Predicting minimum tillage adoption among smallholder farmers using micro-level and policy variables

Motivation

Due to dwindling arable land and ongoing unfavourable climatic changes, the need for sustainable intensification of agriculture has gained considerable urgency (The Royal Society 2009). This has increased interest in climate-smart agriculture (CSA), redirecting agricultural development in ways that recognize the present realities of climate change (Lipper et al. 2014). FAO defines CSA as agriculture that sustainably increases productivity, enhances resilience (adaptation), reduces greenhouse gases (GHGs mitigation) where possible and enhances the achievement of national food security and development goals (FAO 2013). This context has generated a growing interest in conservation agriculture whose key agronomic principles are as follows: minimization of soil disturbance; continuous maintenance of soil cover; and crop diversification (by rotations, intercropping or other crop associations) among other practices. Conservation agriculture is thereby posited to conserve soils and in the long run to improve soil properties, conserve moisture, stabilize and even enhance crop yields (Hobbs et al. 2008; FAO 2012). The principles of resource conservation and the envisaged outcomes are consistent with the basic notions of CSA. Around the globe, the adoption of conservation agriculture has been successful on larger farms in Australia, Canada, the USA and Latin America—most notably Brazil and Argentina—whereas in Africa where some evidence of adoption has been offered, these have been largely modest when measured in terms of widespread diffusion, for example in Ghana, India, Pakistan and Bolivia (Ekboir 2010).

The productivity and environmental benefits of the minimum- and no-till aspects of conservation agriculture have been variously documented in the literature (Erenstein 2003; Hobbs 2007). Minimum tillage (or no-till) is a seedbed preparation method that involves minimizing the number of tillage operations by making only limited slots for placing the seed, either by using specialized machines (seeders) that open up small furrows in which seed is placed or by manually using ‘jab planters’, hoes or dip sticks to bore holes into which seeds are placed. In such cases, there is minimal soil disturbance since the rest of the field is left untilled. Mulching typically involves retaining crop residues and other plant materials on the surface of the plot to cover as much soil surface as possible. The mulch protects the soil from erosion, sun and wind, and the superficial biomass enhances infiltration and provides nutrition to the biota that aerate the soil. Erenstein (2003) enumerated the beneficial impacts of mulching on soil conservation, soil ecology, crop yields and the environment. If implemented properly, the combination of minimum tillage (or no-till) and mulching can alleviate or even halt soil erosion and lead to water conservation and reduced runoff, groundwater replenishment and enhancement of soil biota. In some cases, yield stabilization and increases have also been observed (Erenstein 1999, 2003).

Conservation agriculture for smallholder farmers raises challenging questions regarding its adoption, diffusion and scaling at farm level and beyond. The multiple component technologies that need to be applied simultaneously are one complicating factor. Here, we focus on the subset of minimum tillage combined with mulching (MTM).¹ MTM is likely to face impediments to adoption associated with the inherent substitution of practices, e.g. seasonal labour scarcities for weeding, given that absence of herbicides can make weed-management more labour-demanding in minimum-tillage systems (Nyamangara et al. 2014). Other constraints relate to the opportunity cost of crop residues for feed rather than mulch (Pannell et al. 2014, Jaleta et al. 2013, 2015), and possible short term yield depression in the initial years of implementing no-till and mulching (Pannell et al. 2014). Smallholders often have short-term planning horizons due to high discount rates and are often unwilling to accept yield penalties for a few years to improve soil conditions and yields in later years (Pannell et al. 2014; Erenstein 2003).

Objectives

The objective of this paper is to examine the contribution of micro-level household and policy variables in conditioning the simultaneous adoption² of two important conservation agriculture component technologies: minimum tillage combined with mulching (hereinafter, MTM). The first hypothesis we test is that MTM adoption is conditioned by the characteristics of the smallholder farm, particularly its demographic characteristics, plot characteristics and market and credit access. The second hypothesis is that MTM adoption is also conditioned by macro-level policy variables beyond the farm. We therefore include multiple countries (Ethiopia, Kenya, Malawi and Tanzania) so as to include country-level policy variables obtained from secondary sources: (1) extension personnel-to-farmer ratios (EFR) and (2) subsidy expenditure as a percentage of agricultural budgets spent on farm input subsidies (SER).

The four study countries (Ethiopia, Kenya, Tanzania and Malawi) span a north-south gradient in eastern and southern Africa. They were purposively selected for this study and the overarching research-for-development project that collected the data used in this study. The project was implemented by the International Maize and Wheat Improvement Center in collaboration with national research partners in the four countries. The socioeconomic component of the project collected data based on household surveys intended for use in analysing socio-demographic factors that could be associated with adoption or non-adoption of conservation agriculture-based practices such as MTM.

The rest of this paper is organized as follows: the rest of the introductory section outlines the background on the four study countries with focus on policy issues on agricultural extension and subsidies, followed by a brief overview of the history of conservation agriculture and ending with an outline of the possible drivers of MTM adoption and diffusion in each country. The introductory section is followed by a section that offers a detailed description of the sampling and data-collection methods, the variables and the analytical approach frameworks. The ‘Methods’ section is followed by a presentation of the results from a probit estimation using the pooled and individual country samples as well as those from the post-estimation simulations. The final section concludes this paper by summarizing the key findings and stating the policy implications and suggestions for future research.

Background of study countries: extension and subsidies

In the eastern and southern Africa (ESA) region, agricultural extension services were somewhat neglected in the years after the 1980s, largely as a consequence of structural adjustment programmes (Rivera and Alex 2004; Pye-Smith 2012). In years since that period, there is now some renewed policy focus on agricultural extension. In 2010 for example, Ethiopia was estimated to be on track to have one of the most favourable ratios of extension-personnel-to-farmer ratios in Africa (projected to rise to 21 per 10,000 in short order (Davis et al. 2010)). This compares favourably with the recommendation in Pye-Smith (2012): a ratio of 33 extension agents for every 10,000 farmers was reported as the minimum ratio needed for an effective extension system. In Kenya, smallholder farmers have traditionally benefited from the government extension system supplemented by commodity-based programmes run by parastatal and private agribusiness corporations and farmer cooperatives. However, government budget allocation for extension services declined from 6% of the annual government budget in the 1980s to only 2% (in real terms) in the years since then (Muyanga and Jayne 2006). A new extension policy is now being implemented (Government of Kenya, 2012), reflecting a re-focus on extension. In Tanzania, like the rest of the study countries, agricultural extension services are provided by the public sector through the Ministry of Agriculture. Recently, the National Agriculture Policy of 2007 stated the government’s intention to increase the number of extension workers by about 15,000 (Daniel 2013). The government extension services also remain the major provider of extension services in Malawi (Masangano and Mthinda 2012). In 2000, a white paper with a bottom-up and participatory focus for planning was published to guide the extension system in the country (Masangano and Mthinda 2012).

In recent years, there has been a return of fertilizer subsidies in sub-Saharan Africa (SSA) after a period of their absence in the wake of the structural adjustment programmes of the 1980s and 1990s. For example, Malawi spent about 72% of its agricultural budget in 2008/2009 on agricultural input subsidies (Dorward and Chirwa 2010). The subsidized distribution of fertilizer and improved seeds (maize and beans) at 90% subsidy has played a big role in the wider use of these inputs in the country. In 2003, the government of Tanzania reinstated the fertilizer transport subsidy, and this was still operational in 2007 and in 2008 through the National Agricultural Input Voucher Scheme (NAIVs) (Pan and Christiaensen 2011). Although Kenya has a long history of relying on the private sector as a key driver in the fertilizer and seed markets, in 2013, the Kenyan Government introduced a fertilizer subsidy programme administered by the state cereal marketing board. In Ethiopia, the agricultural input sectors are mainly controlled by the state. Since 1994, aiming to reach many poor farmers who do not have cash to pay for inputs (fertilizer and seed), the government has been providing credit guarantees to help lower trade-financing costs and the national extension programme is often the major buyer and distributor of seeds and fertilizer in order to boost supplies of these (Kherallah et al. 2002).

Background of conservation agriculture

The history of conservation agriculture in the four countries suggests that its promotion is variously attributable to agricultural development projects (Milder et al. 2011). The emergence of conservation agriculture as a component of sustainable agriculture in ESA can be traced in the literature to the 1990s. At the 1996 World Food Summit, the Soil Fertility Initiative was launched, followed by the Better Land Husbandry approach. In 1998, a conservation tillage workshop was held in Zimbabwe, and in 2000, the African Conservation Tillage network was formed (Bishop-Sambrook et al. 2004). One of the early efforts to introduce minimum tillage in Ethiopia was done by Sasakawa-Global (SG2000) in South Achefer district (Matsumoto et al. 2004). In Kenya, conservation agriculture has been formally promoted since 1998 under the Kenya Conservation Tillage Initiative (KCTI), and by 2005, KCTI had projects in five districts in the country. The 3rd World Congress on Conservation Agriculture was held in Kenya, and the government (represented by the vice-president of Kenya at that time) expressed its commitment to conservation agriculture in its strategy to revitalize agriculture (ACT 2008 p. 9).

Interest in and research on conservation agriculture in Malawi date back to the 1980s at Bunda College (Mloza-Banda 2002). This interest has been sustained over the years, as evidenced by wide-ranging literature such as Munthali et al. (2008) on socioeconomic factors that affect adoption of conservation practices, Ngwira et al. (2013) on the estimated benefits of conservation agriculture for smallholders in terms of economic and biophysical outcomes, Ngwira et al. (2012) on the evaluation of conservation agriculture-based maize-legume intercropping systems and Kaczan et al. (2013) which provides a review of conservation agriculture in Malawi. In Tanzania, Marietha et al. (2011) report on a joint programme in 2004 run by the German Ministry of Agriculture and FAO-supported conservation agriculture practices in Northern Tanzania—one of the earliest concerted efforts to promote conservation agriculture in Tanzania. The project used farmer field schools as entry points for extension and farmer education on conservation agriculture, and the private sector was encouraged to participate by retailing and developing custom hire services for conservation agriculture equipment.

Possible drivers of MTM adoption and diffusion

At the farm level, an oft-cited advantage of MTM is that it lowers the costs involved in land preparation (Fowler and Rockstrom 2001). For example, reduced tractor and fuel costs have been a major driver of adoption in mechanized large-scale systems, especially if herbicides are available to manage increased weed pressure (Erenstein et al. 2012). There is mixed evidence regarding yield increases due to reduced or minimum tillage (Giller et al. 2009). The evidence on mulching is also mixed, being complicated by the critical trade-offs between mulch, feed and other uses (Jaleta et al. 2013, 2015). In Morocco, Magnan et al. (2012) calculated the opportunity costs of crop residue in no-till systems and found that the shadow value (its value in livestock feeding as opposed to soil mulch) was 25% of the total value of the crop produced during normal rainfall and 75% during drought (when crop residues become most valuable as animal feed). Similarly, a study by Valbuena et al. (2012) comparing conservation agriculture practices in South Asia and sub-Saharan Africa showed that the use of crop residues for mulching was most feasible in ‘high-potential areas’ where biomass production levels are likely to be high enough to meet the demands of both livestock and mulching.

The implementation of conservation agriculture can represent significant learning costs for farmers who can ill afford extended periods of trial and perhaps error. For resource poor farmers, the financial resources needed for complementary external inputs such as herbicides or fertilizers to implement MTM properly, and the increased labour needed for weed management, can still represent serious challenges if their main resource is family labour. In situations where opportunities for accessing credit or earning cash incomes are limited, and where MTM requires significant commitment of learning time, cash or labour, adoption may prove sluggish (Jack 2013).

Data and data sources

This study focuses on maize-legume-based production systems, the prevailing system across the four countries. The study regions were selected to generally coincide with the dominant smallholder maize production areas identified as important for smallholder productivity growth and poverty reduction by the national agricultural development authorities. The study areas were part of a multidisciplinary project designed to test various options for encouraging the diffusion of conservation agriculture-based practices in the smallholder farming systems of eastern and southern Africa. The project focused on (among other things) generating information on maize-legume value chains and policies and to inform farmers, extension agencies, non-governmental organizations and agribusinesses along the maize-legume value chain about opportunities for the adoption of conservation agriculture based practices leading to sustainable agricultural intensification.

In Ethiopia, the survey was carried out in the maize-legume-based farming systems in three regions of the country: SNNP,⁴ Benshangul and Oromia regions. Multi-stage sampling was employed to select households to be included in the survey. In the first stage, 39 districts were purposively selected based on the importance of maize. In the second stage, 69 Peasant Associations (PAs) were randomly picked from a list of PAs in each district. In the final stage, a total of 2187 households were randomly selected from the various PAs, with the number of households selected from each PA proportional to the number of the households in the PA as per the latest official national population census figures.

In Kenya, two districts were selected from western Kenya (Bungoma and Siaya) and three districts from eastern Kenya (Embu, Meru South and Imenti South). Multi-stage sampling was employed to select lower-level sampling clusters: divisions, locations, sub-locations and villages. In total, 30 divisions were selected—17 from western Kenya and 13 from eastern Kenya. In total, 613 households were included in the Kenya survey with the two regions being assigned approximately 300 households each. Efforts were made to ensure representativeness of the sample depending on the population of the sampling units. Proportionate random sampling was designed to select divisions from each district, sub-locations from each division, villages from each sub-location and number of households from each village. Therefore, the households in a region were distributed across the districts according to the census of farm households per district.

In Malawi, purposive sampling was also used in the first stage to select regions of the country where smallholder maize-farming was important. Stratified sampling was used to select six districts, five in the central region (Lilongwe, Kasungu, Mchinji, Salima and Ntcheu) and one (Balaka) in the south. Eventually, 64 Extension Planning Areas (EPA’s), 89 sections and 235 villages were selected using multi-stage random sampling combined with probability to proportional size. Using the same process, 1870 households from the 235 villages were selected for this study.

In Tanzania, the survey targeted two maize-legume-based farming systems: Kilosa and Mvomero in the eastern zone and Mbulu and Karatu districts in the northern zone. These districts were purposively selected, followed by multi-stage random sampling to arrive at a total sample of 60 villages, and using probability proportional to size, a final sample tally of 681 households was considered in the survey. From the four countries, the total data set comprised 5356 farm households and 11,188 maize plots from 700 villages in 43 districts. Figure 1 shows the location of the study districts across the four countries.

Variables included in the adoption model

When deciding whether or not to implement MTM, the initial decision can be construed as making a binary choice. The subsequent decisions such as the area on which MTM is implemented would be seen as continuous. Several elements in this process condition the final choice: the environment in which farmers operate; observed household and farm characteristics (e.g. gender, educational attainment, age, plot characteristics); unobserved attributes (risk attitudes, motivation, etc.), among others (Kassie et al. 2015; Marenya and Barrett 2007). We used a set of farm-household and environment indicators in a probit model to estimate the factors that affect plot-level adoption in the study sites. We describe below the groups of variables and the hypotheses informing their use in the adoption model. The variables range from self-reported assessments of plot-level characteristics to household demographics such as education level of the household head and main income activities. Finally, we explain the two policy variables included in the adoption model and used in the simulation framework.

Policy variables

Greater investment in agricultural extension to increase the availability of information and subsidies designed to relieve financial and credit constraints to adoption are two policy strategies that can be used to encourage widespread adoption and upscaling of MTM and associated practices. The consensus in much of the literature is that carefully designed subsidies that lower the overall cost of inputs can enable liquidity-constrained farmers to overcome short-term financing constraints, and make it easier for them to use herbicides, fertilizer and hire equipment to implement MTM. The key design principles of a sustainable subsidy policy that could help in achieving these objectives will depend on careful targeting and reliance on existing input-market channels as much as possible (Smale et al. 2011).

Policy simulations

Table 1 shows the EFR and subsidy expenditure as a percentage of agricultural budgets (SER) in each of the four countries. We chose data for the year 2010 because the household data were also collected during 2010. In cases where the data for EFR and SER for 2010 were not available, we chose the data for the year nearest to 2010. In terms of EFR—measured as the number of frontline extension staff per 10,000 farmers—Ethiopia had the highest EFR at 16, followed by Kenya (10), Malawi (6) and Tanzania (4). Malawi spent the most on input subsidies (58.9% of the whole agricultural ministry budget) between 2009 and 2011, compared to Ethiopia (10%), Kenya (19%) and Tanzania (46%) of total agricultural ministry budgets.

Household variables descriptive statistics

Regarding the demographics of the farming population, this sample shows that on average, a Kenyan farmer was older (50 years) compared to an average Tanzanian (45 years), Ethiopian (43 years) and Malawian (42 years) farmer. The Kenyan farmer had on average 7.5 years of formal schooling compared to 3.0 in Ethiopia, 5.7 in Malawi and 5.4 in Tanzania. Only a minority of households had any non-farm sources of income: 23% in Kenya, 13% in Malawi, 6% in Tanzania and 5% in Ethiopia. The average household size ranged from five members in Malawi to seven in Ethiopia. As expected, households in Ethiopia had the highest livestock numbers (6.24 TLUs or tropical livestock units) and Malawi had the lowest (0.72 TLUs). Farm size cultivated during the major agricultural season was the smallest in Kenya (1.32 ha) and the highest in Tanzania (4.7 ha) followed by Malawi (3.4 ha) and Ethiopia (2.6 ha). The value of non-livestock assets was highest in Ethiopia ($883) and lowest in Tanzania ($152). The credit-constraint data show that, typically, about 50% of households reported needing credit and not finding it. In Ethiopia, Kenya and Malawi, 56, 45 and 49% of households, respectively, were credit constrained. In Tanzania, relatively few households reported needing credit and not finding it. Overall, 25% of the farmers belonged to various types of farmers’ groups: approximately 20% in Ethiopia, 20% in Kenya, 39% in Malawi and about 25% in Tanzania.

Results from adoption model

Results from policy simulations

In this section, we present the results of five simulations involving extension-to-farmer ratio and subsidy expenditure as a percentage of agricultural budgets. The aim is to test how an increase in these two indices would affect the predicted probability of adoption all else being equal. As explained in the section above that describe the policy variables, this is followed by varying the level of the other policy as well as that of credit availability. The five policy simulation results are as follows (See also Appendix Tables 3 and 4):

1. 1a)

   Increasing the individual country’s extension-to-farmer ratio (EFR) to Ethiopian level: The first simulation of the impact of EFR on the predicted probability of CA adoption is reported in Fig. 2. In Kenya, the probability of adoption increases from 3.9% (base level) to 6.5% when EFR is increased from 10 to 16 (Ethiopian level). Similarly, in Malawi and Tanzania, the probability of adoption increased from about 3.4 to about 5%, and in Tanzania from 10 to 21.4%.

    
2. 1b)

   Reducing extension (EFR) but increasing subsidies (SER) instead: In Fig. 2, we report simulation results of what would happen if the extension were reduced (by setting the EFRs of the other three countries at Tanzanian level) and SER simultaneously increased to Malawi’s rate of 58.9. The results suggest the powerful impact of subsidy expenditure on the probability of adoption; despite reducing EFR in Ethiopia by 75%, the probability of adoption increased by approximately 4% (from 26 to 30%). The result for Malawi provides a ‘counterfactual’ in this case. When the SER remained the same and the EFR was reduced to Tanzanian level, the probability of adoption reduced in Malawi’s case from about 34 to 31%. For the other three countries’ simulations, the increased SER appears to more than compensate for the reductions in EFR, so that probability of adoption increases despite reductions in EFR (Fig. 2).

    
3. 1c)

   Increasing extension (EFR) with no credit availability: In these simulations (Fig. 2), the compensatory effect of high extension with a lack of credit is demonstrated. This was achieved by setting the EFR at the highest (Ethiopian) level and making credit unavailable for all (100%) of the farmers. The results show that in all cases (except Ethiopia), predicted adoption increased from base levels by approximately 2% in Kenya, 13% in Malawi and 9% in Tanzania. The decrease in the probability of adoption in Ethiopia provides a useful benchmark for demonstrating the effect of credit constraint on the probability of adoption: the likelihood of adoption fell from 26 to 18% when all the households did not have credit. In the other three countries, the lack of credit was somewhat compensated for by an increase in EFR, while in the Ethiopian simulation, the EFR remained unchanged as credit availability was eliminated. These results indicate that increases in the reach of extension systems and availability of information can make up for lack of credit in fostering some level of adoption even though in themselves, these effects are modest, independent of other factors.

    
4. 2a)

   Changing an individual country’s input subsidies (SER) to Malawian level: In Fig. 3, the results show that setting SER at the Malawian level increases probability of adoption by more than 100% in Ethiopia (from 26 to 57%) and Kenya (from 4 to 14%) and by about 40% in Tanzania (10 to 14%). These strong and significant impacts of SER and the earlier reported results for EFR are indicative of the potential that still exists for better information delivery and cost reduction in encouraging MTM adoption.

    
5. 2b)

   Increasing subsidies (SER) with no credit availability: In the simulations that assumed no credit was available, but with a compensatory increase in subsidy at the highest (Malawian) level (also Fig. 3), the probability of adoption increased in all cases except in Malawi where credit availability was eliminated but SER was unchanged, leading to a reduction in probability of adoption from 34 to 31%. The high levels of SER more than made up for the lack of credit. As a suggestion for policy or future research, a more specific policy would provide subsidies for the adoption of MTM, especially as the adoption of MTM can be considered to infer social benefits in terms of better environmental impacts such as soil loss mitigation and reduction of pollution from soil erosion.