You are on page 1of 58

Cropping system

diversification in Eastern
and Southern Africa
Identifying policy options to enhance
productivity and build resilience

FAO AGRICULTURAL DEVELOPMENT ECONOMICS


z

WORKING PAPER 18-05


ISSN 2521-1838

September 2018
Cropping system
diversification in Eastern
and Southern Africa
Identifying policy options to enhance
productivity and build resilience

Giuseppe Maggio, Nicholas J. Sitko and Ada Ignaciuk

Food and Agriculture Organization of the United Nations


Rome, 2018
Required citation:

Maggio, G., Sitko, N. & Ignaciuk, A. 2018. Cropping system diversification in Eastern and Southern Africa: Identifying policy
options to enhance productivity and build resilience. FAO Agricultural Development Economics Working Paper 18-05. Rome,
FAO. Licence: CC BY-NC-SA 3.0 IGO.

The designations employed and the presentation of material in this information product do not imply the expression of any
opinion whatsoever on the part of the Food and Agriculture Organization of the United Nations (FAO) concerning the legal or
development status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or
boundaries. The mention of specific companies or products of manufacturers, whether or not these have been patented, does
not imply that these have been endorsed or recommended by FAO in preference to others of a similar nature that are not
mentioned.

The views expressed in this information product are those of the author(s) and do not necessarily reflect the views or policies
of FAO.

ISBN 978-92-5-130981-0

© FAO, 2018

Some rights reserved. This work is made available under the Creative Commons Attribution-NonCommercial-ShareAlike
3.0 IGO licence (CC BY-NC-SA 3.0 IGO; https://creativecommons.org/licenses/by-nc-sa/3.0/igo).

Under the terms of this licence, this work may be copied, redistributed and adapted for non-commercial purposes, provided that
the work is appropriately cited. In any use of this work, there should be no suggestion that FAO endorses any specific
organization, products or services. The use of the FAO logo is not permitted. If the work is adapted, then it must be licensed
under the same or equivalent Creative Commons license. If a translation of this work is created, it must include the following
disclaimer along with the required citation: “This translation was not created by the Food and Agriculture Organization of the
United Nations (FAO). FAO is not responsible for the content or accuracy of this translation. The original [Language] edition
shall be the authoritative edition.

Any mediation relating to disputes arising under the licence shall be conducted in accordance with the Arbitration Rules of the
United Nations Commission on International Trade Law (UNCITRAL) as at present in force.

Third-party materials. Users wishing to reuse material from this work that is attributed to a third party, such as tables, figures
or images, are responsible for determining whether permission is needed for that reuse and for obtaining permission from the
copyright holder. The risk of claims resulting from infringement of any third-party-owned component in the work rests solely with
the user.

Sales, rights and licensing. FAO information products are available on the FAO website (www.fao.org/publications)
and can be purchased through publications-sales@fao.org. Requests for commercial use should be submitted via:
www.fao.org/contact-us/licence-request. Queries regarding rights and licensing should be submitted to: copyright@fao.org.
Contents

Abstract ............................................................................................................................... v
Acknowledgements ............................................................................................................. vi
Introduction ......................................................................................................................... 1
Unpacking diversification: a conceptual framework ............................................................. 2
Data and descriptive statistics ............................................................................................. 7
1.1 Data sources ......................................................................................................... 7
1.2 Summary statistics by cropping system in Malawi, Mozambique and Zambia ........ 8
1.3 The geography of cropping systems .................................................................... 14
Research design ............................................................................................................... 16
1.4 Adoption equation ................................................................................................ 16
1.5 Outcome equation ............................................................................................... 17
Results and discussion ...................................................................................................... 18
1.6 Determinants of cropping system adoption: policy lessons from cross country
comparison .......................................................................................................... 18
1.7 Country specific insight of cropping systems adoption ......................................... 23
1.8 Impact on productivity .......................................................................................... 24
1.9 Impact on crop income volatility ........................................................................... 27
Conclusions....................................................................................................................... 29
References ........................................................................................................................ 31
Annex 1 ............................................................................................................................. 35

iii
Tables
Table 1. Crops within each crop categories ...................................................................................... 4
Table 2. Most frequent crop combinations within each cropping system ........................................ 10
Table 3. Summary statistics by cropping system for Malawi ........................................................... 11
Table 4. Summary statistics by cropping system for Mozambique ................................................. 12
Table 5. Summary statistics by cropping system for Zambia .......................................................... 13
Table 6. Drivers of cropping systems adoption in Malawi ............................................................... 20
Table 7. Drivers of cropping systems adoption in Mozambique ...................................................... 21
Table 8. Drivers of cropping systems adoption in Zambia .............................................................. 22
Table 9. Effect of farming systems on crop income volatility........................................................... 28

Figures
Figure 1. Prevalent cropping system at district level ........................................................................ 15
Figure 2 Effect of different cropping systems on maize yield compared to maize mono-cropping
in Malawi, Mozambique and Zambia ................................................................................. 25

iv
Cropping system diversification in Eastern and Southern Africa
Identifying policy options to enhance productivity and build resilience

Giuseppe Maggio1, Nicholas Sitko1, Ada Ignaciuk1

1 Food and Agriculture Organization of the United Nations, Agricultural Development Economics
Division (ESA), Viale delle Terme di Caracalla, 00153 Rome, Italy

Abstract
Crop diversification is an important policy objective to promote climate change adaptation,
yet the drivers and impacts of crop diversification vary considerably depending on the specific
combinations of crops a farmer grows. This paper examines adoption determinants of seven
different cropping systems in Malawi, Zambia and Mozambique, and the impact of their
adoption on maize productivity and income volatility – using a multinomial endogenous
treatment effect model. These cropping systems consist in different combinations of four
categories of crops: dominate staple (maize), alternative staples, legumes, and cash-crops.
The study finds that relative to maize mono-cropping systems, the vast majority of systems
have either neutral or positive effects on maize productivity, and either reduce or have neutral
effects on crop income volatility. In particular, cropping systems that include legumes produce
better outcome in most cases than those that feature cash crops. From a policy perspective,
three recurrent determinants of diversification are found. First, private sector output market
access is an important driver of diversification out of maize mono-cropping. Policies crowding
in private output market actors can help to promote a wide range of more diverse cropping
systems. Second, proximity to public marketing board buying depots discourages the
adoption of more diverse cropping systems. Therefore, reforms to these institutions must be
part of any diversification strategy. Finally, in all countries and for all systems, land size is a
key determinant of adopting more diverse systems. Thus, land policy is an integral element
of any boarder diversification strategy.

Keywords: adoption, cropping systems, crop income volatility, diversification, yield,


East Africa.
JEL codes: O13, Q12, Q18, Q54, R20.

v
Acknowledgements
This paper is produced in the context of the programme of work on economics and policy
innovations for climate-smart agriculture within FAO’s Agricultural Development Economics
Division (ESA). The Food and Agriculture Organization of the United Nations would like to
thank the Flanders Cooperation for the financial support.
We greatly appreciate the precious inputs, help, reviews, and comments received from
Solomon Asfaw, Antonio Bubbico, Valentina Conti, Uwe Grewer, Antonio Scognamillo and
Stefanija Veljanoska. This paper reflects the opinions of the authors and not the institution
which they represent or with which they are affiliated. We are solely responsible for
any errors.
Corresponding author: nicholas.sitko@fao.org

vi
Introduction
Many governments in sub-Saharan Africa (SSA) identify crop diversification as a key element
of their climate change adaption strategies. Of the 31 countries in sub-Saharan Africa that
have submitted a National Adaptation Programmes of Action to the United Nation Framework
Convention on Climate Change (UNFCC), eight have projects dedicated specifically to
promoting crop diversification. Crop diversification is thought to enhance adaptation to climate
change in three different ways. First, by spreading agriculture production and income risk over
a range of crops with different degrees of sensitivity to climate induced crop loss and market
conditions (Hahn, Riedered and Foster, 2009; Dercon, 1996). Second, where diversification
entails the adoption of more commercially oriented crops, crop diversification may help to
increase household’s income, leading to a reduction of livelihood vulnerability (Krupinsky et
al., 2002). Finally, crop diversification may provide agronomic benefits, including differential
nutrient uptake, disease management and soil nitrogen fixation, which may help to enhance
or stabilize crop productivity in the presence of climate change (King and Hofmockel, 2017).
To provide evidence that policy makers may find useful for identifying specific crop
diversification strategies and policies to promote these, this study makes use of nationally
representative panel and cross-sectional data collected from smallholder farmers in Malawi,
Mozambique and Zambia. Using econometric techniques, this work moves beyond abstract
definitions of crop diversification, such as the Gini Simpson’s Index (Arslan et al., 2018), to
assess the determinants of adopting seven different cropping systems and the effects of their
adoption on maize productivity and crop income volatility. By so doing, the study is able to
provide concrete insights on the effects of different diversification strategies of famer’s welfare
and resilience, as well as on possible policy strategies for promoting beneficial cropping
systems.
The study finds that relative to maize mono-cropping systems, the vast majority of cropping
systems adopted in these countries have ether neutral or positive effects on maize
productivity, and either reduce or have neutral effects on crop income volatility. In particular,
cropping systems that include legumes produce better outcomes in most cases than those
that feature cash crops. From a policy perspective, three recurrent features are found to help
or hinder the adoption of more diverse systems in all three countries. First, private sector
output market access is an important driver of diversification. Policies that crowd in private
sector output market actors can help to promote a wide range of more diverse cropping
systems. Second, proximity to public marketing board buying depots discourages the adoption
of more diverse cropping systems. Therefore, reforms to these institutions must be part of any
diversification strategy. Finally, in all countries and for all systems, land size is a key
determinant of adopting more diverse cropping systems relative to maize mono-cropping.
Thus, land policy is an integral element of any boarder diversification strategy.
In support of these findings this paper is organized as follows. Section 2 outlines the
conceptual framework for understanding crop diversification. Section 3 presents the data
sources. Section 4 presents the research design. Section 5 introduces summary statistics and
results. Section six concludes.

1
Unpacking diversification: a conceptual framework
Diversification is often considered an important ex-ante strategy deployed by farmers to
manage production and income variability in the context of climate change. Through crop
diversification, farmers spread the risk of crop failure and productivity loss due to weather
events, thus helping to smooth consumption when these events occur. Economic theory,
however, suggests that there are potential trade-offs between managing risks and maximizing
returns (Arslan et al., 2018). Whether or not diversification entails foregoing returns in order
to limit risk is likely a function of whether or not households are pulled into a particular
diversification strategy or pushed (Barrett and Reardon, 2000).
Push factors for crop diversification include high transactions costs in input and output
markets, adverse shocks, including weather shocks, imperfect or absent credit and insurance
markets and stagnation in the agriculture sector (Barrett, Reardon and Webb, 2001; Arslan et
al., 2018). Risk adverse households with limited resources to cope with production variability
associated with climate shocks may be more likely to be pushed into diversification strategies
than better endowed households (Arslan et al., 2018). Where push factors dominate, crop
diversification may not improve productivity or incomes, but may help to stabilize them (Arslan
et al., 2018; Barrett, Reardon and Webb, 2001).
Conversely, diversification pull factors predominate under conditions of crop sector dynamism,
where new market opportunities or practices/technologies induce risk-taking farmers to adopt
new crops into their farm systems. Pull factors may be associated with household strategies
to capture the synergistic benefits associated with, for example, legume intercropping with
maize. Under these conditions, diversification is likely associated with improvements in
average incomes and reduced variability, and may contribute to higher productivity (Arslan et
al., 2018; Reardon et al., 2007). Where pull factors for diversification exist, they are often more
pronounced for wealthier, better off households with reasonably good market access
conditions (Davis et al., 2010).
To better understand the causes and consequences of crop diversification, this study develops
a taxonomy of cropping systems for maize producing households based on four functional
crop categories (see Table 1). These crop categories are maize, legumes, non-maize staples
and cash crops. These crop categories appear in different combinations within smallholder
systems in Malawi, Zambia and Mozambique. The observed combination of crop categories
we refer to as a cropping system. We hypothesize that the crop categories are associated with
differing degrees of commercialization potential, climate risk management potential and
agronomic benefits for maize producing households. If this is the case, then different
combinations of these crop categories—i.e. the cropping systems— should produce different
outcomes in terms of productivity and income variability and adoption will be concentrated
among households with shared attributes.

Crop categories
Maize: national statistics show that maize is cultivated by 50 percent of the population in sub-
Saharan Africa, with 46 out of 53 countries growing maize. In Zambia, Malawi and
Mozambique it is the primary staple and most widely grown crop by resource poor smallholder
farmers (Mazunda and Droppelman, 2012; Mafongoya and Jiri, 2015). Maize is grown by 73
and 79 percent of farmers in Zambia and Mozambique, respectively, while this share increase
to 97 percent for Malawi, where maize accounts for about 60 percent of total caloric
consumption (Denning et al., 2009; JAICAF, 2008). Maize productivity in the region is low by

2
global standards, hovering barely over two tonnes per hectare, and is hindered by a range of
factors, including degrading soil conditions, labour constraints, insufficient use of improved
maize varieties, and inappropriate and limited application of organic and inorganic fertilizer.
As the primary agricultural product and food crop in the region, maize functions as both a cash
and subsistence crop, depending on the scale of production. Despite its prevalence in rain-
fed smallholder systems, maize productivity is sensitive to low and variable rainfall conditions,
while maize mono-cropping can rapidly degrade soils (Kumwenda et al., 1997; Rockström et
al., 2009). Thus, while maize production offers opportunities for commercialization and income
earnings for farmers that produce a surplus, maize production, particularly in mono-cropping,
exposes farmers to significant climate risks and can undermine future growth by degrading a
farm’s soil resource base.
Legumes: this category comprises beans, soybeans, pigeon peas, groundnuts and green
beans among others. Literature suggests that the incorporation of legumes into a cropping
systems will produce beneficial effects on both maize yields and income stability relative to
maize mono-cropped systems. However, some differences persist between different types of
legumes. Short-season legumes, such as groundnuts, are more easily marketable and
therefore, more suitable for commercial-oriented systems. In contrast, long-season legumes,
such as pigeon-pea, help improving maize yields by enhancing phosphorus and fixing a
significant larger amount of nitrogen, a key nutrient in maize systems (Kerr et al., 2007). In
addition, annual legume cover crops have a determinant role in weed suppression (Becker
and Johnson, 1999; Sileshi et al., 2008). Legumes can also improve crop income stability by
stabilizing production through improvements in soil quality (Sileshi et al., 2008).
Non-maize staples: this category comprises common alternative carbohydrate sources such
as cassava, sweet potato, rice, millet and sorghum. These alternative-staples will likely have
a neutral to negative effect on maize yields, as they do not have obvious agronomic benefits
for maize systems, but may divert scarce production factors, namely capital and labour, away
from maize. However, non-maize staples are likely to have a positive or neutral effect on
income stabilization either by providing additional market opportunities for farmers, particularly
where commercial markets for non-maize staples exist, or through their tolerance to weather
shocks relative to maize. For example, cassava is a drought-resistant perennial crop which
survives to rainfall shortfalls that often disrupt maize production. Similarly, sorghum and millet
are often less vulnerable to the effect of increasing weather volatility than maize and thus may
help to increase income stability relative to maize mono-cropping (Schlenker et al., 2010).
Cash Crops: this category is comprised of major crops in the region, such as cotton and
tobacco, as well as more minor crops, such as sunflower, cashews and sugar cane. A priori,
the effects of adoption of this crop category on maize yields and income stability are uncertain.
Cash crop production is often carried out under input provisioning contracts, which may enable
resource-poor households to enter into more intensive, commercially oriented production
(Govereh and Jayne, 2003). By facilitating access to inputs and formal output markets, cash
crop production may have positive spill-over effects on maize intensification for farmers
(Chapoto et al., 2013). However, when labour is scarce, cash crop production may divert
labour away from the sensible management practices for maize, such as weeding (Haggblade,
Plehoples and Kabwe, 2011). In terms of income stability, some authors argue that cotton
production, one of the main cash-crops in this analysis, is expected to increase with future
climate change due to the plant’s relative tolerance of drought conditions (Agrawala et al.,
2003; Morton, 2007). Nevertheless, cash crops are typically globally traded and thus are more

3
exposed to global market prices’ fluctuations, which can produce high crop-income variability
(Fafchamps, 1992).

Table 1. Crops within each crop categories

Crops Crop Categories

Bean, soybean, pigeon pea, groundnut, green bean Legumes

Cotton, tobacco, sunflower, sugarcane, sesame and tea Cash crops

Cassava, rice, millet, sorghum, wheat, Irish potato and sweet potato Non-maize staples

Maize Maize

Cropping systems
The four cropping categories described above are combined within smallholder maize-based
systems in seven different combinations, what we refer to as cropping systems. These
combinations represent different diversification strategies and are likely adopted by different
farm household types, due to different combinations of push and pull factors, and produce
different outcomes in terms of maize productivity and income stability.

Maize mono-cropping (MM)


This system is the reference system in this study. It is hypothesized that a large share of
farmers in this cropping system is resource poor, with limited access to off-farm income, and
has been pushed into maize mono-cropping in an effort to meet household’s staple food
consumption requirements (Nielson, 2009). Adopters, however, may also be pulled in mono-
cropping by the presence of public demand for maize, as in the case of marketing boards and
food reserves in Zambia or Malawi, as well as input subsidy programmes that support input
access for maize (Sitko et al., 2017).

Maize-legume (ML)
This system is likely dominated by subsistence-oriented farmers, focused primarily on legume
production for home consumption. Legume harvesting and planting are more labour intensive
than maize, thus household size, access to hired labour or social networks to access labour
may help to pull farmers into this system (Nhemachena and Hassan, 2007). In addition, due
to the well-established agronomic benefits of incorporating legumes into maize-based
systems, access to extensions information may be an important pull factor driving adoption of
this system. Finally, although most legumes production is likely for home consumption, in
Mozambique and Malawi export markets for pigeon peas and groundnuts (in the case of
Malawi) are important. Similarly, in Zambia, growth in the animal fed and oil expelling
industries is driving demand growth for soybeans (Sitko, Burke and Jayne, 2017). Thus, the
presence of a private agricultural market may be an important pull factor for this category.
Because of the properties of legume when intercropped, ML system are expected to show
higher yield and more income stability than in the maize mono-cropping system.

4
Maize and staple (MS)
Alternative staples in this system likely serve to lower the production and market risks
associated with maize mono-cropping. There are several reasons for this. First, they provide
a more drought-tolerant source of carbohydrate than maize. In places where droughts are
frequent, households may be pushed into diversifying carbohydrate production to meet
household food security needs. This may be particularly the case for households in more
remote market locations, for whom access to retail food markets and off-farm income is more
limited. Second, alternative staples offer some opportunities for income generation and
income stabilization, with potential markets tied to the commercial beer and livestock feed
industries (Garrity et al., 2010). Compared to maize mono-cropping, the impact of MS systems
on productivity is expected to be between negative and neutral, because of high factors
competition between the two crops. In terms crop income, this system is expected to show
higher stability compared to maize mono-cropping due to the drought tolerance of most
alternative staples (Nielson, 2009; Thierfelder et al., 2016).

Maize and cash crops (MC)


Adopters of this system are likely commercially oriented or seeking to transition from
subsistence to commercially oriented production (Chapoto et al., 2013). Resource poor
farmers are likely pulled into this system by the availability of input credit for production and
commercial output markets. Yet, due to significant price volatility caused by global supply and
demand conditions, the inclusion of cash crops entails substantial risks, particularly when
production risks are not offset by a more diverse production system. Access to input credit
and commercial markets may enable resource poor farmers to overcome capital constraints
to maize intensification. However, competition for labour and other resources between maize
and cash crops may translate into lower maize yields relative to mono-cropped systems.

Maize-legume-staples (MLS)
Of the more diversified cropping systems, with three or more crop categories involved, farmers
adopting the MLS system are likely motivated primarily by production and price risk
management, rather than productivity and income growth. Low market access conditions,
access to extension services, historical experiences with weather shocks, combined with
larger land holdings to enable a three-crop system may combine to push and pull farmers into
this system. However, in some cases, an alternative pathway may dominate. In higher market
access regions, the level of commercialization of certain legumes and alternative staples
market opportunities may pull farmers into the Maize-Legume-Staple (MLS). Trade-offs
between the agronomic benefits of incorporating legumes into a maize-based system and
competition between production factors makes it difficult to anticipate the effects of this system
on maize yields. However, a positive effect of income stabilization can be expected by the
combination of legume with alternative staples (Dapaah et al., 2003).

Maize-legume-cash crop (MLC)


Adopters of this crop system are likely to be more commercial-oriented and risk taking relative
to the MLS system. The presence of marketing infrastructures, credit institutions, informal
social networks, and agribusiness firms are all potential pull factors for adopters of this system.
These factors, indeed, may help farmers in coping with the potential risk of crop failure, by
relying access to credit, but also in accessing to the high level of inputs associated with this
commercial production. The diffusion of agricultural extension services and the availability of
large landholdings may represent other pull factors for the adoption of this cropping system

5
(Gladwin, 1983). Because of the crops’ characteristics, this system is expected to increase
maize productivity, while having a neutral to positive effect on crop income stability (Yadav,
1981; Bantinal and Rao, 1999).

Maize-legume-staples-cash crops (MLSC)


This is the most diversified cropping system, which balances opportunities for
commercialization with price and production stabilization benefits derived from the
incorporation of alternative staples and legumes. Adopters of this farming system are likely
characterized by a lower propensity to risk than MLCs but higher than the baseline. They likely
have significant land and labour assets, which help enable a diverse production system.
Climate stressors and market opportunities are the main push and pull factors for this system.
Depending on the relative allocation of each crop in this system, it is possible to expect high
benefits in terms of yield and stability of income.

6
Data and descriptive statistics

1.1 Data sources


This study uses household survey data collected from smallholder farmers in Malawi,
Mozambique and Zambia merged with geo-referenced climatic records to study the impact of
the cropping systems on yield level and crop income variability. The analysis relies on four
sources: 1) socio-economic panel household, field and community data from a) Malawi
Integrated Household Panel Survey (IHPS); b) Zambia Rural Agricultural Livelihoods Survey
(RALS); c) cross-sectional data from the Mozambique Inquérito Agrário Integrado (2015); and
2) historical data on rainfall and temperature from the National Oceanic and Atmospheric
Administration (NOAA), Climate Hazard Infrared Precipitation with Station Data (CHIRPS) and
the European Centre for Medium Range Weather Forecasts (ECMWF).
Malawi’s IHPS is a national representative survey conducted by the Government of Malawi
with the support of the World Bank Living Standards Measurement Study. For Malawi, the
panel sample includes 1 989 households cultivating maize, interviewed during first the rainy
season 2009/2010 and then during 2012/2013, for a total of 3 978 observations.
The RALS is a panel survey conducted jointly by the Zambian Central Statistical Office (CSO),
Michigan State University (MSU), the Indaba Agricultural Policy Research Institute (IAPRI)
and the Ministry of Agriculture and Livestock. This survey is national representative of rural
farm households cultivating less than 20 hectares for farming purposes and/or raising of
livestock. RALS covers 10 Provinces, 150 constituencies and 476 enumeration areas. For
Zambia, the panel includes 5 948 households cultivating maize and observed during
2010/2011 and 2013/2014 agricultural seasons.
Mozambique’s IAI survey derives from the Trabalho de Inquerito Agricola (TIA), a yearly
survey project developed by the Michigan State University (MSU) together with the Ministry of
Agriculture of Mozambique (MINAG). IAI 2015 counts 5 055 observations, it is representative
both at national and provincial level, and it includes observations for 11 provinces, 147 districts
and 780 enumeration areas.
For all the surveys, the enumerators have administered a multi-topic questionnaire. Data are
available on household characteristics, salary (apart from Mozambique), food and non-food
consumption, food security, assets and agricultural production. Since maize mono-cropping is
the baseline category, the final sample includes only households that have cultivated maize
during both waves.
The analysis on determinants and impact of diversification considers three groups of factors:
institutional, biophysical, resource endowment and other socio-economic factors. The
institutional factors are captured through proxies of private input and output market and public
output markets. Public markets are measured as the median distance from all the households
in a given village to the closest marketing board buying depot for Malawi (ADMARC) and
Zambia (Food Reserve Agency).1 In Mozambique there is no public marketing board, therefore
this variable is excluded from analysis. The extent of the private input market is measured
using the median seed prices at village level (not available in Mozambique). In terms of private
output markets, available information differs substantially across the three surveys. For
Zambia, private market is measured in terms of the number of private traders that buy grain

1 We consider the median rather than the average to exclude that the presence of any outlier is driving the estimated
effect of this variable.

7
in a given community. Malawi’s specification includes the median distance from the weekly
market computed at village level. Unfortunately, Mozambique’s questionnaire does not include
any relevant information on the presence of an agricultural market. In this case, therefore, the
private market indicator is constructed using the distance of the households’ district of
residence from the main regional city obtained from the GRUMP dataset (Balk et al., 2006).2
The biophysical factors are operationalized as the lagged value of the positive and negative
SPI rainfall indexes as indicators, respectively, of drought and flood occurring in the
agricultural season before the interview. The set of variables also includes dummies on agro-
ecological zones to control for local agro-ecological differences. Among the resource
endowment and other socio-economic factors, we consider the agricultural asset wealth index,
calculated using a principal component analysis (PCA) on dummies indicating households’
ownership of a range of agricultural tools.3 The second and third resource indicators relate to
the size of land and of number of livestock owned, measured in hectares and TLU units,
respectively. The remaining subset of controls involves a) household demographics, such as
household size, age and gender of the head of the household, average years of education;4
b) share of households accessing to credit and Farmer Input Subsidies program, measured
at community level; c) a dummy capturing the usage of inorganic fertilizer.5
We extract rainfall records for Malawi and Mozambique from the Africa Rainfall Climatology
version 2 (ARC2), a platform managed by the National Oceanic and Atmospheric
Administration’s Climate Prediction Centre (NOAA-CPP). ARC2 provides rainfall records
elaborated on a 10-day interval from January 1983 to July 2015, with a spatial resolution of
0.1 degrees. Rainfall data for Zambia derives from the Climate Hazards Group InfraRed
Precipitation with Station data (CHIRPS), with 0.5 degree of spatial resolution. The European
Centre for Medium Range Weather Forecast (ECMWF) delivers information on the surface
maximum, minimum and average temperature on a 10-day interval with a spatial resolution of
0.25 degrees. The sources of temperature data are both the operational database (1989-
2010) and the interim database (2011-2013). GPS location from the Malawi and Zambia
surveys allows the merge with climatic data. For Mozambique, the merge is at district level,
the only public available geographical location in the data.

1.2 Summary statistics by cropping system in Malawi, Mozambique


and Zambia
In this section, descriptive data on households in each of the seven cropping systems is
examined in order to identify important difference between households in the different
systems, as well as commonalities and difference between countries. What are the primary

2 The list of cities employed is available in Table A8 in Annex 1. To test the robustness of Mozambique’s results
and exclude that the coefficient on private market would capture the effect of population, urbanization, and
infrastructural development, we tested alternative specifications including controls on level of population at district
level (Linard et al., 2012), nightlight index (Cecil, Buechler and Blakeslee, 2014), and road density. The results are
consistent with the ones presented in the main part of the paper.
3 In terms of farm assets, there is not a unique evidence, but literature suggests that households with smaller land

owned are more likely to diversify their crops while livestock ownership is positively associated to multiple adoption
as it represents store value and a source for manure (Deressa et al., 2009).
4 For example, education is expected to have a positive impact on multi-cropping and adaptation, as better

educated households hold more knowledge of the potential benefit deriving from these (Dolisca et al., 2006; Anley
et al., 2007; Tizale 2007).
5 Fertilizer and compost often occur during intercropping (Silberg et al., 2017) but some evidence exists about the

crowding out effect of subsidized fertilizer on non-legume intecropping (Levine et al., 2016).

8
crops grown in each of the cropping systems and how do these vary by country? Three most
frequent crop combinations within each cropping system in each country are summarized in
Table 2. The value in parenthesis refers to the percentage of households adopting a given
combination within the system, but these combinations are not mutually exclusive. It shows
substantial cross-country differences in the type of crops adopted within the same system in
different countries. For the ML system, groundnut is the most widely cultivated crop in Malawi
(66 percent) and Zambia (84 percent), while Mozambican farmers combine more often maize
with beans (89 percent). Among the staples crops, cassava is planted by about 63-64 percent
of MS adopters in Mozambique and Zambia, but not in Malawi and it is widely grown also by
MLS households. Nevertheless, in Malawi cassava has not a prevalent role in this system, as
it represents only the third choice for MS farmers, and it is absent in the three most common
MLS and MLSC combinations.
The analysis shows three important commonalities between the three countries, which are
summarized in Table 3, Table 4 and Table 5. First, maize mono-cropping in all three countries
is dominated by households with the smallest average land sizes and the lowest average
levels of crop income. This suggests that the majority of maize mono-cropping households
are pulled into this system in an effort to meet household food security needs given scarce
resources, and may be locked into this system by limited commercialization and crop income
earning opportunities.
Second, the presence of cash crops, either in combination with maize (MC) or in more diverse
systems (MLC and MLCS), is associated with higher crop incomes in all countries. This is
likely due to the combination of more structured private marketing arrangements for these
crops, and the prevalence of outgrower arrangements in cash crop systems. In all three
countries, participation by female-headed households is lowest in cash cropping systems,
suggesting that female household heads are either structurally excluded from entering into
cash cropping arrangements, prioritizing the production of consumable agricultural products
more than male-headed households, or some combinations of the twos.
Finally, more diverse cropping systems (with three or more crop categories) are associated
with higher crop incomes, larger land sizes, and in many cases higher average maize yields
than less diverse systems. Land size is likely an important determinant of beneficial
diversification. Smallholder farmers with larger landholdings have greater flexibility to
incorporate a range of crops into their production systems, while still dedicating sufficient land
to staple food production. By so doing they are able to capture potential agronomic and
economic advantages associated with alternative crops, thus pushing them into a beneficial
cycle of improved crop income, greater investment capital for farm activities, and higher
productivity.
Several important differences between countries are worth highlighting. First, in land
constrained Malawi, maize-mono-cropping is the most prevalent cropping system.
Conversely, in Zambia and Mozambique, the most prevalent system is the MLS system.
Average land holdings in Malawi are substantially lower for all systems than in the other two
countries, which hinders the capacity of farmers in Malawi to adopt more diverse systems.
The dominance of maize mono-cropping in Malawi suggests that farmers in that country are
particularly vulnerable to climatic or market related shocks that adversely affect the maize
sector.
Second, maize yields in Mozambique are significantly lower for all systems than yields
obtained in Zambia and Malawi. This is likely tied to the extremely low levels of inorganic

9
fertilizer application in the country. Unlike Malawi and Zambia, Mozambique does not maintain
an input subsidy programme and private agro-dealers have only recently begun to expand in
the country following the formal end to the civil war (Sitko, Burke, and Jayne, 2017). Crop
incomes for each of the systems are quite similar between Malawi and Mozambique,
suggesting than low maize productivity in Mozambique relative to Malawi is compensated by
some combinations of higher productivity in non-maize crops, higher total production due to
larger land holdings, or higher farm gate prices. Average crop income in Zambia is higher for
all cropping systems. This is likely a function of larger average land holdings, higher
productivity levels and higher farm gate prices. Zambia has witnessed an impressive
expansion in investment in maize and oilseed markets by large, often multinational firms,
which provide farmers with higher farm gate prices than traditional small-scale traders (Sitko
et al. 2017).
Finally, although the average family sizes, levels of education, and share of households that
are female headed are similar in the three countries, the heads of household in Zambia are,
on average, considerably younger. High crop incomes and greater levels of land availability in
Zambia may help to maintain young people in agriculture, rather than being pushed into urban
migration.

Table 2. Most frequent crop combinations within each cropping system

Malawi Mozambique Zambia


Groundnut (66%) Bean (89%) Groundnut (84%)
ML Pigeon pea (28%) Groundnut (47%) Bean (21%)
Soybean (16%) Pigeon Pea (16%) Soybean (12.4%)
Rice (51%) Cassava (64%) Cassava (63%)
MS Sweet potato (32%) Sorghum (33%) Sweet potato (28%)
Cassava (25%) Rice (18%) Rice (17%)
Tobacco (51%) Sesame (59%) Cotton (69%)
MC Cotton (30%) Cotton (16%) Sunflower (37%)
Sunflower (3%) Sugar Cane (4%) Tobacco (8%)
Pigeon pea-sorghum (43%) Bean-cassava (62%) Groundnut-cassava (54%)
MLS Groundnut-sorghum (18%) Groundnut-cassava (50%) Groundnut-sweet potato (38%)
Groundnut-sweet potato (17%) Bean-sorghum (25%) Bean-cassava (34%)
Groundnut-tobacco (58%) Bean-sesame (38%) Groundnuts-cotton (55%)
MLC Soybean-tobacco (13%) Groundnut-sesame (20%) Groundnuts-sunflower (53%)
Pigeon pea-nkhwani (12%) Pigeon pea-sesame (16%) Soybean-sunflower (12%)
Pigeon pea-sorghum-nkhwani Bean-cassava-sesame Groundnut-sweet potato-
(38%) (37%) sunflower (29%)
Groundnut-sorghum-nkhwani Bean-sorghum-sesame Groundnut-sweet potato-
MLSC
(17%) (27%) cotton (23%)
Groundnut-sorghum-tobacco Groundnut-cassava- Groundnut-rice-
(13%) sesame (27%) cotton (13%)
Notes: the table displays statistics on the combination of crops more frequently adopted within each cropping system divided by
country. The value in parenthesis refers to the percentage of households adopting that given combination within the categories.
Crops combinations are not mutually exclusive within each category, thus percentage values do not necessary sum to 100.
Source: Authors’ own elaboration.

10
Table 3. Summary statistics by cropping system for Malawi

Maize-cash Maize- Maize-


Maize mono- Maize- Maize-staple Maize-legume-
crops legume-cash legume-cash
cropping legume staple
crops crops-staple

(N = 1 342) (N = 1 142) (N = 174) (N = 322) (N = 282) (N = 359) (N = 113)


Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd
HH size 5.10 2.27 5.13 2.19 5.38 2.25 5.60 2.47 5.26 2.39 5.92 2.44 5.65 2.35
Age 45.72 16.61 47.3 16.33 44.3 15.13 43.31 15.1 45.22 16.55 45.83 14.47 46.35 16.17
Female-headed HH
0.26 0.44 0.27 0.44 0.21 0.41 0.15 0.36 0.30 0.46 0.16 0.37 0.26 0.44
(1=yes)
Education (years) 5.03 2.88 5.11 2.50 5.28 2.7 4.89 2.29 4.93 2.53 5.32 2.21 4.85 2.32
Share of credit
0.17 0.17 0.17 0.15 0.15 0.16 0.15 0.12 0.16 0.16 0.18 0.12 0.17 0.14
recipients
Share of FISP
0.52 0.25 0.55 0.22 0.49 0.26 0.6 0.23 0.59 0.22 0.60 0.21 0.61 0.2
Recipients
Inorganic fertilizer
0.75 0.44 0.79 0.41 0.71 0.45 0.84 0.37 0.87 0.34 0.91 0.29 0.87 0.34
(1=yes)
Maize yield
1 454 1 702 1 850 1 952 1 597 1 672 1 533 1 355 1 743 1 770 2 131 1 904 1 672 2 077
(Kg/ha)
Crop income
299 385 462 484 494 532 831 1267 472 494 1 240 1 483 641 798
(USD 2010)
Crop income
per hectare 947 4 376 757 907 1 174 3 163 1055 1 636 635 6 90 1 328 1 684 770 963
(USD 2010)
Agricultural Asset
0.11 0.96 0.30 1.02 0.48 1.03 0.40 0.95 0.35 1.03 0.73 1.09 0.45 0.98
Wealth Index
Livestock
0.51 1.31 0.8 2.07 1.72 2.72 0.83 2.02 1.01 2.01 1.67 2.95 0.91 1.71
(TLU units)
Land owned (ha) 0.5 0.43 0.70 0.48 0.68 0.47 0.78 0.57 0.78 0.55 1.02 0.66 0.78 0.57

Notes: the table reports summaries of dependent and explanatory variables for Malawi. All the currencies are expressed in real
US Dollars 2010. Full summaries are available in Table A1 in Annex 1.
Source: Authors’ own elaboration.

11
Table 4. Summary statistics by cropping system for Mozambique

Maize-cash Maize- Maize-


Maize mono- Maize- Maize-staple Maize-legume-
crops legume-cash legume-cash
cropping legume staple
crops crops-staple

(N = 286) (N = 962) (N = 466) (N = 88) (N = 2 351) (N = 283) (N = 619)


Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd
HH size 5.49 2.99 5.59 3.29 5.12 2.78 5.63 2.98 5.55 3.14 5.64 2.58 5.66 2.68
Age 45.76 16.45 45.60 16.37 44.41 15.42 41.51 13.94 45.73 15.49 42.90 14.48 43.05 14.84
Female-headed HH
0.26 0.44 0.32 0.47 0.29 0.45 0.13 0.33 0.29 0.45 0.15 0.36 0.16 0.37
(1=yes)
Education (years) 6.01 3.88 5.67 3.80 5.32 3.55 5.47 3.57 5.66 3.55 5.08 3.57 4.91 3.33
Share of credit
0.01 0.02 0.02 0.04 0.01 0.03 0.03 0.07 0.01 0.03 0.03 0.07 0.01 0.04
recipients
Share of FISP
0.04 0.20 0.04 0.20 0.02 0.15 0.23 0.42 0.03 0.17 0.28 0.45 0.10 0.30
Recipients
Inorganic fertilizer
487 570 712 803 602 669 771 642 699 831 927 804 892 818
(1=yes)
Maize yield
220 1 243 397 2 493 448 4 767 741 1 924 361 2 285 1 100 4 659 688 3 252
(Kg/ha)
Crop income
98 195 173 550 108 327 208 231 157 941 307 1 517 180 374
(USD 2010)
Crop income
per hectare 0.24 0.98 0.33 1.07 -0.09 0.75 0.49 1.15 0.10 0.89 0.70 1.28 0.05 0.83
(USD 2010)
Agricultural Asset
4.64 9.82 5.24 10.63 2.33 6.05 5.17 11.84 3.33 8.48 5.68 12.04 3.08 8.97
Wealth Index
Livestock
1.68 1.55 2.14 2.23 2.15 2.65 3.55 5.18 2.36 2.55 4.03 5.58 3.41 2.76
(TLU units)
Land owned (ha) 5.49 2.99 5.59 3.29 5.12 2.78 5.63 2.98 5.55 3.14 5.64 2.58 5.66 2.68

Notes: the table reports summaries of dependent and explanatory variables for Mozambique. All the currencies are expressed in
real US Dollars 2010. Full summaries are available in Table A2 in Annex 1.
Source: Authors’ own elaboration.

12
Table 5. Summary statistics by cropping system for Zambia

Maize-cash Maize- Maize-


Maize mono- Maize-staple Maize-legume-
Maize-legume crops legume-cash legume-cash
cropping staple
crops crops-staple

(N = 1 705) (N = 2 296) (N = 1 468) (N = 690) (N = 3 202) (N = 2 063) (N = 472)


Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd
HH size 6.28 2.84 6.75 3.12 6.53 2.76 6.27 2.77 6.91 2.78 6.87 2.96 7.39 3.19
Age 35.46 12.2 36.55 11.96 35.15 10.94 33.29 9.85 34.95 10 34.64 9.9 34.33 9.38
Female-headed HH 0.22 0.42 0.26 0.44 0.18 0.39 0.13 0.34 0.17 0.38 0.14 0.35 0.14 0.34
(1=yes)
Education (years) 6.55 4.22 6.06 4.03 6.04 3.52 5.03 3.73 6.64 3.46 5.48 3.59 6.31 3.3
Share of credit 0.13 0.17 0.16 0.2 0.05 0.09 0.43 0.27 0.07 0.1 0.44 0.26 0.29 0.26
recipients
Share of FISP 0.43 0.25 0.48 0.25 0.35 0.29 0.43 0.24 0.56 0.26 0.46 0.23 0.5 0.24
Recipients
Inorganic fertilizer 0.64 0.48 0.72 0.45 0.45 0.5 0.59 0.49 0.75 0.43 0.75 0.44 0.77 0.42
(1=yes)
Maize yield 2 096 1 586 2 309 1 646 1 769 1 494 2 086 1 586 2 595 1 647 2 482 1 698 2 638 1 602
(Kg/ha)
Crop income 1 908 5 133 2 918 11 375 2 433 3 830 2 704 4 264 3 615 4 552 3 684 4 222 4 583 4 078
(USD 2010)
Crop income 915 1514 793 1 164 1009 921 853 1 143 894 723 842 569 909 758
per hectare
(USD 2010)
Agricultural Asset 0.11 0.19 0.12 0.2 0.08 0.15 0.11 0.17 0.1 0.18 0.14 0.18 0.16 0.21
Wealth Index
Livestock 3.3 12.24 5.56 15.41 3.15 10.52 4.88 10.06 3.45 11.52 6.79 11.15 6.72 10.52
(TLU units)
Land owned (ha) 2.97 8.17 4.65 9.35 3.61 5.72 3.71 3.93 6.13 12.26 4.92 4.85 6.66 8.61

Notes: the table reports mean and standard deviation of dependent and explanatory variables for Zambia. All the currencies are
expressed in real US Dollars 2010. Full summaries are available in Table A3 in Annex 1.
Source: Authors’ own elaboration.

13
1.3 The geography of cropping systems
Figure 1 shows the spatial distribution of the dominant cropping systems in each country at
district level, where the dominant system is defined as the one occupying the greatest share of
cultivated land. Several important points come out of this map. First, the MLS system covers
the largest geographic area in Zambia and Mozambique, while maize mono-cropping and ML
systems predominate in Malawi. In Zambia, the MLS system is concentrated in the north and
north-west of the country, where cassava is an important food source, and is often grown in
combination with beans. As similar cropping system is found in the north-east of Mozambique.
In Zambia the maize mono-cropping and ML systems are concentrated around the urban
centres of Lusaka and the Copperbelt, and extend into the more drought prone regions of
southern Zambia. These are regions of the country where population densities are high and
land availability is often constrained (Sitko and Chamberlin, 2016). Thus, despite the fact that
households in these regions likely have reasonably good access to urban markets, and could
make better use of scarce land by cultivating higher value crops, many households in these
areas prioritize maize. Indeed, spatial clustering of cropping systems is evident within each
country, indicating internal spatial spillovers in adoption, which is likely associated with
variations in infrastructural development, dietary preferences and market investments.6
In eastern Zambia, the dominant system is the MLC system. In this region, cotton outgrower
arrangements are widespread, as is the cultivation of sunflower. Cross-border spill overs of this
system are observable between eastern Zambia and north-west Mozambique, but not between
Zambia and Malawi, despite having a similar ethno-linguistic population. This is likely due to a
combination of poor infrastructural connectivity in Mozambique, relative to Malawi, and stricter
enforcement of cross border movements between the Malawi and Zambia border. As a result,
farmers in north-western Mozambique are reliant on markets in eastern Zambia, while markets
in western Malawi may be more oriented to the nearby capital of Lilongwe. Infrastructural
connectivity between southern Malawi and segments of north-western Mozambique are also
likely the reason for observed prevalence of ML systems in both regions.

6The result from Moran’s I test (MI) on the share of adopters indicates the presence of significant positive spatial
correlation in adoption for Mozambique and Zambia, while it is negative and slightly less significant in Malawi (Malawi:
MI=-0.24, p=0.06; Mozambique: 0.19, p=0.00 Zambia: MI=0.18, p=0.00).

14
Figure 1. Prevalent cropping system at district level

Notes: Prevalent cropping system is the one with the highest relative share of land cultivated at the district level.
Source: Authors’ own elaboration.

15
Research design

1.4 Adoption equation


The analysis is conducted using a multinomial endogenous treatment effect, as this approach
presents several advantages relative to other approaches for this analysis. Other methodologies
such as multinomial logit, OLS with fixed effects or random effect with Mundlak correction are
only able to estimate separately either the determinants of adoption or the impact of these
systems on the output dimension (Mundlak, 1978; Di Falco and Veronesi, 2013). The
multinomial endogenous treatment effect, in contrast, allows for a simultaneous estimation of
these two components, while also accounting for endogeneity in the selection of the cropping
systems, as a result of the inclusion of one or more instrumental variables. The strategy is
divided into two inter-dependent stages. The first stage models adoption decision of a given a
farming system j using a multinomial logit, as follows:
′ ′ ′
Pr(𝑓𝑗 |𝑧𝑖 𝑙𝑖 ) = 𝑔(𝑧1𝑒 𝛼1 + 𝛿1 𝑙𝑖1 , 𝑧𝑒2 𝛼2 + 𝛿2 𝑙𝑖2 , … , 𝑧𝑒𝑖 𝛼𝑗 + 𝛿𝑗 𝑙𝑖𝑗 ) (1)

Where 𝑓𝑗 denotes the variable on the type of farming system and where 𝑗 = 0 refers to the
benchmark category of maize mono-cropping adopters. In addition, 𝒁𝑖𝑒𝑡 is a matrix of control
variables, 𝛼𝑗𝑡 their linked coefficients and 𝑙𝑖 = (𝑙𝑖1 , 𝑙𝑖2 … 𝑙𝑖𝐽 ) a vector of latent variables.7 Deb and
Trivedi (2006) suggests using instrumental variables to obtain a more robust identification. The
instrument involves at least an explanatory variable entering exclusively in the adoption
equation. The matrix of explanatory variables involves four instrumental variables in the first
stage: 1) drought risk exposure; 2) flood risk exposure; 3) informal and 4) formal diffusion of
agricultural practices. For the validity of the first two instruments, the hypothesis is that farmers
are more likely to adopt a cropping system different from maize mono-cropping when residing
in areas that have been more historically exposed to weathers shocks. Being a subjective belief
of the household, the postulate is that this will not affect the yield or crop income volatility through
other channels. Flood and drought risk exposures are computed for household i residing in the
area e as the probabilities of being exposed to these events conditional on area e historical
weather distribution. A drought shock occurrs if the Rainfall Standard Precipitation Index (SPI)
in a given area and agricultural season takes a value lower than -2 (Guttman, 1998; McKee,
Doesken and Kleist, 1993). A flood shock occurs when the SPI index takes value above 2.
Following Scognamillo, Asfaw and Ignaciuk (forthcoming) and generalizing for a shock S
occurring in a given area e in the year t, the risk exposure index takes the following form:
∑𝑡−2
1 𝑆𝑒 𝑡−2
𝑃𝑖𝑒𝑡 = (2)
(𝑡−2)−1983

Where the probability of experiencing a shock is computed as the ratio between the total
numbers of seasons in which the shock occurred, divided by the total number of seasons
available in our weather source. Drought and flood indicators are lagged by two years to exclude
any recent shocks likely affecting household’s adoption of a cropping system through the
production channel. Following the work from Manda et al. (2016), the remaining two instruments
are proxies of information availability at village/district level, under the assumption that
agricultural information can be transmitted through institutional channels, such as extension
services and informal institutions, such as family and friends network. We compute the average
access to these two sources of information at a higher geographical level than the one observed

7 These are not observable variables but are inferred from other observed variables in of the model.

16
in the data, hypothesizing that the average access to these sources is not correlated to the
household’s outcome. In addition to these exogenous covariates, the matrix of control
variables 𝒁𝑖𝑒𝑡 includes the determinants introduced in section 3.1.

1.5 Outcome equation


To estimate the impact of adoption on maize yield and crop income volatility, the second stage
of the multinomial treatment effect model consists in an OLS with the selectivity correction, as
follows:
′ 𝑗 𝑗
𝐸(𝑦𝑖𝑡 |𝑑𝑖𝑡 𝑥𝑖𝑡 𝑙𝑖𝑡 ) = 𝑥𝑖𝑡 𝛽 + ∑𝑗=1 𝛾𝑗𝑡 𝑓𝑖𝑗𝑡 + ∑𝑗=1 𝜆𝑗𝑡 𝑙𝑖𝑗𝑡 (3)

Where the expected outcome depends on a set of observable determinants 𝑥𝑖𝑡 with linked
parameters 𝛽, on the adoption of a given farm system 𝑓𝑖𝑗𝑡 and associated coefficients 𝛾𝑗𝑡 . In
addition to the set of variables in the first-stage, the matrix of determinants 𝑥𝑖𝑡 comprises the
seasonal positive and negative SPI indexes, capturing the effect of seasonal rainfall distribution
on the productivity and crop income volatility.8 To investigate the effect on maize productivity
we use the natural log of maize yield. To address the effect of farm systems on crop income
volatility, we adopt a two steps approach following the extant literature (Antle, 1983; Di Falco
and Chavas, 2009). First, the approach predicts the error term from a regression with the crop
gross income as dependent variable. Secondly, it involves the addition of the square of the error
term as new dependent variable in the main empirical specification, keeping fixed the controls.

8 The technical note in the appendix introduces more details about the empirical approach.

17
Results and discussion

1.6 Determinants of cropping system adoption: policy lessons from


cross country comparison
What factors explain the adoption of different cropping systems, and what are the commonalities
and difference between countries? Three policy relevant commonalities are apparent in the
three countries (see Tables 6, Table 7 and Table 8). First, private sector output market access
is an important driver of diversification. In Zambia, for example, the number of private
agricultural traders in a village has a significant and positive effect on the adoption of all cropping
systems, apart from the MS system. The MS system is primary a subsistence orient system. As
competition from private traders in the village increases, farmers move away from maize mono-
cropping and the MS system to more commercially oriented and diverse systems. In Malawi,
output market access is measured in terms of distance to a weekly market. Results show that,
as the distance to these markets decreases, farmers are significantly more likely to adopt MC
and MSL systems. Similarly, in Mozambique, proximity to urban markets positively drives
adoption of more diverse cropping systems. In particular, as the distance to urban markets
increases, farmers are less likely to adopt MS, MLS and MLCS systems.
Second, proximity to parastatal marketing boards, which operate in Zambia and Malawi and
predominantly provide output market subsidies for maize, discourages the adoption of more
diverse cropping systems. For example, in Zambia, as the distance from a Food Reserve
Agency depot increases, the probability of adopting the ML, MS, MLS, MLC and MLCS
increases significantly, all else equal. Similarly, in Malawi as the distance from an ADMARC
depot increases, the probability of adopting the three most diverse systems, MLS, MLC and
MLCS, all increases significantly relative to maize mono-cropping.
Finally, in all countries and for all systems, land size is a key determinant of adopting more
diverse cropping systems relative to maize mono-cropping. For example in Zambia, all else
equal, a 10 percent increase in average land size, roughly 0.5 hectares, correlates with an
increase of 29 percent in the probability of adopting the MLSC system.
Taken together several key policy insights emerge. First, policies that crowd in private sector
investments or improve the access of farmers to private markets are critical for achieving crop
diversification policy objectives. This includes adopting policies that improve the predictability of
government action in agricultural output markets, including in terms of cross border trade and
the release of government held grain stocks onto the market. A lack of predictability related to
government action in output markets and the resultant effects on prices, is identified as a key
constraint to increased private investments in output markets (Jayne et al. 2014). Policies that
affect the price and liquidity of bank lending rates are also important for driving investment in
agricultural markets. Finally, public investments in market infrastructure, including roads and
electrification, may help to improve investments in output markets and farmers access to them.
Second, public investments in supporting maize output prices through direct purchases from
farmers must be reconsidered if governments are committed to promoting more diverse
production systems. National food reserve agencies could, for example, use first refusal
contracts with private sector buyers or call options on the Johannesburg commodity exchange
to help manage retail maize price volatility, without taking physical control of large stocks of
maize (Devereux, 2007; Dana et al., 2006).

18
Finally, land policy reform must be part of any strategy to promote crop diversification. There is
emerging evidence from Eastern and Southern Africa that land policies are encouraging
significant urban investment in relatively large-scale land acquisitions (Sitko and Jayne, 2014;
Jayne et al., 2016; Ansueeuw, 2016). These acquisitions may be contributing to increased land
scarcity in smallholder production areas, leading to more rapid land fragmentation than would
be the case otherwise, and increased challenge for small-scale producers to expand their
landholdings. The strong and significant effects of land size on the adoption of more diverse
cropping systems in all countries suggest that policies supporting land access and protect the
land rights of existing smallholder farmers are essential for achieving government objectives
related to crop diversification.

19
Table 6. Drivers of cropping systems adoption in Malawi

Maize- Maize- Maize- Maize-


Maize- Maize-
cash legume- legume- legume-cash
legume staple
crops staple cash crops crops-staple
(1=yes) (1=yes)
(1=yes) (1=yes) (1=yes) (1=yes)
Institutional Maize Seeds Price
0.342** -0.150 -1.071*** -0.902*** 0.302 -1.058**
(EA's level)
(0.166) (0.331) (0.259) (0.259) (0.241) (0.474)
Distance from
Weekly Market 0.005 0.071 -0.117* -0.157** -0.042 -0.023
(EA's level)
(0.047) (0.085) (0.067) (0.072) (0.068) (0.102)
Distance from FRA
-0.115 0.066 0.209 0.269** 0.344*** 0.254*
(ln, EA's level)
(0.088) (0.182) (0.139) (0.132) (0.128) (0.154)
HH Agricultural Asset
0.124** 0.089 0.146* 0.124 0.256*** 0.192
resource Wealth Index
(0.059) (0.105) (0.084) (0.098) (0.080) (0.124)
Livestock
0.049 0.180*** 0.051 0.119*** 0.151*** 0.077
(TLU units)
(0.041) (0.041) (0.055) (0.044) (0.042) (0.058)
Land owned
1.627*** 1.416*** 2.218*** 2.735*** 3.368*** 2.324***
(ln, ha)
(0.206) (0.365) (0.319) (0.357) (0.298) (0.468)
Biophysical SPI Negative (lag) 0.450 0.379 -0.023 1.238** -0.121 -0.569
(0.320) (0.595) (0.482) (0.507) (0.472) (0.703)
SPI Positive (lag) 0.170 0.578** 0.504** -0.634** 0.530** 0.202
(0.192) (0.271) (0.257) (0.320) (0.263) (0.393)
Instruments Flood Probability
0.051** -0.022 0.004 -0.039 0.114*** 0.103**
Index (%)
(0.024) (0.051) (0.031) (0.045) (0.033) (0.051)
Drought
Probability Index 0.028 0.020 -0.028 0.068** 0.012 0.076
(%)
(0.022) (0.041) (0.030) (0.034) (0.030) (0.050)
Formal Diffusion
0.303 0.953** 0.169 0.645* 0.597* 0.443
(EA's level)
(0.238) (0.451) (0.337) (0.381) (0.352) (0.559)
Informal Diffusion
0.565 -2.129** -1.810* 0.905 0.652 2.804***
(EA's level)
(0.479) (0.992) (0.977) (0.705) (0.701) (0.987)
Other controls Yes Yes Yes Yes Yes Yes
Agro-ecological
Yes Yes Yes Yes Yes Yes
dummies
Year dummy Yes Yes Yes Yes Yes Yes
Observations 3 977 3 977 3 977 3 977 3 977 3 977
Notes: the table displays the determinants of adoption of different farm systems, compared to the baseline scenario of maize mono-
cropping for the panel pooled sample (2010 and 2013). Full list of controls and coefficients are available in Table A4 in Annex 1.
The specification is the first stage of a multinomial treatment effect model (multinomial logit). The specification controls also for
household-size (ln), age (ln), dummy for female-headed households, years of education of the household's head (ln), asset wealth
index, livestock owned (as TLU units), hectares of land owned (ln), share of credit and FISP recipients at EA level, use of inorganic
fertilizer, mean seasonal maximum temperature, positive and negative SPI indexes for both seasonal and lagged, agro-ecological
and year dummy. Standard errors are clustered at EA's level. Significant levels are * p<0.10, ** p<0.05, *** p<0.001.
Source: Authors’ own elaboration.

20
Table 7. Drivers of cropping systems adoption in Mozambique

Maize- Maize- Maize- Maize-


Maize- Maize-
cash legume- legume- legume-cash
legume staple
crops staple cash crops crops-staple
(1=yes) (1=yes)
(1=yes) (1=yes) (1=yes) (1=yes)
Distance from the
-0.019 -0.159** -0.244** -0.143** -0.215** -0.342***
Institutional main market cities
(0.075) (0.074) (0.099) (0.070) (0.088) (0.082)
Agricultural Asset
0.054 -0.493*** 0.036 -0.329*** 0.184 -0.582***
Wealth Index
(0.092) (0.114) (0.167) (0.093) (0.142) (0.126)
Livestock
0.003 -0.016 -0.013 -0.017** -0.023** -0.030**
Resource (TLU units)
(0.007) (0.010) (0.014) (0.008) (0.011) (0.012)
Land owned
0.114 0.395*** 0.807*** 0.462*** 1.019*** 1.473***
(ln, ha)
(0.079) (0.104) (0.189) (0.079) (0.124) (0.130)
SPI Negative (lag) -1.049 1.723** 2.127* 0.645 0.698 1.002
(0.682) (0.843) (1.173) (0.740) (0.867) (0.872)
Biophysical
SPI Positive (lag) -0.243 -0.228 0.111 -0.361 -0.395 0.349
(0.233) (0.257) (0.348) (0.251) (0.330) (0.272)
Flood Probability
0.061 -0.003 -0.066 0.091 0.017 0.043
Index (%)
(0.055) (0.061) (0.083) (0.058) (0.074) (0.064)
Drought
Probability Index 0.035 -0.048 0.065 -0.050 0.061 0.034
(%)

Instruments (0.034) (0.039) (0.065) (0.035) (0.055) (0.044)


Formal Diffusion
0.193 2.438 5.158** 0.225 5.251** 3.646*
(EA's level)
(2.016) (2.253) (2.273) (2.046) (2.121) (2.216)
Informal Diffusion
2.693*** 0.538 2.819*** 2.565*** 2.964*** 2.907***
(EA's level)
(0.505) (0.487) (0.861) (0.474) (0.769) (0.629)
Other controls Yes Yes Yes Yes Yes Yes
Regional dummies Yes Yes Yes Yes Yes Yes
Observations 5 055 5 055 5 055 5 055 5 055 5 055
Notes: the table displays the determinants of adoption of different farm systems, compared to the baseline scenario of maize mono-
cropping for the Mozambique’s sample. The specification is the first stage of a multinomial treatment effect model (multinomial
logit). Full list of controls and coefficients are available in Table A5 in Annex 1.
The specification controls also for household-size (ln), age (ln), dummy for female-headed households, years of education of the
household's head (ln), asset wealth index, livestock owned (as TLU units), hectares of land owned (ln), use of inorganic fertilizer,
mean seasonal maximum temperature, positive and negative SPI indexes for both seasonal and lagged, regional dummies.
Standard errors are clustered at EA's level. Significant levels are * p<0.10, ** p<0.05, *** p<0.001.
Source: Authors’ own elaboration.

21
Table 8. Drivers of cropping systems adoption in Zambia

Maize- Maize- Maize- Maize-


Maize- Maize-
cash legume- legume- legume-cash
legume staple
crops staple cash crops crops-staple
(1=yes) (1=yes)
(1=yes) (1=yes) (1=yes) (1=yes)
Institutional Maize Seeds
0.030 -1.394*** 0.161 -1.082*** 0.117 -1.028**
Price (EA's level)
(0.176) (0.240) (0.258) (0.232) (0.227) (0.411)
Total Traders
0.346*** -0.266** 0.242* 0.423*** 0.510*** 0.314**
(ln, EA's level)
(0.075) (0.115) (0.125) (0.101) (0.106) (0.136)
Distance
from FRA 0.195** 0.551*** 0.105 0.650*** 0.200* 0.379**
(ln, EA's level)
(0.078) (0.119) (0.122) (0.109) (0.104) (0.193)
HH resources Agricultural
Asset Wealth -0.493* -0.671* -0.057 -0.962*** -0.458 -0.274
Index
(0.263) (0.343) (0.334) (0.263) (0.305) (0.370)
Livestock
0.007 -0.005 -0.000 -0.001 0.001 -0.001
(TLU units)
(0.005) (0.006) (0.006) (0.004) (0.006) (0.006)
Land owned
0.614*** 0.569*** 0.860*** 1.155*** 1.472*** 1.513***
(ln, ha)
(0.060) (0.071) (0.086) (0.069) (0.080) (0.097)
SPI Negative
-0.617** 0.470 -2.017*** 0.767** -3.560*** -1.320***
(lag)
Biophysical (0.246) (0.381) (0.364) (0.340) (0.331) (0.440)
SPI Positive (lag) 0.051 -0.285* -0.139 -0.546*** 0.954*** -0.532*
(0.129) (0.172) (0.243) (0.158) (0.189) (0.274)
Instruments Flood Probability
0.068* -0.007 -0.035 0.054 0.024 0.124**
Index (%)
(0.036) (0.048) (0.063) (0.047) (0.052) (0.062)
Drought
Probability Index -0.023 -0.037 0.048 -0.083** -0.011 -0.023
(%)
(0.029) (0.040) (0.062) (0.035) (0.044) (0.056)
Formal Diffusion
0.624** 0.025 0.114 1.550*** -0.314 0.830
(EA's level)
(0.306) (0.438) (0.515) (0.384) (0.438) (0.579)
Informal
Diffusion (EA's 0.001 -1.406*** -0.673 -1.589*** -0.358 -1.206**
level)
(0.308) (0.417) (0.509) (0.369) (0.402) (0.550)
Other Controls Yes Yes Yes Yes Yes Yes
Agro-ecological
Yes Yes Yes Yes Yes Yes
dummies
Year dummy Yes Yes Yes Yes Yes Yes
Observations 11 496 11 496 11 496 11 496 11 496 11 496
Notes: the table displays the determinants of adoption of different farm systems, compared to the baseline scenario of maize mono-
cropping for the panel pooled sample (2011 and 2014). Full list of controls and coefficients are available in table A6 in Annex 1. The
specification is the first stage of a multinomial treatment effect model (multinomial logit). The model controls also for the total number
of traders (ln), the distance from FRA (ln), household-size (ln), age (ln), dummy for female-headed households, years of education
of the household's head (ln), asset wealth index, livestock owned (as TLU units), hectares of land owned (ln), share of credit and
FISP recipients at EA level, use of inorganic fertilizer, mean seasonal maximum temperature, positive and negative SPI indexes for
both seasonal and lagged, agro-ecological and year dummies (only for the pooled sample). Standard errors are clustered at EA's
level. Significant levels are * p<0.10, ** p<0.05, *** p<0.001.
Source: Authors’ own elaboration.

22
1.7 Country specific insight of cropping systems adoption

Malawi
Maize seed prices, which represent the main target of the government’s input subsidy
programme have significant, yet complex effects on cropping system adoption. An increase in
maize seed price has a positive and significant effect on adopting the ML system. This suggests
that for many maize producing households, legumes act as a substitute for maize, either
because they can be consumed directly, or because they can be cultivated and sold in more
competitive markets to acquire maize. Thus, input subsidy programmes that push down maize
seed prices may inadvertently discourage farmers from adopting an agronomically more
beneficial ML system. Conversely, as maize seed prices increase, farmers are less likely to
adopt MC, MLS, or MLCS systems.
The effect of agricultural asset wealth on cropping system adoption is largest and most
significant for the MLC system. This system is dominated by tobacco-groundnut and tobacco-
soybean production (Table 2), which require significant management and labour allocations.
Drought shocks, measured in terms of negative lagged SPI, only have a significant and positive
effect the adoption of the MLS. This is a system dominated by pigeon pea and sorghum (Table
2), both of which withstand droughts reasonably well. This suggests that as droughts increase,
farmers are receptive to moving away from maize mono-cropping and adopting cropping
systems that lower their risk of crop failure.9 Lagged flood shocks induce adoption of MS, MC
and MLC, while surprisingly reduce the probability of MLS adoption. In terms of long-term
weather distribution, results on flood probability index suggest that an increase by 1 percent in
the flood index raises likelihood of diversification by pushing farmers into ML (+0.5 percent),
MLC and MLCS (+0.11 percent) systems. Formal institutions, such as public extension systems,
have a moderate positive effect on the adoption of some systems (MS and MLS), while large
informal networks may discourage some diversification pathways (MS and MC) and encourage
others (MLCS).

Mozambique
In Mozambique, the dominant alternative staple is cassava, which is more commercialized than
in other countries of the region. As a result, farm households endowed with more land and less
agricultural wealth are pulled into the production of MS, MLS, and MLCS, with likely beneficial
effects in terms of productivity and crop income stability (See Figure 2b and Table 9). Low
lagged drought shocks are significant determinants, both for their statistical power and
magnitude, of complex farming systems adoption, such as MS, MC, MLC and MLCS. Despite
this, the variables on long-term climatic risk do not correlate significantly with adoption on any
cropping system.
The access to information measured in terms of formal and informal diffusion networks, such
as fellow farmers and extension agencies, is a significant driver of most forms of cropping
system diversification. Increasing access to agricultural advice from public or private extension
providers increases the likelihood of MS, MC, MLS, MLC and MLCS adoption. MLS coefficient
presents the highest magnitude, with 41 percent probability of adoption when the share variable

9This is consistent with what found on the impact on of drought shocks on consumption in Malawi (Asfaw and Maggio,
2018).

23
increases by 1 percent. Increasing investment in public extension provision, combined with
policies that crowd in private extension providers appears to be a key lever for driving
diversification in Mozambique.

Zambia
Like in Malawi, maize seed prices are important drivers of diversification. In particular, as maize
seed prices increase, farmers are less likely to adopt alternative staple systems (MS, MLS and
MLCS). This finding is surprising and suggests that prioritize maize over other staples. Thus, as
input budgets are stretched by an increase in maize prices, smallholders will invest resources
in acquiring maize seeds before investing in the production of alternative staple crops.
The effect of agricultural asset wealth on cropping system adoption suggests that poorer
households are more likely to adopt more subsistence oriented systems, such as ML, MS and
MLS, while increased assets do not have a positive effect on the adoption of any system.
The lagged effects of drought are important determinants of adopting the MLS system, which is
Zambia is dominated by a cassava-groundnut mix and negatively effects the adoption of
systems containing cash crops, as well as the ML system. Drought risk likely discourages
households from assuming the risk of cash crop production, which often entails entering into an
input credit arrangement with agro-business firms. Finally, formal information networks strongly
and positively influence the adoption of the MLS system, and marginally increases the adoption
of the ML system. This is a positive finding and suggests that improving formal extension
networks can drive beneficial changes in farm behaviour. Interestingly, large informal networks,
which include large extended kinship arrangements, discourage most forms of diversification.
This is perhaps due to the social and cultural function maize serves in terms of building social
bonds through gifts of maize or in-kind piecework to needy relatives (Sitko, 2012).

1.8 Impact on productivity


The choice of the cropping system influences significantly the level of maize productivity. In all
three countries, more diverse systems (three crop categories or more) have positive and
significant effects on maize yields (Figure 2a, Figure 2b and Figure 2c).10 These more diverse
systems likely enable farmers to capture a combination of economic and agronomic benefits
from a range of crops, which they translate into improved yields for maize. For example,
although the dominant crops in the MLS system vary between countries (Table 2), adoption of
this system has similar positive effects in all three countries: +30.85 percent increase in maize
yield relative to mono-cropping in Zambia, +30.55 percent increase in Malawi and +20.28
percent increase in Mozambique. Similarities in the magnitude of the impact for MLS, makes
this system suitable for a wide range of agro-ecological and geographic zones. For the MLC
system, the effects of adoption are also positive and significant for all three countries. However,
there is more variability in terms of the magnitude of the effect and confidence interval range
between the countries than for MLS. This variability suggests that these systems are more
regionally variable in their positive effects than the MLS system.
Compared to maize mono-cropping, ML cropping systems are found to have positive and
significant effects on maize yield in Malawi, but no significant impact on maize productivity in
Mozambique and Zambia. This may be a function of differences between countries in terms of

10Figure 2a, Figure 2b and Figure 2c show the point estimates of the multinomial treatment effect, the 10 percent
confidence interval, and the percentage change in maize yield as a result of adoption of a system relative to maize
mono-cropping. The only exception is MLSC in Malawi.

24
the type of legumes grown. In Malawi, for example, pigeon-peas are widely grown, while they
are virtually absent in Zambia. Relative to other grain legumes, pigeon pea is found to have
significant long-term positive impacts on soil fertility (Kerr et al., 2007).
Consistent with our hypotheses, the MS system has neutral to negative effects on maize yield
in all the three countries. Alternative staples do not offer any obvious benefits in terms of
agronomic impact on maize yields, but may divert some labour away from maize cultivation.
The MC system has a neutral effect on maize yields in Mozambique and Zambia. Thus, in less
diverse, two crop settings, income generated from cash crop production is not translating into
increased maize intensification. From a productivity standpoint, adoption of these systems does
not offer a clear path to improvements in maize yields.11

Figure 2 Effect of different cropping systems on maize yield compared to maize


mono-cropping in Malawi, Mozambique and Zambia

A: Malawi

11 As a further robustness test, we run the main specification considering labour productivity in agriculture as
dependent variable, defined as the natural logarithm of crop income per adult worker. As Table A10 in appendix
reports, diversification through cropping systems based on three and four crop has a positive and significant effect
in the three countries.

25
B: Mozambique

C: Zambia

Notes: The figure displays the second-stage coefficients of the multinomial treatment effect. The blue dot represents the point
estimate while the blue bars report 95 percent confidence interval. Dots’ labels report the conversion of the log coefficient x into
percentage change as x=exp(x)-1.
Source: Authors’ own elaboration.

26
1.9 Impact on crop income volatility
The level of crop income volatility provides insights into how adoption of a particular system
effects livelihood resilience and risk, relative to maize mono-cropping. Table 9 shows that there
is no single system that decreases crop income volatility significantly in all three countries. This
suggests that there is no one size fits all policy option for increasing resilience through promotion
of a particular cropping system.
In Malawi, only the ML system significantly lowers income volatility relative to maize mono-
cropping, although the coefficients are negative for all systems. The competitiveness and
stability of legume markets in Malawi likely explains the positive effect of the ML system, while
volatility in cash crop and alternative staple markets may be comparable to the volatility of maize
markets.
In Mozambique, the three crop combinations of MLS and MLC significantly decrease crop
income volatility relative to maize mono-cropping. Prominent crops in these systems, namely
cassava and sesame, respectively, are both drought tolerant and have vibrant commercial
markets. The combination of these attributes likely helps to lower income volatility relative to
mono-cropping.
In Zambia, all forms of diversification decrease crop income volatility relative to maize mono-
cropping. Interestingly, the magnitude of the effect is higher for more diverse systems. Indeed,
in Zambia farmers that adopt the MLCS system are predicted to have on average a 43 percent
lower volatility in crop income than a maize mono-cropping household.

27
Table 9. Effect of farming systems on crop income volatility

Dependent: Crop income volatility (u2)


Malawi Mozambique Zambia (Wave 1+2)
Maize-legume (1=yes) -1.477*** -0.122 -0.251***
(0.450) (0.139) (0.037)
Maize-staple (1=yes) 0.220 -0.011 -0.187***
(0.814) (0.163) (0.051)
Maize-cash crops (1=yes) -0.542 0.092 -0.273***
(0.691) (0.326) (0.052)
Maize-legume-staple (1=yes) -0.196 -0.255* -0.363***
(0.646) (0.132) (0.039)
Maize-legume-cash crops (1=yes) -1.110 -0.414*** -0.392***
(0.683) (0.154) (0.043)
Maize-legume-cash crops-staple (1=yes) -0.914 -0.168 -0.430***
(0.779) (0.139) (0.046)
Other controls Yes Yes Yes
Agro-ecological/Regional dummies Yes Yes Yes
Number of observations 3 732 5 395 11 884
Log-likelihood -19 078.90 -18 788.02 -31 588.78
Chi2 / F / Wald 814.937 773.119 6 686.129
Notes: the table displays the determinants of crop income volatility, compared to the baseline scenario of maize mono-cropping for
Malawi (column 1), Mozambique (column 2) and Zambia (column 3). The specification is the second stage of a multinomial treatment
effect model (MTM) where the dependent is the squared residual from a MTM including crop income as dependent variable. The
baseline category is maize mono-cropping. The instrumental variables in the first stage are the flood and drought shock probability
and the average diffusion of formal and informal institutions at EA's level. For the entire set of explanatory variables refer to table
A8 in Annex 1. Significant levels are * p<0.10, ** p<0.05, *** p<0.001 and errors are clustered at EA's level.
Source: Authors’ own elaboration.

28
Conclusions
Crop diversification is a common agricultural policy objective in eastern and southern Africa.
Yet, the type of diversification targeted and the policy measures utilized for its achievement are
rarely well-defined. This study shows that not all diversification pathways are the same, both in
terms of their effects on farmers’ welfare and resilience, and the factors that influence their
adoption.
From the analysis, several stylized facts emerge regarding cropping system diversification.
First, diversification through a legume-based approach is typically more beneficial than a cash-
crop based diversification pathways. As this study shows, maize-cash crop systems do not
positively affect productivity and crop income volatility in most cases, but when combined in with
legumes they do. Conversely, legume-based systems, either in combination with maize or in a
maize-legume-staple system produce beneficial income and productivity effects in most cases.
The heterogeneity of these results is determined by a combination of market and agronomic
factors affecting these crops and their combination. Indeed, crops such as legumes fix
atmospheric nitrogen, which is often a limiting nutrient for cereal crops, and are characterized
by more established output market, especially in the case of Malawi. Cash crops, by contrast,
can be more profitable for farmers but they are effected by considerable market volatility due to
international price movements and exchange rate fluctuations. Moreover, they do not provide
complementary agronomic benefits to cereals. In absence of proper investments and policies,
diversification into these crops can expose households to considerable risk and lead to a
potential deterioration of household’s welfare.
This finding suggests that legumes should be prioritized in crop diversification strategies in these
countries. The specific policy support required to achieve this objective will vary by country. In
general terms, development of the legume sector can be supported through improvements in
legume seeds availability, improved regional trade conditions, and improvements in local
processing capacity. In terms of legume seeds, forecasting smallholders’ demand is a serious
challenge for seed companies. This leads to a chronic under provision of legume seeds on the
market. To improve this condition, financial option, such as a first-loss guarantee mechanism
for specific quantities of legume seeds, can be considered. This mechanism allows seed
companies to forecast to maximum risk exposure they have when multiply seeds. Second, input
subsidy programmes dedicated to legume production can help firms to forecast future demand.
In terms of improved trade conditions, a major obstacle in the legume sector is effective
compliance and monitoring of sanitary and phytosanitary requirements to export, particularly
aflatoxins. Investment to improve compliance and monitoring can help to open new export
markets for legumes. Finally, concessional loans or other financial arrangements to support
local investment in legume processing should be considered to help improve formal market
conditions for legumes.
The second major finding, which lends additional support to the recommendations above, is that
policies that crowd in the private sector, such as by improving the predictability of government
actions in agricultural output markets, or improve market access conditions are important drivers
of diversification. Conversely, policies that exclusively support maize output markets, such as
buying maize from farmers through parastatal marketing boards, hinders diversification.
Reforms to parastatal marketing boards are particularly important, as these boards often absorb
significant amounts of agricultural sector budgets, and therefore come at opportunity costs to
other investment areas. Reforms to marketing boards must be carried out in ways that allow
policy makers to achieve national food security and staple food price stabilization objectives.
Reform options include using private sector buyers to procure and store national grain reserves,

29
where the government has the right of first refusal on these stocks. Alternatively, governments
can use call options on local or regional commodity exchanges, which will only be exercised at
predefined price points. In both cases, the outcome is greater predictability of government
actions in food markets, less public expenditure on grain purchases, and greater private
investment in markets (Jayne, Zulu and Nijhoff, 2006).
Finally, land policy, which is both complex and politically sensitive, cannot be ignored when
developing diversification strategies. Many land policies in the region favour consolidation and
appropriation of land by well-placed elites and may be hastening processes of land
fragmentation and land scarcity for smallholders (Jayne et al., 2014). Where land constraints
are severe and land sizes are small, diversification is difficult.
Taken together, this study suggests that crop diversification must be promoted in a holistic
fashion and should leverage private sector investment interest in output and input markets as
much as possible.

30
References
Anseeuw, W. 2006. Can We Speak about Effective Land Reform in South Africa?. IFAS
Working Paper Series/Les Cahiers de l'IFAS, no. 8.
Arslan, A., Cavatassi, R., Alfani, F., Mccarthy, N., Lipper, L. & Kokwe, M. 2018.
Diversification under climate variability as part of a CSA strategy in rural Zambia. The Journal
of Development Studies, 54(3): 457–480.
Asfaw, S. & Maggio, G. 2018. Gender, weather shocks and welfare: Evidence from Malawi.
The Journal of Development Studies, 54(2): 271–91.
Agrawala, S., Moehner, A., Hemp, A., Van Aalst, M.V., Hitz, S., Smith, J., Meena, H. et al.
2003. Development and climate change in Tanzania: focus on Mount Kilimanjaro. Paper
prepared for the “Working Party on Global and Structural Policies, Working Party on
Development Co-operation and Environment“ (unpublished).
Anley, Y, Bogale, A. & Haile-Gabriel, A. 2007. Adoption decision and use intensity of soil and
water conservation measures by smallholder subsistence farmers in Dedo district, Western
Ethiopia. Land Degradation and Development, 18(3): 289–302.
Antle, J.M. 1983. Testing Approach the Stochastic Structure of Production : Flexible. Journal
Business & Economic Statistics, 1 (3): 192–201.
Balk, D.L., Deichmann, U., Yetman, G., Pozzi, F., Hay, S.I. & Nelson, A. 2006. Determining
Global Population Distribution: Methods, Applications and Data. Advances in Parasitology, 62:
119–156.
Barrett, C. & Reardon, T. 2000. Asset, activity, and income diversification among African
agriculturalists: Some practical issues (unpublished).
Barrett, C.B., Reardon, T. & Webb, P. 2001. Nonfarm income diversification and household
livelihood strategies in rural Africa: concepts, dynamics, and policy implications. Food
policy, 26(4): 315–331.
Becker, M. & Johnson, D. 1999. The role of legume fallows in intensified upland rice-based
systems of West Africa. Resource Management in Rice Systems: Nutrients, 76(2–3): 105–20.
Cecil, D.J., Buechler, D. & Blakeslee, R. 2014. Gridded Lightning Climatology Data Collection.
In: NASA Global Hydrology Resource Center DAAC [online]. Huntsville, USA. [Cited 2 February
2017]. http://dx.doi.org/10.5067/LIS/LIS-OTD/DATA311
Chapoto, A., Haggblade, S., Hichaambwa, M., Kabwe, S., Longabaugh, S., Sitko, N. &
Tschirley, D. 2013. Institutional models for accelerating agricultural commercialization:
Evidence from maize, cotton and horticulture in Zambia, 1965 to 2012. In Hillbom E and
Svensson P, eds. Agricultural Transformation in a Global History Perspective. London,
Routledge Press.
Dana, J., Gilbert, C.L. & Shim, E. 2006. Hedging grain price risk in the SADC: Case studies
of Malawi and Zambia. Food Policy, 31(4): 357–371.
Dapaah, H. K., Asafu-Agyei, J. N., Ennin, S. A. & Yamoah, C. 2003. Yield stability of cassava,
maize, soya bean and cowpea intercrops. The Journal of Agricultural Science, 140(1): 73–82.
Davis, B., Winters, P., Carletto, G., Covarrubias, K., Quiñones, E.J., Zezza, A., Stamoulis,
K., Azzarri, C. & DiGiuseppe, S. 2010. A cross-country comparison of rural income generating
activities. World development, 38(1): 48–63.
Deb, Partha & Trivedi, K.P. 2006. Maximum Simulated Likelihood Estimation of a Negative
Binomial Regression Model with Multinomial Endogenous Treatment. The Stata Journal, 6(2):
246–55.

31
Denning, G., Kabambe, P., Sanchez, P., Malik, A., Flor, R., Harawa, R., Nkhoma, P. et al.
2009. Input subsidies to improve smallholder maize productivity in Malawi: Toward an African
Green Revolution. PLoS biology, 7(1): 2–10.
Dercon, S. 1996. Risk, crop choice, and savings: Evidence from Tanzania. Economic
development and cultural change, 44(3): 485–513.
Deressa, T.T., Hassan, R. M., Ringler, C., Alemu, T. & Yesuf, M. 2009. Determinants of
farmers’ choice of adaptation methods to climate change in the Nile Basin of Ethiopia. Global
environmental change, 19(2): 248–255.
Devereux, S. 2007. The impact of droughts and floods on food security and policy options to
alleviate negative effects. Agricultural Economics, 37(1): 47–58
Di Falco, S. & Chavas J.P. 2009. On Crop Biodiversity, Risk Exposure, and Food Security in
the Highlands of Ethiopia. American Journal of Agricultural Economics, 91(3): 599–611.
Di Falco, S. & Veronesi, M. 2013. How Can African Agriculture Adapt to Climate Change? A
Counterfactual Analysis from Ethiopia. Land Economics, 89(4): 743–66.
Dolisca, F., Carter, R.D., McDaniel, J.M., Shannon, D.A. & Jolly, C.M. 2006. Factors
influencing farmers’ participation in forestry management programs: A case study from Haiti.
Forest Ecology and Management, 236(2): 324–31.
Fafchamps, M. 1992. Cash crop production, food price volatility, and rural market integration
in the third world. American journal of agricultural economics, 74(1): 90–99.
Feder, Gershon. 1982. Adoption of Interrelated Agricultural Innovations : Complementarity and
the Impacts of Risk, Scale and Credit. American Journal of Agricultural Economics, 64(1): 94–
101.
Garrity, D.P., Akinnifesi, F.K., Ajayi, O.C., Weldesemayat, S.G., Mowo, J.G., Kalinganire,
A., Larwanou, M. & Bayala, J., 2010. Evergreen agriculture: a robust approach to sustainable
food security in Africa. Food security, 2(3): 1–18
Govereh, J. & Jayne, T.S., 2003. Cash cropping and food crop productivity: synergies or trade‐
offs? Agricultural economics, 28(1): 39-50.
Guttman, N.B. 1998. Comparing the Palmer drought index and the standardized precipitation
index. Journal of the American Water Resources Association, 34(1): 113–121.
Haggblade, S., Plerhoples, C. & Kabwe, S. 2011. Productivity impact of conservation farming
on smallholder cotton farmers in Zambia (unpublished).
Hahn, M.B., Riederer, A.M. & Foster, S.O. 2009. The Livelihood Vulnerability Index: A
pragmatic approach to assessing risks from climate variability and change - A case study in
Mozambique. Global Environmental Change, 19(1): 74–88.
Hassan, R. & Nhemachena, C. 2008. Determinants of African farmers’ strategies for adapting
to climate change: Multinomial choice analysis. African Journal of Agricultural and Resource
Economics, 2(1): 83–104.
Honig, L. & Mulenga, B.P. 2015. The status of customary land and the future of smallholder
farmers under the current land administration system in Zambia. Indaba Agricultural Policy
Research Institute.
JAICAF (Japan Association for International Collaboration of Agriculture and Forestry),
2008. One Village One Product Movement in Africa: Issues and Feasibility in Ethiopia. Tokyo,
JAICAF.
Jayne, T.S., Chapoto, A., Sitko, N., Nkonde, C., Muyanga, M. & Chamberlin, J. 2014. Is the
scramble for land in Africa foreclosing a smallholder agricultural expansion strategy?. Journal
of International Affairs, 67(2): 35–53.

32
Jayne, T.S., Zulu, B. & Nijhoff, J. J., 2006. Stabilizing food markets in eastern and southern
Africa. Food Policy, 31(4), 328–341.
Kerr, R.B., Snapp, S., Chirwa, M., Shumba, L. & Msachi, R. 2007. Participatory research on
legume diversification with Malawian smallholder farmers for improved human nutrition and soil
fertility. Experimental Agriculture, 43(4): 437–453.
King, A. E. & Hofmockel, K. S. 2017. Diversified cropping systems support greater microbial
cycling and retention of carbon and nitrogen. Agriculture, Ecosystems & Environment, 240(1):
66-76. Chicago, USA.
Krupinsky, J.M., Bailey, K.L., McMullen, M.P., Gossen, B.D. & Turkington, T.K. 2002.
Managing plant disease risk in diversified cropping systems. Agronomy Journal, 94(2): 198–
209.
Kumwenda, J.D.T., Waddington, S., Snapp, S.S., Jones, R.B. & Blackie, M.J. 1997. Soil
fertility management in the smallholder maize-based cropping systems of eastern and southern
Africa. In: Eisher, C. (Ed.), The Emerging Maize Revolution, pp. 153-72. Lansing, USA,
Michigan State University.
Levine, N.K., Mason, N.M. & Morgan, S.N. 2016. Do input subsidies crowd in or crowd out
other soil fertility management practices? Panel survey evidence from Zambia (unpublished).
Linard C., Gilbert M., Snow R.W., Noor A.M. & Tatem A.J. 2012. Population Distribution,
Settlement Patterns and Accessibility across Africa in 2010. PLoS ONE, 7(2): 1–8.
Mafongoya, P.L. & Jiri, O. 2015. Impact of conservation agriculture on weed dynamics and
maize grain yield in eastern Zambia. African Journal of Agricultural Research, 10(46): 4231–
4240.
Manda, J., Alene, A. D., Gardebroek, C., Kassie, C. & Tembo, G. 2016. Adoption and Impacts
of Sustainable Agricultural Practices on Maize Yields and Incomes: Evidence from Rural
Zambia. Journal of Agricultural Economics, 67(1): 130–153.
Mazunda, J. & Droppelmann, K. 2012. Maize consumption estimation and dietary diversity
assessment methods in Malawi (unpublished).
McKee, T. B., Doesken, N.J. & Kleist, J. 1993. The relationship of drought frequency and
duration to time scales. Proceedings of the 8th Conference on Applied Climatology, 17(22):
179–183.
Morton, J.F. 2007. The impact of climate change on smallholder and subsistence agriculture.
Proceedings of the national academy of sciences, 104(50): 19680–19685.
Mundlak, Y. 1978. On the pooling of time series and cross section data. Econometrica: journal
of the Econometric Society 46(1): 69–85.
Nielson, H.H. 2009. The role of cassava in smallholder maize marketing in Zambia and
Mozambique (Doctoral dissertation, Michigan State University. Agricultural, Food, and
Resource Economics).
Nhemachena, C. & Hassan, R. 2007. Micro-level analysis of farmers adaption to climate
change in Southern Africa. Washington, DC, International Food Policy Research Institute.
Nyangena, W. 2007. Social determinants of soil and water conservation in rural Kenya.
Environment, Development and Sustainability, 10(6): 745–767.
Reardon, T., Berdegue, J., Barrett, C.B. & Stamoulis, K. 2006. Household income
diversification into rural nonfarm activities. In: Haggblade, S., Hazell, P., Reardon, T. eds.,
Transforming the Rural Nonfarm Economy. Baltimore, USA, Johns Hopkins University Press.

33
Rockstrom, J., Kaumbutho, P., Mwalley, J., Nzabi, A.W., Temesgen, M., Mawenya, L.
Barron, J. & Mutua, J., Damgaard-Larsen S. 2009. Conservation farming strategies in East
and Southern Africa: Yields and rain water productivity from on-farm action research. Soil &
Tillage Research, 103(1): 23–32
Schlenker, W. & Lobell. D. 2010. Robust negative impacts of climate change on African
agriculture. Environmental Research Letters, 5(1): 1–8.
Scognamillo, A., Asfaw, S. & Ignaciuk, A. forthcoming. One is not enough: Adoption and the
impact of sustainable agricultural practices in rural Malawi. FAO Agricultural Development
Economics Working Paper. Rome, FAO.
Silberg, T.R., Richardson, R.B., Hockett, M. & Snapp, S.S. 2017. Maize-legume intercropping
in central Malawi: determinants of practice. International Journal of Agricultural Sustainability,
15(6): 662–680.
Sileshi, G., Akinnifesi, F.K., Ajayi, O.C. & Place, F. 2008. Meta-analysis of maize yield
response to woody and herbaceous legumes in sub-Saharan Africa. Plant and soil, 307(1–2):
1–19.
Sitko, N.J. 2012. Hunger for Money: Maize, Food-Insecurity, and Identity Formation Within
Polygynous Households in Rural Southern Zambia. Food, Culture & Society, 15(3): 415–435.
Sitko, N.J. & Chamberlin, J. 2016. The geography of Zambia’s customary land: assessing the
prospects for smallholder development. Land Use Policy, 55(1): 49–60.
Sitko, N.J., Chamberlin, J., Cunguara, B., Muyanga, M. & Mangisoni, J. 2017. A
comparative political economic analysis of maize sector policies in eastern and southern
Africa. Food Policy, 69(1): 243–255.
Sitko, N.J., Burke, W.J. & Jayne, T.S. 2017. Food System Transformation and Market
Evolutions: An Analysis of the Rise of Large-scale Grain Trading in Sub-Saharan Africa.
International Development Working Paper, 153.
Thierfelder, C., Rusinamhodzi, L., Setimela, P., Walker, F. & Eash, N.S. 2016. Conservation
agriculture and drought-tolerant germplasm: Reaping the benefits of climate-smart agriculture
technologies in central Mozambique. Renewable Agriculture and Food Systems, 31(5):
414–428.
Tizale, C.Y. 2007. The dynamics of soil degradation and incentives for optimal management in
the Central Highlands of Ethiopia. Department of Agricultural Economics, University of Pretoria.
(PhD dissertation).
Yadav, J.S.P. 1981. Reclamation and crop production in alkali soils. Current Science, 50(9):
387–393.

34
Maize mono- Maize-legume- Maize-legume- Maize-legume-
cropping Maize-legume Maize-staple Maize-cash crops staple cash crops cash crops-staple
(N = 1 342) (N = 1 142) (N = 174) (N = 322) (N = 282) (N = 359) (N=113)
Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd
Maize Yield (Kg/ha) 1 453 1 702 1 849 1951 1 596 1 672 1 533 1355 1 742 1 769 2 130 1 904 1 672 2 077
Dependents Crop Income (USD
Table A 1.
Annex 1

2010) 299 384 461 483 494 531 831 1266 472 493 1 239 1482 641 798

US Dollars 2010.
Maize Seeds Price/Kg
(USD 2010) 1.31 0.95 1.39 0.88 1.23 0.99 1.04 0.63 1.10 0.65 1.32 0.90 1.04 0.65
Staple Seeds Price/Kg
(USD 2010) 46.91 239.37 42.63 234.10 3.57 10.20 16.02 115.66 29.18 193.55 42.71 228.89 14.22 137.81
Cash Crop Seeds
Institutional
Price/Kg (USD 2010) 297.26 555.13 296.07 501.29 593.98 915.67 337.74 635.36 287.95 559.72 360.22 591.98 317.70 666.06

Source: Authors’ own elaboration.


Distance from the
5.14 6.92 4.90 6.93 5.47 6.55 4.39 5.94 4.24 6.46 4.42 5.87 5.02 6.97
market (EA's level)
Distance from FRA
7.94 5.88 7.41 5.00 8.23 5.94 8.2 5.32 8.83 6.00 8.2 4.85 8.44 5.35
(EA's level)
Seasonal Max
26.76 1.29 26.63 1.23 26.73 1.43 26.89 1.17 26.95 1.13 26.55 1.1 27 1.08
Temperature
Negative Seasonal SPI
-0.37 0.43 -0.38 0.5 -0.29 0.39 -0.39 0.45 -0.3 0.41 -0.42 0.52 -0.22 0.38
Index
Positive Seasonal SPI
Biophysical 0.08 0.15 0.09 0.15 0.14 0.23 0.09 0.15 0.13 0.19 0.1 0.17 0.18 0.22
Index
Negative Seasonal SPI
-0.17 0.21 -0.16 0.21 -0.15 0.2 -0.15 0.2 -0.15 0.21 -0.15 0.21 -0.21 0.22
Index (lag)

35
Positive Seasonal SPI
0.21 0.32 0.21 0.33 0.38 0.49 0.27 0.39 0.21 0.35 0.29 0.39 0.21 0.37
Index (lag)
Agricultural Asset
0.11 0.96 0.3 1.02 0.48 1.03 0.4 0.95 0.35 1.03 0.73 1.09 0.45 0.98
Wealth Index
Wealth Livestock (TLU units) 0.51 1.31 0.8 2.07 1.72 2.72 0.83 2.02 1.01 2.01 1.67 2.95 0.91 1.71

Land owned (ha) 0.5 0.43 0.7 0.48 0.68 0.47 0.78 0.57 0.78 0.55 1.02 0.66 0.78 0.57
HH size 5.1 2.27 5.13 2.19 5.38 2.25 5.6 2.47 5.26 2.39 5.92 2.44 5.65 2.35
Age 45.72 16.61 47.3 16.33 44.3 15.13 43.31 15.1 45.22 16.55 45.83 14.47 46.35 16.17
Summary statistics by cropping system in Malawi

Female-headed HH
0.26 0.44 0.27 0.44 0.21 0.41 0.15 0.36 0.3 0.46 0.16 0.37 0.26 0.44
(1=yes)
Education (years) 5.03 2.88 5.11 2.5 5.28 2.7 4.89 2.29 4.93 2.53 5.32 2.21 4.85 2.32
Socio-
Economic Share of Credit
0.17 0.17 0.17 0.15 0.15 0.16 0.15 0.12 0.16 0.16 0.18 0.12 0.17 0.14
Recipients
Share of FISP
0.52 0.25 0.55 0.22 0.49 0.26 0.6 0.23 0.59 0.22 0.60 0.21 0.61 0.20
Recipients
Inorganic fertilizer
0.75 0.44 0.79 0.41 0.71 0.45 0.84 0.37 0.87 0.34 0.91 0.29 0.87 0.34
(1=yes)
Flood Probability Index 1.35 2.26 1.52 2.46 1.25 2.04 1.53 2.24 1.13 2.01 2.07 2.61 1.65 2.00
Drought Probability
Instruments 3.54 2.86 3.57 2.85 3.19 2.78 3.42 2.57 3.5 2.94 3.26 2.57 3.16 2.09
Index
First Stage

Notes: the tables reports summaries of dependent and explanatory variables for Malawi. All the currencies are expressed in real
Formal Diffusion Index 0.52 0.27 0.57 0.26 0.54 0.26 0.51 0.26 0.55 0.26 0.57 0.25 0.56 0.21
Informal Diffusion Index 0.07 0.11 0.09 0.13 0.06 0.1 0.05 0.11 0.09 0.12 0.09 0.13 0.11 0.15
Maize-legume-
Maize mono- Maize-staple Maize-cash crops Maize-legume- Maize-legume-
Maize-legume cash crops-staple
cropping staple cash crops

(N = 286) (N = 962) (N = 466) (N = 88) (N = 2 351) (N = 283) (N = 619)


Table A 2.

Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd


Maize Yield (Kg/ha) 487 570 712 803 601 669 771 642 698 831 927 803 892 818
Dependents Crop Income (USD
219 1 243 397 2 493 448 4 767 740 1 924 360 2 284 1 099 4 659 687 3 251
2010)

in real US Dollars 2010.


Distance from Market
Market 505 410 455 401 354 343 212 150 404 383 208 211 209 219
cities (km, District level)
Seasonal Max

Source: Authors’ own elaboration.


29.94 1.45 29.70 1.78 29.89 1.21 29.77 1.41 29.99 1.25 29.28 1.44 30.06 1.10
Temperature
Negative Seasonal SPI
-0.10 0.19 -0.12 0.19 -0.04 0.13 -0.01 0.04 -0.08 0.16 -0.01 0.07 -0.02 0.08
Index
Positive Seasonal SPI
Biophysical 0.50 0.49 0.59 0.57 0.51 0.42 0.72 0.37 0.47 0.49 0.86 0.42 0.67 0.44
Index
Negative Seasonal SPI
-0.05 0.13 -0.09 0.19 -0.04 0.10 -0.08 0.14 -0.06 0.14 -0.12 0.17 -0.06 0.14
Index (lag)
Positive Seasonal SPI
0.54 0.41 0.46 0.41 0.53 0.50 0.48 0.48 0.48 0.47 0.34 0.43 0.56 0.58
Index (lag)
Agricultural Asset
0.24 0.98 0.33 1.07 -0.09 0.75 0.49 1.15 0.10 0.89 0.70 1.28 0.05 0.83
Wealth Index
Wealth Livestock (TLU units) 4.64 9.82 5.24 10.63 2.33 6.05 5.17 11.84 3.33 8.48 5.68 12.04 3.08 8.97

36
Land owned (ha) 1.68 1.55 2.14 2.23 2.15 2.65 3.55 5.18 2.36 2.55 4.03 5.58 3.41 2.76
Hh size 5.49 2.99 5.59 3.29 5.12 2.78 5.63 2.98 5.55 3.14 5.64 2.58 5.66 2.68
Age 45.76 16.45 45.60 16.37 44.41 15.42 41.51 13.94 45.73 15.49 42.90 14.48 43.05 14.84
Female-headed HH
0.26 0.44 0.32 0.47 0.29 0.45 0.13 0.33 0.29 0.45 0.15 0.36 0.16 0.37
(1=yes)
Socio-
Economic Education (ln) 6.01 3.88 5.67 3.80 5.32 3.55 5.47 3.57 5.66 3.55 5.08 3.57 4.91 3.33
Share of Credit
0.01 0.02 0.02 0.04 0.01 0.03 0.03 0.07 0.01 0.03 0.03 0.07 0.01 0.04
Recipients
Inorganic fertilizer
0.04 0.20 0.04 0.20 0.02 0.15 0.23 0.42 0.03 0.17 0.28 0.45 0.10 0.30
(1=yes)
Flood Probability Index 1.71 1.76 1.89 1.86 1.68 1.98 1.25 1.62 2.04 1.99 1.45 1.66 1.63 1.87
Summary statistics by cropping system in Mozambique

Drough Probability
Instruments 3.57 2.22 3.85 2.66 3.31 2.65 4.51 2.95 3.32 2.67 4.66 2.60 3.88 2.92
Index
First Stage
Formal Diffusion Index 0.05 0.06 0.06 0.06 0.06 0.06 0.11 0.14 0.05 0.05 0.12 0.14 0.08 0.09
Informal Diffusion Index 0.76 0.21 0.86 0.17 0.78 0.20 0.85 0.13 0.84 0.18 0.87 0.16 0.85 0.16
Road density
0.14 1.28 0.20 1.43 0.10 1.02 0.01 0.00 0.23 1.56 0.01 0.00 0.03 0.24
(#roads/area)
Infrastructure
Nightlight index (0-63) 1.25 2.63 1.24 2.94 1.26 2.12 0.92 0.32 1.42 3.20 0.97 0.42 1.17 0.67
development
District's Population
2 745 13 542 2 464 8 212 2 389 2 481 2 336 2 353 2 604 5 609 2 060 2 077 2 879 3 059
Density

Notes: the tables reports summaries of dependent and explanatory variables for Mozambique. All the currencies are expressed
Maize mono- Maize-cash crops Maize-legume- Maize-legume- Maize-legume-
Maize-legume Maize-staple
cropping staple cash crops cash crops-staple
(N = 1 705) (N = 2 296) (N = 1 468) (N = 690) (N = 3 202) (N = 2 063) (N = 472)
Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd
Maize Yield (Kg/ha) 2 096 1 585 2 309 1 645 1 769 1 494 2 086 1 586 2 595 1 646 2 482 1 697 2 637 1 602
Dependents Crop Income (USD
1 908 5 132 2 917 11 375 2 432 3 830 2 704 4 264 3 614 4 552 3 683 4 222 4 582 4 077
Table A 3.

2010)
Maize Seeds Price/Kg
3.54 1.11 3.59 1.06 2.59 1.38 3.49 1.17 3.20 1.25 3.47 1.14 3.33 1.20
(USD 2010)

in real US Dollars 2010.


Legume Seeds
2.10 1.25 1.93 1.07 1.84 0.72 1.82 1.60 1.82 0.74 1.82 1.86 1.98 1.82
Price/Kg (USD 2010)
Staple Seeds Price/Kg
0.27 0.14 0.27 0.14 0.23 0.10 0.34 0.19 0.22 0.13 0.33 0.19 0.26 0.16
(USD 2010)
Institutional

Source: Authors’ own elaboration.


Cash Crop Seeds
4.30 3.78 3.96 3.44 3.41 1.84 3.22 2.72 2.95 2.33 3.48 3.09 2.93 2.25
Price/Kg (USD 2010)
Distance from the
7.72 16.48 8.27 16.13 6.12 10.65 6.26 8.73 11.3 89.5 7.51 9.88 9.99 22.54
market (EA's level)
Distance from FRA
7.24 9.92 7.70 9.92 13.05 20.41 7.38 9.89 8.92 13.72 7.19 9.11 9.57 12.78
(EA's level)
Seasonal Max
26.87 1.22 26.9 1.29 26.93 1.27 27.14 1.21 26.23 1.29 27.25 1.23 26.74 1.33
Temperature
Negative Seasonal SPI
-0.11 0.20 -0.15 0.24 -0.08 0.22 -0.23 0.26 -0.20 0.35 -0.22 0.24 -0.30 0.40
Index
Positive Seasonal SPI
Biophysical 0.35 0.52 0.34 0.52 0.86 0.81 0.18 0.29 0.67 0.77 0.19 0.29 0.31 0.47
Index

37
Negative Seasonal SPI
-0.11 0.25 -0.17 0.3 -0.07 0.19 -0.3 0.41 -0.08 0.2 -0.32 0.4 -0.15 0.29
Index (lag)
Positive Seasonal SPI
0.63 0.57 0.57 0.61 0.61 0.51 0.73 0.76 0.54 0.53 0.79 0.81 0.73 0.69
Index (lag)
Agricultural Asset
0.11 0.19 0.12 0.20 0.08 0.15 0.11 0.17 0.10 0.18 0.14 0.18 0.16 0.21
Wealth Index
Wealth Livestock (TLU units) 3.3 12.24 5.56 15.41 3.15 10.52 4.88 10.06 3.45 11.52 6.79 11.15 6.72 10.52
Land owned (ha) 2.97 8.17 4.65 9.35 3.61 5.72 3.71 3.93 6.13 12.26 4.92 4.85 6.66 8.61
HH size 6.28 2.84 6.75 3.12 6.53 2.76 6.27 2.77 6.91 2.78 6.87 2.96 7.39 3.19
Age 35.46 12.2 36.55 11.96 35.15 10.94 33.29 9.85 34.95 10.00 34.64 9.90 34.33 9.38
Female-headed HH
Summary statistics by cropping system for Zambia

0.22 0.42 0.26 0.44 0.18 0.39 0.13 0.34 0.17 0.38 0.14 0.35 0.14 0.34
(1=yes)
Education (years) 6.55 4.22 6.06 4.03 6.04 3.52 5.03 3.73 6.64 3.46 5.48 3.59 6.31 3.30
Socio-
Economic Share of Credit
0.13 0.17 0.16 0.20 0.05 0.09 0.43 0.27 0.07 0.10 0.44 0.26 0.29 0.26
Recipients
Share of FISP
0.43 0.25 0.48 0.25 0.35 0.29 0.43 0.24 0.56 0.26 0.46 0.23 0.50 0.24
Recipients
Inorganic fertilizer
0.64 0.48 0.72 0.45 0.45 0.5 0.59 0.49 0.75 0.43 0.75 0.44 0.77 0.42
(1=yes)
Flood Probability Index 1.12 1.72 1.54 1.85 1.25 1.74 1.73 1.93 1.53 1.84 2.07 2.03 2.29 1.94
Drought Probability
Instruments 2.29 2.45 2.26 2.42 1.56 2.1 2.45 2.27 1.69 2.13 2.4 2.24 2.18 2.48
Index

Notes: the table reports mean and standard deviation of dependent and explanatory variables for Zambia. All the currencies are expressed
First Stage
Formal Diffusion Index 0.38 0.23 0.42 0.22 0.32 0.25 0.38 0.24 0.44 0.23 0.38 0.23 0.39 0.23
Informal Diffusion Index 0.34 0.24 0.36 0.24 0.27 0.22 0.34 0.24 0.28 0.20 0.36 0.24 0.29 0.21
Table A 4. Determinants of adoption of different farm systems in Malawi, pooled
sample and complete list of coefficients
Maize- Maize- Maize-cash Maize-legume- Maize-legume- Maize-legume-cash
legume staple crops staple cash crops crops-staple

Maize Seeds Price (EA's level) 0.342** -0.150 -1.071*** -0.902*** 0.302 -1.058**
(0.166) (0.331) (0.259) (0.259) (0.241) (0.474)
Staple Seeds Price (EA's level) -0.045 -0.232*** -0.128** -0.252*** -0.040 -0.528***
(0.039) (0.083) (0.059) (0.058) (0.058) (0.135)
Cash-crop Seeds Price (EA's
-0.027 0.105** -0.110*** -0.038 -0.082** -0.214***
level)
(0.024) (0.050) (0.036) (0.038) (0.033) (0.063)
Distance from Weekly Market
0.005 0.071 -0.117* -0.157** -0.042 -0.023
(EA's level)
(0.047) (0.085) (0.067) (0.072) (0.068) (0.102)
Distance from FRA (ln, EA's level) -0.115 0.066 0.209 0.269** 0.344*** 0.254*
(0.088) (0.182) (0.139) (0.132) (0.128) (0.154)
Agricultural Asset Wealth Index 0.124** 0.089 0.146* 0.124 0.256*** 0.192
(0.059) (0.105) (0.084) (0.098) (0.080) (0.124)
Livestock (TLU units) 0.049 0.180*** 0.051 0.119*** 0.151*** 0.077
(0.041) (0.041) (0.055) (0.044) (0.042) (0.058)
Land owned (ln, ha) 1.627*** 1.416*** 2.218*** 2.735*** 3.368*** 2.324***
(0.206) (0.365) (0.319) (0.357) (0.298) (0.468)
HH size (ln) -0.160 0.044 0.266 -0.029 0.325* 0.434*
(0.106) (0.198) (0.184) (0.191) (0.172) (0.234)
Age (ln) 0.010 -0.622** -0.653*** -0.607** -0.477** -0.225
(0.153) (0.308) (0.235) (0.252) (0.240) (0.360)
Female-headed HH (1=yes) 0.211* -0.008 -0.473** 0.429** -0.056 0.120
(0.123) (0.243) (0.205) (0.190) (0.193) (0.257)
Education (years, ln) 0.198** -0.034 -0.034 -0.088 0.209 -0.189
(0.098) (0.195) (0.147) (0.156) (0.155) (0.223)
Share of Credit Recipients 0.045 -0.459 0.320 -0.192 0.528 -0.527
(0.362) (0.737) (0.495) (0.571) (0.477) (0.728)
Share of FISP Recipients 0.634*** -0.570 0.988*** 0.445 1.044*** 0.721
(0.235) (0.434) (0.346) (0.376) (0.319) (0.504)
Inorganic fertilizer (1=yes) 0.127 -0.126 0.388** 0.728*** 0.873*** 0.689**
(0.122) (0.229) (0.184) (0.215) (0.223) (0.312)
Negative Seasonal SPI Index (lag) 0.450 0.379 -0.023 1.238** -0.121 -0.569
(0.320) (0.595) (0.482) (0.507) (0.472) (0.703)
Positive Seasonal SPI Index (lag) 0.170 0.578** 0.504** -0.634** 0.530** 0.202
(0.192) (0.271) (0.257) (0.320) (0.263) (0.393)
Probabilities of flood (SPI 6
0.051** -0.022 0.004 -0.039 0.114*** 0.103**
months April)
(0.024) (0.051) (0.031) (0.045) (0.033) (0.051)
Probabilities of drought (SPI 6
0.028 0.020 -0.028 0.068** 0.012 0.076
months April)
(0.022) (0.041) (0.030) (0.034) (0.030) (0.050)
Formal Diffusion (EA's level) 0.303 0.953** 0.169 0.645* 0.597* 0.443
(0.238) (0.451) (0.337) (0.381) (0.352) (0.559)
Informal Diffusion (EA's level) 0.565 -2.129** -1.810* 0.905 0.652 2.804***
(0.479) (0.992) (0.977) (0.705) (0.701) (0.987)
Constant -1.844** -1.806 -0.236 -1.960 -5.467*** -4.974***
(0.741) (1.583) (1.107) (1.206) (1.143) (1.798)
Year dummy Yes Yes Yes Yes Yes Yes
Agro-ecological dummy Yes Yes Yes Yes Yes Yes
Observations 3 977 3 977 3 977 3 977 3 977 3 977
Year Dummy Yes Yes Yes Yes Yes Yes
Notes: the table displays the determinants of adoption of different farm systems, compared to the baseline scenario of maize mono-
cropping for the panel pooled sample (2010 and 2013). The specification is the first stage of a multinomial treatment effect model
(multinomial logit). The specification controls also for household-size (ln), age (ln), dummy for female-headed households, years of
education of the household's head (ln), asset wealth index, livestock owned (as TLU units), hectares of land owned (ln), share of
credit and FISP recipients at EA level, use of inorganic fertilizer, mean seasonal maximum temperature, positive and negative SPI
indexes for both seasonal and lagged, agro-ecological and year dummy. Standard errors are clustered at EA's level. Significant
levels are * p<0.10, ** p<0.05, *** p<0.001.
Source: Authors’ own elaboration.

38
Table A 5. Determinants of adoption of different farm systems in Mozambique, pooled
sample and complete list of coefficients

Maize- Maize- Maize-cash Maize-legume- Maize-legume- Maize-legume-


legume staple crops staple cash crops cash crops-staple
Distance from the main market
-0.019 -0.159** -0.244** -0.143** -0.215** -0.342***
cities (EA's level)
(0.075) (0.074) (0.099) (0.070) (0.088) (0.082)
Agricultural Asset Wealth Index 0.054 -0.493*** 0.036 -0.329*** 0.184 -0.582***
(0.092) (0.114) (0.167) (0.093) (0.142) (0.126)
Livestock (TLU units) 0.003 -0.016 -0.013 -0.017** -0.023** -0.030**
(0.007) (0.010) (0.014) (0.008) (0.011) (0.012)
Land owned (ln, ha) 0.114 0.395*** 0.807*** 0.462*** 1.019*** 1.473***
(0.079) (0.104) (0.189) (0.079) (0.124) (0.130)
HH size (ln) -0.006 -0.105 -0.024 0.053 0.095 0.152
(0.152) (0.164) (0.281) (0.144) (0.190) (0.166)
Age (ln) -0.042 -0.046 -0.781** 0.201 -0.673** -0.591**
(0.251) (0.276) (0.368) (0.239) (0.306) (0.271)
Female-headed HH (1=yes) 0.417** 0.079 -0.659* 0.210 -0.333 -0.437**
(0.184) (0.206) (0.354) (0.175) (0.252) (0.208)
Education (years, ln) 0.012 0.004 -0.180 0.111 -0.303** -0.254**
(0.107) (0.112) (0.188) (0.104) (0.146) (0.119)
Share of Credit Recipients 6.231*** 0.369 6.981** 1.117 6.288** 1.149
(2.411) (2.569) (2.725) (2.702) (3.078) (2.864)
Inorganic fertilizer (1=yes) -0.330 -0.478 1.332** -0.396 1.425** 0.785
(0.519) (0.492) (0.614) (0.540) (0.560) (0.560)
Negative Seasonal SPI Index (lag) -1.049 1.723** 2.127* 0.645 0.698 1.002
(0.682) (0.843) (1.173) (0.740) (0.867) (0.872)
Positive Seasonal SPI Index (lag) -0.243 -0.228 0.111 -0.361 -0.395 0.349
(0.233) (0.257) (0.348) (0.251) (0.330) (0.272)
Probabilities of flood
0.061 -0.003 -0.066 0.091 0.017 0.043
(SPI 6 months April)
(0.055) (0.061) (0.083) (0.058) (0.074) (0.064)
Probabilities of drought
0.035 -0.048 0.065 -0.050 0.061 0.034
(SPI 6 months April)
(0.034) (0.039) (0.065) (0.035) (0.055) (0.044)
Formal Diffusion (EA's level) 0.193 2.438 5.158** 0.225 5.251** 3.646*
(2.016) (2.253) (2.273) (2.046) (2.121) (2.216)
Informal Diffusion (EA's level) 2.693*** 0.538 2.819*** 2.565*** 2.964*** 2.907***
(0.505) (0.487) (0.861) (0.474) (0.769) (0.629)
Constant -1.380 1.121 -0.377 -0.082 -0.074 0.996
(1.155) (1.271) (1.646) (1.088) (1.400) (1.282)
Year dummy Yes Yes Yes Yes Yes Yes
Regional Dummies Yes Yes Yes Yes Yes Yes
Observations 5 055 5 055 5 055 5 055 5 055 5 055
Year Dummy Yes Yes Yes Yes Yes Yes
Notes: the table displays the determinants of adoption of different farm systems, compared to the baseline scenario of maize mono-
cropping for the Mozambique’s sample. The specification is the first stage of a multinomial treatment effect model (multinomial
logit). The specification controls also household-size (ln), age (ln), dummy for female-headed households, years of education of the
household's head (ln), asset wealth index, livestock owned (as TLU units), hectares of land owned (ln), use of inorganic fertilizer,
mean seasonal maximum temperature, positive and negative SPI indexes for both seasonal and lagged, regional dummies.
Standard errors are clustered at EA's level. Significant levels are * p<0.10, ** p<0.05, *** p<0.001.
Source: Authors’ own elaboration.

39
Table A 6. Determinants of adoption of different farm systems in Zambia, pooled
sample and complete list of coefficients

Maize- Maize- Maize-cash Maize- Maize-legume- Maize-legume-


legume staple crops legume-staple cash crops cash crops-staple
Maize Seeds Price (EA's level) 0.030 -1.394*** 0.161 -1.082*** 0.117 -1.028**
(0.176) (0.240) (0.258) (0.232) (0.227) (0.411)
Cash-crop Seeds Price (EA's level) -0.404** -0.222 -0.432** -0.689*** -0.550*** -0.210
(0.180) (0.278) (0.215) (0.248) (0.208) (0.270)
Legume Seeds Price (EA's level) 0.494 -1.112 2.515*** -0.344 1.858*** -0.131
(0.554) (0.736) (0.638) (0.770) (0.603) (0.878)
Staple Seeds Price (EA's level) -0.196* -0.101 -0.955*** -0.685*** -0.681*** -1.202***
(0.114) (0.159) (0.204) (0.161) (0.168) (0.192)
Total Traders (ln, EA's level) 0.346*** -0.266** 0.242* 0.423*** 0.510*** 0.314**
(0.075) (0.115) (0.125) (0.101) (0.106) (0.136)
Distance from FRA (ln, EA's level) 0.195** 0.551*** 0.105 0.650*** 0.200* 0.379**
(0.078) (0.119) (0.122) (0.109) (0.104) (0.193)
Agricultural Asset Wealth Index -0.493* -0.671* -0.057 -0.962*** -0.458 -0.274
(0.263) (0.343) (0.334) (0.263) (0.305) (0.370)
Livestock (TLU units) 0.007 -0.005 -0.000 -0.001 0.001 -0.001
(0.005) (0.006) (0.006) (0.004) (0.006) (0.006)
Land owned (ln, ha) 0.614*** 0.569*** 0.860*** 1.155*** 1.472*** 1.513***
(0.060) (0.071) (0.086) (0.069) (0.080) (0.097)
HH size (ln) 0.125 0.115 -0.113 0.248** 0.199* 0.553***
(0.090) (0.112) (0.146) (0.104) (0.109) (0.157)
Age (ln) 0.501*** 0.205 -0.918*** 0.529*** -0.228 0.053
(0.133) (0.178) (0.199) (0.145) (0.163) (0.241)
Female-headed HH (1=yes) 0.495*** -0.236* -0.707*** 0.289*** -0.340*** 0.053
(0.104) (0.128) (0.190) (0.105) (0.126) (0.175)
Education (years, ln) -0.039 -0.050 -0.271*** 0.127* -0.095 -0.005
(0.060) (0.077) (0.075) (0.067) (0.075) (0.106)
Share of Credit Recipients 0.253 -4.068*** 4.342*** -2.849*** 4.108*** 2.302***
(0.314) (0.594) (0.492) (0.512) (0.450) (0.546)
Share of FISP Recipients 0.498* 0.051 0.188 1.435*** 0.260 1.446***
(0.288) (0.409) (0.513) (0.335) (0.423) (0.452)
Inorganic fertilizer (1=yes) 0.178 -0.640*** -0.049 -0.135 0.565*** 0.252
(0.109) (0.127) (0.151) (0.126) (0.139) (0.184)
Negative Seasonal SPI Index (lag) -0.617** 0.470 -2.017*** 0.767** -3.560*** -1.320***
(0.246) (0.381) (0.364) (0.340) (0.331) (0.440)
Positive Seasonal SPI Index (lag) 0.051 -0.285* -0.139 -0.546*** 0.954*** -0.532*
(0.129) (0.172) (0.243) (0.158) (0.189) (0.274)
Probabilities of flood (SPI 6 months
0.068* -0.007 -0.035 0.054 0.024 0.124**
April)
(0.036) (0.048) (0.063) (0.047) (0.052) (0.062)
Probabilities of drought (SPI 6
-0.023 -0.037 0.048 -0.083** -0.011 -0.023
months April)
(0.029) (0.040) (0.062) (0.035) (0.044) (0.056)
Formal Diffusion (EA's level) 0.624** 0.025 0.114 1.550*** -0.314 0.830
(0.306) (0.438) (0.515) (0.384) (0.438) (0.579)
Informal Diffusion (EA's level) 0.001 -1.406*** -0.673 -1.589*** -0.358 -1.206**
(0.308) (0.417) (0.509) (0.369) (0.402) (0.550)
Constant -3.754*** 0.597 0.911 -3.369*** -4.128*** -4.356***
(0.793) (1.007) (1.101) (1.004) (0.959) (1.417)
Year dummy Yes Yes Yes Yes Yes Yes
Agro-ecological dummy Yes Yes Yes Yes Yes Yes
Observations 11 496 11 496 11 496 11 496 11 496 11 496
Year Dummy Yes Yes Yes Yes Yes Yes
Notes: the table displays the determinants of adoption of different farm systems, compared to the baseline scenario of maize mono-
cropping for the panel pooled sample (2011 and 2014). The specification is the first stage of a multinomial treatment effect model
(multinomial logit). The model controls also for the total number of traders (ln), the distance from FRA (ln), household-size (ln), age
(ln), dummy for female-headed households, years of education of the household's head (ln), asset wealth index, livestock owned
(as TLU units), hectares of land owned (ln), share of credit and FISP recipients at EA level, use of inorganic fertilizer, mean seasonal
maximum temperature, positive and negative SPI indexes for both seasonal and lagged, agro-ecological and year dummies.
Standard errors are clustered at EA's level. Significant levels are * p<0.10, ** p<0.05, *** p<0.001.
Source: Authors’ own elaboration.

40
Table A 7. Impact of different farm systems on maize yield, pooled sample and
complete list of coefficients

Dependent: Maize Yield (ln)


Malawi Mozambique Zambia (Wave 1+2)
Market index: distance from Market index: distance from Market index: total
weekly market main market cities traders (ln)
Maize-Legume (1=yes) 0.285*** 0.080 -0.020
(0.061) (0.103) (0.045)
Maize-Staple (1=yes) -0.075 0.222* -0.181***
(0.097) (0.116) (0.053)
Maize-Cash Crops (1=yes) 0.155* 0.104 0.043
(0.083) (0.196) (0.063)
Maize-Legume-Staple (1=yes) 0.267*** 0.185* 0.269***
(0.086) (0.094) (0.042)
Maize-Legume-Cash Crops
0.468*** 0.569*** 0.114**
(1=yes)
(0.069) (0.113) (0.053)
Maize-Legume-Cash Crops-
0.002 0.776*** 0.236***
Staple (1=yes)
(0.106) (0.114) (0.052)
Maize Seeds Price (EA's level) 0.125** NA 0.141***
(0.063) NA (0.035)
Legume Seeds Price (EA's
NA NA -0.050
level)
NA NA (0.033)
Staple Seeds Price (EA's
0.034** NA -0.361***
level)
(0.014) NA (0.103)
Cash-crop Seeds Price (EA's
0.000 NA -0.046*
level)
(0.008) NA (0.024)
Market index 0.024 0.054*** 0.081***
(0.016) (0.019) (0.015)
Distance from FRA (ln, EA's
-0.075** NA -0.019
level)
(0.029) NA (0.016)
Agricultural Asset Wealth
0.115*** 0.092*** 0.407***
Index
(0.018) (0.027) (0.046)
Livestock (TLU units) 0.026*** 0.006*** 0.004***
(0.008) (0.002) (0.001)
Land owned (ln, ha) -0.027 -0.452*** -0.094***
(0.063) (0.023) (0.011)
Hh size (ln) -0.010 0.085*** 0.011
(0.036) (0.028) (0.017)
Age (ln) 0.037 0.133*** -0.035
(0.053) (0.043) (0.026)
Female-headed HH (1=yes) -0.066 -0.079** -0.068***
(0.041) (0.035) (0.020)
Education (ln) 0.180*** 0.064*** 0.059***
(0.035) (0.021) (0.012)
% Credit Recipients (average
0.059 -0.004 0.144**
EA's level)
(0.118) (0.424) (0.065)
% FISP Recipients (average
-0.130* NA 0.206***
EA's level)
(0.070) NA (0.053)

41
Dependent: Maize Yield (ln)
Malawi Mozambique Zambia (Wave 1+2)
Market index: distance from Market index: distance from Market index: total
weekly market main market cities traders (ln)
Inorganic fertilizer (1=yes) 0.385*** 0.328*** 0.545***
(0.047) (0.074) (0.022)
Mean Seasonal Maximum
-0.075*** -0.030* 0.009
Temperature (C)
(0.020) (0.018) (0.012)
SPI Negative -0.112*** -0.856*** -0.058
(0.039) (0.174) (0.048)
SPI Positive -0.322*** 0.160*** 0.005
(0.119) (0.054) (0.018)
SPI Negative (lag) -0.067 0.033 -0.169***
(0.105) (0.144) (0.044)
SPI Positive (lag) 0.074 0.058 0.064**
(0.063) (0.045) (0.028)
Constant 8.155*** 5.851*** 6.305***
(0.608) (0.610) (0.386)
Agro-ecological/Regional
Yes Yes Yes
dummies
Number of observations 3 732 5 055 11 858
Log-Likelihood -10 573.14 -14 279.67 -28 319.58
chi2 / F / Wald 1 468.986 1 747.862 9 323.872

Notes: the table displays the determinants of maize yield, compared to the baseline scenario of maize mono-cropping. The
specification is the second stage of a multinomial treatment effect model (multinomial logit). Column 1-2-3 report results Malawi,
Mozambique, and Zambia. The instrumental variables are the flood and drought shock probability, and the average diffusion of
formal and informal institutions at EA's level (not reported). The shocks probability are measured as the share of seasons where a
flood or drought shock occurred (SPI>2). Column 5-6 are OLS-FE and GLS-RE. The specifications include the total number of
traders (ln), the distance from FRA (ln) household-size (ln), age (ln), dummy for female-headed households, years of education of
the household's head (ln), asset wealth index, livestock owned (as TLU units), hectares of land owned (ln), share of credit and FISP
recipients at EA level, use of inorganic fertilizer, mean seasonal maximum temperature, positive and negative SPI indexes for both
seasonal and lagged, agro-ecological and year dummies (only for the pooled sample). Significant levels are * p<0.10, ** p<0.05,
*** p<0.001 and errors are clustered at EA's level.
Source: Authors’ own elaboration.

42
Table A 8. Impact of different farm systems on crop income volatility, pooled sample
and complete list of coefficients

Dependent: Maize Yield (ln)


Malawi Mozambique Zambia (Wave 1+2)
Market index: distance from Market index: distance from main Market index: total
weekly market market cities traders (ln)
Maize-Legume (1=yes) -1.477*** -0.122 -0.251***
(0.450) (0.139) (0.037)

Maize-Staple (1=yes) 0.220 -0.011 -0.187***


(0.814) (0.163) (0.051)

Maize-Cash Crops (1=yes) -0.542 0.092 -0.273***


(0.691) (0.326) (0.052)

Maize-Legume-Staple (1=yes) -0.196 -0.255* -0.363***


(0.646) (0.132) (0.039)
Maize-Legume-Cash Crops -1.110 -0.414*** -0.392***
(1=yes)
(0.683) (0.154) (0.043)
Maize-Legume-Cash Crops- -0.914 -0.168 -0.430***
Staple (1=yes)
(0.779) (0.139) (0.046)

Maize Seeds Price (EA's level) 1.361** NA -0.053*


(0.624) NA (0.032)

Legume Seeds Price (EA's level) NA NA 0.088**


NA NA (0.040)

Staple Seeds Price (EA's level) -0.210 NA -0.172*


(0.141) NA (0.089)
Cash-crop Seeds Price (EA's -0.294*** NA -0.007
level)
(0.094) NA (0.019)

Market index 0.170 0.050* -0.047***


(0.153) (0.026) (0.017)
Distance from FRA (ln, EA's -0.417 NA 0.022
level)
(0.269) NA (0.017)

Agricultural Asset Wealth Index -0.212 0.013 -0.018


(0.163) (0.048) (0.077)

Livestock (TLU units) -0.020 0.008 0.002*


(0.050) (0.005) (0.001)

Land owned (ln, ha) -1.593** 0.094** 0.131***


(0.623) (0.044) (0.020)

Hh size (ln) 0.159 -0.022 -0.058**


(0.266) (0.059) (0.024)

Age (ln) 0.881** -0.054 -0.065**


(0.395) (0.094) (0.031)

Female-headed HH (1=yes) 0.151 -0.020 0.057**


(0.330) (0.055) (0.026)

Education (ln) 0.199 -0.049 -0.028**


(0.366) (0.041) (0.013)
% Credit Recipients (average -1.459 1.019 -0.016
EA's level)
(1.080) (1.009) (0.052)

43
Dependent: Maize Yield (ln)
Malawi Mozambique Zambia (Wave 1+2)
Market index: distance from Market index: distance from main Market index: total
weekly market market cities traders (ln)
% FISP Recipients (average -2.413*** NA -0.110**
EA's level)
(0.855) NA (0.047)

Inorganic fertilizer (1=yes) 0.896** 0.520** -0.099***


(0.385) (0.258) (0.024)
Mean Seasonal Maximum 0.730** 0.050** -0.018*
Temperature (C)
(0.330) (0.024) (0.011)

SPI Negative -0.362 -0.536* 0.072


(0.397) (0.309) (0.048)

SPI Positive 3.529*** -0.183* 0.035


(1.157) (0.100) (0.031)

SPI Negative (lag) 0.432 -0.139 0.168***


(0.856) (0.263) (0.049)

SPI Positive (lag) 1.244* -0.049 -0.026


(0.640) (0.085) (0.028)

Constant -16.018* -0.589 -108.411***


(8.689) (0.777) (24.251)
Agro-ecological/Regional Yes Yes Yes
dummies
Number of observations 3 732 5 395 11 884

Log-Likelihood -19 078.90 -18 788.02 -31 588.78

chi2 / F / Wald 814.937 773.119 6 686.129

Notes: the table displays the determinants of crop income volatility, compared to the baseline scenario of maize mono-cropping for
Malawi (column 1), Mozambique (column 2), and Zambia (column 3). The specification is the second stage of a multinomial
treatment effect model (MTM) where the dependent is the squared residual from a MTM including crop income as dependent
variable. The instrumental variables are the flood and drought shock probability, and the average diffusion of formal and informal
institutions at EA's level (not reported). The shocks probability are measured as the share of seasons where a flood or drought
shock occurred (SPI>2). Column 5-6 are OLS-FE and GLS-RE. Significant levels are * p<0.10, ** p<0.05, *** p<0.001 and errors
are clustered at EA's level.
Source: Authors’ own elaboration.

44
Table A 9. Main market cities employed for the private market proxy in Mozambique

District in Provinces Maket cities


Niassa, Cabo Delgado, Nampula, Zambezia Nampula
Tete, Manica Tete
Sofala, Inhambane Beira
Gaza, Maputo Maputo
Source: Authors’ own elaboration.

Table A 10. Main market cities employed for the private market proxy in Mozambique

Dependent: Labour productivity in agriculture


Malawi Mozambique Zambia (Wave 1+2)
Maize-Legume (1=yes) 0.247*** 0.308** 0.298***
(0.091) (0.127) (0.071)
Maize-Staple (1=yes) 0.297*** -0.256 0.609***
(0.105) (0.158) (0.098)
Maize-Cash Crops (1=yes) 0.502*** 0.266 0.528***
(0.097) (0.196) (0.107)
Maize-Legume-Staple (1=yes) 0.260** 0.141 0.720***
(0.105) (0.125) (0.065)
Maize-Legume-Cash Crops (1=yes) 0.882*** 0.652*** 0.436***
(0.100) (0.139) (0.073)
Maize-Legume-Cash Crops-Staple (1=yes) 0.452*** 0.848*** 0.642***
(0.107) (0.128) (0.056)
Other controls Yes Yes Yes
Agro-ecological/Regional dummies Yes Yes Yes
Number of observations 3 732 5 395 11 884
Log-Likelihood -8 967.18 -18 788.02 -30 252.89
chi2 / F / Wald 3 476.197 773.119 17 610.565
Notes: the table displays the determinants of labour productivity in agriculture, measured as crop income per unit of adult worker,
compared to the baseline scenario of maize mono-cropping for Malawi (column 1), Mozambique (column 2) and Zambia (column
3). The specification is the second stage of a multinomial treatment effect model (MTM). The baseline category is maize mono-
cropping. The instrumental variables in the first stage are the flood and drought shock probability and the average diffusion of formal
and informal institutions at EA's level. The entire set of coefficients are available upon request. Significant levels are * p<0.10, **
p<0.05, *** p<0.001 and errors are clustered at EA's level.
Source: Authors’ own elaboration.

45
Technical note on the conceptual and empirical approach
Following Feder (1982), Di Falco and Veronesi (2013) and Manda et al., (2016), we model the
adoption of different cropping systems using the random utility framework. In this context, we
posit that risk-adverse farmers select the cropping system maximizing their utility conditional on
land and labour availability, input costs and other ecological, demographic, and institutional
constraints. Formally, the assumption is that a given farmer i will maximise his/her expected
indirect utility Vij comparing the utility provided by alternative systems. This assumption implies
the selection of any practice j over any alternative practice k if Vij > Vik.
Adoption within any category is a non-random event, as farmers are pushed/pulled into each
system according to both observable factors, such as socio-economic characteristics and level
of resources and unobservable ones, such as risk aversion and expectations on future
weather’s distribution. For example, it is probable that complex cropping systems will be mainly
adopted by high skilled farmers with technical abilities. If the empirical model fails to account for
this, the coefficient linked to the farm system will be upward biased, as it will likely capture the
effect of the innate ability of the farmer independently from the farm system adopted.
This study adopts a multinomial endogenous treatment effect model to address the endogeneity
of farming selection while accounting for the categorical nature of the cropping systems’
variable, the heterogeneity arising from a different levels of exposure of farmers to the climate
risk, and the interdependence of cropping system’ adoption ((Deb and Trivedi, 2006, Manda et
al. 2016; Di Falco and Veronesi 2013).

Adoption equation
This model requires setting a benchmark for the sake of comparison which in this case is maize
mono-cropping.12 Let 𝑗 = 0 be the control group category, with Vie0=0, adopted by household i
residing in village e. Denoting as 𝑓𝑗 the variable on the type of farming system, 𝒁𝑖𝑒𝑡 a matrix of
control variables, 𝛼𝑗𝑡 their linked coefficients and 𝑙𝑖 = (𝑙𝑖1 , 𝑙𝑖2 … 𝑙𝑖𝐽 ) a vector of latent variables,
the probability of adopting a farming system j takes the following form:

′ ′ ′
Pr(𝑓𝑗 |𝑧𝑖 𝑙𝑖 ) = 𝑔(𝑧1𝑒 𝛼1 + 𝛿1 𝑙𝑖1 , 𝑧𝑒2 𝛼2 + 𝛿2 𝑙𝑖2 , … , 𝑧𝑒𝑖 𝛼𝑗 + 𝛿𝑗 𝑙𝑖𝑗 ) (1)

With 𝑔 being a multinomial probability distribution characterized by a Mixed Multinomial Logit


(MMNL) structure. The non-linear form of the selection equation embedded in the semi-
structural model guarantees the identification of the parameters even when the explanatory
variables in the two stages coincide. However, Deb and Trivedi (2006) suggests using an
instrumental variable for the exclusion restriction to obtain a more robust identification. The
instrument involves at least an explanatory variable entering exclusively in the adoption
equation and not in the outcome one. The identification derives from the exclusion restriction,
implying that the instrumental variable does not impact the outcome equations through any other
channel. The matrix of explanatory variables involves four instrumental variables in the first
stage: 1) drought risk exposure; 2) flood risk exposure; 3) informal and 4) formal diffusion of
agricultural practices. The first two variables follow the work from Di Falco and Veronesi (2013),

12The selection of maize mono-cropping as benchmark derives from theoretical and statistical considerations. From
a theoretical perspective, literature often associates maize mono-cropping to the poorest part of the population
(Hassan and Nhemachena, 2008).

46
who consider the experience of extreme weather events such as droughts, flood and hailstorms,
to instrument the adoption of different adaptation strategies. For the validity of the exclusion
restriction, the underlying hypothesis is that farmers are more likely to adopt a cropping system
different from maize mono-cropping when residing in areas that have been more historically
exposed to weathers shocks. Being a subjective belief of the household, the postulate is that
this will not affect the yield or crop income volatility through other channels. Flood and drought
risk exposure are computed for household i residing in the area e as the probability of being
exposed to this event conditional on area e historical weather distribution. A drought shock if
the Rainfall SPI index in a given area and agricultural season takes a value lower than -2
(Guttman, 1998; McKee et al., 1993). A flood shock occurs when the SPI index takes value
above 2. Following Scognamillo, Asfaw and Ignaciuk (forthcoming) and generalizing for a shock
S occurring in a given area e in the year t, the risk exposure index takes the following form:
∑𝑡−2
1 𝑆𝑒 𝑡−2
𝑃𝑖𝑒𝑡 = (2)
(𝑡−2)−1983

Where the probability of experiencing a shock for household i, residing in area e, at time t, is
computed as the ratio between the total number of seasons in which the shock occurred, divided
by the total number of seasons available in our weather source. Drought and flood indicators
are lagged by two years to exclude any recent shocks likely affecting household’s adoption of a
cropping system through the production channel, condition that would violate the exclusion
restriction. Following the work from Manda et al. (2016), the remaining two instruments are
proxies of information availability at village/district level, under the assumption that agricultural
information can be transmitted through institutional channels, such as extension services and
informal institutions, such as family and friends network. To guarantee the exclusion restriction,
we compute the average access to these two sources of information at a higher geographical
level than the one observed in the data, hypothesizing that the average access to these sources
is not correlated to the household’s outcome (Scognamillo, Asfaw and Ignaciuk, forthcoming).
In addition to these exogenous covariates allowing the robust identification of the model, the
matrix of control variables 𝒁𝑖𝑒𝑡 includes the determinants introduced in section 3.1.

Outcome equation
To estimate the impact of adoption on maize yield and crop income volatility, the second stage
of the multinomial treatment effect model consists in an OLS with the selectivity correction, as
follows:
′ 𝑗 𝑗
𝐸(𝑦𝑖𝑡 |𝑑𝑖𝑡 𝑥𝑖𝑡 𝑙𝑖𝑡 ) = 𝑥𝑖𝑡 𝛽 + ∑𝑗=1 𝛾𝑗𝑡 𝑓𝑖𝑗𝑡 + ∑𝑗=1 𝜆𝑗𝑡 𝑙𝑖𝑗𝑡 (3)

Where the expected outcome 𝐸(𝑦𝑖𝑡 |𝑑𝑖𝑡 𝑥𝑖𝑡 𝑙𝑖 ) depends on a set of observable determinants 𝑥𝑖𝑡
with linked parameters 𝛽, on the adoption of a given farm system 𝑓𝑖𝑗𝑡 and associated
coefficients 𝛾𝑗𝑡 . In addition to the set of variables in the first-stage, the matrix of determinants
𝑥𝑖𝑡 comprises the seasonal positive and negative SPI indexes, capturing the effect of seasonal
rainfall distribution on the productivity and crop income volatility. The latent components 𝑙𝑖𝑗𝑡
suggest that unobserved characteristics are explaining variation both in the selection and
outcome (Deb and Trivedi, 2006). The latent factor parameters 𝜆𝑗𝑡 allow for correlation between
the selection and the outcome equation (Deb and Trivedi, 2006). Where the main dependent
variables vary according to the objective of the analysis.

47
To investigate the effect on maize productivity is the natural log of maize yield. To address the
effect of farm systems on crop income volatility, the empirical framework takes advantage from
the work developed by Antle (1983) and Di Falco and Chavas (2009), based on two steps. First,
the approach predicts the error term from a regression with the crop gross income as dependent
variable. Secondly, it involves the addition of the square of the error term as new dependent
variable in the main empirical specification, keeping fixed the controls. This approach provides
a consistent indication of the production uncertainty measured as volatility of the error, assuming
no omitted variable bias. As the resulting specification is likely to exhibit heteroskedasticity, the
final regression is weighted using the inverse of the variance of the error terms (Di Falco and
Chavas, 2009).13

13 This procedure relies on the assumption of no measurement error. To exclude that large measurement errors will
affect the estimation on volatility, we have imputed outliers using the median value in each country.

48
FAO AGRICULTURAL DEVELOPMENT ECONOMICS
WORKING PAPERS
This series is produced by the Agricultural Development Economics Division (ESA) of the Food and Agriculture
Organization of the United Nations (FAO) since 2001 to share findings from research produced by the Division
and elicit feedback for the authors.
It covers different thematic areas, such as food security and nutrition global trends and governance; food security
and resilience; sustainable markets, agribusinesses and rural transformations; and climate-smart agriculture.
The analysis and conclusions are those of the authors and do not indicate concurrence by FAO.

The complete series is available at:


www.fao.org/economic/esa/working-papers

The Agricultural Development Economics Division (ESA) is the focal point for FAO’s research and policy analysis on
agricultural and economic development. The Division produces evidence-based policy analysis and strengthens
the capacity of member countries to improve decision-making on food security and nutrition, resilience,
climate-smart agriculture, sustainable markets, agribusinesses and rural transformations.

CONTACTS
Agricultural Development Economics Division (ESA) ISBN 978-92-5-130981-0

Food and Agriculture Organization of the United Nations (FAO)


www.fao.org/economic/agricultural-development-economics
www.fao.org/economic/esa 9 7 8 9 2 5 1 3 0 9 8 1 0
CA1562EN/1/09.18
ESA-Director@fao.org

You might also like