logarithmic form. Because we want to analyze intensifi- and some mechanization. It is medium labor intensive, uses some cation in SSA, the country data are pooled and a country dummy is fertilizer application and chemical pest, disease and weed control, included to account for country-specific fixed effects. adequate fallows and some conservation measures.” (Tóth et al., 2012, p 18). In light of the limited irrigation in Africa, we are using the data for the rain-fed category. To summarize, we will use the 4. Definition of the variables used and descriptive statistics agro-ecological level for the current climate conditions at intermediate levels of input use under rain-fed conditions. 4.1. Agro-ecological potential per hectare The data in the GAEZ system is for the potential yield of individ- ual crops. However, we want to characterize the aggregate agro- We calculate the agro-ecological potential from the currently climatic potential in the communities being analyzed. Therefore, available Global Agro-Ecological Zones (GAEZ) data portal9 of the we need to assign a value to each of the potential crop yields. In International Institute for Systems Analysis and the Food and Agri- order for the calculations to be comparable across countries, we culture Organization (Tóth et al., 2012). first converted the yields into dollar values using average world For each 5 arc-minute grid cell of agricultural land of the World, market prices for the past three years during which the first rounds the data set uses crop models to calculate agro-climatic yields, for of the LSMS-ISA studies were carried out. The commodities include 280 crops and land-use types.10 These are progressively aggregated the 15 crops mentioned above, for which we have found interna- to 49 crops. Agro-climatic yield takes into account climate-related tional price data.14 constraints and uses and optimum crop calendar. GAEZ then calcu- To get a unique value for the AEP of a location, we aggregate the lates Agro-ecological suitability and productivity that takes into individual potential crop values into an aggregate potential crop account the grid-specific soil and terrain conditions and fallow value. This is best done by using as weights the proportion of each requirements.11 Because the crop yield estimates that have been crop in the crop mix being produced in the enumeration areas or used in computing AEP include the known impacts of soil degrada- close to them. We use the average cropping pattern across all tion, today’s estimates are possibly a slight underestimation of past households in the EAs as weights to aggregate the potential crop AEPs. However, much of the AEP is explained by innate characteris- values into the overall agro-ecological potential of the EA. For the tics of the soils that have not changed and a relatively stable climate aggregation of the potential crop values to AEP, we only take into over the past. Therefore, the current and past agro-ecological suit- account the value of the main product, and not any by-products. ability and productivity are likely to be highly correlated. Let Siz denote the average share (across farmers j) of crop i in the A limitation of the proposed AEP measure has to be signaled: EAz and let Aijz be the area under crop i of famer j (i = 1. . ..M, Population density, market access and intensification variables j ¼ 1 . . . N Þ. The denominator in Eq. (5) is the total area under crop that are observed today reflect not just the potential today, but i in EAz divided by the total cropped area in EAz . past potentials at the time that public investment and migration PN decisions were made. But the AEP measure reflects international j¼1 Aijz Siz ¼ PM PN ð5Þ prices for three very recent years, and the present cropping pat- i¼1 j¼1 Aijz tern, and therefore are AEPs for the current period. When analyzing the influence of the AEP on current farming systems variables, such Let Pi be the international price of crop i. And Let X iz be the as cropping intensity, value of production or input use, the current agroecological potential of crop i in the EA j. Then, the agro- AEPs are the right variables to use. However, when we analyze the ecological potential in the EA z is impact of AEP on population density and road investments, the AEP 9 The Global Agro-Ecological Zones website can be found here: http://www.fao.org/ 12 nr/gaez/en/ In principle, the AEP should include livestock production possibilities, but such 10 For the coordinates of the community to be matched to the geographic units of data do not exist. 13 the GAEZ data, we calculate the central point of each of the enumeration areas, using The GAEZ data also include a low input level and a high input level. There are the geo-location of the households in the EA. We select the corresponding grid cell many countries in Africa in which the low input level is no longer practiced. The high from the IIASA-FAO data set, as well as the adjacent grid cells. We average across such input level is only practiced in the commercial farm sectors for example of South geographic units by weighing the values for the adjacent grid cells by their Euclidian Africa or Kenya and therefore does not reflect what smallholders do or can aspire to. 14 distance from the central point of the EA for which we calculate the AEP. Another way to aggregate across crops is to use calories per kilogram of each crop, 11 If there is little use of fertilizer or manure, soil fertility has to be restored by and then aggregate them as discussed in the next paragraph. However, the market leaving land fallow. The fallow requirement may be one year or more. The fallow value of calories from different crops is very different, as exemplified by calories from adjustment converts the model result to the average number of growing years in the tubers relative to calories from grains. Moreover, farmers are not interested in the crop-fallow cycle. calories they can produce per ha, but in the revenues that they generate. H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 31 X AEPz ¼ Siz Pi X iz ð 6Þ cross-border cities by the composite index of the difficulty of move- i ment of people, goods and information across the respective borders, We want to stress here that our estimate of the AEP may not using the higher difficulty of cross-border movement of the two adequately capture the ‘‘true” underlying AEP, and that the latter respective countries. The result is the aggregate UG to which each is therefore estimated with error. EA is exposed. As in the case of AEP, we assume that today’s urban gravity is 4.2. Agro-ecological potential per person correlated with UG over the past, during which migration, fertility, infrastructure investment decisions were made, and therefore the In this study we use two measures of population pressure: the coefficients of today’s urban gravity capture both current and past traditional population density (persons/ sq. km), and what we impacts of UG. Since urban populations and incomes have changed define the agro-ecological population pressure or the agro- very rapidly over the past decades, the errors in variable problem ecological potential per person computed via Eq. (7) associated with past UG being imperfectly correlated with current UG is more severe than in the case of AEP. Again, for the intensifi- AEPP ¼ AEP Â 100=PD ð 7Þ cation variables that changes more quickly over time, the problem We have developed this new measure to take into account the will be less. vast differences in agro-ecological potential across EAs, regions and countries that are not captured by the traditional measure of 4.4. Public infrastructure population density. As the population for each EA has not been collected in the We used distance to the main road as a proxy of public infras- LSMS-ISA surveys, we use the data for rural population density col- tructure, and also included distance to nearest major market lected by the Harvest Choice project15 which are disaggregated to (which is an additional measure of market access embedded in the level of communities contained in the EAs of the LSMS-ISA. This the concept of UG). Both variables are included in the set of GEO external variable includes farmers and people who are not engaged variables collected under the LSMS-ISA project by means of house- in agriculture, and since peri urban EAs are likely to have a higher holds’ GEO coordinates. The former is the distance in kilometers to non-farm population, the overall population pressure computed the nearest trunk road, while the latter is the household’s distance according to Eq. (7) for peri-urban areas will most likely go down. to the nearest major market. 4.3. Urban gravity 4.5. Owned and operated land We follow Henderson et al. (2009), Gallup et al. (1999), and There are two measures of plot sizes in the data, the area Kiszewski et al. (2004) in using the measures of intensity of light reported by the farmers, (the self-reported area), and the area mea- emitted which is available for each pixel on earth. While light sured by the enumerators using GPS. The measured areas are avail- intensity is not a direct measure of economic activity, it is highly able for a large share of plots, but not for all of them. For the correlated with it.16 A great advantage of light intensity data is that missing areas that would correspond to an estimate via GPS, they can be used for cities for which GDP data are unavailable, as for regression analysis was used to relate self-reported area to area most cities in Africa. The data for light intensity come from the measured by GPS for the households that had both measures. Fol- Defense Meteorological Satellite Program (DMSP) of the National lowing Kilic et al. (2013), the estimated regression coefficients Geophysical Data Center.17 were then used to impute a predicted GPS area for plots with only To measure the aggregate emission of light at night from a city, self-reported areas. the light intensities of each urban pixel are aggregated over all pix- Operated area is defined as owned area, plus rented in area, els of the city. The light intensities of the cities are converted to minus rented-out area. urban gravities (UG) by weighting them by travel time in hours to the EAs, using a negative exponential function (Deichmann, 4.6. Land use intensity 1997). We then aggregate the resulting UGs to a national UG, sep- arate for each enumeration area, by summing it over all cities in The cropping intensity (CI) of cropped land is used, rather than the country or across the border of neighboring countries with Boserup’s and Ruthenberg’s R-value. This is because in most coun- population above 500,000.18 We adjust the light intensity of tries fallow rates are now very low, and they are no longer in the transition from long or short fallow systems to permanent agricul- 15 Harvest Choice data are based on calculations from data from the Center for ture. The R-value is best suited for these earlier stages.19 Cropping International Earth Science Information Network (CIESIN), Columbia University; International Food Policy Research Institute (IFPRI); The World Bank; and Centro intensity takes account of multiple cropping, which is the use of the Internacional de Agricultura Tropical (CIAT), 2004. We cannot use the population land for more than one crop a year. Cropping intensity CI is defined computed from LSMS-ISA as the EAs have been chosen with probability proportional as to their population at the last census giving preference to EAs with higher population. If we had aggregated our measure of population at the EA level, low population G density areas would have been underrepresented. CI ¼ ð8Þ 16 N If GDP data are flawed, they may be a superior measure of economic activity at the national level. The authors present estimates across countries and over time of where G is gross cropped, the sum of the areas cropped in the main reported GDP and light intensity, and also present an estimator of GDP which season plus the areas cropped in the second season, and N is net optimally combines the two. 17 cropped area, the area cropped in the main cropping season. If there The data source is the DMSP F16, the U.S. Air Force’s research project called the Defense Meteorological Satellite Program (DMSP), established in 1960. Since 1994, is only single cropping, CI is 1. It rises to 2 when all cropland is used DMSP have produced a time series of annual cloud-free composites of DMSP in both seasons, and can go higher when some land is used more nighttime lights. Together with the NGDC – EOG (the National Geographic Data than 2 times in a year. The cropping intensity is calculated as the Center – Earth Observation Group). mean over households in an EA, while the population density is 18 We choose cities with populations over 500,000 at the current time, because smaller cities often are centers of agricultural services and markets, and therefore 19 their population is influenced by the AEP of the surrounding area, which makes it Ruthenberg’s (1980a,b) R-value RV = N ⁄ 100/(N + F), where N is net cropped area endogenous. (also called cultivated area) and F is fallow area. 32 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 the mean over communities, as defined in the Harvest Choice data differences in definitions therefore are of little relevance. Among sets. the EAs, the distances to roads and markets vary little, suggesting that most of the variation is associated with the countries, rather 4.7. Irrigation and technology variables than the agro-ecological potential. Fig. 1 also shows the Urban Gravities for the six countries that For irrigation, we use the share of cropped land that is irrigated. not only reflect light intensities of cities, but also travel time. The Data on inputs and outputs are collected at the plot level, which is distribution of UGs across the countries are shown in Map 2. Urban a subdivision of the parcel. The data do not contain the area of each gravity is the highest in Malawi, at the value of 169, and the con- plot. Because different plots may use different inputs and tech- centration of red dots suggests that UG is high near the urban cen- niques, this means that we cannot estimate area under a particular ters of Blantyre and Lilongwe, but then tapers off quickly in the technique in this data set. Instead, we have to focus on whether a north. Then comes Nigeria, where the highest UGs are in the south, farmer does use, or does not use, a particular technique. We esti- and much lower ones in the north. UG is the lowest in Ethiopia at mate the proportion of households in each EA that are using only 7. improved seeds, chemical fertilizers, organic manure and Table 2 shows that the rural poverty rate is the highest in Tan- pesticides. zania (92%) and the lowest in Niger (41%). Tanzania has not been able to take advantage of its high AEP per person to foster sufficient agricultural growth to reduce poverty. Nor has the low AEP per ha 5. Descriptive results resulted in high poverty in Niger. In the other countries, the pov- erty rates vary between 52 and 75%. 5.1. Agro-ecological potential, AEP per person, and urban gravity In Table 1, Row 1, we see that the average AEP per ha across all 5.2. Land and land use intensity the countries is 740 dollars per ha, evaluated at international com- modity prices prevailing between 2005 and 2008.20 The totals Area operated per household is owned area plus rented in area, across countries are population weighted. From Fig. 1, it is clear that less rented-out area. Across countries, it is on average 1.57 ha per it is the highest in Uganda, because of its good climate conditions,21 farm (Table 3). It varies from the lowest in Malawi, at 0.74 ha, to and the lowest is in Niger, in the very dry Sahelian zone. Map 1 also the highest in Niger, at 5.1 ha (Fig. 2). Malawi’s AEP per ha is twice illustrates that high potential areas are most prevalent in Uganda the one in Niger, which partly compensates for its low operated and Central and Southern Malawi. In other countries, it is mostly area. What is surprising is that Uganda, one of the high population light green22 areas, with potentials between 478 and 786 dollars density countries, has an operated area quite close to Tanzania’s per ha, rather than the darker green areas with higher potential. In 2.4 ha. Since Tanzania has a much lower population pressure, we Ethiopia and Nigeria, there are also many brown areas that have would expect farm sizes there to be significantly larger. It appears low potential, mainly in the dry northern parts of each of these coun- that Tanzanian farmers are unable to make use of the larger land tries. In Niger, low potential areas dominate in the entire country. endowment per person, perhaps because they are labor con- In the second row, the AEP/km2 has been divided by the rural strained and unable, or unwilling, to make the investments population density to arrive at the AEP per person. Across all the required for animal draft or tractor plowing that would allow them countries, it is only $394.23 Fig. 1 shows that there are many rever- to operate larger areas. sals between AEP per ha and per person: Tanzania has the highest Cropping intensity is gross cropped area divided by net cropped AEP/person at 1314$, almost twice as high as that of Uganda, a rever- area. It is greater than one in all countries, therefore the stage of sal with respect to AEP per ha which is on account of them having permanent cropping has been reached everywhere. Cropping the highest and the lowest rural population densities among the intensity is especially low in Malawi (1.01) and Tanzania (1.07): countries considered. Given its dry climate, it is surprising that Niger For Tanzania this is not a surprise, as is AEP per person is by far has the third-highest AEP/person. This is on account of its high oper- the highest in the sample of countries, indicating a low population ated area per farm (Table 3) and its low population density. pressure on the agro-ecological resources. However, in Malawi it Average distance of households to the nearest tarred road the AEP per person is less than half that of Tanzania, yet its crop- across all countries is 15 km, while to the nearest market it is much ping intensity is the lowest among the six countries. The BR model higher, at 66 km. Distances to roads are the lowest in Uganda, at suggests that Malawi’s high population pressure would have led to 8 km, followed by Malawi, which also has the lowest distance to high land and irrigation investment, allowing for high cropping markets. The farthest distances to markets occur in Nigeria and intensities. We therefore find another inconsistency with the pre- Tanzania, at 70 km. That Nigeria, among the highest per capita dictions of the BR framework. Crop intensity is by far the highest income countries, should do so poorly in market access, suggests in Uganda, at 1.89, which is on account of the bimodal rainy sea- that they may have used larger markets as a reference, while son. The other countries have cropping intensities between 1.19 Malawi may have chosen very small markets. In the regression and 1.23. analysis, we use the log of the variables and also include a country In light of permanent cropping, on average the rate of fallow in dummy, so that only the within country variation is used to the six countries is only 1.2%, and therefore fallow can no longer estimate the relationships to the dependent variables, and the contribute to soil fertility maintenance and restoration. It is clear that the high population growth rates and growth in urban 20 This seems low. Note that famers will obtain less than this value because most of demand have virtually eliminated fallows in the countries. The them are far from intermediate input levels used to calculate the AEP. 21 highest proportion of land under current fallow is found in Tanza- Note that the AEP only takes account of the first season, and that Uganda in many areas is able to grow two rain-fed crops, and therefore has by far the highest cropping nia, at 7.5%. While that is consistent with Tanzania’s low popula- intensity (Table 6). Therefore, the real advantage of Uganda is even more striking. tion density and AEP per person, one would have thought that 22 For interpretation of color in Map 1 and 2, the reader is referred to the web Tanzanian farmers could make more use of fallow to restore soil version of this article. 23 fertility. The lowest rate of fallow is in Nigeria, at only 0.1%. Past Recall that this is the average population density across the EAs in each sample, which will be higher than the rural population density reported in national statistics, fallow rates are derived from the data on whether a plot had been as it comes from a sample of EAs chosen with probability proportional to population fallowed in the year before the current year. For the four countries size, rather than from all EAs in the Census. where we have the data, current and past fallow rates are similar. H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 33 Table 1 Countries’ Endowments. Source: Authors’ computation from LSMS-ISA surveys. ETH MWI NER NGA TZA UGA Total 1. Value of agro-ecological potential (US$/ha) 691.2 999.1 478.7 657.0 786.4 1877.9 739.6 2. Agroecological potential per person 396.7 547.6 792.4 301.0 1313.5 703.7 393.8 3. Rural population density (pers./sq. km) (2005) 174.2 182.5 60.4 218.3 59.9 266.9 187.8 4. UG* 7.4 169.3 22.8 134.6 30.1 63.6 82.9 5. Distance (in Kms) to the nearest major road 14.4 10.6 11.5 16.0 17.8 7.9 15.3 6. Households’ distance (in Kms) to the nearest market 64.5 7.7 56.3 70.1 70.4 31.6 66.3 * UG travel time in hours to cities with 500 K population. Fig. 1. Agro-ecological potential, agro-ecological population pressure and urban gravity. 5.3. Irrigation and technology tion and urban demand, suggests that farmers have not responded to these trends by increasing irrigation, as the BR framework would Across the six countries, the average area irrigated per farm is predict. Is it possible that this lack of response is caused by excep- only 0.03 ha, and the share of irrigated area in total area is only tionally poor availability of groundwater, which farmers might 4.4% (Table 4), which in Ethiopia, Malawi and Nigeria appears to have tapped via bore-wells? be inconsistent with the low AEPs per person observed. Surpris- Except for Malawi, the proportion of households using ingly, the area under irrigation is higher in Tanzania, at 0.045 ha improved seeds is less than 18% of the households. Malawi is doing compared to Malawi, at 0.030 ha (Fig. 3). Given the previous dis- by far the best, at 61% of households. It also has the highest propor- cussions, this is particularly inconsistent with the BR hypotheses. tion of households using inorganic fertilizers, at 76%. Given its high On irrigation, we also have the data by agro-ecological zones population pressure, this is consistent with the BR hypothesis. across the countries. (Online Annex). The area of land irrigated is However, only 16% of its households use organic manure, which by far the highest in the warm arid areas (0.11 ha). This is not sur- according to BR should have become an important technology for prising because the payoff to irrigation is higher, the dryer the cli- soil fertility maintenance in this country. Moreover, agro- mate. In all other climate zones, it is around 0.01–0.05 ha. This is chemicals are used by only 3% of farmers. Malawi is doing far bet- also not surprising in the cool or warm humid and sub-humid ter with respect to seeds and fertilizers than with respect to crop areas, because the payoff to irrigation is lower in such areas than intensity, irrigation, organic fertilizer and pesticides. Malawi in more arid zones. What is surprising is that the cool and the appears to be a major puzzle for the BR framework, according to warm semi-arid tropics have such low irrigation levels, as here which we should have seen higher levels of all intensification the payoffs to irrigation are higher than in more humid areas. Irri- levels, rather than the very uneven pattern across them. gation, with the promise of a secure crop in the first season and a In terms of inputs, Ethiopia appears to have a more even perfor- crop in the second season, should long have been a favored invest- mance than Malawi. In Ethiopia, 53% of its farmers use organic fer- ment for farmers in these zones. Even if groundwater resources in tilizer, 41% use inorganic fertilizer, and 18% and 23% use improved Africa are less than in South and East Asia, for some farmers, they seeds and agrochemicals, respectively. Ethiopia has a strong agri- are still available. Many of these could probably have used bore- cultural extension system and also subsidizes fertilizer. In terms wells to install irrigation. of the BR intensification variables, Ethiopia conforms well to BR. That irrigation, even in the semi-arid and arid zones where pay- Niger does very well in terms of use of organic fertilizer too, at offs to irrigation are very high, is so low despite growth in popula- 48% of the farmers. This may be because in the arid areas cattle 34 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 Map 1. AEP/ha for the enumeration areas in of each of the six study countries. herding is very important and manure more easily available, while ated with the unavailability of significantly improved varieties of in Ethiopia it may be caused by the widespread use of animal draft. sorghum and millet in the Sahel. However, Niger’s chemical fertilizers are used only by 18% of farm- Forty-one percent of households in Nigeria use inorganic ers, and the use of improved seeds and agrochemicals are also very fertilizer and 34% use agro-chemicals. However, the use of organic low. The low use of improved seeds in Niger is likely to be associ- fertilizer is the lowest among the countries, at only 3%. This low H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 35 Map 2. Urban gravities for the six countries. use in the country with the lowest agro-ecological potential per In Uganda, the use of improved seeds is at 18%, while that of person is again inconsistent with BR. inorganic fertilizer is at only 3% of households. Organic fertilizer Tanzania’s use of the four inputs varies between 12% for agro- and agro-chemicals fall in between, at around 12%. Even though chemicals and 18% for improved seeds. That the use of these inputs its agro-ecological potential per person is far lower than for is low in the country with the highest agro-ecological potential per Tanzania, it is doing worse than Tanzania in terms of inputs, again capita is consistent with BR. a challenge for BR. 36 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 Table 2 Households’ characteristics. Source: Authors’ computation from LSMS-ISA surveys. Income variables computed in US$ at 2009 constant prices. ETH MWI NER NGA TZA UGA Total 1. Head’s age 43.0 44.5 51.2 48.5 45.8 48.8 2. Share of female head 0.2 0.1 0.1 0.2 0.3 0.2 3. Gross income from crop per ha (US$/ha) 500.5 179.6 1144.6 519.9 495.3 983.4 4. Gross household income = Ag wage + Non-ag. wage + Crop + 622.2 1235.7 1413.9 1072.8 1164.4 1333.6 Livestock + Self employment + Transfer (US$) 5. Gross income per capita (US$/pc) 130.99 181.19 234.87 188.54 192.78 227.42 6. Poverty headcount ratio below PPP $1.25/day (2005) 75.2 40.8 65.5 91.5 52.5 66.6 Data on income and consumption for ETH not available. As in Deininger, Xia, and Savastano income figures are doubtful for Nigeria where there are some data issues (Oseni et al., 2014) therefore the descriptive statistics should be interpreted carefully. Table 3 Land and fallow. Source: Authors’ computation from LSMS-ISA surveys. ETH MWI NER NGA TZA UGA Total 1. Area owned (ha) 1.2 0.68 4.5 1.1 2.41 1.8 1.3 2. Area operated (ha) 1.3 0.74 5.1 1.4 2.45 2.0 1.6 3. Gross cropped area (ha) 0.6 0.74 5.8 1.6 2.03 2.4 1.5 4. Net cropped area (ha) 0.3 0.67 4.9 1.3 1.95 1.0 1.1 5. Crop intensity 1.21 1.02 1.19 1.23 1.07 1.89 1.23 6. Proportion of current fallow NA 0.0 0.1 0.0 0.3 0.1 0.0 7. Proportion of past fallow in current fallow NA 0.01 0.03 NA 0.08 0.05 0.01 Fig. 2. Area operated, crop intensity and fallow. Table 4 Irrigation and technology by country. Source: Authors’ computation from LSMS-ISA surveys. ETH MWI NER NGA TZA UGA Total 1. Irrigated area (ha) 0.016 0.003 0.036 0.033 0.045 0.02 0.029 2. Dummy improved seeds 0.18 0.61 0.03 NA 0.18 0.18 0.09 3. Dummy inorganic fertilizer 0.41 0.76 0.18 0.41 0.16 0.03 0.38 4. Dummy organic fertilizers 0.53 0.16 0.48 0.03 0.17 0.12 0.25 5. Dummy agro-chemicals 0.23 0.03 0.07 0.34 0.12 0.11 0.27 6. Regression results to road and markets, and (b) on a range of agricultural intensi- fication variables. AEP and UG are exogenous to the conditions In this section, we report on (a) the estimates of the causal in the enumeration areas and, apart from issues of measurement impact of AEP/ha and UG on population density and distances error, should estimate causal links. As discussed, the regressions H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 37 Fig. 3. Input use and irrigation. under (b) estimate the total impact of AEP and UG on the Table 5 variables on the intensification variables, including the effect Population density and infrastructure. that goes via population density and market access. (1) (2) (3) For all variables, the individual observations are aggregated Log Pop. Log Dist. Log Distance to their mean at the EA level. There are 1993 EAs located in Dens. To Road to Mrkt six countries. The regressions are estimated in double log form, Log Value of AEP $/ha 0.056⁄ À0.146⁄⁄⁄ 0.001 and, apart from the two variables of interest, AEP, UG and their UGa 0.066 À0.309⁄⁄⁄ À0.061⁄ interaction, include only country dummies.24 By doing so, only Interaction Log UG and Log AEP À0.001 0.024⁄⁄⁄ À0.006 Country dummy ETH 0.393⁄⁄⁄ À0.274⁄⁄ À0.325⁄⁄⁄ the within-country variations are used to estimate the equations Country dummy MWI 0.289⁄⁄ À0.069 À1.960⁄⁄⁄ and differences in policies, and other country-specific factors are Country dummy NER À0.947⁄⁄⁄ À0.670⁄⁄⁄ À0.705⁄⁄⁄ therefore left out. Country dummy TZA À0.971⁄⁄⁄ À0.219⁄ À0.376⁄⁄⁄ In Table 5, the R-squares for the three equations population Country dummy UGA 0.508⁄⁄⁄ À0.472⁄⁄⁄ À0.935⁄⁄⁄ density and distances to road and markets are between 0.12 and Constant 4.092⁄⁄⁄ 3.453⁄⁄⁄ 4.292⁄⁄⁄ 0.14. Population density and road investments have responded Observations 1993 1993 1993 over the past to AEP, but not the distance to markets. In absolute R-squared 0.118 0.136 0.122 terms, the coefficient of AEP for distance to road is almost three Nigeria is the baseline for the country dummy. a times than that of population density. While road investment has UG: travel time negative exponential, with borders restriction to cities with 50. *** been responsive to AEP, market distance has not, suggesting that p < 0.01. ** p < 0.05. factors other than AEP determine investments in, or emergence * p < 0.1. of, markets. Urban gravity, on the other hand, does not affect population density, perhaps because the growth of urban areas has been too recent for population density to respond. But instead, it has a can be overcome using panel data with fixed effects, as done in strong impact on distance to roads, with an elasticity of 0.31, more Binswanger et al. (1993). than twice as high as that of AEP. Market distance is also reduced Table 6 looks at area farmed, crop intensity and current fallow. for EAs subject to more urban gravity, suggesting that market The R-squares or Pseudo R-squares vary between 0.16 for crop and investments respond to urban gravity. perennial area, to 0.77 for area under fallow. AEP does not influ- It is therefore clear that both population and public investment ence any of the five variables, while UG affects all, except for the in the past have responded significantly to AEP and UG, which is as fallow variable. Own area, cropped area and crop and perennial we expected. Therefore, cross section regressions explaining any area decrease with elasticities from À0.05 to À0.09, while crop intensification variable (or any other agricultural variable that intensity has a smaller absolute elasticity of 0.03. This is the only stems from a public or private decision), with population density variable for which the interaction term of UG and AEP is statisti- and infrastructure variables, will lead to upwardly biased coeffi- cally significant. The elasticity of AEP with respect to crop intensity cients of the independent variables. As discussed, the problem at the mean of AEP is only 0.003, but still statistically significant. Unless there are left out variables with opposing impacts on these 24 Square terms were insignificant for all dependent variables and therefore were variables, these total impacts suggest that they are unresponsive to dropped it from the regression. AEP in general, and therefore may also be unresponsive to 38 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 Table 6 Land areas and intensification. OLS Tobit Log Own Area Log Crop Area Log Crop and Crop intensity Proportion of land Perennial Area under current fallowb Log Value of AEP $/ha 0.016 0.006 À0.002 0.001 À0.002 UGa À0.086⁄⁄⁄ À0.054⁄⁄ À0.062⁄⁄⁄ 0.029⁄⁄⁄ 0.0005 Interaction Log UG and Log AEP 0.003 À0.001 0.000 À0.004⁄⁄⁄ À0.001 Country dummy ETH À0.067 À0.602⁄⁄⁄ À0.161⁄⁄⁄ À0.086⁄⁄⁄ Country dummy MWI À0.094⁄⁄⁄ À0.185⁄⁄⁄ À0.229⁄⁄⁄ À0.102⁄⁄⁄ 0.122⁄⁄⁄ Country dummy NER 0.761⁄⁄⁄ 0.801⁄⁄⁄ 0.761⁄⁄⁄ À0.019 0.131⁄⁄⁄ Country dummy TZA 0.291⁄⁄⁄ 0.166⁄⁄⁄ 0.128⁄⁄⁄ À0.090⁄⁄⁄ 0.295⁄⁄⁄ Country dummy UGA 0.238⁄⁄⁄ 0.121⁄⁄⁄ 0.197⁄⁄⁄ 0.205⁄⁄⁄ 0.248⁄⁄⁄ Constant 0.694⁄⁄⁄ 0.848⁄⁄⁄ 0.934⁄⁄⁄ 0.796⁄⁄⁄ À0.250⁄⁄⁄ Observations 1993 1993 1993 1993 1750 R-squared 0.256 0.320 0.159 0.158 0.771 Elasticity of AEP taking account of both À0.0032 the linear and the interaction term P-value 0.543 Elasticity of UG taking account of both 0.0241 the linear and the interaction term P-value 0.001 Nigeria is the baseline for the country dummy. a UG: travel time negative exponential, with borders restriction to cities with 50. b Information on Proportion of land under current fallow is NA in ETH. *** p < 0.01. ** p < 0.05. * p < 0.1. Table 7 Irrigation and technology variables, Tobit regression. Variables Tobit Regressions Probit Regression b Share of Land irrigated Share organic fertilizer Share inorganic fertilizer Share agro-chemicals Share of Improved seeds Log Value of AEP $/ha À0.054 0.030⁄⁄⁄ 0.071⁄⁄⁄ 0.048⁄⁄ 0.059⁄⁄⁄ UGa À0.169 À0.021 À0.027 À0.038 0.122⁄⁄ Interaction Log UG and Log AEP 0.021 0.000 0.001 À0.003 À0.019⁄⁄⁄ Country dummy ETH 0.457⁄ 0.946⁄⁄⁄ 0.182⁄⁄⁄ À0.237⁄⁄⁄ À0.720⁄⁄⁄ Country dummy MWI À0.302⁄ 0.450⁄⁄⁄ 0.455⁄⁄⁄ À0.731⁄⁄⁄ Country dummy NER À0.325 0.848⁄⁄⁄ À0.245⁄⁄⁄ À0.533⁄⁄⁄ À0.626g Country dummy TZA À0.022 0.312⁄⁄⁄ À0.484⁄⁄⁄ À0.590⁄⁄⁄ À0.750⁄⁄⁄ Country dummy UGA À0.444⁄⁄ 0.264⁄⁄⁄ À0.818⁄⁄⁄ À0.448⁄⁄⁄ À0.675⁄⁄⁄ Constant À0.879⁄ À0.465⁄⁄⁄ À0.081 0.070 Observations 1993 1993 1993 1993 1633 R-squared 0.0356 0.486 0.185 0.0917 0.0256 Nigeria is the baseline for the country dummy in all other regressions. a UG: travel time negative exponential, with borders restriction to cities with 50. b Regressions on Improved seeds does not include NGA as the variable is not available. MWI is the baseline in this case. *** p < 0.01. ** p < 0.05. * p < 0.1. population density. On the other hand, land areas decline with impacts for an impact via population density and market access. urban gravity, while crop intensity increases. On the other hand, except for the use of improved seeds, UG has In Table 7, the share of land irrigated is unresponsive to either very little to do with use of inputs. This stands in contrast to the AEP or UG and seems to be determined by other factors, such as significant impact of UG on cropping intensity. availability of canal or groundwater. However, AEP has a signifi- cant impact on all four technology variables, with the largest elas- 7. Summary and conclusions ticity of 0.07 for inorganic fertilizer and the lowest one at 0.03 for organic fertilizer. As discussed in the introduction, this means that 7.1. New measures of agro-ecological potential and of urban gravity the regressions are not inconsistent with the BR predictions. Note also that the results are consistent with our constructed AEP mea- This is the first paper to develop internationally comparable sure being a valid proxy for the ‘‘true” underlying AEP. measures of agro-ecological potential and urban gravity. These The interpretation of these finding is that higher input use has measures impact positively on population densities, public invest- significantly higher payoffs in areas of high AEP than of low AEP. ments in road and markets, and on some indicators of agricultural This, of course, is well known, but it is interesting to see that our intensification. AEP variable and the household data can capture this effect. The We find that AEP per person ranks countries quite differently estimated coefficients suggests that there is room in these total than with respect to AEP per ha. The AEP/ha of Uganda is by far H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 39 the highest among the countries, the lowest being Niger, with Tan-  AEP increases population density and road investment, but not zania close to the average across countries. However, in terms of distance to markets, while UG does not affect population den- AEP per rural person, this is the highest in Tanzania, followed by sity, but reduces both the distance to roads and to markets. Niger, and then only Uganda. The lowest potential per person is  AEP has no impact at all on key characteristics of the farming in Nigeria. These reversals of the measures of potential arise system, such as areas farmed, crop intensity and fallow areas, because of the sharply different population densities in the while UG reduces all area measures and increases cropping countries. intensity.  While neither AEP nor UG have an impact on irrigation invest- ment, AEP affects the use of all four inputs, while UG only 7.2. Descriptive results increases the use of improved seeds. Given the rise in population pressure in all these African coun- We have provided a few hints as to why the response patterns tries, the improvements in infrastructure and the growing urban with respect to AEP and UG differ so significantly, but a full under- demand land use intensity, consistent with BR predictions, has standing will undoubtedly require more sophisticated research reached permanent cropping in all of the countries. Fallow areas approaches. In terms of testing BR with respect to UG, we see that have virtually disappeared. Under permanent agriculture, high it increases crop intensity and improved seeds, but not the other doses of organic and inorganic fertilizers are required to maintain technology variables, which does not provide much support for or restore the soil nutrients taken out by the plants. Except for the operation of the BR predictions in Africa. Malawi and Ethiopia, the proportion of households using chemical fertilizers is clearly too low to do so. Nor is this compensated for by 7.4. Implications the high proportion of households using organic fertilizer, which is relatively high only in Niger and Ethiopia. The BR theory also pre- The facts described in this paper are only partially consistent dicts that, under pressure from population growth and market with the BR framework. In particular, and in line with other find- access, irrigation investment and other modern technologies ings in the literature, the use of organic and chemical fertilizers, would be used more intensively to increase yields. However, these except perhaps in Malawi and Niger, appears far too low to main- factors did not trigger significant irrigation investments, even in tain soil fertility. Except for Ethiopia, this also applies to the use of semi-arid areas where the payoff to irrigation is high. Unfortu- organic fertilizers. In addition, investments in irrigation also seem nately, we do not have data on other land investments, or on mech- to fall far short of what the high population densities and signifi- anization, to judge whether expected intensification responses cant market access would require. This last finding is consistent have occurred with respect to these important investments. How- with Headey and Jayne (2014), who stress that other investments, ever, the descriptive analysis suggests that the BR impacts of pop- such as mechanization, also have responded inadequately to rising ulation pressure and market access have triggered an inadequate population pressure. The implication of these results, and of the response of the farming systems with respect to irrigation and observations of many other observers of African agriculture, is that technology use. the process of intensification over much of these African countries An additional inconsistency arises when comparing Tanzania appears to have been less beneficial to farming systems and farm- with Malawi, with Tanzania having about 2.4 times the AEP/person ers than what could have been expected according to the BR as Malawi. Yet, cropping intensity is about the same and so is the hypothesis. intensity of use of manure. Use of agro-chemicals is more prevalent in Tanzania than in Malawi. The only area where Malawi has greater intensity of input use than Tanzania is in the use of inor- Acknowledgement ganic fertilizers and improved seeds. In addition to being triggered by the forces of intensification, these higher uses are consistent We are grateful to the team of the World Bank project ‘‘Agricul- with the long-standing effort of Malawi to increase the use of these ture in Africa– Telling Facts from Myths” for helpful comments two factors, including the significant subsidies that have been pro- and, in particular, to Luc Christiaensen for his continuous support, vided again in recent years. and Siobhan Murray for computing the agro-ecological potential As stressed all along, while the descriptive analysis can and the urban gravity variables. We thank two anonymous review- uncover apparent inconsistencies of cross-country patterns with ers for their comments, which helped at improving the quality of the BR framework, the descriptive analysis provides no rigorous the paper. tests of it. First of all, there are variations in soils, crops and other biological variables that are likely to have a significant Appendix A. Supplementary material impact on the degree of intensification. These have been ignored so far. In addition, there are sharp differences in policies and Supplementary data associated with this article can be found, in infrastructure investments that have not been taken into the online version, at http://dx.doi.org/10.1016/j.foodpol.2016.09. account. It is therefore important that the theory be tested with 021. panel data, where these other variations can be aggregated into fixed country effects. References Baltenwick, I., Staal, S., Ibrahim, M.N.M., Herrero, M., Holdermann, F., Jabbaar, M., 7.3. Regression analysis Manyong, V., Thornton, Patil P., Willioams, T., Waithaka, M., de Wolff, T, 2003. Crop Livestock Intensification and Interactions Across Three Continents: Main Report. International Livestock Research Institute, Nairobi, Kenya. We found significant responses of population density and Binswanger, Hans P., McIntire, John, 1987. Behavioral and material determinants of infrastructure, farming systems characteristics, farm technology production relations in land-abundant tropical agriculture. Econ. Dev. Cult. and profits per ha to our measures of AEP and UG, and the signs Change. University of Chicago Press 36 (1), 73–99. are all according to expectations. However, there is a sharp divide Binswanger, Hans P., Khandker, Shahidur R., Rosenzweig, Mark R., 1993. How infrastructure and financial institutions affect agricultural output and between the nature of the impact of AEP and UG across the investment in India. J. Dev. Econ. 41, 337–366. North-Holland. variables: Boserup, Ester, 1965. The Conditions of Agricultural Growth. Aldine, Chicago. 40 H.P. Binswanger-Mkhize, S. Savastano / Food Policy 67 (2017) 26–40 Deichmann, U., 1997. Accessibility Indicators in GIS. Department for Economic and Oseni, G., Corral, P., Goldstein, M., Winters, P., 2014. Explaining Gender Differentials Social Information and Policy Analysis, United Nations Statistics Division, New in Agricultural Production in Nigeria. The World Bank, Policy Research Working York. Paper Series: 6809. Gallup, John Luke, Sachs, Jeffrey, Mellinger, Andrew, 1999. Geography and economic Otsuka, K., Place, F., 2014. Changes in Land Tenure and Agricultural Intensification development. Int. Reg. Sci. Rev. 22 (2), 179–232. in Sub-Saharan Africa. WIDER Working Paper 2014/051. Geertz, Clifford, 1963. Agricultural Involution. University of California Press, Pender, John, Place, Frank, Ehui, Simeon (Eds.), 2006. Strategies for Sustainable Land Berkeley, Los Angeles, and London. Management in the East African Highlands. IFPRI, Washington, DC. Hayami, Yijiro, Ruttan, Vernon, 1985. Agricultural Development: An International Pingali, Prabhu, Bigot, Yves, Binswanger, Hans P., 1987. Agricultural Mechanization Perspective. Baltimore. The Johns Hopkins Press, 1971 (1st ed.) and 1985 (2nd and the Evolution of Farming Systems in Sub-Saharan Africa. Johns, Baltimore, ed.). Md. Headey, D., Jayne, T.S., 2014. Adaptation to land constraints: is Africa different? Ruthenberg, Hans P., 1980a. Farming Systems in the Tropics. Clarendon, Oxford. Food Policy 48 (October), 18–33. Ruthenberg, Hans P., 1980b. Farming Systems in the Tropics. Clarendon, Oxford. Henderson, J. Vernon, Storeygard, Adam, Weil, David N., 2009. Measuring Economic Tiffen, M., Mortimore, M., 1992. Environment, population growth and productivity Growth From Outer Space, Working Paper 15199. National Bureau of Economic in Kenya: a case study of Machakos District. Dev. Policy Rev. 10, 359–387. Research, Cambridge, Massachusetts. University Press. . Ecological Zones (GAEZ v3.0), GAEZ Data Portal, User Guide, Laxenburg, Austria. Kilic, Talip, Zezza, Alberto, Carletto, Calogero, Savastano, Sara, 2013. Missing(ness) Turner, B.L., Hyden, G., Kates, R.W. (Eds.), 1993. Population Growth and Agricultural in Action: Selectivity Bias in GPS-based Land Area Measurements. Policy Change in Africa. University Press of Florida, Gainesville. Research Working Paper Series 6490, The World Bank. Turner II, B.L., Fischer-Kowalskic, Marina, 2010. Ester Boserup: an interdisciplinary Kiszewski, Anthony, Mellinger, Andrew, Spielman, Andrew, Malaney, Pia, Sachs, visionary relevant for sustainability. Proc. Natl. Acad. Sci. 107 (51), 21963– Sonia E., Sachs, Jeffrey A., 2004. A global index representing the stability of 21965. malaria transmission. Am. J. Trop. Med. Hyg. 70 (5), 486–498. World Bank, 2008. Agriculture for Development, World Development Report, 2008, Lele, U., Stone, S., 1989. Population Pressure, the Environment and Agricultural Washington DC. Intensification. MADIA Discussion Paper 4. World Bank, Washington, DC. World Bank, 2009. Living Standard Measurement Study (LSMS) Surveys. LSMS – McIntire, J., Bourzat, D., Pingali, P., 1992. Interaction in Sub-Saharan Africa. World Integrated Surveys on Agriculture (LSMS–ISA). . Bank Regional and Sectoral Studies. World Bank, Washington.