Food Policy xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Food Policy
journal homepage: www.elsevier.com/locate/foodpol

Can small farms benefit from big companies’ initiatives to promote

mechanization in Africa? A case study from Zambia
Ferdinand Adu-Baffour, Thomas Daum , Regina Birner
⁎

Hans-Ruthenberg-Institute of Agricultural Science in the Tropics, University of Hohenheim, Wollgrasweg 43, 70599 Stuttgart, Germany

ARTICLE INFO ABSTRACT

Keywords: After years of neglect, there is a renewed interest in agricultural mechanization in Africa. Since government
Agricultural mechanization policy initiatives to promote mechanization are confronted with major governance challenges, private-sector initiatives
Agricultural intensification may offer a promising alternative. However, given limited scientific studies on such private-sector options such
Private business approaches are often viewed skeptically. One concern is that multi-national agribusiness companies take ad-
Zambia
vantage of smallholder farmers. Another concern is that mechanization causes rural unemployment. To shed
Employment effects
light on these concerns, this paper analyzes an initiative of the agricultural machinery manufacturer John Deere
to promote smallholder mechanization in Zambia through a contractor model. The analysis focuses on the
impact of this initiative on farmers who receive tractor services using Propensity Score Matching. The results
indicate that farmers can almost double their income by cultivating a much larger share of their land. The
analysis suggests that the increased income is used for children’s education and more food, but does not result in
increased food diversity. The demand for hired labor increases due to land expansion and due to a shift from
family labor, including that of children, to hired labor. Questions that require further investigation are identified,
including strategies to incentivize tractor owners to provide services, to also increase land productivity, and to
avoid new forms of dependency of agricultural laborers that may result from a shift in the timing of the labor
demand.

1. Introduction There were substantial efforts to promote mechanization in Africa’s

agriculture in the 1960s and 1970s, but these efforts were state-led and
During the last decade, agriculture has emerged as a top priority on they largely failed (Pingali, 2007). This negative experience led to a
Africa’s development agenda. Even though there is some new scope for subsequent neglect of agricultural mechanization in development ef-
large-scale farming, especially in the land abundant countries on the forts, except for some efforts to introduce animal traction. Likewise,
continent (Deininger and Byerlee, 2012), smallholder farming systems research on the mechanization of smallholder farming systems in Africa
will have to play the key role for agricultural development in Africa became a rather neglected field in the 1990s and 2000s (Diao et al.,
(Birner and Resnick, 2010; World Bank, 2007; Davis et al., 2017). Al- 2012). Research conducted in the 1990s had shown that machinery has
most 70% of the farms in Sub-Saharan Africa operate less than two an important role to play in improving farmers’ crop management
hectares (Deininger and Byerlee, 2011: 28) and they typically do not practices, especially by allowing for better tillage, weed control and
realize more than 25% of their potential yields (Deininger and Byerlee, moisture management (Anderson and Dillon, 1992; Byerlee and Husain,
2011: xxxviii). Substantial efforts have been made to close this yield 1993). The institutional dimension of mechanization had always re-
gap, but in recent years, there has been an increasing recognition that it mained a neglected field of research, in spite of overwhelming historical
also important to increase the labor productivity in African agriculture evidence that institutions such as rental markets and cooperative ex-
in order to reduce poverty (Diao et al., 2018). In most countries of change have played a key role in the history of the countries that are
Africa, population density is relatively low, and the theory of induced now industrialized. As shown by Olmstead and Rhode (1995) for the
innovation would predict that mechanization should play an important case of the USA, such institutions were essential to facilitate the access
role in the early phases of agricultural development (Hayami and of smallholder farmers to mechanization.
Ruttan, 1985). Yet African farming systems remain the least mechan- Following the food price crisis of 2008, there has been a renewed
ized of all continents (Pingali, 2007; Sheahan and Barrett, 2017). emphasis on agricultural development as a top priority in Africa’s

⁎
Corresponding author.
E-mail address: thomas.daum@uni-hohenheim.de (T. Daum).

https://doi.org/10.1016/j.foodpol.2019.03.007
Received 29 May 2018; Received in revised form 21 March 2019; Accepted 24 March 2019
0306-9192/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).

Please cite this article as: Ferdinand Adu-Baffour, Thomas Daum and Regina Birner, Food Policy,
https://doi.org/10.1016/j.foodpol.2019.03.007
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

development agenda. This new interest in agriculture has also revived business models that allow smallholder farmers to access tractor ser-
the interest in agricultural mechanization (FAO and UNIDO, 2008; vices. The approach is to support small and medium-sized enterprises
Kienzle et al., 2013; Mrema et al., 2008). Governments in several and “emerging farmers” that is medium-size farmers who own between
African countries subsidize the provision of tractor services, often by approximately 20 and 200 ha and can afford to purchase a tractor. In
importing tractors that are then provided at subsidized prices to private our sample, their median farm size was 66 ha. The main form of support
sector operators who are expected to provide tractor services to is facilitating the financing of the tractor. For 10 of the 21 tractor
smallholder farmers. A study of such a subsidy scheme in Ghana found owners in our sample this has been their first tractor. This has been
that it was not a viable business model for private tractor service pro- done through different mechanisms since 2010, including a loan pro-
viders, in spite of substantial subsidies provided by the government to vided by AFGRI with an interest rate below the market rate or by fa-
private operators (Houssou et al., 2013). There is evidence that the cilitating the linkage with a private bank, using the tractor as collateral.
often neglected governance challenges of mechanization contribute to John Deere’s dealer AFGRI provides after-sales services such as main-
the failure of such government-sponsored programs (Daum and Birner, tenance services, spare part supply and repairs. While such approaches
2017). to provide a dealer credit or facilitate a linkage with a bank are not
Against this background, the question arises as to whether private unusual in the tractor business, the remarkable feature of the initiative
sector models that do not rely on government support are economically by AFGRI and John Deere in Zambia is that they applied this approach
more promising and suitable to benefit smallholder farmers. Based on to the rather risky customer segment of emerging and medium-size
field observations in Ghana and a review of the international experi- farmers with the explicit goal to facility smallholder mechanization.
ence, Diao et al. (2014) hypothesized that private sector models have These customers are more likely to provide tractor services on a con-
more potential than those that involve state interventions. Many ser- tract basis1 to smallholder farmers than large-scale farmers, who can
vices and inputs for smallholders, such as agricultural extension, re- fully utilize the capacity of a tractor on their own land. However, the
quire public sector involvement due to market failures (Feder et al., experience of John Deere and AFGRI indicates that service provision by
2011). In contrast, considering that agricultural machinery is a pure tractor owners does not simply happen without further efforts, even
private good in which innovations are embodied, machinery services though it is a common practice in countries where mechanization is
offers specific opportunities for the private sector. However, since well established. According to Sheahan and Barrett (2017), only around
tractors are indivisible (unlike other inputs such as seeds and fertilizer), 1% of households across all countries of the Living Standard Measure-
business models such as hire markets are required for smallholders to ment Study-Integrated Surveys on Agriculture (LSMS-ISA) engage in
benefit from mechanization. In recent years, major international agri- tractor rental market. A recent study by Baudron et al. (2019) suggests
cultural machinery companies, such as John Deere and AGCO, have that the demand for mechanization may be much higher than often
recognized the new business opportunities in smallholder agriculture in assumed and that current low levels of mechanization may be due to
Africa, and they have started to invest in developing their own business market imperfections, such as failures to make mechanization acces-
models to access this market. There is limited evidence in the literature sible. To encourage service provision to smallholder farmers, John
on the opportunities and limitations of such purely private-sector Deere and AFGRI linked up with two non-governmental organizations
driven options. Expectedly, civil society organizations are highly (NGOs). One is MUSIKA, an NGO focused on linking smallholders with
skeptical of such initiatives. One reason is a general skepticism that business enterprises, and the other is the Conservation Farming Unit
multi-national agribusiness companies may take advantage of small- (CFU), an NGO focused on promoting conservation agriculture. MU-
holder farmers (see, e.g., Martínez-Torres and Rosset, 2010). Another SIKA has supported service provision by providing business training to
reason is the fear that mechanization may lead to rural unemployment. the tractor owners, where the need to reach high machinery utilization
Such concerns are not new. As Juma shows in his book on “Innovation rates was highlighted. CFU facilitated the formation of groups of
and Its Enemies” (Juma, 2016), farm mechanization has been one of the smallholder farmers who wanted to access tractor services, thus redu-
most controversial of all agricultural innovations – not only in con- cing the transaction costs that arise for reaching smallholders. In our
temporary times, but also historically. During early waves of state- sample of 21 tractor owners, 12 provided services to smallholder
driven and often subsidized agricultural mechanization in newly in- farmers. On the average, they served approximately 60 smallholder
dependent countries of Sub-Saharan-Africa, the International Labor farmers in an average maximum radius of 40 km).2
Organisation warned against potential unemployment effects but such At the current stage, the smallholder farmers typically use tractor
concerns were constrained by a lack of empirical data (ILO, 1973). The services to mechanize the most labor-intensive activity in crop pro-
concerns continue to be voiced but lack of empirical evidence has re- duction, which is ploughing. Alternatively, if farmers practice con-
mained a constraint (Massey et al., 1993; Bhandari and Ghimire, 2016). servation agriculture, they use tractor services for ripping.3 The tractors
Based on theory and insights from historical experience, Binswanger are often also used for a labor-intensive post-harvest activity: maize
(1986) hypothesized that mechanization can - depending on the access shelling. Other steps in crop production, such as weeding, pest control
to land and output markets - raise or reduce employment. In the same and harvesting continue to rely on hand labor or animal traction.
study, he highlighted that farming operations are typically mechanized The overall objective of this study was to assess the economic and
step-wise - starting with the biggest labor bottlenecks - which may ei- social impact of providing tractor services on smallholder farmers
ther raise or reduce the labor requirements for subsequent farming under this initiative by John Deere, AFGRI, MUSIKA and CFU (hereafter
steps. In addition, one should differentiate between different types of referred to as the JD initiative) and to calculate the effect on total labor
labor: The loss of (well-paid) work opportunities for laborers constitutes
a problem, but a reduction of child labor and unpaid family work (with
1
opportunity costs) would have to be seen as an advantage. Research- This is mostly on cash basis but some also provide services on credit basis
based evidence is, thus, very important to better understand whether (either cash or in-kind). This is similar to the provision of animal draught
smallholder farmers can benefit from private-sector led mechanization services.
2
According to the tractor owners who do not provide services one reason
initiatives and how such initiatives affect rural employment. Yet, there
where high transaction costs of offering services to smallholder farmers.
is a lack of empirical evidence on this topic.
Another reason is the synchronic timing of farming. Tractor owning rippers
The goal of this paper is to contribute to filling this knowledge gap have a larger time window to offer ripping services but tractor owners owning
by presenting a case study of a private-sector led smallholder me- ploughs stated to focus on their own field first.
chanization initiative in Zambia. We analyze an initiative where the 3
Ripping is a technique of conservation agriculture where tillage is reduced
company John Deere, the largest manufacturer of agricultural ma- to using a narrow implement, similar to a spike, which creates a planting furrow
chinery worldwide, worked with its dealership AFGRI to develop without turning the soil.

2
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

requirements, taking into account that farmers may expand crop pro- group discussions were conducted, in which Participatory Impact
duction when they access tractor services. Since a randomized control Diagrams were constructed (Kariuki and Njuki, 2013). Participatory
trial approach was not feasible, we used Propensity Score Matching Impact Diagrams are a technique that relies on visualizing the per-
(PSM) to assess the effects of accessing tractor services on smallholder ceived impacts of the participants using a large sheet of paper. Positive
farms. We focus on the effect of the initiative on smallholder farmers, as well as negative impact chains are indicated on the paper in the form
considering that the effects are contested in the policy debate, as of tree structure (similar to a mind map), which then serves as a basis
highlighted above. Examining different models under which private for further discussion. The team held 13 such focus group discussions
sector companies can foster the sale of tractors to promote the con- with men and 12 with women. The discussions focused on the impact of
tracting of tractor services was beyond the scope of this study, but will mechanization at the community level. Therefore, the impact of
be an interesting topic for future research. households who do not use mechanization services was captured as
well. Such households may be affected indirectly, especially though
2. Background information changes in the demand for agricultural labor.
Consequently, the research design for the study was based on the
With an average population density of 22 inhabitants per km2, following combination of methods: (1) a survey among a sample of farm
Zambia is one of the most sparsely populated countries in Sub-Saharan households that receive and did not receive tractor services; and (2)
Africa.4 Agriculture supports the livelihoods of 60–70% of the popu- focus group interviews in selected communities, where smallholders
lation (Tembo and Sitko, 2013: 2). On the average, Zambian farmers had used tractor services provided under the Initiative. These two
own 3.2 ha (ha) of land (Tembo and Sitko, 2013: 20), but due to labor methods constituted the trunk of the data collection for this paper. In
and other constraints, they usually do not cultivate all their land. addition, we used: (3) semi-structured interviews with representatives
Overall, agriculture is dominated by smallholder farmers as 95% of the of the organizations involved in the JD Initiative and (4) in-depth in-
farms cultivate less than 5 ha (Sitko and Jayne, 2014: 194). However, terviews with a sample of farmers who had purchased a tractor, which
during the past decade, there has been a rapid increase in the number of were used to obtain additional background information.
medium-scale farmers who cultivate between 5 and 20 ha of land. They
are referred to in Zambia as “emergent farms.” A recent study found 3.1. Sampling strategy and data collection
that “between 2001 and 2011 the population of emergent farmer
households in Zambia grew by 62.2%, vastly outstripping the 33.5% The following sampling strategy was applied: The tractor owners
growth rate of the total smallholder population.” (Sitko and Jayne, were randomly sampled from the six (out of the eight) Zambian pro-
2014: 194). vinces, where the JD Initiative was implemented. A total of 21 tractor
So far, access to agricultural machinery such as tractors and pro- owners were interviewed, the number per province was proportional to
cessing machines is very low in Zambia. According to a nationally re- the total number of farmers who had participated in the Initiative. The
presentative survey conducted by IAPRI in 2015, only 1.8% of all interviews with the selected tractor owners revealed that 12 out of the
households used mechanical power in their farm operations. On the 21 selected tractor owners provided services to smallholders. The
average, 36.5% use animal traction. The underutilization of the coun- smallholders for the household survey were selected as follows: In each
try’s agricultural potential results in widespread poverty among the location, eight farmers were selected who received services from a
rural population. 78% of the rural households live below the poverty tractor owner who had participated in the JD Initiative. They are re-
rate of 1.25 USD per day, and for female-headed households, the rate is ferred to as “participants” here. For the control group, five farmers who
almost 85% (IAPRI, 2015: 114-115). As in other African countries, do not receive services were randomly selected from the same locations.
there has been an increasing interest in agricultural mechanization in The five households from the control group could use mechanization
Zambia in recent years. For example, in 2011, the Ministry of Agri- services offer by other service providers, but this was rarely the case. As
culture and Livestock started a Tractor Mechanization Fund in colla- the control households are located within the service area of the
boration with the FAO and the Zambian National Farmers Union emerging farmers, they are potentially affected by the mechanization
(ZFNU)5 and the country hosts AgriTech Expo Zambia, a major trade scheme in an indirect way (spill-over). To assess this effect, three ad-
fair for agricultural machinery. ditional control group households from a close-by community were
selected, as well. In total, 121 households that use tractor services
under the Initiative (“participants”) and 129 households that do not use
3. Methods
tractor services were included in the household survey. The survey was
conducted by the research team in face-to-face interviews with the
According to current standards of program evaluation, a rando-
farmers using hand-held computer devices.
mized control trial would be the preferred approach to assess the im-
pact of the JD Initiative on smallholders. Since the Initiative was not
3.2. Analysis
implemented in such way, a survey was conducted and the Propensity
Score Matching (PSM) approach was used to assess the impact of par-
To assess the impact of the mechanization scheme on farm house-
ticipation in the Initiative on the income and the use of the income by
hold income and food consumption, a Propensity Score Matching (PSM)
smallholder farmers (Khandker et al., 2010; Caliendo and Kopeinig,
approach was used (cf. Khandker et al., 2010; Caliendo and Kopeinig,
2008). PSM allow to reduce self-selection bias due to observed char-
2008). The main impact measure of interest, the average treatment
acteristics but cannot address bias due to unobserved characteristics
effect on the treated (ATTJ), is estimated according to:
(see De Janvry et al., 2010; Janvry and Sadoulet, 2016 for a discussion
on PSM). To addressed the later point, by testing different model spe- ATTJ = E [y1j |JDMechj = 1] E [yoj |JDMechj = 1] (1)
cifications and matching algorithms as well as supporting our findings
where y1j is the value of the outcome of farm household j after bene-
with economic theory, an extensive literature review and using quali-
fiting from the John Deere (hereafter JD) tractor service provider and
tative methods.
yoj is the outcome of the same farm household j if the household did not
To better understand social dynamics within households and com-
benefit from the JD Initiative.
munities, the team also used qualitative methods. Specifically, focus
The underlying estimation problem of Eq. (1) can be represented as
a treatment-effects model of the form:
4
http://data.worldbank.org/indicator/EN.POP.DNST. '
5 yjt = + + xjt + JDMechj + (2)
See http://www.znfu.org.zm/tractor_mechanization. j t jt

3
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Table 1
Outcome and explanatory variables used for impact assessment.
Variable name Variable description

Outcome variables
Net on-farm income Farm gross margin
Yield Per hectare seasonal crop output
Land ownership increment Increase in land size owned
Farm input used (fertilizer, herbicides, seeds) Changes in the quantities of farm inputs used
Household expenditure (Food, non-food household needs, Average amount of money (in ZMW) spent on daily needs over stipulated periods
education, health, recreation)
Food Intake (Food diversity, Food consumption frequency) Quantity, quality and frequency of food consumed by respondent household. The frequency weighted
diversity score is calculated using the frequency of consumption of different food groups consumed by the
household the day before the survey (WFP, 2008)

Explanatory variables
Farming experience Number of years of farming
Off-farm business participation Farmer’s involvement in off-farm businesses: 1 = Yes, 0 = No
Size of household Total count of household members above age 5 years of age
Gender of household head Gender of the household: 1 = male, 0 = female
Education level of household head Years of schooling
Land cultivated Cultivated land per capita – total cultivated land divided by total members of household
Access to extension service Farmer’s has access to private, public or third sector extension service: 1 = Yes, 0 = No
Access to credit facilities Farmer’s access to credit/loan facility: 1 = Yes, 0 = No
Market access Amount of travel time (in minutes) required to access nearest village market
Group membership Farmer’s membership in a social or political group such as a farmer cooperative: 1 = Yes, 0 = No
Household asset index Total count of household assets, e.g., solar panels, bicycles owned by farmer
Livestock ownership Total number of cattle owned by farmer before mechanization scheme. Weighted using Tropical Livestock
Unit conversion factors (Jahnke, 1983)
Farmer willingness to invest Percentage of an amount of money that a farmer is willing to invest in any venture of choice considering
potential losses and gains

JDMechj = '
wj + uj considered acceptable (Caliendo and Kopeinig, 2008).
In calculating the treatment effects and their standard errors, the
JDMechj = {1, ifJDMechj > 0 and 0 if otherwise} (3) bootstrapping method (with 500 replications) was employed, as used in
most of the literature. The ATT of participating in the JD Initiative is
Prob (JDMechj = 1) = F ( 'wj ) (4) defined by the use of a John Deere tractor at least for land preparation.
The ATTs of the program were obtained by estimating the models using
F ( 'wj ) data from the sample described above, which included 121 tractor
Prob (JDMechj = 0) = 1 (5)
service users and 129 households who do not use these services. The
where JDMechj* is a latent unobserved variable whose counterpart, data refer to the 2014–2015 cropping season. The outcome variables
JDMechj, is observed in dichotomous form only; JDMechj = 1 represents and the explanatory variables used for the assessment are shown in
a user (i.e. a farmer who decides to hire services) of JD tractor service Table 1.
provider (that is, treatment) and JDMechj = 0 represents non-user of
the facility (that is control); xj is the vector variable determining the 4. Results
outcome of the JD Initiative, wj is the vector variable determining the
probability of being a user of the JD mechanization facility which in- The first subsection of this section presents descriptive statistics,
cludes the list of explanatory variables given in Table 1 below; αj and τt comparing the treatment and the control group. Since this comparison
respectively captures the individual and time-specific effects; β and γ does not control for a possible sample selection bias, the findings of
are the vectors of parameters measuring the relationships between the Section 4.1 should be seen as background information for the PSM
dependent and independent variables; ε and u are the random compo- analysis, which is presented in Section 4.2. Section 4.3 presents a
nents of the respective equations. The functional form (F) may take the matching quality and sensitivity analysis and Section 4.4 deals with the
form of a normal, logistic or probability function. results of the focus group discussions.
A two-stage weighted estimation approach was used. In stage one,
Eq. (3) is estimated using a probit model to obtain the propensity 4.1. Descriptive results
scores, which are then used as weights in a second stage estimation of
Eq. (2), based on matched treatment and control observations identified Table 2 presents information about the socio-economic character-
in stage one. Of the 4 matching algorithms commonly proposed in lit- istics of the of the surveyed smallholder farmers. The table indicates
erature (Caliendo and Kopeinig (2008), for a detailed overview), the that smallholders who receive tractor services have similar character-
variant of radius matching (Dehejia and Wahba, 2002) was applied for istics as those who do not receive services. The differences shown in the
the second stage estimation. This method has an advantage of using table were not statistically significant. The findings indicate that par-
only as many units as are available within a caliper (c), allowing for ticipation in the JD Initiative was not biased towards the larger ones
more matching options, hence improving matching quality (Caliendo among the smallholder farms. However, the data suggest that the
and Kopeinig, 2008). Rosenbaum and Rubin (1985) recommends ca- schemes are implemented in areas where smallholder farmers tend to
liper (c) used to be one-fourth the share of the standard deviation (s.d) have somewhat larger holdings and higher education levels than on the
of the probability model of the propensity score (c = 0.25*s.d). national average (cf. IAPRI, 2015).
The matching procedure must be able to balance the distribution of Table 3 shows that the farmers who receive mechanization services
the relevant variables in both control and treatment groups. Rosenbaum cultivate almost the entire arable land that they own, whereas the
and Rubin (1985) suggest calculating the standardized bias (SB) before farmers in the control group cultivate only 60%. According to the
and after matching. A bias reduction below 3% or 5% after matching is farmers this is due to labor shortages. According to the farmers

4
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Table 2
Socioeconomic characteristics of surveyed smallholder farmers.
Variablea Participants (N = 121) Control group (N = 129) Total (N = 250)

Age of household head (years) 50.0 47.0 48.4

Farming experience (years) 20.1 21.0 20.6
Off-farm business participation (yes/no) 46% 40% 43%
Number of household members 7.4 7.4 7.4
Female household heads (%) 22% 18% 20%
Education of household head (years of schooling) 8.3 7.2 7.7
Total land owned (ha) 10.8 9.2 10.0
Access to extension service (percent) 74% 65% 70%
Access to credit facilities (yes/no) 13% 15% 14%
Access to markets (minutes of walking time) 30.9 30.5 30.7
Indicator of farmers’ willingness to invest 79% 81% 80%

a
See Table 1 for an explanation of the variables.

Table 3
Differences in agricultural practices and outcomes.
Participants (n = 121) Control group (N = 129) Differencea Statistical significanceb

Arable land owned (ha) 7.1 6.1 16% no

Arable land cultivated (ha) 6.5 3.7 76% yes
Percent of owned land cultivated 92% 60% 53% –
Beginning of land preparation 30. Sept 6. Nov – –
Use of fertilizer for maize (kg/ha) 260 190 37% yes
Percent of farmers using herbicides 63% 24% 162% –
Use of herbicides for maize (litres/ha) 2.3 2.4 −4% no
Maize yields (metric tons/ha) 3.1 2.5 24% yes

a
Difference is calculated as the difference between the values for participants and control group divided by the value of the control group.
b
Yes indicates that difference in mean values is statistically significant at the 5% level.

cultivating more land now, this land was mostly fallow land before and expenditure, but the difference was not statistically significant. However,
a small share was land that was rented out before. Moreover, the par- service users had significantly higher expenditures on education and
ticipants are able to start land preparation much earlier than the control food. Based on the survey data, a food diversity score was calculated,
group. We did not ask about planting dates during the household survey which is an indicator of nutritional quality. More diverse diets provide
but the results from the focus group discussions suggest that this allows more micro-nutrients, which is important to combat “hidden hunger.”
farmers to plant more timely.6 The amount of fertilizer that the parti- The findings indicate that households that access mechanization services
cipants use is almost 40% higher than that of the control group. The do not consume a significantly more diverse diet than the control group.
share of farmers who apply herbicides is 63% among the participants as This finding suggests that the additional income that the participants
compared to 24% in the control group. earn is mostly spent on food staple crops. Nutrition education may be
The data also show that the participants achieve maize yields that required to encourage households to invest their additional income in
are 24% higher than those of the control group. This can be because of increased diet diversity.
mechanized tillage enhance timeliness and tillage quality, the later Table 5 reports differences regarding labor hours between house-
which can contribute to better weed control, better soil moisture holds that access tractor services and those that do not. As indicated
management and a higher germination rate of seeds. All of these aspects above, the differences do not show causal effects, but they give im-
were mentioned during the focus group discussions (see Section 4.4). In portant clues. Interpreting the figures, one needs to keep in mind that
addition, mechanized tillage may encourage fertilizer use and better the participating households cultivate on the average 76% more land
weed control but these are indirect effect that we will control for with (see Table 3) than the non-participating households. As indicated
the PSM.7 above, the only two activities for which tractor services are used are
As shown in Table 4, farmers who use mechanization services had a land preparation and processing (i.e. maize shelling). Expectedly, the
significantly higher total farm income than the control group, whereas participating households use significantly less labor for land prepara-
the difference in income per hectare was not significant. This finding tion and significantly more labor for harvesting. The table suggests that
suggests that the main income effect from accessing tractor services may access to tractor services reduces the labor burden for family labor,
be due to the increase in cultivated land area, which is made possible by including the labor burden of children and women, while it increases
mechanizing soil preparation. The finding from recall questions posed to the opportunities for hired labor during the harvesting season as a
the treatment group (not reported in the table) suggest that they were consequence of the expansion in cultivated area. Weeding time is also
indeed able to increase the cultivated land area. Farm households that significantly reduced overall (and increased for hired labor), which
use tractor services spend, on the average, slightly less on health reflects the better land preparation (reducing weed growth) and the
higher percentage of farmers using herbicides (see Table 3). In our
sample, farmers practicing conservation farming, which is associated
6 with higher weed pressure, do use significantly more herbicides than
Optimal sowing dates can have a big implication on yield (Low and
Waddington, 1991). Sallah et al. (1997), for example, found that a delay in those who do not.8
planting of 14 days reduces maize yield by 30% in the Guinea Savanna.
7
In general the yield response to mechanized tillage depends on the soil type
8
(for example, it is higher for clay soil than sandy soil), on the topo-sequence of Authors such as Giller et al. (2009) focus more explicitly on the relation
the location and the crop (Pingali et al., 1987). between conservation farming and labor dynamics.

5
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Table 4
Differences in farm income, expenditure and nutrition.
Income and expenditures in ZMW Participants (N = 121) Control group (N = 129) Differencea Statistical significanceb

Farm income total 16,999 7323 132% yes

Farm income per hectare 2839 2045 39% no
Farm income per household member 2528 2045 24% yes
Health expenditure per year 270 340 −21% no
Education expenditure per term 1730 842 105% yes
Food expenditure per month 561 299 88% yes
Food diversity score 6.4 5.8 10% no

Note: 1 USD equals approx. 10 Zambian Kwacha (ZMW).

a
Difference is calculated as the difference between the values for participants and control group divided by the value of the control group.
b
Yes indicates that difference is statistically significant at the 5% level.

Table 5
Differences in labor hours for cultivating and processing all crops.
Total labor hours Hired labor hours Family labor hours Female family labor hours Children family labor hours Male family labor hours

Land preparation −374*** –22 −348*** −93*** −24*** −231***

Planting 106* 131** −28 −50** 4 19
Fertilizer application 10 44*** −37 −29** −6 −2
Weeding −313*** 86** −418*** −207*** −28** −183***
Pests/disease control −16 2 −18 −1 0 −17*
Harvesting 423** 488** −49 −36 −2 −11
Processing −218 −117 −88** −51 −7 −29

Note: Mean difference is the difference between mean values of participant group members and non-participant groups.
* Statistical significance at the 10% level.
** Statistical significance at the 5% level.
*** Statistical significance at the 1% level.

Table 6
Factors influencing participation in mechanization schemes.
Explanatory variables Average marginal Standard Error
effect (dF/dx)

Off-farm business (yes/no) 0.061 0.0678

Gender of family head (male/female) 0.116 0.0828
Years of schooling 0.022** 0.0097
Access to credit −0.075 0.0952
Access to extension services 0.094 0.0757
Network group membership (yes_no) 0.198** 0.0936
Access to market 0.001 0.0013
Livestock owned before participation 0.004* 0.0026
in scheme
Farmer’s investment behavior 0.033 0.1305 .2 .4 .6 .8 1
Propensity Score
LR chi2(12) 19.54
Prob > chi2 0.029 Untreated Treated: On support
Pseudo R-square 0.056 Treated: Off support

1 Fig. 1. Estimated propensity score distribution and common support area by

See Table 1 for an explanation of these variables.
pairwise comparison.
Note: *Statistical significance at the 10% level, ** at the 5% level.

4.2. Results of the propensity score matching (PSM) analysis Fig. 1 below displays the distribution of the propensity scores and
the overlap between the groups. For this pairwise comparison, the
The first step in the PSM analysis is the construction of a probit figure also shows the cases that were dropped from the analysis in order
regression model, which identifies the factors that are significantly as- to avoid bad matches. 3 out of the 121 treated assignments had to be
sociated with the decision of a farm household to access tractor ser- excluded from the analysis.
vices. The results displayed in Table 6, which indicates that better For the matched sample, the bootstrapping method was applied
educated farmers and farmers who are members in social, religious and with 500 repetitions to estimate the standard errors and hence check for
political groups are more likely to access tractor services. Farmers who distinct variations. Table 7 reports the estimates of the ATT. It indicates
owned more livestock (an indicator of wealth) before the start of the that using tractor services has a significant positive effect on the on-
mechanization scheme were more likely to use tractor services, but the farm income of the entire household and on the on-farm income per
magnitude of this effect was negligible. household member. This effect is not only significant, but also large.
Using a probit model, the balancing scores for each pairwise com- The difference in household income of approx. 10,000 ZMK per year
parison of service users with their matching counterfactuals were esti- indicates that the use of tractor services allowed smallholder farmers to
mated. The model was used to predict the probability of opting for more than double their income.
using tractor services. The model’s predictive power can generally be The ATT for yield was also significant, which confirms a causal ef-
judged to be high and the variables show the expected signs. fect of using tractor services on yield. The magnitude of the effect

6
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Table 7 The residual mean bias of 3.9% is within the range of 3–5%, which is
Causal effects of using mechanization services. suggested in Caliendo and Kopeinig (2008) as an acceptable threshold
Outcome variable Average Treatment Standard for remaining bias after matching. The low remaining SB and the high
effect of the Treated Error reduction rate of mean SB indicates a good balancing power and hence,
(ATT) good matching results.
We can, however, not rule out the problem that unobservable fac-
Net on-farm income (ZMW) 10,000*** 3,460
On-farm income per hectare (ZMW) 685 493
tors could influence these findings (hidden bias). We are confident that
On-farm income per household (ZMW) 1500*** 557 this influence is limited since the choice of variables was based on
Yield (metric tons/ha) 0.41* 0.25 economic theory and an extensive literature review. Moreover, the re-
Yearly Expenditure on food (ZMW) 225*** 69 sults are supported by the qualitative findings, as shown below.
Expenditure on education per term 850** 305
(ZMW)
Expenditure on basic household non- 770*** 251 4.4. Results of the focus group discussions
food household needs (ZMW)
Health expenses (ZMW) −48 119 We used qualitative methods to triangulate the findings from the
Expenditure on recreation (alcohol, −40 25
quantitative study and to analyze aspects that cannot be addressed with
tobacco, etc.) (ZMW)
Skipping meals −0.16** 0.06
a purely quantitative study design. As indicted in Section 3, we orga-
Food Diversity Count −0.08 0.7 nized focus group discussions where participants were asked to con-
Before and after JD mechanization 0.3*** 0.1
struct Participatory Impact Diagrams. Table 9 displays the main posi-
difference in fertilizer used (MT) tive impacts that were identified. Since the researchers did not prompt
the respondents to discuss any specific theme, the number of groups
Note: *Statistical significance at the 10% level, **
at the 5% level, ***
at the 1% who identified a specific impact can be seen as an indicator for the
level. relevance of the respective impact in the areas where the schemes were

Table 8
Indicators of matching quality and sensitivity analysis.
Source: own data.
SB (%) SB (%) % |SB| reduction Residual Bias Cases lost to critical selection Critical levels of gamma

Participants/Control Group 14.5 1.10 92.4 3.9 3 1.1–1.15

Note: Calculation using pstest and rbounds.

(approx. 0.4 metric tons//ha) was also substantial, which supports the implemented. As can be derived from Table 9, the communities strongly
findings above on yield effects. However, higher yields did not result in associate the use of tractors with increased yields. They consider more
a higher income per hectare, because the treatment effect was not timely land preparation and the retention of soil moisture due to rip-
significant for the parameter “on-farm income per hectare”. The reason ping as major benefits that contribute to increased yields.9 According to
could be that the yield increase was not sufficient to cover the increased the respondents, these factor and in combination with better land
costs per hectare arising from using more inputs. This finding confirms preparation (which reduces weed pressure) reduces the risk of bad
the results reported above, which suggest that the main causal impact harvest, which makes them more willing to invest in complementary
of accessing mechanization services is allowing smallholders to culti- inputs such as fertilizers. The cultivation of more land was identified as
vate a larger share of the land that they own. another positive impact of mechanization.
The PSM analysis also shows that the increased expenditure in Table 9 also lists the positive socioeconomic affects that are seen to
education and food found in the descriptive statistics can be attributed be the result of the agronomic effects identified above. The majority of
the use of tractor services. The households did not significantly change both the male and the female focus discussion groups (FDGs) identified
their expenditure on alcohol or tobacco, which indicates that the increased income as a positive impact of mechanization, which con-
farmers used their additional income for the benefit of their families. firms the finding of the quantitative assessment. The Participatory Im-
The findings also indicate the increased income from accessing tractor pact Diagrams suggest that the increased income is also used for pur-
services allows farm families to skip fewer meals. However, the findings chasing improved seeds, fertilizers and herbicides, which strengthens
also show that they do not diversify their diets. the effect of mechanization on yields. The FDGs also pointed out that
As indicated above, the survey included recall data from re- the increased income is used for education, as indicated by the quan-
spondents on selected outcome variables, which include changes in titative assessment. According to the FDGs, the increased income is also
input use, yield and livestock that occurred after accessing mechan- used for buying household and farm assets as well as personal supplies.
ization services. For these variables, a double difference ATT technique The reduced work load during the time of land preparation was seen
was used to estimate the differences in mean outcomes for these vari- as a positive impact by half of the male FDGs and a quarter of the
ables. The only significant effect was identified for the use of fertilizer. female FDGs. According to the Participatory Impact Diagrams, the main
The ATT analysis also confirmed that the farmers who use mechan- positive effect was that children, who previously had to work on the
ization services did not purchase additional land; they rather expanded fields, can now go to school. This is an indication that mechanization
cultivation on the land they already owned. contributes to reduced child labor in agriculture. The time saving

4.3. Matching quality and sensitivity analysis

9
Respondents did not mention that tractor plowing increase moisture holding
As indicated in Section 3, the quality of matching was assessed by capacity with deep plowing. However, a meta-study by Schneider et al. (2017)
calculating the standardized bias (SB) before and after matching. The suggests that deep tillage can increase plant-available water reservoir. How-
results (Table 8) show that a very good matching quality was attained. ever, effects will depend on soil types and local climates and both optimal
The standardized bias was reduced from 14.5% before matching to ploughing depth and whether or not to plough are debated topics (see Knowler
1.1% after matching, which corresponds to a bias reduction of 92.4%. & Bradshaw, 2007).

7
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Table 9
Positive impacts.
Positive impacts identified Percent of male groups Percent of female groups Quotes from the interviews that illustrate the perceptions of the
identifying this impact identifying this impact community members
(N = 13) (N = 12)

Agronomic
Higher yield 92% 100% “If you do early planting (…) you are likely to get a high yield”
Early planting und retention of soil 92% 75% “When you use a tractor…, the moisture content is kept for longer, the
moisture due to ripping germination of maize is good, …”
Improved land preparation 69% 42% “When using a tractor, the depth is better than when using animals…. Even
when the rain goes, the plants don’t dry up…”
Cultivation of more land 38% 83% “When we use a tractor, we can cultivate a bigger portion of land compared to
using animals”

Socio-economic
Increased income 92% 100% “When you have better yields, you provide for own consumption, you will be
able to find money for the children’s school fees and for other things… you
also have money to buy farming inputs….”
Reduction of labor demand 54% 25% “When using a tractor, just one person is needed, when using animals, lots of
people are supposed to do the work”
Time saved during land preparation 38% 58% “It is faster when you use a tractor, …”
Improved human and animal health 38% 0% “When you use a tractor, cattle have enough time for grazing… but when you
use them for farming, you might use them from 7 to 11 or 12- they won’t have
enough time for grazing and resting…”

Table 10
Negative impacts.
Impacts Percent of male groups Percent of female groups Quotes from the interviews that illustrate the perceptions of the
identifying this impact identifying this impact community members
(N = 13) (N = 12)

Agronomic
Yield losses because services were 31% 17% “At the time we need the tractors, they are not available, and so we are
delivered late forced to plant late.”
Soil degradation (in case of using the 31% 8% “Soil fertility is reduced after repeatedly turning the soil surface season
plough) after season.”

Socio-economic
Fewer jobs for agricultural laborers 54% 8% “Before starting hiring a tractor, you used to hire people to come and help
during the land preparation season you in the fields. Now you have tractors so you won’t be hiring the people
(…) so that person you used to hire will have a problem because there is no
income for him.”
Migration to other areas 8% 0% “The leaders of the household migrate to towns and communities where the
farm land has been expanded.”
More work load for women 0% 17% “Women are doing more work…. because there are more activities after
using the tractor, more activities like weeding…”

during the time of land preparation was also seen as a benefit, espe- The Participatory Impact Diagrams were also used to identify pro-
cially by female community members According to the Participatory blems that the communities had identified with regard to mechaniza-
Impact Diagrams, the saved time was mostly used for vegetable gar- tion. The results are displayed in Table 10. In general, the percentage of
dening, performing household chores, engage in off-farm work and FDGs that identified problematic impacts was comparatively low. Only
attend social events. Five of the 13 male FDGs felt that mechanization two agronomic problems were identified: late service provision and soil
improved their either their own health or the health of their animals. degradation. Four of the 13 male FDGs reported problems because the
This impact was not identified in female FDGs. The reason might be that tractor services were provided too late. In these cases, delayed land
male household members have to bear the main drudgery of labor for preparation resulted in late planting, which in turn led to a sharp yield
land preparation, which is the activity that is mechanized. Crop hus- decrease and thus lower farm incomes. Soil degradation was mentioned
bandry and harvesting activities, which are mainly carried out by in four of the 13 male FDGs and in one of the 12 female FDGs. This
women, are not yet mechanized, as shown above. problem was associated with the use of the disc plough rather than the

8
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

ripper. 5.2. Social benefits for participating households

The main socioeconomic problem associated with mechanization
identified by the community members were reduced job opportunities The study provides strong evidence that the smallholder farmers
for agricultural laborers at the beginning of the farming season. Farmers who accessed tractor services were able to generate additional income.
who use oxen to provide ploughing services were also seen as being While increasing income is generally welfare enhancing, it depends on
disadvantaged. The community members reported that working op- the intra-household division of labor and power relations whether in-
portunities for agricultural laborers in particular dropped during the come translates in benefits such as improved nutrition and education,
months of land preparation. However, it was also acknowledged that and this division is not independent of how the income is earned
agricultural laborers benefitted from a higher demand for labor during (Tavenner et al., 2019; Quisumbing and Maluccio, 2000). While this
weeding, fertilizer application and harvesting times. These findings study provides no evidence on how mechanization affects the control
confirm the results of the quantitative analysis on labor use (Table 5). over resources, the results show that income was use for the education
Two out of 12 female FDGs mentioned increasing workload for women of children and food security. Their expenses for food were higher and
from land expansion. they were less likely to skip meals, which is an important finding
considering high levels of undernutrition in Zambia. It is also worth
5. Discussion noting that, according to the survey findings, the participating house-
holds did not increase the consumption of alcohol or tobacco. The
As indicated above, this study aimed to assess the impact of the JD qualitative findings from the focus group discussions indicate that some
Initiative, as an example of a private-sector business model, on small- smallholders were able to invest their income into off-farm businesses,
holder farmers. In view of the criticism of such initiatives by NGOs, such as trading livestock or running grocery stores.
special attention was paid to a range of potential effects, including in-
come, nutrition, child labor and the potential displacement of labor. As
a general disclaimer to the following discussion, one needs to take into 5.3. Use of farm inputs and land productivity
account that the study was based on a PSM analysis of cross-sectional
data and not on a randomized control trial, which has become the “gold The findings suggest that the participating farmers purchased more
standard” in impact evaluation. We still believe that the results are of farm inputs, in particular, fertilizer. This may be because better land
interest, considering that empirical studies that deal with pure private- preparation (which reduces weed pressure) and more timely land pre-
sector initiatives are scarce. paration, yields risks decline and the willingness to invest in com-
plementary inputs become higher. However, the fact that tractor users
5.1. Income effects apply higher levels of fertilizer may also reflect the fact that they can no
longer fallow land to restore fertility since they cultivate nearly all their
One of the most important findings of this study is the evidence that, land. This may also explain the fact that net income per ha does not
on the average, the smallholders who used tractor services were able to increase. The use of tractor services was also found to be associated
double their income because they were able to cultivate a much larger with an increased use of herbicides. Partly, this may be due to the fact
share of the land that they own. The focus group discussions largely that CFU promoted herbicide use in connection with the introduction of
confirmed this finding. According the results of this study, accessing conservation farming. Another reason could be labor shortages during
mechanization services also increased labor productivity quite sub- the weeding time that were due to the increase in the area under cul-
stantially. This is an expected benefit, but nevertheless important, tivation. It was beyond the scope of this study to assess to what extent
considering the concerns about low labor productivity in African agri- herbicides were used appropriately and safely by the smallholders but
culture mentioned in the introduction. this should be carefully monitored.
The potential of the JD Initiative is particularly promising if one The study provides evidence that the smallholders were able to in-
takes the number of smallholders into account that can potentially crease their yields. According to the PSM analysis this effect was in the
benefit from one single tractor. One of the tractor owners included in range of 0.5 metric tons per ha, which corresponds to a yield increase of
this study served more than 150 smallholders, indicating that, under approximately 25%. This is less than the frequently mentioned claim
the conditions in which the JD Initiative was implemented, facilitating that mechanization can double or triple yields but consistent with
access to one single tractor can potentially help to double the income of Pingali et al. (1987) which report yield increases of between 20 and
approx. 150 smallholder farmers. However, this potential was not fully 30% depending on the crop. However, the results indicate that the
utilized. Altogether, the 21 emerging farmers included into the sample smallholders who use mechanization services were not able to achieve a
served 693 smallholders, which corresponds to an average of 33 higher income per hectare. This finding suggests that farmers may
smallholders per tractor. This result indicates the need to conduct more benefit from extension services to use their inputs more effectively.
research on the factors that can increase the incentives of tractor
owners to provide services to smallholder farmers.

9
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

5.4. Expansion of the cultivated area follow-up study on the effects of mechanization on labor use in
households. For this study, a picture-based smartphone app was de-
The study provides strong evidence that the major mechanism be- veloped that allows household members to record the time they spend
hind the remarkable income increase among the smallholder farmers on their daily activities in real time.
was the expansion of the land area that they cultivate. In the locations The findings presented in Table 5 indicate that mechanization did
where the evaluation was conducted, smallholders typically own, ac- not reduce the demand for hired labor. To the contrary, the results
cording to the survey results, between 6 and 7 ha of land. There were no suggest that the demand for hired labor increased for two reasons. One
statistically significant differences in land size owned between the reason is the expansion of the cultivated area, which increased the labor
farmers who accessed tractor services and the control group. The demand for all activities that are not mechanized. The second effect is a
findings indicate that due to labor constraints, farmers without access shift from family labor to the use of hired labor, which may be due to
to tractor services are not able to cultivate the entire land that they the income effect of mechanization. This finding indicates that me-
own. chanization increases the demand for hired labor under conditions
The finding that the income effect was mostly achieved by land where land expansion is possible. The historical experience analyzed by
expansion has important implications for the up-scaling of the JD Binswanger (1986: 33) is well in line with this finding. However, the
Initiative. In general, land is not scarce in Zambia, as has been pointed demand for additional labor will be reduced once crop husbandry and
out in Section 2. In view of the debate about large-scale land acquisi- harvesting activities also become mechanized.
tions and “land grabbing”, it is important to note that access to me- The findings from the focus group discussions suggest that the shift
chanization services allows smallholders to make better use of Zambia’s in the timing of the labor demand may, however, involve problems.
underutilized land resources so that this potential is not only left to Smallholder farmers who work as laborers used to purchase inputs for
large-scale investors. their own farm with the money they earned at the time of land pre-
However, one also needs to take into account that not all small- paration. If they work for farmers who use tractor services for land
holders can easily expand the land that they cultivate. If they are not preparation, they have to borrow money from those farmers to pur-
able or willing to resettle, they need land resources that are located chase their inputs and pay it back in form of labor provided for crop
sufficiently close to the villages in which they are residing. In a recent husbandry and harvesting. This shift has introduced a new type of de-
nationally representative survey, more than 54% of the rural popula- pendency of agricultural laborers, a finding that calls for further in-
tion said that there is no more additional land available to them, despite vestigation.
the existence of underutilized arable land in Zambia (Chisinga and
Chopoto, 2015: 36). In the areas where the study was conducted, land 6. Policy implications
availability did not yet seem to be a main constraint yet. The reason
may well be that service provision was directed towards locations Overall, the findings indicate that private-sector driven initiatives to
where land availability for smallholders is still relatively high. These promote smallholder mechanization in Africa have a considerable po-
insights suggest that going forward, the mechanization initiatives tential to increase farm incomes. In line with the literature - and con-
should not only focus on the expansion of land, but also on increasing trary to concerns of the critics of such initiatives, smallholder me-
the profitability of the land that is already cultivated. This is also im- chanization increases rather than reduces the demand for hired labor in
portant in view of growing concerns that the expansion of land culti- situations where an expansion of the cultivated area is feasible. This
vation in Savanna regions can have negative environmental and climate expansion in labor demand will come to an end once land expansion is
effects (Ceballos et al., 2010; Searchinger et al., 2015). not feasible any longer. Therefore, a stronger focus on using mechan-
ization to increase land productivity rather than promoting land ex-
5.5. Use of intra-household and hired labor pansion will be required. As pointed out above, limiting the expansion
of land cultivation is also necessary to ensure environmental sustain-
Two types of concerns regarding labor use are associated with me- ability. Therefore, it is recommended to assist smallholders in in-
chanization, one referring to the intra-household division of labor and creasing revenues per ha. This goal requires complementary efforts.
one referring to hired labor. The first concern stems from the fact that, Providing agricultural extension services to a large number of small-
initially, only very labor-intensive farming activities, such as ploughing holder farmers can hardly be considered the task of agricultural ma-
which are mostly carried out by men, are mechanized, whereas other chinery manufacturers or dealers. Other actors, such as government
activities, which are mostly carried out by women and children, such as extension services, need to play a role to reach this goal.
weeding, are not mechanized. If households expand the area cultivated, To ensure that smallholder farmers benefit from contractor models,
this may well result in an increase of the burden of labor for women and it is essential to better understand the economics of tractor service
children. The evidence provided by the study suggests that this was not provision. Some of the tractor owners interviewed for this study pointed
the case (Table 5). This is because agricultural activities were much less out that the transaction costs of providing services to smallholders are a
gendered than assumed by the literature. Land preparation is more major reason for limited service provision. Tractor owners who pro-
done by males and weeding more by females but the difference is, al- vided services to smallholders benefitted from the support of an NGO,
though significant, not a major one (see Tables 11 and 12 in the Ap- which organized smallholders in groups and linked them to tractor
pendix). Thus, households with access to tractor services used on the owners. ICT tools that follow the “Uber” model such as “Hello Tractor”
average significantly less household labor from men, women and chil- may help to reduce the transaction costs of providing and accessing
dren than households without access to tractor services. Two factors tractor service.10 In addition, tractor owners need the capacity to
may account for this result. One factor may be the increased use of manage tractors and there may be a threshold of own farm size above
herbicides, which reduced the labor requirements for weeding. The which tractor owners do not provide any services any more. These
other factor may be the use of hired labor for harvesting, as further questions open the room for follow up studies, including randomized
discussed below. It appears that the increased income achieved by control trials testing different interventions. For example, such studies
mechanization allowed farm households with access to mechanization could assess what kinds of financial interventions could encourage
services to hire more labor for the non-mechanized activities. However, medium-holding famers who would otherwise not purchase tractors to
the findings regarding the labor effects of mechanization have to be purchase tractors? Would they need to be trained to understand that
interpreted with care, since data on labor use in smallholder farm
households are difficult to collect in interviews with recall questions. To
address this challenge, Daum et al. (2018, 2019) have conducted a 10
https://www.hellotractor.com/.

10
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

higher utilization rates contribute to larger profits? And finally, does Upreti for their support during the field research. We also thank Tim
organizing farmers reduce transaction costs enough to encourage those Loos, Heike Baumüller, Zaneta Kubik and Oliver Kirui for their valuable
who would not have otherwise provided or rented services? comments. The financial support of Deere & Company for the fieldwork
Going forward, it will also be important to pay attention to avoiding is gratefully acknowledged. We are also grateful for financial support
potential negative environmental effects of mechanization. In the case provided by the “Program of Accompanying Research for Agricultural
considered here, problems of increased soil erosion have been limited Innovation” (PARI), which is funded by the German Federal Ministry of
because the ripper rather than the disc plough was promoted due to the Economic Cooperation and Development (BMZ).
involvement of CFU. In situations where farmers select the implements
without the advice by CFU, they may, however, prefer the plow to the
Funding
ripper. Extension services to smallholder farmers, as mentioned above,
could play an important role in ensuring appropriate soil fertility
The Deere & Company partly funded the fieldwork for this study.
management on mechanized smallholder farms. This could also be
While writing this paper, Ferdinand Adu-Baffour and Thomas Daum
supported via lower tax and tariffs for soil-protecting implements.
have received financial support from the “Program of Accompanying
Overall, the findings indicate smallholder farmers can benefit from
Research for Agricultural Innovation” (PARI), which is funded by the
private-sector driven mechanization initiatives, especially when they
German Federal Ministry of Economic Cooperation and Development
place efforts on tractor service provision. This does not imply that
(BMZ) under grant number 2014.0690.9.
government have no role to play. The public sector is essential to
promote mechanization by providing complementary services, in-
cluding providing training and agricultural extension to build the ca- Declaration of interest statement
pacity of small and medium-size farmers to own and manage tractors
and ensuring the environmental sustainability of mechanization This project was funded partly by the “Program of Accompanying
through applied research on soil conserving mechanization. Research for Agricultural Innovation” (PARI), which is funded by the
German Federal Ministry of Economic Cooperation and Development
Acknowledgements (BMZ) under grant number 2014.0690.9 and partly by Deere &
Company, Moline, Illinois, United States of America. Neither BMZ nor
The authors would like to express their gratitude to all persons who Deere & Company had any influence on the results of this study as they
took part in the interviews and focus group discussions. Special thanks were not involved in study design, data collection, analysis and inter-
are due to the Indaba Agricultural Policy Research Institute and Suchita pretation of data, or writing of this publication.

Appendix

See Tables 11 and 12.

Table 11
Per hectare (per 50 kg bag) hour differences for cultivating and processing of all crops.
Total Labor hours/ha Hired Labor hours/ Family labor hours/ha Female family labor hours/ Children family labor/ha Male family labor/ha
ha ha

Land preparation −219*** −9 −196*** −70*** −14* −111***

Planting −11 25** −35** −29** −3 −4
Fertilizer application −13 6** −21** −11** −4 −5
Weeding −200*** 27** –222*** −110*** −16*** −97***
Pests/disease control −29** 1 −31** −2 0 −6
Harvesting −14 45*** −34 −11 −3 −20
Processing (per 50 kg) −4*** −2 −3** −2*** −1 −1

Note: Mean difference is the difference between mean value of participant group members of the scheme and non-participant groups.
* Statistical significance at the 10% level.
** Statistical significance at the 5% level.
*** Statistical significance at the 1% level.

11
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Appendix A. Supplementary material

Children family Supplementary data to this article can be found online at https://
labor/ha doi.org/10.1016/j.foodpol.2019.03.007.

15

24

18

74
7
8

2
References
Female family
labor/ha

Anderson, J.R., Dillon, J.L., 1992. Risk Analysis in Dryland Farming Systems. Farm
Systems Management Series 2, Food and Agriculture Organization of the United
172

135

469
73
55
29

Nations (FAO), Rome.

2

3
Baudron, F., Misiko, M., Getnet, B., Nazare, R., Sariah, J., Kaumbutho, P., 2019. A farm-
level assessment of labor and mechanization in Eastern and Southern Africa. Agron.
Male family

Sustain. Dev. 39 (2), 17.

labor/ha

Bhandari, P., Ghimire, D., 2016. Rural agricultural change and individual out-migration.
Rural Sociol. 81 (4), 572–600.
115

148

125

479
24
23

41

Binswanger, H., 1986. Agricultural mechanization: a comparative historical perspective.

3

World Bank Res. Observer 1 (1), 27–56.

Birner, R., Resnick, D., 2010. The political economy of policies for smallholder agri-
Family labor

culture. World Dev. 38 (10), 1442–1452.

hours/ha

Byerlee, D., Husain, T., 1993. Agricultural research strategies for favoured and marginal
areas. Exp. Agric. 29, 155–171.
1020
201

344

278

Caliendo, M., Kopeinig, S., 2008. Some practical guidance for the implementation of
86
60

43

propensity score matching. J. Econ. Surveys 22 (1), 31–72.

Ceballos, G., Davidson, A., List, R., Pacheco, J., Manzano-Fischer, P., Santos-Barrera, G.,
Hired labor

Cruzado, J., 2010. Rapid decline of a grassland system and its ecological and con-
hours/ha

servation implications. PLoS One 5 (1).

Chisinga, B., Chapoto, A., 2015. Under-appreciated Facts about Zambia’s Agriculture: In
10

24

18
12

69

Chapoto, A., Sitko, J., Nicholas (Eds.), Agriculture in Zambia: Past, Present, and
4
1

0
Non-participants
Per hectare (per 50 kg bag) labor use in absolute terms for cultivating and processing of all crops for participants and non-participants.

Future. Indaba Agricultural Policy Research Institute (IAPRI), Lusaka, pp. 32–46.
Daum, T., Buchwald, H., Gerlicher, A., Birner, R., 2019. Times have changed: using a
Total Labor

pictorial smartphone app to collect time-use data in rural Zambia. Field Meth. 31 (1),
hours/ha

3–22.
1133

Daum, T., Buchwald, H., Gerlicher, A., Birner, R., 2018. Smartphone apps as a new
226

375

328
92
60

43

method to collect data on smallholder farming systems in the digital age: a case study
9

from Zambia. Comput. Electron. Agric. 153, 144–150.

Daum, T., Birner, R., 2017. The neglected governance challenges of agricultural me-
Children family

chanisation in Africa–insights from Ghana. Food Security 9 (5), 959–979.

Davis, B., Di Giuseppe, S., Zezza, A., 2017. Are African households (not) leaving agri-
labor/ha

culture? Patterns of households’ income sources in rural Sub-Saharan Africa. Food

Policy 67, 153–174.
15

Dehejia, R.H., Wahba, S., 2002. Propensity score-matching methods for nonexperimental
31
1
4
3
8
0

causal studies. Rev. Econ. Stat. 84 (1), 151–161.

Deininger, K., Byerlee, D., 2011. Rising Global Interest in Farmland - Can it Yield
Female family

Sustainable and Equitable Benefits? Management. World Bank, Washington, D.C.

Deininger, K., Byerlee, D., 2012. The rise of large farms in land abundant countries: do
labor/ha

they have a future? World Dev. 40 (4), 701–714.

De Janvry, A., Sadoulet, E., 2016. Development Economics: Theory and Practice.
124

236
26
18
63
3

Routledge.
De Janvry, A., Dustan, A., Sadoulet, E., 2010. Recent advances in impact analysis methods
for ex-post impact assessments of agricultural technology. A discussion on estimating
Male family

treatment effects, organized by the CGIAR Standing Panel on Impact Assessment

labor/ha

(SPIA), 2 October, 2010, Berkeley, California, USA.

Diao, X., Cossar, F., Houssou, N., Kolavalli, S., 2014. Mechanization in Ghana: emerging
105

212
20
18
51
12
4

demand, and the search for alternative supply models. Food Policy 48, 168–181.
Diao, X., Cossar, F., Houssou, N., Kolavalli, S., Jimah, K., Aboagye, P., 2012.
Mechanization in Ghana: Searching for Sustainable Service Supply Models. IFPRI
Family labor

Discussion Paper 01237, International Food Policy Research Institute, Washington,

hours/ha

D.C.
Diao, X., McMillan, M.S., Mwangwe, S., 2018. Agricultural labour productivity and in-
122

244

481
51
39

13

dustrialisation: lessons for Africa. J. Afr. Econ. 27 (1), 28–65.

8

FAO & UNIDO, 2008. Agricultural Mechanization in Africa. Time for Action. United
Nations Industrial Development Organization (UNIDO) and Food and Agriculture
Hired labor

Organization (FAO), Vienna.

hours/ha

Feder, G., Birner, R., Anderson, J.R., 2011. The private sector’s role in agricultural ex-
tension systems: potential and limitations. J. Agribus. Dev. Emerg. Econ. 1 (1),
163
29

51

63
11

31–54.
1

Giller, K.E., Witter, E., Corbeels, M., Tittonell, P., 2009. Conservation agriculture and
smallholder farming in Africa: the heretics’ view. Field Crops Res. 114 (1), 23–34.
Participants

Total Labor

Hayami, Y., Ruttan, V., 1985. Agricultural Development: An International Perspective.

hours/ha

Johns Hopkins University Press, Baltimore, MD.

Houssou, N., Diao, X., Cossar, F., Kolavalli, S., Jimah, K., Aboagye, P.O., 2013.
174

314

645
10
81
47

14

Agricultural mechanization in Ghana: is specialized agricultural mechanization ser-

vice provision a viable business model? Am. J. Agric. Econ. 95 (5), 1237–1244.
https://doi.org/10.1093/ajae/aat026.
Processing (hours per

IAPRI, 2015. Rural Agricultural Livelihoods Survey: 2015 Survey Report. Indaba
Pests/disease control
Fertilizer application

Agricultural Policy Research Institute (IAPRI), Lusaka.

Land preparation

ILO, 1973. Mechanization and employment in agriculture: case studies from four con-
tinents. Geneva.
Harvesting

50 kg)

Jahnke, H.E., 1983. Livestock production systems and livestock development in tropical
Weeding
Table 12

Planting

Africa. Am. J. Agric. Econ. 65.

Total

Juma, C., 2016. Innovation and its enemies: why people resist new technologies. Oxford
University Press, New York.

12
F. Adu-Baffour, et al. Food Policy xxx (xxxx) xxx–xxx

Kariuki, J., Njuki, J., 2013. Using participatory impact diagrams to evaluate a community Rosenbaum, P., Rubin, D., 1985. Constructing a Control Group Using Multivariate
development project in Kenya. Dev. Pract. 23 (1), 90–106. Matched Sampling Methods That Incorportate the Propnsity Score. American
Khandker, S.R., Koolwal, G.B., Samad, H.A., 2010. Handbook on Impact Evaluation: Statistical Association, 30. No. 1.
Quantitative Methods and Practices. World (Vol. 41). World Bank, Washington, DC. Sallah, P., Twumasi-Afriyie, S., Kasei, N., 1997. Optimum planting dates for four maturity
Kienzle, J., Ashburner, J.E., Sims, B.G. (Eds.), 2013. Mechanization for Rural groups of maize varieties grown in the Guinea savanna zone. Ghana J. Agric. Sci. 30,
Development: A Review of Patterns and Progress from around the World, vol. 20. 63–69.
Food and Agricultural Organization (FAO), Rome. Schneider, F., Don, A., Hennings, I., Schmittmann, O., Seidel, S.J., 2017. The effect of
Knowler, D., Bradshaw, B., 2007. Farmers’ adoption of conservation agriculture: a review deep tillage on crop yield–What do we really know? Soil Tillage Res. 174, 193–204.
and synthesis of recent research. Food Policy 32 (1), 25–48. Searchinger, T.D., Estes, L., Thornton, P.K., Beringer, T., Notenbaert, A., Rubenstein, D.,
Low, A., Waddington, S.R., 1991. Farming systems adaptive research: achievements and Herrero, M., 2015. High carbon and biodiversity costs from converting Africa’s wet
prospects in southern Africa. Exp. Agric. 27 (2), 115–125. savannahs to cropland. Nat. Clim. Change 5 (5), 481.
Martínez-Torres, M.E., Rosset, P.M., 2010. La vía campesina: the birth and evolution of a Sheahan, M., Barrett, C.B., 2017. Ten striking facts about agricultural input use in Sub-
transnational social movement. J. Peasant Stud. 37 (1), 149–175. Saharan Africa. Food Policy 67, 12–25. https://doi.org/10.1016/j.foodpol.2016.09.
Massey, D.S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., Taylor, J.E., 1993. 010.
Theories of international migration: a review and appraisal. Popul. Dev. Rev. Sitko, N.J., Jayne, T.S., 2014. Structural transformation or elite land capture? The growth
431–466. of “emergent” farmers in Zambia. Food Policy 48, 194–202.
Mrema, C.G., Baker, D., Kahan, D., 2008. Agricultural Mechanization in Sub-Saharan Tavenner, K., van Wijk, M., Fraval, S., Hammond, J., Baltenweck, I., Teufel, N., Kihoro, E.,
Africa: Time for a New Look. Food and Agricultural Organization (FAO), Rome. de Haan, N., van Etten, J., Steinke, J., Baines, D., Carpena, P., Skirrow, T.,
Olmstead, A.L., Rhode, P.W., 1995. Beyond the threshold: an analysis of the character- Rosenstock, T., Lamanna, Ng'endo, M., Chesterman, C., Namoi, N., Manda, L., 2019.
istics and behavior of early reaper adopters. J. Econ. History 55 (1), 27–57. Intensifying Inequality? Gendered trends in commercializing and diversifying
Pingali, P., 2007. Agricultural mechanization: adoption patterns and economic impact. In: smallholder farming systems in East Africa. Front. Sustain. Food Syst. 3, 10. https://
Evenson, R., Pingali, P. (Eds.), Handbook of Agricultural Economics, Volume 3. doi.org/10.3389/fsufs.
North-Holland, pp. 2779–2805. Tembo, S., Sitko, N., 2013. Technical Compendium: Descriptive Agricultural Statistics
Pingali, P., Bigot, Y., Binswanger, H.P., 1987. Agricultural Mechanization and the and Analysis for Zambia. Lusaka: Working Paper 76, Indaba Agricultural Research
Evolution of Farming Systems in Sub-Saharan Africa. John Hopkins University, Institute (IAPRI).
Baltimore and London. WFP, 2008. Food consumption analysis: Calculation and use of the food consumption
Quisumbing, A.R., Maluccio, J.A., 2000. Intrahousehold Allocation and Gender Relations: score in food security analysis. Retrieved from www.wfp.org/odan/senac.
New Empirical Evidence from Four Developing Countries. International Food Policy World Bank, W.B., 2007. World Development Report 2008: Agriculture for Development.
Research Institute, Washington, DC. World Bank, Washington, DC.

13