DETERMINANTS OF CROP DIVERSIFICATION AMONG MOZAMBICAN
SMALLHOLDERS: EVIDENCE FROM HOUSEHOLD PANEL DATA
By
Eleanor Catherine Turner

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food, and Resource Economics – Master of Science

2014

ABSTRACT

DETERMINANTS OF CROP DIVERSIFICATION AMONG MOZAMBICAN
SMALLHOLDERS: EVIDENCE FROM HOUSEHOLD PANEL DATA

By
Eleanor Catherine Turner

More than half of Mozambique’s population lives below the poverty line and
more than a third are malnourished. Poverty and hunger are intrinsically linked to
agriculture in Mozambique, a sector dominated by nearly four million smallholder
families primarily growing food for themselves. As its government and international
donors turn their focus to improving smallholder agriculture and nutrition, it is crucial to
understand those farmers’ behaviors and how they make production decisions. One such
decision—how to allocate land between different crops—changes frequently between
years, and there is no evidence as to what drives the changes in that decision. I use
household panel data collected in 2008 and 2011 to investigate the determinants of crop
diversification. Using fixed effects, I eliminate unobservable village level factors and
isolate the change in diversification from year to year. I find that expected crop prices,
access to roads and mobile networks, household and farm size are all significant
determinants of household level diversity. I employ a two-stage decision model using
correlated random effects to explore the recent upsurge in pigeon pea cultivation, finding
market prices to be significant predictors of a farmer’s decision to plant pigeon peas,
while the presence of communication infrastructure in a village increase the amount of
land allocated to pigeon peas.

ACKNOWLEDGEMENTS
This research was made possible by the Mozambican Ministry of Agriculture
(MINAG), the United States Agency for International Development (USAID),
Technoserve, and the Michigan State University (MSU) Food Security Group. I am
deeply indebted to the Mozambique research team for making these data available to me.
I would specifically like to thank David Mather and Benedito Cunguara for their help
with the data; but I am also aware that it is a result of years of work by enumerators and
analysts in Mozambique. I would also like to thank my advising committee—Cynthia
Donovan, Eric Crawford, Andrew Dillon, and Robert Richardson—for all of their help
and support during this process.
My work is largely based on the work of Samuel Benin, Melinda Smale, John
Pender, Berhanu Gebremedhin, and Simeon Ehui in Ethiopia in 2004; without their
model of crop diversification, mine would not exist. Finally, I would like to acknowledge
Bill Burke, MSU AFRE alum and author of the program craggit used in my second
model, and Valentin Verdier, MSU econometrics expert, without whom I could never
have adapted the craggit Stata code for my model.

iii

TABLE OF CONTENTS

LIST OF TABLES

vi

LIST OF FIGURES

vii

INTRODUCTION

1

CHAPTER ONE: MOZAMBICAN CONTEXT
1.1 Agriculture
1.1.1 Crop Diversification
1.1.2 Cash Crops
1.1.3 Perennial Crops
1.2 Diet and Nutrition
1.3 Infrastructure
1.3.1 Road Infrastructure
1.3.2 Telecommunications
1.4 Data
1.4.1 Survey
1.4.2 Partial Panel
1.4.3 Panel Attrition
1.5 Research Questions

8
9
10
15
17
18
19
19
20
22
22
24
24
25

CHAPTER TWO: LITERATURE REVIEW
2.1 Agricultural Household Decision Models
2.1.1 Crop Choices
2.2 Markets, Infrastructure and Agricultural Household Decisions
2.3 Determinants of Crop Diversification

27
27
29
29
32

CHAPTER THREE: EMPIRICAL MODEL
3.1 Agricultural Household Model
3.1.1 Single Crop Model
3.1.2 Multiple Crop Model
3.2 Dependent Variables
3.2.1 Aggregate measures of farm diversification
3.2.2 Individual crop analysis
3.3 Estimation Strategy
3.3.1 Indices of Diversity: Fixed Effects Model
3.3.2 Pigeon Pea Cultivation and Land Allocation: Two Stage Decision Model
3.4 Explanatory Variables and Hypotheses
3.4.1 Market Prices
3.4.2 Market Constraints
3.4.3 Household and farm-level characteristics

35
35
35
36
38
38
40
40
40
41
47
47
49
51

CHAPTER FOUR: RESULTS AND DISCUSSION
4.1 Determinants of Crop Diversification
4.1.1 Infrastructure
4.1.2 Prices

53
53
53
57

iv

4.1.3 Household and Farm Characteristics
4.2 Pigeon Pea Two Stage Model
4.2.1 Infrastructure and prices
4.2.2 Household and Farm Characteristics
4.3 Conclusions

58
59
59
60
61

APPENDIX

68

REFERENCES

71

v

LIST OF TABLES

Table 1. Mozambique at a Glance

8

Table 2. Household Land Allocation by Crop

11

Table 3. Household Food and Cash Crops in Mozambique

12

Table 4. Common Crop Portfolios by Province

14

Table 5. Household Food and Cash Crops in Mozambique

17

Table 6. Anthropometric and Nutrition Indicators, Mozambique 2008

18

Table 7. Panel Attrition Summary Statistics, 2008 Mean

25

Table 8: Selected Results from the Literature

34

Table 9. Dependent Variables used in Analysis of Aggregate Crop Diversity

39

Table 10. Summary Statistics, Explanatory Variables

48

Table 11. Correlations between Infrastructure and Diversification

49

Table 12. Household Fixed Effects Regression Results, 2008 – 2011

54

Table 13. Household Two Stage Regression Results, 2008 – 2011

61

Table 14. Average Household Land Allocated to Crops by Province

69

Table 15. Household FE Regression Results with Infrastructure Index

70

vi

LIST OF FIGURES

Figure 1. Motivations for Diversification

2

Figure 2. Maize Prices in Mozambique, 2007-2012

5

Figure 3. Average Land Allocation by Province, 2008

13

Figure 4. Mozambique Partial Panel Survey Coverage Map

23

vii

INTRODUCTION

Almost one billion people in the world currently suffer from chronic hunger, and as
population growth continues to exceed food production, more and more families will find
themselves without enough food in the future (WFP 2013). This issue is especially
prominent in rural areas in developing countries; as such, international development
organizations increasingly fund programs that target improving agricultural productivity
and nutritional outcomes in rural areas.
Paramount to solving this problem is understanding the behaviors of poor
households. In an agrarian society such as rural Mozambique, a family’s access to food is
largely dependent on what that household grows, either because they consume what they
grow, or they purchase food with the income earned from what they grow. Physical
access to food became a key constraint to food security following the destruction of the
country’s infrastructure during the civil war. Therefore, to reduce hunger in this scenario,
it is essential to improve our understanding of a farmer’s decision of which crops to
cultivate. This research will focus on factors relating to that decision.
In this paper, I will refer to this decision—how a farmer allocates his or her land
among different crops—as diversification. Diversification can range from complete
specialization, in which a household grows only one crop and sells it to earn income, to a
high level of diversity with many crops more evenly distributed over the farm.
Alternatively, complete specialization could be indicative of an extreme food insecure
scenario in which a family focuses on just one staple to ensure a minimum level of food

1

from their own production. On-farm diversity is thus a spectrum, and is neither
universally positive nor negative. As Figure 1 shows, farmers may engage in different
levels of diversification for different reasons, and outcomes will depend on the unique
situation of each farm. For example, subsistence farmers may diversify to protect
themselves from risk. By growing some of everything that they need to eat, a family
gains some assurance that they will be able to eat even in the event of crop failure, food
price shocks, or lack of food availability in local markets. Others may diversify for
income purposes; while there is no direct link proven between levels of diversification
and income, some farmers may add new crops to their mix with the hope of earning more
from an emerging or more marketable crop. Alternatively, farmers may choose a low
level of diversity (specialize in just one marketable crop) for the same reason.
Figure 1. Motivations for Diversification
Vulnerable to food insecurity;
Focus on one crop to ensure
basic calorie needs

Reliant on own harvest for
diet diversity; Grow multiple
Subsistence Oriented crops to balance diet, insure
against risk

High Specialization
Uneven portfolio

High Diversity
Even portfolio

Able to rely on market for food;
Market Oriented
Specialize in one crop for income
generation

Grow new or different crops
to sell some in the market;
consume some; insure
against risk

For interpretation of the references to color in this and all other figures, the reader is
referred to the electronic version of this thesis.

2

This thesis examines farmers’ behavior related to land allocation decisions in
rural Mozambique, using panel household survey data collected in a Ministry of
Agriculture (MINAG) and Michigan State University (MSU) survey of 1,186 households
in the northern and central regions in 2008 and 2011. These regions were selected due to
their relatively higher agricultural potential and productivity. The outcome is important as
it has implications for total agricultural productivity, household food security and
nutrition, and labor availability. For example, if a household grows only one or two crops
and is dependent on its own harvest for sustenance, its members are unlikely to consume
the full variety of nutrients they need. This is the case in most of rural Mozambique
where low-protein staples such as maize and cassava constitute most of the diet.
Alternatively, if a household grows a wide range of different crops that requires more
labor, there may be less labor available for work in commercial agriculture or rural nonfarm employment. Furthermore, understanding smallholders’ behavior and risk attitudes
related to agricultural production decisions will be useful for other potential policies or
interventions that target smallholders such as technology adoption and insurance or credit
markets. A farmer’s decision to adopt a new crop or try a new portfolio is dependent on
his or her risk preferences, and therefore similar to the decision to adopt, for example, a
new higher yielding hybrid seed or purchase crop insurance, should those options become
available in the future.
Mozambique is a fascinating place to study land allocation behavior because
Mozambican smallholders regularly engage in the reallocation of their land to different
crops. In some countries, land allocation decisions are a function of tradition; farmers
grow maize because their fathers and neighbors only grew maize, and so on. However, in
3

Mozambique diversification is much more dynamic, so it is evident farmers are not just
stuck in their ways. The average number of crops grown per farm doubled between 1995
and 2003 (World Bank 2006) and during the three-year duration of this study, 82% of
households surveyed changed the number of annual crops they grew. Additionally,
Mozambique experienced a dramatic change in diversification into pigeon peas; between
2008 and 2011 average land allocated to pigeon peas increased by over 50 percent, and
the percentage of farmers growing them nearly doubled.
There is no observable causality associated with this behavior so it remains to be
understood why farmers decide to add and remove different crops from their mixes each
year. In past studies from other developing countries, researchers have found that farm
size and fragmentation, market prices, productive capacity (labor and capital), farmers’
age and education levels, and extension visits are all associated with higher levels of
diversification. However, none of these studies has investigated changes by the same
household across time. To understand why a household grows three crops one year and
eight the next, as in Mozambique, we must have panel data on the same household in
multiple years.
Because of the year-to-year nature of the land allocation changes, it follows that
external or environmental factors—rather than characteristics of the farm itself—may be
driving them. Two significant such developments affected the lives of Mozambican
smallholders during our panel study. First, Mozambique was hit hard by the global food
price crisis. Figure 2 shows the volatility of maize prices in Mozambique beginning in
2008. As most families are net buyers of staple foods, this left many unable to purchase

4

sufficient food. This volatile price environment could influence a farmer’s crop
diversification decision as it relates to protecting his or her family from risk.

Apr-12

Sep-11

Feb-11

Jul-10

Dec-09

May-09

Oct-08

Mar-08

Aug-07

Jan-07

Metical (MZN)/kg

Figure 2. Monthly Retail Maize Prices, Nominal, Selected
Markets in Mozambique, 2007-2012.
18
16
14
12
10
Manica
8
6
Chimoio
4
Tete
2
0

Source: SIMA (Sistema De Informação De Mercados Agrícolas
De Moçambique/Agricultural Market Information System)
Second, infrastructure has rapidly developed in Mozambique since 2008. The
government has spent over $350 million on road improvements, and cell phone coverage
has expanded dramatically. Smallholders in previously isolated areas are becoming better
connected to markets for inputs, information, and sales. It is unclear how this could affect
land allocation decisions. Previous studies have shown a positive association between
physical market isolation and overall levels of farm diversity, while access to market
information is also associated with higher diversity. As one might expect, farms lacking
access to transport infrastructure do not allocate land to marketable or cash crops.
Within the context of this environment, I seek to answer the following questions:
What are the determinants of household-level crop diversification in Mozambique?

5

Specifically, what drives diversification into an emerging, mixed-purpose crop such as
pigeon peas? Is improved infrastructure associated with diversification levels?
Diversity is inherently difficult to evaluate because it can be measured in so many
ways, such as total quantity of crops grown, evenness of distribution, the ratio of cash to
subsistence crop area, or myriad other combinations or ratios of individual crops or crop
categories. I select three different indices of overall farm diversity from the literature for
my analysis, and then do a crop-specific investigation into the growth of pigeon pea
cultivation in recent years.
Several studies have shown associations between crop diversification and
infrastructure (Ibrahim et al. 2009; Bittinger 2010; Mesfin et al. 2011; Rahman 2008).
However, none measure a household’s change in diversification level and therefore fail
to account for the possibility that infrastructure—such as paved roads—actually follows
agricultural productivity, in which case the error term of the estimation would be
correlated with the infrastructure variables themselves. I fill this gap by using villagelevel fixed effects to correct for the potential endogeneity of this relationship. In so doing,
I find that that expected crop prices are the strongest determinants of land allocation
decisions. Infrastructure, however, is not insignificant when price variation and villagelevel fixed effects are controlled for; households far from paved roads are likely to grow
fewer crops, and those with year round roads and mobile networks have more evenly
distributed crop portfolios. Using a two stage model for my crop-specific research, I find
no infrastructure variables to affect the probability of deciding to cultivate pigeon peas,
but both mobile and radio coverage are associated with higher land allocated to pigeon
peas, conditional on a farmer’s decision to cultivate them.
6

The research is presented as follows: in Chapter One, I provide context for rural
agriculture and nutrition in Mozambique with specific focus on recent infrastructure
developments. I also provide an overview of the data used for the research. In Chapter
Two, I review relevant literature on agricultural household decision-making and
determinants of crop diversification. In Chapter Three, I present my models and
estimation strategy. In Chapter Four, I discuss the results of my estimation. Finally, I
conclude with brief comments on the implications of my study and suggestions for future
research.

7

CHAPTER ONE: MOZAMBICAN CONTEXT

Mozambique’s current economic and social conditions are largely influenced by its
history of continuous conflict, first for liberation from the Portuguese and then a brutal
civil war that killed over a million people (Cunguara et al. 2011). Since the peace
agreement was signed in 1992, the country has remained a peaceful democracy and
experienced impressive growth. Mozambique is well endowed with natural resources, has
an abundance of unused land and water, and has an advantageous coastal location with
major trade corridors and ports (USG 2011). Table 1 provides an overview of key
demographic and economic figures. In the past 15 years the economy has grown 7.7
percent per year, driven primarily by the service sector, light industry, and agriculture,
but it remains one of the poorest countries in the world (Dominguez-Torres and BriceñoGarmendia 2011).
Table 1. Mozambique at a Glance
Total Population
24,096,669
Climate
Tropical to subtropical
Population Under 14
45.5%
Population Growth Rate
2.44%
Urban Population (2011)
31.2%
Infant Mortality Rate
74.63 deaths/1,000 live births
Life Expectancy at Birth
52.29 years
Fertility Rate
5.34 children born/woman
Access to improved drinking water (2010)
77% urban/29% rural
Access to improved sanitation (2010)
38% urban/5% rural
Literacy Rate (2010)
56.1% (>15 years old)
Average schooling (Primary – Tertiary, 2011)
10 years
Gross Domestic Product (GDP) (Purchasing
$26.69 billion/ $14.6 billion
Power Parity/Official Exchange Rate, 2012)
GDP Growth Rate (2012)
7.5%
GDP—service sector (2012)
46.5%
GDP—industry sector (2012)
23.9%
GDP—agriculture sector (2012)
29.5%
Source: CIA World Factbook. Figures are 2013 unless specified.
8

Despite its progress over the past two decades, 54 percent of Mozambicans are
below the poverty line (Dominguez-Torres and Briceño-Garmendia 2011) and 46 percent
of children under five years old are chronically malnourished (Cunguara et al. 2011).
Access to basic infrastructure services such as water and sanitation are below average for
the region (Dominguez-Torres and Briceño-Garmendia 2011). This poverty is
intrinsically linked to agriculture; the sector employs most Mozambicans, yet the country
is still dependent on its neighbors for food imports, leaving families vulnerable to food
price volatility and often unable to meet nutritional needs. In this chapter, I provide an
overview of Mozambique today, specifically its agriculture and infrastructure, to give
context for my research.
1.1 Agriculture
Agriculture is the livelihood of most Mozambicans. The sector employs 80 percent of the
population and constitutes 29 percent of GDP (CIA 2013). Mozambican agriculture is
dominated by almost four million smallholder families; the majority grows food crops,
two-thirds of which is for home consumption, and 16 percent produce cash crops such as
cotton and tobacco (World Bank 2006). Crop income accounted for 73 percent of rural
income in 2002 (Mather 2012). Despite its relative importance, agriculture has not
received sufficient investment from the government and has not improved with the pace
of population and GDP growth (Cunguara et al. 2011).
Agricultural productivity has not increased in Mozambique (Cunguara et al. 2011),
despite overall agricultural growth averaging 8 percent (USG 2011). On a per capita basis,
most food production has shown negative growth. Cereal production per capita has
declined, leaving Mozambicans dependent on cereal imports and therefore more
9

susceptible to global price shocks. This lack of improvement in agriculture is mainly a
result of lack of access to improved technologies, markets and services (Cunguara et al.
2011). Very few smallholders in Mozambique use fertilizer, pesticides, or irrigation, and
most use reserved seeds from the previous year (World Bank 2006).
In addition to low productivity, agriculture in Mozambique is also characterized
by low market participation. Rural smallholders are generally subsistence-oriented,
although when pressed and need cash for health or other spending, they turn to the market.
Preliminary findings from the MSU/MINAG 2011 survey indicate that market
participation rates have increased since international food prices began soaring in 2008
(Benfica and Tschirley 2012).
1.1.1

Crop Diversification

Maize dominates the smallholder crop portfolio in Mozambique, particularly in the
central and northern provinces surveyed by MSU/MINAG. Almost every household
surveyed grew maize in 2011 and on average households allocate 40 percent of their
farmland to it1. The most important food crops in Mozambique are maize and cassava;
more than 50 percent of cultivated land is devoted to these low value crops to ensure
household food security (World Bank 2006). Table 2 shows the average share of land
allocated to each crop at a household level in 2008 and 2011.

1

Farmland refers to land cultivated with annual crops. See Section 1.1.3.
10

Table 2. Household Land Allocation by Crop
Crop
Mean Share of Farmland Allocated
2008
2011
Maize
44.0%
41.2%
Cassava
10.3%
11.0%
Sorghum
8.7%
6.6%
Horticulture
5.2%
5.6%
Cowpea
4.8%
4.3%
Rice
3.7%
4.3%
Common Bean
3.4%
3.5%
Pigeon Pea
3.3%
5.4%
Groundnut (small)
3.1%
3.2%
Sesame
3.1%
2.4%
Groundnut (large)
2.4%
2.3%
Tobacco
1.6%
1.7%
Cotton
1.4%
1.6%
Non-Orange-Fleshed Sweet Potato
0.9%
1.3%
Jugo Bean
0.8%
0.9%
Potato
0.8%
1.0%
Millet
0.7%
0.4%
Soybean
0.6%
0.9%
Sugar
0.5%
0.7%
Sunflower
0.3%
0.4%
Orange-Fleshed Sweet Potato
0.2%
0.5%
N=1,175. Percentage of every household surveyed, including those not
cultivating the crop. Source: MSU/MINAG Survey 2008 & 2011.
The World Bank (2006) finds evidence that crop diversification may be used as a
coping mechanism for persistent low income and productivity and the high risk
environment of farming in Mozambique; the average number of household crops nearly
doubled from 5 to 9 between 1995 to 2003, but this did not significantly affect farm
income.
Our survey shows a trend of increasing the total number of crops cultivated,
however many households also decreased as well. The two other indices used to measure
diversity (described in Chapter Two) show no clear trend; indeed the majority of farmers
11

changed their portfolios between 2008 and 2011 though the average diversification levels
remained very similar.
Figure 3 and Tables 3 and 4 provide a snapshot of cropping patterns by province.
Table 3. Average Number of Crops Cultivated per Household, by Province
2008
2011
Nampula
3.6
4.1
Zambezia
4.1
4.6
Tete
4.6
4.4
Manica
3.4
4.2
Sofala
4.2
6.0
Source: MSU/MINAG Survey. N=1,186.
In Nampula, the northernmost province surveyed, cassava is the dominant crop in terms
of land allocation, and households grow fewer total crops. Most households grow cassava
and maize, as well as some integration of sesame, a cash crop (see Table 4, and next
section for a description of cash crops.) Nampula also has the highest rates of child
malnutrition in the country (FAO 2011). Maize is the dominant crop in the other four
provinces studied. In Manica and Sofala, the southernmost provinces studied, most
households grow a combination of maize and sorghum, and approximately 15 percent
and 40 percent grow sesame, respectively. Pigeon peas are most frequently grown in
Zambezia, where they commonly appear as one of the top three crops cultivated by
households, along with maize, cassava, and sorghum. In Tete, the westernmost province,
surrounded primarily by Zambia, Malawi, and Zimbabwe, maize dominates but sorghum
and cassava are not common; instead households grow common beans and horticulture,
and some produce tobacco, typically grown under contract as described in the following
section.

12

Figure 3. Average Land Allocated to Common Crops by Province, 2008
Zambezia

Nampula

Sorghum
6%

Other
19%

Sorghum
7%

Cassava
32%

Other Cassava
16%
17%

Rice
6%

Cowpea
7%

Maize
27%

Small
Groundnut
9%

Pigeon
Pea
13%

N=198

Other
13%
Tobacco
6%

Cowpea
5%

3%

Sofala

Other
20%
Maize
47%

Cowpea
5%

Maize
59%

Horticulture
7% Sesame

N=248
Tete, 2008

Manica, 2008

Sorghum
13%

Maize
41%

Common
Bean
Horticulture
N=210 11%
11%

N=254

Other
17%
Sorghum
16%

Horticulture
5%

Maize
44%

Source: TIA 2008. Information
on all crops and years available
in the appendix.

Rice
9%
Sesame
9%

N=265
13

Table 4. Common Crop Portfolios by Province
2008
2011
Nampula (n=200)
Cassava, Maize
10%
5%
Cassava, Maize, Small Groundnuts
7%
7%
Cassava, Maize, Cowpea
6%
12%
Cassava, Maize, Sesame
5%
1%
Cassava, Maize, Sorghum
5%
11%
Cassava
4%
2%
Other
65%
63%
Zambezia (n=252)
Cassava, Maize, Pigeon Pea
17%
23%
Maize, Pigeon Pea
6%
6%
Cassava, Maize, Sorghum
6%
6%
Cassava, Maize, Rice
5%
6%
Maize, Pigeon Pea, Sorghum
4%
4%
Other
62%
55%
Tete (n=256)
Maize, Horticulture, Common Bean
11%
6%
Maize, Horticulture, Large Groundnut
10%
3%
Maize, Common Bean, Tobacco
8%
7%
Maize, Horticulture, Cowpea
6%
3%
Maize, Common Bean, Large Groundnut
5%
7%
Other
62%
74%
Manica (n=209)
Maize, Sorghum, Horticulture
11%
9%
Maize, Sorghum
11%
3%
Maize, Sorghum, Cowpea
8%
4%
Maize
6%
3%
Maize, Sorghum, Sesame
5%
5%
Maize, Cassava, Horticulture
2%
8%
Maize, Cassava, Sorghum
5%
6%
Other
53%
62%
Sofala (n=268)
Maize, Sorghum, Sesame
11%
3%
Maize, Sorghum
9%
1%
Maize, Sorghum, Horticulture
7%
8%
Maize, Sorghum, Rice
7%
1%
Maize, Sesame, Horticulture
4%
4%
Maize, Sorghum, Cassava
4%
6%
Maize, Rice, Sesame
3%
6%
Other
54%
70%
Percentages reflect the number of households for whom the crops listed are
the three crops using the most land. Households may grow more than those
crops, except in the cases when less than three are listed. Source:
MSU/MINAG Survey 2008 & 2011.
14

1.1.2 Cash Crops

While food security and risk management often drive crop portfolio diversification,
market and earnings potential could also drive diversification from subsistence into cash
crops. For the purpose of this analysis, I define cash crops as those marketed by more
than 75 percent of the farmers cultivating them; the World Bank (2006) identifies them as
cotton, tobacco, cashew, sugar, and tea. Only five percent of cultivated land in
Mozambique is devoted to cash crops (The World Bank 2006). Tea and sugar are
typically grown on plantations and therefore not commonly grown by smallholders on
their own plots. Households that diversify into cotton and tobacco do so under a
contracting arrangement with large firms that also provide inputs and extension advice.
This has increased the number of households participating in cash crops, but is limited by
access to such arrangements. Cashew is a cash crop traditionally grown by smallholders,
but trees tend to suffer from age (most were planted in the early colonial era) and disease,
lowering yield and quality (World Bank 2006).
Although there does not exist much opportunity for the small farmer to diversify
into these primary cash crops without the presence of a cotton or tobacco contractor, in
the past decade Mozambique has experienced growth in the market potential of several
legumes and oilseeeds that are relatively more feasible for a smallholder to adopt than the
plantation crops mentioned above, but still have market potential similar to cash crops.
This is due to a combination of demand and donor support. For example, Mozambican
sesame exports increased from 1,500 to 36,000 tons between 1998 and 2009 with United
States Agency for International Development (USAID) assistance and growth of private
15

sector processing (USG 2011). Smallholder production of soybean has also increased
recently to meet the demands of the growing poultry industry, with assistance from the
Bill and Melinda Gates Foundation (USG 2011). While sesame and soy are produced
primarily to sell, other crops, such as pigeon peas, are increasingly grown by
smallholders for their market potential and for household food security. In addition,
intercropping with legumes such as pigeon peas can improve soil quality and cereal
yields, which can further improve a family’s income, particularly in a case of little to no
fertilizer use.
Unlike soybeans, which were recently introduced to Mozambican smallholders
and are grown almost exclusively as a cash crop, pigeon peas are not new and have been
grown as a subsistence crop for decades. However, their market potential as a cash crop
is emerging which may be the reason for the substantial recent increase in cultivation (see
Table 5); Mozambique’s exports of dried legumes increased from 8,709 metric tons (MT)
in 2007 to 128,127 MT in 2011 (UN Comtrade 2013). It is evident that there is a market
for oilseeds and legumes; in my analysis, I will investigate the characteristics of
smallholders that are taking advantage of the growing market for one such legume, the
pigeon pea.

16

Table 5. Household Food and Cash Crops in Mozambique
%
% Growers
%
% Growers
Growing
Selling
Growing
Selling
2008
2011
Sorghum
37
7
35
6
Jugo Bean
9
10
10
6
OFSP
3
11
8
14
Subsistence
Rice
14
14
14
14
Crops
Cowpea
37
14
43
17
Cassava
41
20
58
20
Non-OF SP
13
22
18
25
Groundnut Sm
20
25
26
20
MixedGroundnut Lg
17
27
19
27
Purpose
Maize
93
32
95
38
Crops
Pigeon Pea
20
32
36
44
Common Bean
23
37
26
44
Sesame
16
73
16
92
Tobacco
6
75
7
84
Cash Crops
Sunflower
3
84
3
93
Soy
4
86
5
77
Cotton
5
93
6
86
N=1,186. Source: MSU/MINAG Survey, 2008 & 2011

1.1.3 Perennial Crops
Smallholders in Mozambique also frequently cultivate tree crops, such as mango, coconut,
and papaya. The nature of a farmer’s decision to plant a tree crop is fundamentally
different from that of annual crops; trees can take years to bear fruit, and land allocation
decisions are not made on a “harvest season” basis. Therefore, for the purpose of
understanding the annual land allocation decisions of a farmer, I exclude orchard crops
from my analysis. When I refer to the percent or share of farmland allocated to a crop in
this paper, the total farmland cultivated constitutes only that cultivated with annual crops.

17

1.2 Diet and Nutrition
Malnutrition is prevalent in Mozambique, and can be largely attributed to lack of dietary
diversity and insufficient health services. In 2008, 44 percent of children under five
suffered chronic malnutrition (FAO 2011); Table 6 provides a summary of other nutrition
indicators for the population. Rural Mozambicans’ diets mirror their agricultural
Table 6. Anthropometric and Nutrition Indicators, Mozambique 2008
Stunting (chronic malnutrition), rural children <5
Wasting (weight-for-height), children <5
Underweight, children <5
Chronic energy deficiency, rural women 15-49
Iodine deficiency (goitre), rural children 6-12
Rural households with adequate iodine level of household salt (>15 ppm)
Vitamin A deficiency , rural children 6 months – 5 years old
Anemia/Iron Deficiency, rural children 6 months – 5 years old
Source: FAO 2011

47%
5%
18%
10%
15.7%
19.7%
73.1%
80.5%

production. In the north, the diet is primarily composed by cassava, a low-protein staple,
while households in the center rely mostly on maize. Cereals and starchy roots constitute
80 percent of the energy supply in Mozambique, making it the country with the lowest
levels of dietary diversification in the region (FAO 2011). Staples are typically consumed
in a porridge form and served with green leafy vegetables or bean stew, and seafood
when available in coastal areas. Consumption of dairy, eggs, and meat is very low in rural
areas; fruits are consumed only when in season. The per capita supply of fruits and
vegetables has decreased by half since the 1970s and is the lowest in the region. As
income increases, meat and fish substitute for greens and beans, while staple
consumption remains the same (FAO 2011).

18

1.3 Infrastructure
Lack of sufficient rural infrastructure is one of the obstacles to improving the
productivity, income, and food security of smallholders in Mozambique. Only 20 percent
of smallholders sell in the market, and only 34 percent of those receive price information
(USG 2011). Previous studies have found infrastructure to significantly impact crop
diversification (see Table 8, next chapter), and the MSU/MINAG survey shows
significant correlations between infrastructure variables and diversity. To provide context
for the analysis of this relationship, in this section I summarize the conditions and recent
development in two areas of infrastructure that may impact those statistics: roads and
telecommunication, specifically mobile phones. Both have experienced recent growth
and represent opportunities to connect rural smallholders to markets. Indeed, DominguezTorres and Briceño-Garmendia (2011) estimate that if Mozambique could improve its
infrastructure to the level of middle-income countries in the region, its growth
performance would increase by up to 2.6 percentage points per capita.
1.3.1 Road Infrastructure
Mozambique’s road quality contrasts sharply between rural and urban areas: densely
populated areas and main transport corridors enjoy good connectivity and road quality,
while rural areas suffer poor quality and low connectivity. Its classified network density
per land area is one of the lowest in the region (Dominguez-Torres and BriceñoGarmendia 2011) and only 18 percent of that network is paved, with many of those paved
roads in terrible condition (Brouwer and Brito 2012). Only one quarter of rural
Mozambicans live within two kilometers (km) of a road in the classified network, and 40

19

percent of rural roads are categorized as poor quality (Dominguez-Torres and BriceñoGarmendia 2011).
Nevertheless, roads constitute the only major area of transport improvement in
Mozambique in recent years, and such changes have reduced travel time to cities for
many (Cunguara et al. 2011). The national road connecting the north and south was
completely paved in 2009 (Brouwer and Brito 2012), though rural roads improvements
still lag behind. The Government of Mozambique spent approximately $347 million per
year on road sector investments from 2007 to 2009, and in 2011 the World Bank
extended its decade-long involvement under the Roads and Bridges Management and
Maintenance Program with an additional $30 million for the objectives of increasing “(a)
percentage of classified roads in good and fair condition; and (b) the percentage of the
rural population within two kilometers of an all- season road” (World Bank 2009, p.2).
As such, the quality of the roads is improving and the road network is expanding.
The villages in our survey reported an average decrease of 12 km to the nearest paved
road, and the percent living over 50 km from a paved road dropped from 33 to 24 (see
Table 10, Chapter 3). Dorosh et al. (2012) predict that reductions in travel time will lead
to large increases in output; I investigate whether they are associated with diversification
levels as well.
1.3.2 Telecommunications
The poor transport infrastructure makes telecommunications critical in
Mozambique. Mobile networks can connect people in greater geographic areas faster and
more cheaply than landlines or transport (Brouwer and Brito 2012). As in most of Africa,

20

the expansion of mobile networks has transformed communications in Mozambique in
the past decade. According to the International Telecommunications Union (ITU), the
number of mobile phone subscriptions per 100 inhabitants increased from 0.28 in 2000 to
32.83 in 2011 (ITU 2012). Mobile growth between 2005 and 2008 was around 40 percent
per year, and the population covered by a global system for mobile communications
(GSM) increased from 14 percent in 2000 to over 80 percent in 2008 (Dominguez-Torres
and Briceño-Garmendia 2011).
Since 2003, two mobile operators, MCel and Vodacom, have shared the market in
Mozambique, and network rollout has generally followed economic development,
beginning in Maputo and expanding main roads to provincial capitals (Brouwer and Brito
2012). A third company, Movitel, was also launched in 2012 (Zita 2012). While both
MCel and Vodacom boast high coverage rates in the northern and central provinces,
research has shown that operators tend to overestimate their number of users. Rural areas
clearly lag behind in access to network coverage. In 2007, 32 percent of the population in
Maputo (city and province) used mobile phones compared with a national average of 7.8
percent. In Zambézia, one of the provinces of focus for this study, that figure was only
1.8 percent (Brouwer and Brito 2012).
Acquiring a mobile phone in Mozambique is relatively easy since the introduction
of prepaid starter packages with SIM cards, retailing for around five dollars or less. The
primary obstacle is the handset itself, the cheapest of which retails for around $40 new.
Because of the large market of used and stolen devices, many are able to obtain a phone
more cheaply. Airtime and SMS are prepaid, unless an annual contract is signed, and
callers only pay for calls or messages that they make or send (Brouwer and Brito 2012).
21

There is evidence that information and communication technology (ICT) drives
growth; Dominguez-Torres and Briceño-Garmendia (2011) state that 0.9 of the 1.9
percentage improvement in per capita growth rates in Africa 2003-2007 were attributable
to improved infrastructure, primarily from the ICT revolution, but the effect of cell
phones on agriculture in Mozambique remains uncertain. Based on the 2007 census,
Brouwer and Brito (2012) found that only four percent of mobile phone users work in
agriculture, and the majority of those in agribusiness, not on household farms. They also
found that only five percent use the phones primarily for professional needs. It is,
however, worth noting that with such rapid expansion, mobile phone data quickly
becomes out of date. Given the system investments in coverage outside Maputo, it is
likely that much has changed since the 2007 census.
Another way in which ICT has integrated rural African farmers into markets has
been through mobile banking; M-Pesa’s widely celebrated popularity in Kenya has
revolutionized rural finance. M-Pesa is expanding across Africa and Vodacom is
launching its mobile money platform in Mozambique, which will undoubtedly change
how mobile coverage impacts rural farmers. However, for the purpose of the following
analysis—which spans 2008 to 2011—mobile money is not included as a component of
rural market infrastructure.
1.4 Data
1.4.1 Survey
This study uses a two-year panel of rural household surveys from the north and
central regions of Mozambique. The 2008 survey was conducted as part of the Trabalho

22

do Inquérito Agrícola (TIA), implemented by the Ministry of Agriculture (MINAG)
Directorate of Economics with support from MSU. The TIA 2008 sample was drawn
from the listing of the 2000 agricultural census and used a stratified, clustered sampling
design to form a nationally representative dataset. The 2011 survey was supported by
USAID and Technoserve, implemented by MSU and MINAG, and represents a follow-up
survey to TIA 2008 farmers, only in five high agricultural production provinces: Tete,
Zambézia, Manica, Nampula, and Sofala (see Figure 4.) The 2008-2011 partial panel is
discussed in the following section.
Figure 4. Mozambique Partial Panel Map

Shaded area
represents
districts
covered by
MSU/MINAG
Survey
Maputo

The household surveys cover a range of topics relating to household economic
activity and wellbeing. They capture agricultural production including costs and revenues,
livestock activities, land use, salaried-employment, and other income-generating
23

activities. Household demographic information was collected in detail, including
members’ ages, genders, and education levels and literacy. The surveys also capture
simple self-reported disease and health information. Community-level questionnaires
were also administered to village leaders. These surveys included information on village
infrastructure, access to input and output markets, climate, and local commodity prices.
1.4.2 Partial Panel
The 2011 survey was administered to only a subset of those households surveyed
in 2008. This survey selected districts in the provinces of Nampula, Zambézia, Tete,
Manica, and Sofala based on their production of maize, sesame, soybean, and sunflower.
The panel data cover 1,186 farm households; only those households surveyed in both
years are used for my analysis. The results are therefore not nationally representative, but
rather representative of small- and medium-holders in the region specified.
1.4.3 Panel Attrition
Like most longitudinal household surveys, our panel suffered some attrition
between 2008 and 2011. Eighteen percent of the households in the 2011 selected districts
that were originally surveyed in 2008 did not complete the interviews. This rate is
approximately what is expected for rural household surveys in sub-Saharan Africa;
households move, dissolve, or become unreachable over time. In some specific cases,
village chiefs declined to participate in the follow up due to conflicts resulting from
cholera riots. Refusal is rarely a bias for attrition in this context. Attrition can bias results
if it is not random; for example, if households dissolve because the head dies, attrition
could be strongly correlated with other characteristics, such as family health status or the
age of the head, which may also be correlated with farming behaviors. Summary statistics
24

for the attritted households (see Table 7) show that wealthier or more advanced farms, as
proxied by ownership of livestock, use of fertilizer, cropped area of maize, and total farm
size, were less likely to be interviewed in 2011. To minimize potential attrition bias, I use
inverse probability weighting (Wooldridge 2010); this predicts a household’s probability
of completing a second interview in 2011 based on a set of characteristics. Mather and
Donovan (2007) tested this method for the TIA 2002/2005 panel and confirmed that
many observable characteristics were significant predictors of the probability of reinterview, concluding that inverse probability weighting, when combined with population
sample weights, is a reasonable approach for minimizing attrition bias.
Table 7. Panel Attrition Summary Statistics, 2008 Mean
Households Reinterviewed in 2011
Household head female (1=yes)
0.21
Household head’s age
39.42
Maximum education in household
3.93
Household size
4.80
Cropped area of maize
0.75
Cropped area of pigeon peas
0.09
Total household landholding
2.18
Household has latrine
0.57
Household has oil lantern
0.41
Tropical livestock units (TLU)
1.57
Household uses fertilizer on any crop (1=yes)
0.02
Village has electricity (1=yes)
0.20
Village has mobile network (1=yes)
0.79
Number of Households
1,186
Source: TIA 2008.

Households
Attritted in 2011
0.17
41.83
4.40
5.48
1.13
0.06
2.81
0.66
0.43
2.73
0.10
0.14
0.80
268

1.5 Research Questions
Between 2008 and 2011, Mozambique experienced improvements in road quality,
expansion of paved road networks, and a rapid increase in mobile network coverage and
mobile phone users. Over the same period, smallholders in the northern and central
25

provinces changed their land allocation strategies and overall levels of on-farm crop
diversity. However, we do not know whether the infrastructure itself has affected
cropping practices. I hypothesize that infrastructure may be associated with crop
diversification patterns via three pathways: 1) price changes, by decreasing transportation
and/or transaction costs, 2) access to information, such as market prices, crop information,
or agricultural extension, and 3) access to inputs, such as seeds for crops previously
unavailable in isolated villages. The survey data indicate evidence for improved access to
information; the number of households receiving price information increased from 37
percent in 2008 to 57 percent in 2011 and is significantly correlated with radio and
mobile coverage.
My paper will seek to answer the following research questions: What are the
determinants of household crop diversification in Mozambique? More specifically, is
improved infrastructure associated with changes in crop diversification patterns?
Additionally, I will do a crop-specific investigation of which factors determine a
household’s probability of cultivating pigeon peas, an emerging mixed-use crop, and
what factors are associated with the amount of land allocated to cultivating them. In the
following section, I review literature that has investigated these and similar questions
around farm household decision-making.

26

CHAPTER TWO: LITERATURE REVIEW
The body of literature on crop diversification has grown over the past decade; however it
builds on literature about agricultural household theory and decision-making that has
formed the basis of most studies and policy decisions relating to small farmers in
developing countries in the last half century. In the following sections, I trace the origin
of the agricultural household model in the literature and its extensions into farm decisionmaking, particularly with respect to crop choice decisions. I then summarize literature
relating infrastructure and market access to household decisions, and conclude with a
review of empirical works that have investigated the determinants of crop diversification.
2.1 Agricultural Household Decision Models
The agricultural household requires a unique theoretical model because it combines
producer and consumer theory. The farm simultaneously behaves both as a profitmaximizing firm and a utility-maximizing household. Farm profits include explicit
profits from selling products in the market and implicit profits of consuming some of
some of those products, while consumption includes that of both purchased and selfproduced goods (Taylor and Adelman 2003). Singh et al. (1986) formalized the
agricultural household model, which I use as the basis of my own model in Chapter Three,
to understand agricultural household behavior and its implications for policy. They noted
that any exogenous shock such as a price policy or change in the market would
simultaneously affect production, consumption, and labor supply, and therefore any
attempt to estimate the impact of such a shock must recognize the interdependence of
these components.
27

This integration of producer and consumer theory explained two fundamental
paradoxes experienced in developing agricultural areas: first, the positive own price
elasticity of demand for food in farm households, and second, the lower-than-expected
supply responses to food price changes (Taylor and Adelman 2003). Taylor and
Adelman (2003) credit agricultural household models as a fundamental component of
research on developing countries and the “building block of economy-wide models,” but
also recognize some limitations. For example, such models assume preferences and
income are shared by household members (though alternatives to the unitary decisionmaking model have been developed and are becoming more commonly used in research),
and their micro-focus can fail to capture indirect or linkage effects between households
and in villages. Additionally, de Janvry et al. (1991) note that agricultural household
models often postulate the existence of markets for all of a farm’s choice variables. This
is explored further in section 2.2.
One component of the agricultural household model that is crucial to the
discussion of crop diversification is uncertainty and risk aversion. A farm household’s
expected utility is dependent on its attitude toward risk. Even in a one-season model, a
farm household’s utility is subject to uncertainty in, inter alia, levels of rainfall, output
prices, and consumption prices. A farmer’s production decisions—including optimal crop
allocation—are therefore dependent on that farmer’s attitude toward risk (Fafchamps
1992) as well as the presence of markets for risk (de Janvry et al. 1991). For example,
markets for price information or crop insurance would decrease a farmer’s perceived
level of risk, affecting crop allocation decisions. The literature suggests that farmers in
developing countries tend to be risk averse and crop diversification may be a strategy to
28

insure against production and price risk. Indeed, Bezabih and Sarr (2012) find that both
rainfall risk and individual risk aversion lead to higher levels of crop diversity among
farmers in Ethiopia.
2.1.1 Crop Choices
The agricultural household model explains farmers’ production decisions and
therefore can be extended to the decision of which crops to produce. The literature often
postulates a simple two-good production function of cash crops, which are sold in the
market, and food or staple crops, which are consumed by the household and sold in the
market if a functioning market exists (de Janvry et al. 1991; Deboer and Chandra 1978).
In the absence of a food market, a household must first be subsistent before allocating
land to a cash crop (de Janvry et al. 1991). It then follows that larger farms will allocate a
higher percentage of land to cash crops (Fafchamps 1992). Fafchamps (1992) showed
that even when food markets do exist, small farmers are unlikely to diversify into cash
crops because farmers prefer to protect themselves against food price risk by remaining
self-sufficient in staple crop production. In order for a household to diversify into cash
crops, it must feel secure in its ability to obtain staple foods elsewhere. This can be
achieved by integrating food markets by investing in roads and transportation and
removing policies that hinder trade (Mellor 1985; Fafchamps 1992).
2.2 Markets, Infrastructure and Agricultural Household Decisions
Infrastructure is widely acknowledged to be crucial for poverty alleviation (Antle
1983); basic water and sanitation infrastructure are essential for improving health,
schools are required for improving education, and transport for trade, and so forth. In the
case of agriculture, much research has been done to identify what kind of infrastructure
29

improvement best supports agricultural development. Antle (1983) used country level
data to estimate total agricultural output and found infrastructure, as measured by the
GDP of the country’s transportation and communication industries, to be a significant
contributor to agricultural production. Binswanger et al. (1993) investigated the
interrelationships of financial institutions, government investment, and agricultural output
in India and found that improved road investment, electrification, and primary education
enhanced agricultural output between 1960 and 1980.
While the above literature estimates the effects of infrastructure at a macro level,
my research will explore the relationship between infrastructure and farmer decisionmaking at a household level. The existence and quality of markets for inputs, outputs, and
information affect a household’s decision to produce and to sell certain goods. De Janvry
et al. (1991) explained the paradox of low market participation through their analysis of
market failure, namely that if the cost of market participation exceeds the value generated
by participation, farmers will consume instead of sell their goods. Poor infrastructure,
lack of information, transaction risk, and uncompetitive marketing systems decrease the
likelihood of a farmer selling in the market (de Janvry et al. 1991). This analysis was
focused on market participation given that a household was already producing a specific
good, but is easily extended to the decision to produce a crop for the purpose of selling it.
Several researchers have investigated the relationship between infrastructure—
such as transport and information—and agricultural household decision-making and
production. There is further evidence that improved road connectivity positively affects
agricultural productivity and income levels by reducing travel time and transaction costs
(Arndt et al. 2012A; Dorosh et al. 2012; Stifel and Minten 2008). To be sure, the
30

causality in this relationship could be in both directions; roads are likely placed in regions
with higher productive potential. Jacoby (2000) uses regional fixed effects to eliminate
such “unobserved regional productivity” and finds that providing road access to markets
increases plot values in rural Nepal, indicating welfare benefits for poor households.
Dercon et al. (2009) use household fixed effects and instrument for household income to
address the same problem and conclude that access to all weather roads reduces poverty
by 6.9 percent.
Information—often made available through mobile phones—can also affect farm
decisions and therefore income. Mittal et al. (2010) argue that the “next great
evolutionary step in agriculture” will require an information-based, decision-making
agricultural system and will depend on mobile-enabled information systems. The authors
categorize information required by small farmers into (A) know-how, such as crop choice
and seed varieties, (B) contextual information, such as weather and best practices for
cultivation, and (C) market information. Improved information can reduce costs and
improve returns for farmers (Muto and Yamano 2009) and mobile phones significantly
reduce the cost of obtaining that information (Aker 2011). Research in this area has
focused on the impact of access to market price information via mobile phones (Aker
2010; Bizimana et al. 2013; Jensen 2007; Muto and Yamano 2009) or the impact of
specific ICT-based extension programs (Fafchamps and Minten 2012; Mittal et al. 2010).
I am not aware of any empirical studies that have linked mobile phone coverage
specifically to crop diversification, though some test extension contact and access to
market information as potential determinants (Rahman 2008; Mesfin et al. 2011). These
works are explored further in the following section.
31

2.3 Determinants of Crop Diversification
Several recent studies have attempted to identify household and village level
characteristics that affect crop diversification. These studies range in their definitions of
diversification and in their empirical methodologies. This section will summarize the
methods used.
Immink and Alarcon (1993) first went beyond the traditional crop
commercialization studies, which examined the complete substitution between cash and
food crops, to investigate changes in crop mix patterns observed during the
commercialization of subsistence farms in Guatemala in the 1980s. Using cross-sectional
data, they classified households into four categories (traditional, potato-growing, wheatgrowing, and vegetable-growing) and estimated the probability of a farmer selecting each
portfolio, finding access to credit and geographic location to be significant determinants
of diversification out of the traditional portfolio of maize and beans and into cash crops.
More recent studies tend to use indices as measures of crop diversification. The
simplest index is a count of the number of crops or varieties cultivated (Benin et al. 2004;
Ibrahim et al. 2009; Van Dusen and Taylor 2005). This provides a general level of overall
diversity on a farm, but gives no insight as to whether the farm is growing cash or staple
crops, and what percentage of resources are allocated to which. In the common example
of a farm dominated by maize or another staple grain, a family often has a kitchen garden,
or small plot used to grow vegetables or other crops for home use, the count index would
fail to accurately capture the diversity of that farm.

32

A variety of other indices are used in the literature to address this problem. Most
commonly used are the Simpson index of proportional abundance (Ibrahim et al. 2009;
Ndhlovu 2010; Aneani et al. 2011) and the Herfindahl index of concentration (Rahman
2008; Bittinger 2010) which both employ a summation of the share of total farmland
allocated to each crop squared. Other indices employed include Margalef (Benin et al.
2004; Rahman 2008), Shannon (Benin et al. 2004; Rahman 2008), Entropy (Mesfin et al.
2011), and Berger-Parker (BP) (Benin et al. 2004), all of which use some variation of the
share of land allocated to each crop. For my analysis, I select two of the above indices
and define them in the following section.
Another common method for measuring the determinants of crop diversification
is to examine the share of total land allocated to each individual crop as its own
dependent variable. This allows for the targeted investigation of what characteristics are
associated with diversification into and expansion of that crop in particular and can be
done with independent ordinary least squares (OLS) regressions (Ndhlovu 2010) or
simultaneously via multinomial logit regression (Allen 2012). Bittinger (2010) uses a
similar method to predict the share of land allocated to a particular category of crops,
rather than an individual crop.
Although no studies have explicitly tested the relationship between mobile phones
and crop diversity, many have tested infrastructure and market access. In Table 8, I
summarize the results of these studies. Results vary widely across the literature, in part
because of the range of indices used to capture diversity and the range of measures of
infrastructure. Additionally, none of these studies addressed the potential endogeneity
between infrastructure placement and crop diversity. Roads may be better in regions with
33

higher agricultural potential and diversity, and farmers may engage in higher levels of
diversification for the same reason. As such, there is no conventional wisdom in the
literature about the relationship between infrastructure and diversification. My research
will contribute to filling this gap.

Table 8: Selected results from the literature
Study
Infrastructure Measure Diversity Measure
Benin et al.
Walking time to
Margalef Index;
2004
nearest all-weather
Shannon Index
road
Benin et al.
Distance to nearest
Margalef Index;
2004
district town
Shannon Index
Ibrahim et al.
Road network
Count
2009
condition (1=good)
Ibrahim et al.
Distance to local
Count
2009
market
Bittinger 2010
Travel time to
Share of land
population of 50,000
allocated to crop
categories

Result
Not Significant
Not Significant
Negative
Not Significant

Bittinger 2010

All-weather road
density

Share of land
allocated to crop
categories

Mesfin et al.
2011
Mesfin et al.
2011
Rahman 2008

Distance to market

Entropy Index

Positive (cereals/pulses,
fruits/vegetables)
Negative (oils/spices,
cash crops)
Positive (cash crops,
fruits/ vegetables)
Negative
(cereals/pulses,
oils/spices)
Not significant

Access to market
information
Infrastructure Index

Entropy Index

Positive

Margalef Index;
Shannon Index;
Herfindahl Index

Negative (Margalef &
Shannon)
Positive (Herfindahl)

34

CHAPTER THREE: EMPIRICAL MODEL
3.1 Agricultural Household Model
Smallholder behavior cannot be understood via producer or consumer theory
independently. The agricultural household is simultaneously both, and therefore decisions
are not as simply profit- or utility-maximizing as those theories assume. As such, the
agricultural household model was developed (Singh et al. 1986) to explain the decisionmaking of smallholder families, and has been the foundation of most such research in the
past three decades. I use this model as the theory for my analysis.
3.1.1 Single Crop Model
The agricultural household is both a producer and consumer. Its objective is to maximize
expected utility from self-produced goods, purchased goods, and leisure subject to
several constraints. Below, I develop a simplified static agricultural household model
based on the work of Singh et al. (1986):
The household is assumed to maximize a utility function U = U(Ga, Gm, Gl) where Ga is
a consumed agricultural product, Gm is a good purchased in the market, and Gl is leisure.
The household also faces three constraints:
(1) Cash income constraint:
pm Gm = pa(Qa - Ga) – pl(L – F) – pvV + E

(1)

This constraint simply explains that a household can only purchase in the market
what it earns from its production. The prices of purchased goods and produced

35

agricultural products are pm and pa are, respectively; Qa is the quantity of
agricultural products produced by the household; pl is the market wage; L is total
labor input; F is family labor input; and V and pv are the variable input and its
price, respectively. E is any income earned not via the farm or via labor.
(2) Time constraint:
Gl + F = T

(2)

where leisure plus family labor cannot exceed total time available, T.
(3) Technology constraint:
Qa = Q(L,V,A,K)

(3)

Where A is the fixed amount of land belonging to the household and K is the stock
of capital.
These three constraints can be substituted to form a single constraint:
pmGm + paGa + plGl =plT + π + E

(4)

where π represents firm profits and equals paQa (L,V,A,K) - plL – pvV.
3.1.2 Multiple Crop Model
The model above assumes that a household only produces one crop, a. To
examine how a household chooses to allocate its land between multiple crop choices, I
will use the model employed by Benin et al. (2004) in which a farmer produces a vector
of outputs, Q, using a vector of inputs, X.

36

Let the farmer’s production function for each crop j be:
Qj = f(Xjk, αj | A, Z)

(5)

A remains the fixed amount of land and Z represents farm and household characteristics.
αj is the share of A allocated to crop j. Therefore, the profit of the farm is given by the
sum of the outputs of each crop less the costs of the inputs used:

!=

!
!!! !! !!    –  

!
!!! !!

!!"

(6)

where pj is the vector of output prices and wj is the vector of input prices.
The household maximizes the expected utility of its profits, EU( π(Q, X, p, w | A, Z) and
via first order conditions defines optimal input levels:
X*jk = X*jk (pj, wk, U | A, Z)

(7)

Then, the optimal output level of each crop j depends on (X*jk ) and is defined by:
Q*j = f(X*j1… X*jk )| A, Z

(8)

To express the demand for crop diversification, D, on the farm, I then use the conceptual
form from Benin et al. (2004). They argue that prices are endogenous to the household
and depend on the cost of market transactions, and therefore exclude prices from the
model. I instead use market prices, P, which capture exogenous price variation because
they are observed at large, regional markets and therefore unlikely to be affected by the
behaviors of an individual farmer, in addition to the exogenous market characteristics, M,
37

used by Benin et al. to capture transaction costs. Then, the household’s optimal choice
can be expressed as a function of farm size, market prices, household, farm, and market
characteristics, and initial wealth, W0:
h*= h*(A, P, Z, M, W0)

(9)

Letting αj equal the share of farmland allocated to crop j so that

!
! !!

= 1, ! = 1,2 … !.

Then, a farmer’s optimal α is α*j = f(A ,P, Z, M, W0) so total farm diversification, D, can
be expressed:
D = D(αj* (A, P, Z, M W0)).

(10)

3.2 Dependent Variables
3.2.1 Aggregate measures of farm diversification
The dependent variables in my fixed-effects estimation of farm diversification
approximate the total level of diversification undertaken by a household farm. Following
my review of those indices employed in the literature in Chapter Two, I select three
measures to evaluate different aspects of total farm diversity (defined in Table 9). The
first, DC, represents the total number of different annual crops cultivated by a household
in a single year. In this measure, individual horticultural crops are counted as separate
crops. DC is a very general measure of diversity, but will capture a household’s
propensity to increase or decrease the level of on-farm diversification.

38

Table 9. Dependent variables used in analysis of aggregate crop diversity
Dependent
Definition
Explanation
Mean
Variable
2008
J = total number of annual
Crop Count
DC = J
4.04
crops cultivated
!
Herfindahl
αj = share of farmland
!! =
!!!
0.42
Index
allocated to the jth crop
!
Berger-Parker
(BP) Index

!!" = 1/max  (!! )

αj = share of farmland
allocated to the jth crop

Mean
2011

2.10

4.72
0.38
2.22

N=1,186. Source: MSU/MINAG Survey, 2008 & 2011.

The second aggregate measure is the Herfindahl Index—a widely used measure of
market concentration, in the context of industrial organization or antitrust (Herfindahl
1950)—applied to crop diversification, as defined by !! =

! !
! !! .

As in the conceptual

model outlined above, αj represents the share of total farmland allocated to crop j. The
Herfindahl index can range from zero to one, where a zero value represents perfect
diversification and one indicates perfect specialization. This index captures the overall
level of concentration on a farm. For both DH and the following index, horticultural crops
are grouped together in a single share of the farm, αj.
The final aggregate measure used as a dependent variable is the Berger-Parker
(BP) index, !!" = 1/max  (!! ) (Parker and Berger 1971). The BP index measures
inverse dominance, or proportional abundance, by capturing the share of farmland
allocated to the crop with the largest share. DBP takes on values greater than or equal to
one, with higher values indicating greater—or more even—diversification. Values closer
to one denote higher relative abundance of the dominant crop.

39

3.2.2 Individual crop analysis
The second model will investigate the determinants of diversification into pigeon peas. I
use a two-stage model to estimate the two stages of a farmer’s decision. The first
dependent variable is a dummy variable to indicate cultivation of pigeon peas. The
second stage measures the amount of land allocated to pigeon peas, given that a farmer
has decided to grow them. The model is described in detail in section 3.3.2.

3.3 Estimation Strategy
3.3.1 Indices of Diversity: Fixed Effects Model
To investigate the determinants of each of the three indices, I estimate the following
fixed-effects model:

!!" = !! + !! !! + !! !!" + !! !!" + !! + !!"

(11)

where Pt is a vector of regional market prices, Mvt represents the village-level market
constraints and Zit represents farm-level characteristics of household i in time period t.

λ t represents a dummy variable for 2011 to allow for time-varying intercepts.
Household-level fixed effects capture village level fixed-effects given that no households

€

moved villages between the panel years. This will yield the same results as the firstdifferencing model, or regressing the change in diversity on the change in explanatory
variables (Wooldridge 2010).

40

The estimation method I use does not account for unobserved factors that affect
crop diversification, influence decisions on road access and also change over time. For
instance, if new roads systematically targeted areas where there are local initiatives to
increase productivity and crop diversification, the FE estimator would be positively
biased. However, road improvement decisions are much more likely to follow past
agricultural performance than anticipated future initiatives. Therefore, most change in
road access over time should be exogenous to changes in other unobserved factors so that
bias in the FE estimates I present, if any, should be small.
The fundamental assumption of this model is that the unobserved heterogeneity in
question is time constant. I argue that this is a function of a region’s inherent
agroecological potential and is therefore unchanging between 2008 and 2011. One
limitation, however, is that the direct effect of that agroecological potential cannot be
estimated in the model, only eliminated. Additionally, all explanatory variables in the
model must be strictly exogenous and time-varying in order to be identified. I discuss
each explanatory variable in section 3.4.
3.3.2 Pigeon Pea Cultivation and Land Allocation: Two Stage Decision Model
A farmer deciding to cultivate pigeon peas actually faces two decisions: first, whether to
grow them at all, and second, how much land to allocate to them. Some studies of crop
diversification have used the share of land allocated to a given crop as the dependent
variable (Allen 2012; Ndhlovu 2010); however in the case of Mozambique, only a
minority of cases grows pigeon peas and therefore the distribution of land allocation is
lumped at zero. Hence I model these two decisions separately as part of a two-step
process using Cragg’s double-hurdle (DH) model (1971), commonly referred to as the
41

tobit alternative model. This method is used to measure the two stage decision of selling
a product and then, conditional on selling, what quantity to sell (Burke 2009; Mather et al.
2011; Cairns 2012; Reyes 2012). One limitation to the DH model is the assumption that
the error terms in each stage are independent; this could be unrealistic because the
decisions are in fact similar. However, the two stages still provide more accurate
estimates than modeling them as a single outcome.
Cragg’s DH is a favorable alternative to the Heckman two-step approach used by
Goetz (1992) because it does not treat the high frequency of zero responses as a censored
sample problem; a zero value is a decision made by the farmer, and therefore constitutes
a valid economic choice, or “corner solution” (Wooldridge 2010; Mather et al. 2011).
Below, I adapt the model employed by Mather et al. (2011) to estimate the factors
determining farmers’ land allocation decisions conditional on the decision to grow pigeon
peas.
The structure of the two decisions is as follows. The probability of household i
planting pigeon peas in time period t is represented by:
1

1

1

!!!∗=! !!! +!! +!!!

1

1

!!! ~(0,1)

(12)

where !!! =1 if !!!∗>0, otherwise !!!=0.
The land allocated to pigeon peas is then:
∗
!
!!"
= ! ! !!"
+ !!! + !!"!

!
!!"! ~!"(− ! ! !!"
+ !!! , 0, ! ! )

*

where α!! = α!! if !!!=1, otherwise α!! = 0.
42

(13)

2

2

2

Where TN denotes the truncated normal with lower bound –(γ xit +ci ), mean 0, and
2

1

2

variance σ , and eit is independent of eit conditional on the explanatory variables and
the unobserved heterogeneity terms. The superscripts denote the stage of the decision.
The variable pit* represents the unobservable latent variable that determines
whether a farmer plants pigeon peas or not which is observable and captured by pit. The
variable αit* is how much land would be allocated to growing pigeon peas by a farmer if
the decided to plant pigeon peas and is observable only when pit =1, and αit is the
observed amount of land allocated to growing pigeon peas. xit ={ xit

1

, xit2} is a vector of

observable explanatory variables assumed to be exogenous. Unobserved heterogeneity
that may be correlated with the explanatory variables and affect farmers’ decisions is
1

2

captured by ci and ci . In my previous set of models, ci was eliminated through the use
of fixed effects. However, as shown by Wooldridge (2010) and Greene (2004), fixed
effects estimation is not typically available for non-linear models; in particular, there is
no fixed effects estimator available for Cragg’s two stage model.
Instead, to address the problem of unobserved heterogeneity, Mather et al. (2011)
employ a correlated random effects (CRE) probit (stage 1) and lognormal (stage 2) using
the Mundlak version of the Chamberlain device (Mundlak 1978; Chamberlain 1982). The
Chamberlain device is an estimation method that postulates that a linear projection of
time constant unobserved factors on strictly exogenous explanatory variables is sufficient
43

to account for all endogeneity due to these unobserved factors. The Mundlak device is a
restricted version of the Chamberlain device that imposes that the coefficients of
covariates in that linear projection be equal across time so that only averages of the
covariates across time need to be added to the list of covariates to account for unobserved
heterogeneity.
I use the same method employed by Mather et al. (2011), with instead a truncated
normal regression for stage two, assuming that the correlation between the unobserved
heterogeneity and the explanatory variables takes the form:
!
!! ! = ! ! + !! !!! + !!"

!!! |!!! ~!(0, !!! )

(14)

!! ! = ! ! + ! ! !!! + !!"!

!!! |!! ! ~!(0, !!! )

(15)

where ! is the household specific time-mean of each explanatory variable. In addition, in
!

!

order to use Cragg’s double hurdle model I assume that !! and !! are independent
conditional on the explanatory variables, which means that unobserved heterogeneity
terms that affect the two decisions are not correlated after conditioning on the
explanatory variables. This is the same kind of assumption is made for the standard
Cragg double hurdle model without CRE where noise terms affecting the two decisions
are not correlated after conditioning on the explanatory variables. This assumption is
parallel to that made for the two stages of the model, and also may not be likely in a case
where the decisions are so closely related. Additionally, the Mundlak device requires the
assumption that all variation in the error term that is correlated with the explanatory
variables is captured in ci. This is the same assumption made in a fixed effects model,
44

and allows the model to account for as much of the endogeneity due to unobserved
factors as possible.
This specification is then added to the right-hand side of the estimation equations
for p* and q*. I add ! to the set of explanatory variables, and the error terms are
2

transformed to uit = εi +eit. The final estimations are as follows:
∗
!
!
!!"
= ! ! + ! ! !!"
+ !! !!! + !!"

!
!
!
!!"
|!!"
~!(0, !!!
)

(16)

!
!
!
!!"
|!!"
~!(0, !!!
)

(17)

where !!! =1 if !!!∗>0, otherwise !!!=0.
∗
!
!
!!"
= ! ! + ! ! !!"
+ ! ! !!! + !!"
*

where α!! = α!! if !!!=1, otherwise α!! = 0.
Because the values of ! vary only across households, but are time constant, this method
controls for any time-constant heterogeneity.
To interpret the results of these two estimations, I calculate the average partial
effects (APE) across all households and time periods. Following Cairns (2012) and using
Burke’s (2009) Stata program, craggit, I calculate household-specific conditional partial
effects (PE) of xm (where xm is explanatory variable m) on the probability of planting
using the maximum likelihood estimated coefficients from the first stage probit:
2

Uit is not independent across time; however, this does not affect the consistency of the
pooled maximum likelihood estimator, and time dependence is accounted for by
reporting clustered standard errors.

45

!"(!∗!" !!|!!! )
!!!

!
!
!
!
!
!
= !!
/!!!
!(!! /!!!
+ !1!!"
/!!!
+ !! !!! /!!!
)

1

(18)

1

Where ϕ is the standard normal probability density function (PDF) and β xit is the
matrix of explanatory variables and their parameters. The variance of the error term is
!
!!!
, as defined in equation (16). By calculating the partial derivative with respect to xit

but not !, I isolate the causal effect of each explanatory variable. For binary variables, I
evaluate the change in the mean outcome of p for xit = 0 and xit = 1.
To calculate the partial effects for the second stage truncated normal regression, I
use the following equation (Wooldridge 2010):

!(
0
!

!!!"

!
!!!"

!
!!

!! !

|!! , !!" >

1−!

!
!! !!! !!"
!!! !!!

!
!! !!! !!"
!!! !!!

where

!!!

!"
!"

!!!

)

!
!! !!! !!!
!!! !!!

!!!

+

(19)

denotes the partial effect of x on α, λ is the inverse mills ratio (PDF/cumulative

density function) and σ is the estimated variance from the second stage regression.
(Again, for binary variables I calculate the change rather than partial derivative.) From
these two equations, I obtain the partial effect of each explanatory variable on each
household in both stages. APEs are then calculated by taking the mean across the sample.
46

3.4 Explanatory Variables and Hypotheses
Following Equation 10, diversification is a function of land endowment, household and
farm characteristics, market prices and constraints, including infrastructure, and initial
wealth. This section will identify specific variables in those categories that could impact a
household’s decision to alter its crop mix. Descriptive statistics for all explanatory
variables are listed in Table 10.
3.4.1 Market Prices
For Mozambican smallholders, expected market prices are the key determinant in a
farmer’s expected profits from growing any given crop and therefore crucial in the
decision to diversify. To capture variation in expected prices across villages and years, I
use market prices collected by the Agricultural Market Information System of
Mozambique (SIMA) in seven major markets. I calculate expected price using the
average monthly price from the previous year. Households are assigned the expected
prices of the market nearest to them. I include prices for three crops commonly grown by
farmers in our sample—maize, common beans, and cowpeas—and use natural logs so
that changes can be interpreted as percentages. Maize prices are much lower than bean
prices per kilogram, therefore to compare them it is more useful to interpret percentages.

47

Table 10. Summary Statistics, Explanatory Variables
2008
2011
Mean
Mean
Expected Maize
Log monthly average price/kg at regional
1.399
2.058
Price
market, previous year
(.141) (.133)
Expected Common Log monthly average price/kg at regional
3.280
3.664
Bean Price
market, previous year
(.238) (.199)
Expected Cowpea Log monthly average price/kg at regional
2.474
2.849
Price
market, previous year
(.177) (.276)
Log district median producer price/kg,
1.636
1.971
Pigeon Pea Price
previous year
(.249) ( .172)
Paved Road, Mid- = 1 if nearest paved road between 10-50 km 0.387
0.403
Distance
from village
(.487) (.491)
Paved Road, Far= 1 if nearest paved road > 50 km from
0.333
0.241
Distance
village
(.472) (.428)
= 1 if road to nearest town is accessible
0.778
0.783
Year Round Road
during rainy season
(.415) (.413)
1.697
1.612
Market
= 1 if village has a market
(.460) (.487)
0.621
0.813
Mobile Network
= 1 if village has mobile phone network
(.485) (.390)
0.735
0.835
Radio
= 1 if village has radio coverage
(.442) (.371)
5.478
6.640
Household Size
Number of people in household
(2.516) (3.043)
Proportion of household members >14
0.502
0.494
Proportion Male
years that are male
(.134) (.173)
2.813
3.618
Land Size
Total size of land owned by household (ha)
(4.037) (5.865)
Household ownership of basic assets, max
0.502
0.565
Asset Index
score = 1
(.306) (.307)
N=1,186. Standard deviation listed in parentheses. Source: MSU/MINAG Surveys,
2008 & 2011.
Variable

Definition

For the pigeon pea model, I incorporate the market price of pigeon peas,
expecting that higher prices would be a key determinant of the decision to cultivate.
Unfortunately, the SIMA price data do not include pigeon peas. Instead, I use districtmedian post-harvest prices from the preceding year as expected market price, as reported
in the TIA 2007 and agricultural census of 2010. I recognize the potential endogeneity in
this relationship; median reported prices depend on local production and therefore the
48

same farmers’ behavior as those the model attempts to predict, albeit in the prior year. I
was unable to access regional market price data for this specific crop, but I believe
including them in the model is still better than the alternative, in which the coefficients on
the infrastructure variables could also be picking up the effects of prices. By using the
district median, I minimize the potential endogeneity of an individual household’s
behavior on sales price, but I still expect some bias given the small number of households
per district selling pigeon peas.
As noted previously, Mozambican smallholders rarely use inputs. Most seed is
retained from previous harvests, and fertilizer use is scarce. Therefore, I do not include
input costs in my model. Expected profits are represented by expected output prices.
3.4.2 Market Constraints

The explanatory variables selected at the village level will attempt to capture major
changes in infrastructure that occurred between 2008 and 2011 and may change the
market constraints faced by farm households. As noted by Rahman (2008), infrastructure
can affect diversification both directly, and indirectly through its impact on prices.
However, in my model, because prices are included at a regional market level, they are
unlikely to be correlated with village-level infrastructure. Therefore village-level
infrastructure may capture variation in expected prices for farmers that are not captured
in the regional market prices.
At a village level, distance from a village to the nearest paved road and year
round road accessibility may lower market constraints for farmers by lowering the
transaction costs of acquiring inputs or selling their harvests. It could also improve the
49

market for information relating to crops and prices, because easier transportation implies
more travel between villages and more efficient agricultural extension. However, in
Mozambique it is also possible that isolation (higher market constraints) encourages
higher diversification because households are required to be self-sufficient (Benin et al.
2004). In a simple t-test (see Table 11), the data show that households farther from paved
roads tend to grow fewer crops, which contradicts this hypothesis. In addition to the
transportation infrastructure, the lack of a market within the village represents a market
constraint for similar reasons. It could mean lower transaction costs for acquiring inputs
and selling outputs, and it could influence a farmer’s decision to diversify into new crops
simply by increasing his awareness of different crops. Indeed, we see significantly higher
levels of diversity in those villages with markets, as measured by the Herfindahl and BP
indices, but not a significantly higher average crop count (Table 11.)

Table 11. Correlations between Infrastructure and Diversification
Mean
Mean
Pr(|T|>|t|)
Village or Household
Diversity
(households
(households
Level Characteristic
Measure
with
without
characteristic) characteristic)
Village >50 km from
Crop Count
4.22
4.44
0.0149
paved road
Mobile network
Crop Count
4.46
4.14
0.0003
coverage
Radio Coverage
Crop Count
4.44
4.15
0.0037
Herfindahl
Market in village
.394
.409
0.0600
Index
Berger-Parker
Market in village
2.22
2.13
0.0077
Index
Year round road
Crop Count
4.31
4.56
0.0114
accessibility
Two-sample t-tests with equal variances. Source: MSU/MINAG Surveys, 2008 &
2011. Means are pooled from both years.

50

Insufficient communication infrastructure is also a constraint in the market for
information, which I capture through the presence of mobile phone network and radio
coverage in a village. Farmers with access to information may be more likely to diversify
into new crops if they learn about prices or technology via agriculture radio programming
or simply by calling friends or family in other villages or districts. Indeed we see higher
crop counts in villages with radio and mobile coverage (Table 11.) Alternatively, access
to information could lower the risks faced by a farmer; diversification is a coping strategy
for some farmers against price or climate risk, so guaranteed access to market prices
could actually lower their need for diversification.
Simple, time-averaged correlations show that smallholders with local markets
tend to allocate their land more evenly among different crops, while those in more
isolated areas grow lower total numbers of crops. However, the presence of markets and
roads may also be correlated with unobservable characteristics of the region; they are
likely to target areas with better agricultural production, and mobile networks follow. By
using fixed effects, I minimize potential endogeneity of infrastructure and diversification.
I argue infrastructure development may depend on a regions inherent agroecological
potential, and therefore is time-invariant between the two survey years. Therefore, the
agroecological potential is captured by village fixed effects.
3.4.3 Household and farm-level characteristics
Following Benin et al. (2004), I include gender composition of the household
measured by proportion of males within the household. This may affect diversification
via labor availability. Household size is also an indicator of labor availability and
heterogeneity of preferences (Benin et al. 2004).
51

Finally, the physical characteristics of a farm may affect crop diversification.
Benin et al. (2004) use slope, erosion, fertility, and distance to plots to capture this;
however with a panel dataset, these features should drop out assuming they remain
constant over the three-year period. I include total landholding as we do see some
variation in this variable across the panel. Total land available should affect the level of
diversification as farmers with a greater amount of land have a greater ability to diversify.
To capture initial wealth of a household, I cannot use income, as that is clearly a
function of a household’s land allocation. Instead, I construct a basic household asset
index; households are scored based on how many of five basic assets they possess (oil
lantern, radio, bicycle, latrine, and table.) A zero value means they own none; a value of
one means they own all five.

52

CHAPTER FOUR: RESULTS AND DISCUSSION
4.1 Determinants of Crop Diversification
Employing a fixed effects regression with robust standard errors clustered by household,
I predict farm levels of crop diversification based on expected market prices, village-level
infrastructure, and household characteristics. Regression results are listed in Table 12 and
described below.
4.1.1 Infrastructure
Households in villages located over 50 kilometers from a paved road are likely to grow
fewer total crops than those close to a paved road (see Table 12.) This result is
inconsistent with the hypothesis that households more isolated from the market tend to
have higher levels of diversity in order to be self-sufficient. That isolated farms grow
fewer crops suggests that either they are self-sufficient, but consume less diverse diets
(possible given that such farms allocated an average of over 50 percent of land to starches,
but would require consumption data to confirm), or that these more isolated farms lack
access to information about different crops; farmers in areas gaining access to paved
roads may also be gaining access to traders or extension workers that encourage
diversification. Therefore, one possible method for improving the diversity of production
in remote areas could be to better connect farmers to traders or extension that are not
typically present in the absence of a paved road; however, more research would be
required to ascertain the nutritional status of such families.
Households with year round road access to the nearest town have a lower
Herfindahl index (see Table 12), which represents a more even distribution across
53

Table 12. Household Fixed Effects Regression Results, 2008 - 2011
Ψ
Crop Count
BP Index
Herfindahl Index
Expected Maize Price
0.574
0.106
-0.394
(1.127)
(0.113)
(0.570)
Expected Common Bean
-1.739***
0.099*
-0.554**
Price
(0.602)
(0.060)
(0.268)
Expected Cowpea Price
3.965***
-0.351***
1.724***
(1.239)
(0.125)
(0.631)
Paved Road, Mid-Distance
-0.243
0.012
-0.086
(0.184)
(0.019)
(0.083)
Paved Road, Far-Distance
-0.501**
0.003
-0.032
(0.208)
(0.023)
(0.088)
Year Round Road
0.200
-0.026*
0.073
(0.172)
(0.014)
(0.078)
Market
0.062
0.004
-0.085
(0.164)
(0.015)
(0.070)
Mobile Network
0.211
-0.031*
0.100
(0.172)
(0.016)
(0.077)
Radio
-0.005
0.015
-0.061
(0.154)
(0.015)
(0.080)
Household Size
0.145***
-0.013**
0.033
(0.053)
(0.006)
(0.026)
Proportion Male
0.164
-0.038
-0.146
(0.496)
(0.062)
(0.362)
Land Size
0.046*
-0.002
0.018**
(0.027)
(0.002)
(0.009)
Asset Index
0.466*
-0.027
0.158
(0.546)
(0.025)
(0.116)
Year 2011
-0.852
0.011
-0.134
(0.546)
(0.053)
(0.265)
R-Squared (within)
0.1100
0.0559
0.0439
N=1,130. Robust standard errors reported in parentheses, clustered at household
level; *** (p<.01), ** (p<.05), * (p<.10). Weighted by population sample weights
Ψ
and inverse probability weighting for attrition. Lower value indicates more even
portfolio/ higher diversity. Higher BP index indicates more domination by one crop.
different crops. Lack of year round road access is also a measure of isolation, but the
index provides more information about the farming behavior than just the crop count. As
expected, this result is similar to that of the paved road and is inconsistent with the
hypothesis that isolation can lead to greater self-sufficiency and affect diversity levels.
Instead, it supports the hypothesis that in scenarios of high food insecurity and risk
54

aversion, families may focus on just one crop to ensure that their staple is available from
their own production.
Households that do not have year round road access allocate their crops less
evenly than those that do. This result indicates one of the following possibilities: (1)
farmers can sell staple crops within their own village, and therefore lack of year round
road access does not affect their ability to sell; (2) farmers only sell once per year, or
during the season in which they have road access; or (3) that they do not sell at all, but
simply grow and consume an uneven distribution of crops. To test (1), I generated an
indicator variable for those villages with no year round road access but with a market in
the village; the coefficients on this variable when included in the estimation were not
statistically significant. However, it is possible that the ability to sell a staple, such as
maize or cassava, exists in villages that do not have a formal market captured by the
survey. Possibilities (2) and (3) are both likely; the harvest typically continues up to two
months after the rainy season ends (FEWSNET 2013), so isolation during the rainy
season would not necessarily prevent a farmer from traveling to market with his or her
harvest. Additionally, many Mozambicans’ diets consist primarily of one staple, such as
maize or cassava, and therefore more land allocated to one crop could be indicative of the
family’s diet or risk profile.
Mobile network coverage is also associated with a lower Herfindahl index, or a
more even allocation of land among different crops (see Table 12.) I hypothesized that
mobile network coverage is a proxy for access to information, but that different
information could affect a farmer’s cropping decisions in different ways. First, mobile
phones could provide smallholders with information about current market prices, which
55

in turn, shape their price expectations in the following year. This could affect their
perceived risk or profitability of diversification and therefore their planting decisions. For
example, if a farmer learns via cell phone that the price of a crop he or she intends to sell
is high, he or she might perceive a lower risk of allocating more land to that crop, and
therefore have a lower measure of on-farm evenness. Alternatively, a farmer could learn
via cell phone that the price of a food he or she plans to purchase is high or very volatile,
and therefore decide to grow everything he or she plans to consume in order to insure
against price risk. My results indicate that the latter is a more likely explanation, if
mobile phones are indeed serving to communicate market price information. However,
there could be a time lag in this relationship, because farmers’ decisions to plant would
be based on expected and not current prices; therefore having cell phones in the previous
year or at the time of planting would be more important than at harvest time.
In addition to market price information, farmers can also benefit from information
relating to know-how, such as crop, seed, and input choices, and context-specific
knowledge such as weather and climate (Mittal et al. 2010). Mobile phones could provide
access to such information; many countries are exploring the effectiveness of mobilebased agricultural extension, but information can also be communicated informally via
phone calls to friends or family outside the village. For example, a household could learn
from a family member about a new crop or a new way to grow a crop, and that could
affect the household’s decision-making around land allocation. Depending on the nature
of the information, this could either increase or decrease diversification. The regression
results show that mobile coverage is associated with more diversity, suggesting the
addition of new crops but only as a small fraction of the total cropped area. Additionally,
56

mobile phone coverage could reduce diversification risk, as phones can connect farmers
with market agents with whom they would not otherwise be in communication.
My model isolates five different infrastructure variables (see Table 11) as
independent features of a village; however, it is also possible that the components of
infrastructure when interacted may have a different coefficient from when they are
isolated. For example, access to information via cell phone and access to markets via year
round roads could have more impact on a farmer’s cropping decisions than each alone. I
therefore tested several combinations of those variables. First, I use a basic index of the
five components where a village can score between zero and one depending on how
many of the components it has. This infrastructure index is a significant predictor of crop
count, indicating that households in villages with better infrastructure—regardless of
type—grow more crops. The other results in the model do not change (see appendix for
full results.)
4.1.2 Prices
Prices are the strongest determinants of crop diversification, as expected; it is surprising,
however, that the expected price of maize—the crop most commonly grown in our
sample—does not have a significant coefficient (see Table 12.) This could be because
the majority of households in the surveyed regions grow maize no matter what the
expected price, so it does not affect their cropping behavior, although it may be very
important in their marketing behavior. The expected price of cowpeas has a positive
association with crop diversity as measured by all three indices and has the highest
magnitude of coefficient. Cowpeas are grown by almost half of our sample, yet marketed

57

by very few. I therefore interpret the decision to grow cowpeas as generally a
subsistence-oriented decision; they are grown primarily to feed the family. It follows,
then, that when the price of such a subsistence crop increases, it represents a threat to
food security and farmers diversify to manage their risk.
Higher expected prices of common beans have the opposite effect; they are
negatively associated with crop diversity levels as measured by both the crop count and
BP index (see Table 12.) This could be because of the three crops with prices included in
this model, common beans are the most likely to be marketed by the farmers that grow
them. Therefore, higher prices represent a market opportunity rather than a threat to food
security, incentivizing farmers toward specialization. It must be noted, however, that
farmers are more likely to specialize in any crop, not common beans in particular, when
common bean prices rise.
4.1.3 Household and Farm Characteristics
Additionally, I find that land size and household size have positive associations with
diversification levels (see Table 12.) This follows the hypotheses that greater land and
labor availability enable higher farm diversity. The proportion of adult males in the
household—another indicator of labor availability—however has no significant
association with diversification. Finally, a higher asset index is associated with a higher
crop count, indicating that wealthier families grow more crops. This is also inconsistent
with the subsistence hypothesis; wealthier households ought to be able to purchase food
to eat and therefore more likely to specialize for the market. It may be the case, however,
that even those families with higher wealth indices do not have such security; the assets

58

are very basic items such as lamps and tables, so these families could still be vulnerable
to food insecurity, despite having relatively more income than their neighbors.
4.2 Pigeon Pea Two Stage Model
To estimate the probability of a household cultivating pigeon peas and the land allocated
to pigeon peas based on prices, village-level infrastructure, and farm-level characteristics,
I use Cragg’s double hurdle model with correlated random effects and bootstrapped
standard errors. Results of this estimation are presented in Table 13 and described in the
following section.
4.2.1 Infrastructure and prices
No infrastructure variables were significant determinants of a household’s binary
decision of whether or not to plant pigeon peas. This is consistent with my hypothesis
that pigeon peas are a “multi-use” crop—ie. in some cases, they are planted only for
home consumption, while in other cases they are grown for the emerging market
opportunities they offer. Therefore, modeling only this first stage of the decision, I would
not expect to detect a significant effect of infrastructure. The two-stage approach allows
for interpretation of this first decision separately from how much land a farmer allocates
to pigeon peas, which may be more indicative of a farmer’s intent to market the crop.
Access to both communication infrastructure variables—mobile network and
radio coverage—positively affect the amount of land that a farmer allocates to pigeon
peas, conditional on deciding to grow them. This indicates that access to information is
also central to a farmer’s decision to grow pigeon peas for the market; the information
could be market prices, linkages with traders, or simply the knowledge that pigeon peas
59

are becoming tradable and are therefore deserving of more land than a pure subsistence
crop. As explained in Chapter 1, pigeon peas have recently become an exportable
commodity in Mozambique, though they have always existed as a subsistence crop. So,
these results indicate that mobile phones and radio coverage may be the channel that
informs farmers of this new market.
The expected price of pigeon peas is only a significant predictor of the first stage
of a farmer’s decision. Higher prices increase the probability that a farmer plants the crop,
as expected, but are not associated with the amount of land allocated. This is surprising,
as I would hypothesize a strong, positive effect as more farmers perceive pigeon peas as
marketable when the price is high, and therefore allocate more land. The coefficient is
indeed positive but is not statistically significant. This could be a result of the data used
to estimate expected prices; as previously explained, these prices are district medians and
are therefore dependent on how much was produced and sold in the district the previous
year.
4.2.2 Household and Farm Characteristics

Household size is strongly associated with both stages of the decision, confirming that
labor availability is a key factor in a family’s decision to grow and market a new crop.
The asset index is also positively associated with both stages, indicating that wealthier
families are more likely to take on a new and market-oriented crop. Finally, the results
confirm that greater total landholdings allow farmers to adopt new crops.

60

Table 13. Household Two-Stage Regression Results, 2008-2011
Land Allocated
(Conditional)
0.022
0.023
Paved Road, Mid-Distance
(0.020)
(0.033)
0.005
-0.047
Paved Road, Far-Distance
(0.017)
(0.030)
0.022
0.020
Year Round Road
(0.015)
(0.032)
-0.038
-0.018
Market
(0.030)
(0.032)
0.016
0.063**
Mobile Network
(0.015)
(0.028)
0.015
0.068***
Radio
(0.012)
(0.021)
0.039**
0.028
Pigeon Pea Price
(0.017)
(0.036)
0.013***
0.034***
Household Size
(0.005)
(0.009)
0.063
-0.025
Proportion Male
(0.047)
(0.088)
0.007**
0.004
Total Landholdings
(0.003)
(0.003)
0.047*
0.075*
Asset Index
(0.025)
(0.045)
Coefficients represent average partial effects (APE). Bootstrapped standard errors in
parentheses, clustered at the household level. *** (p<.01), ** (p<.05), * (p<.10).
N=1,186.
Pr (Cultivate Pigeon Peas)

4.3 Conclusions

Agriculture defines Mozambique; employing 80 percent of the population and
constituting 24 percent of GDP (USG 2011), it has the attention of international donors
and Mozambique’s own government as a lynchpin for alleviating poverty. Indeed,
stakeholders are spending millions to improve the sector and enhance food security; the
Government of Mozambique recently borrowed $150 million from the World Bank “to
promote private sector-led agriculture in order to improve access to food and better

61

nutrition” (World Bank 2013) and USAID has budgeted $18 million for agricultural
development and $5.1 million for nutrition in Mozambique in 2013 (Department of State
2013).
With so much global interest in improving Mozambican smallholder-dominated
agriculture, it is crucial to understand how smallholders make their production decisions
in order to affect change. One decision every smallholder must make is how to allocate
his or her land among different crops, or at what level of diversification to operate. Under
some conditions, greater crop diversification can improve smallholder welfare by
insuring against price and weather risks and improving nutrient intake via higher dietary
diversity. Many agriculture initiatives in Africa, including USAID’s strategy for
Mozambique (USG 2011), promote diversification for these reasons. In other scenarios,
lower diversity—or greater specialization—can improve smallholder welfare by enabling
economies of scale in both the production and marketing of crops. In either case, before
encouraging a shift in diversification, we must understand what drives the decision.
My research contributes to the understanding of smallholder decision-making
around total crops cultivated and crop portfolio evenness, specifically within the context
of Mozambique’s rapidly expanding infrastructure and food price volatility. The
conclusions are based on literature and data alone; I expect that our understanding of
smallholder behavior around this decision could be greatly enhanced through qualitative
research. Interviews and focus groups were not possible for this study, but could
illuminate different farmers’ rationale their changing cropping patterns.

62

From my quantitative analysis, I find that improved infrastructure is associated
with crop diversification, though the effect of changes in infrastructure is relatively less
important than changing prices. Farmers far from paved roads grow fewer crops,
suggesting that if they are growing for their own consumption, they have less diverse
diets than households in areas with better transport connectivity. This finding is
inconsistent with results from the literature; Ibrahim et al. (2009) found that good road
network conditions decreased the number of crops grown and Rahman (2008) found that
an index of improved infrastructure measures was associated with lower diversity as
measured by three indices, including the Herfindahl index. This could be indicative of
farmers specializing for commercial reasons rather than for basic food security, as I
predict is the case in Mozambique. I also find that those with year round road access to
neighboring towns allocate their land more evenly among crops, or have a higher level of
diversity, which is consistent with the above finding that less isolation from the market
leads to more diversity. Both results indicate that in the case of subsistence farmers,
improving access to the information and trading opportunities typically enabled by
infrastructure could also improve diversity of production. In order to confirm these
results, further research should be done with more precise indicators of market access and
isolation, so that households’ ability to cross the border on foot—or to connect with the
market in other ways—is captured.
The percentage of households in our survey growing pigeon peas nearly doubled
between 2008 and 2011. This represents a promising opportunity to improve smallholders’
income by connecting them to the growing export market for legumes. The USAID Feed
the Future strategy for Mozambique does not specifically mention pigeon peas, but its
63

focus is linking smallholders to markets for exportable oilseeds and pulses (USG 2011). I
investigate the characteristics of the farmers that have already decided to adopt pigeon
peas and find that those in areas with the highest market prices for pigeon peas are most
likely to grow them, as is expected and consistent with the basic microeconomic theory
that supply is positively dependent on price. I also find that farmers with greater access to
information—as proxied by communication infrastructure—allocate more land to pigeon
peas. As such, programs seeking to encourage smallholder cultivation of pigeon peas and
similar crops for export should investigate promotion and extension via mobile and radio
networks.
For further research, I would suggest the incorporation of pigeon pea processors
into a model of pigeon pea adoption. I did research in this area and confirmed that new
processing plants have opened since 2007, however I was not able to obtain their exact
locations or dates of inception and therefore did not use them as explanatory variables. I
would hypothesize proximity to a pigeon pea processor to be a strong, positive predictor
of both the probability of planting and the amount of land allocated to pigeon peas.
Similarly, I do not incorporate the presence or magnitude of input markets in
either of my models. This is based on the assumption that very few Mozambican
smallholders purchase inputs, which from the literature I conclude to be true. However,
there could be anomalies in the data; for example, if a program or private company
begins promoting fertilizer or improve seeds in certain villages, this would have an
impact on those farmers’ land allocation decisions and therefore affect both models.
Since it would likely be at a village level, such an impact could be picked up by other
village level characteristics, such as infrastructure.
64

Additionally, I predict that environmental factors such as rainfall, temperatures, or
pests in one year could influence a farmer’s decision to plant different crops in the
following year. For example, if a drought ruined the horticulture harvest, farmers might
be more likely to allocate more land to a drought-tolerant crop such as cassava the next
season. I did not have appropriate data to include these factors in my model; I expect that
significant variations in yields due to such events would be captured by the market prices,
but a useful extension could be to use rainfall variability at the village level to predict
cropping decisions.
My results also show that wealthier families with larger farms and more
household members are more likely to adopt pigeon pea cultivation. This is consistent
with the theory; wealthier households are less vulnerable to crop and price risk (or less
risk averse in general), and therefore more likely to take on a cash crop (Fafchamps 1992).
This would be important to consider if a program were designed to promote pigeon pea
cultivation, and could be extended for other marketable crops such as groundnuts, sesame,
and soy that the Gates Foundation and USAID, inter alia, are encouraging for
smallholders. In order for such programs to be successful with small, poorer farms, they
would first need to address vulnerability, risk attitudes, and access to land.
My research also shows that high prices of staple crops encourage subsistence
farming, if overall crop portfolio evenness is a proxy for subsistence farming. This is
important if we expect that specialization can yield efficiency gains, and therefore would
contribute to growth. Price stabilization—and access to markets—would lower the risk
smallholders take on when they decide to rely on the market for food instead of their own
cultivation, which is required in order for them to expand into more market-oriented and
65

income-generating crops. However, overall evenness could also be indicative of
innovative or market-oriented farmers taking on new crops.
Additionally, my conclusions assume a certain relationship between
diversification and market participation, but I do not use data on market participation
itself because it was not available at the time of the analysis. To gain a more accurate
understanding of the relationship between diversification levels and subsistence or
market-oriented farming, market participation data could be employed using similar
models to those used in this paper. Understanding the relationship between market
participation and land allocation is very difficult because while a farmer has some plans
for the market when he or she decides to allocate land, the decision to participate is
ultimately made after harvest and therefore affected by land allocation.
The use of diversity indices is somewhat limiting as it fails to provide insight on
shifts between different crops or crop categories; for example, a farmer could grow half
groundnuts and half maize one year, then half vegetables and half pigeon peas the next,
and none of the three measures used here would detect a change, though such a change
would have very serious implications for that household’s income, nutrition, and
workload. While there is no simple measure available in the literature that would capture
this, a comprehensive analysis could be done using my model to estimate the probability
of growing each individual crop, crop category, and the land allocated to each. This
would provide a more complete understanding of behaviors as they relate to specific
crops.

66

Diversity indices are also limiting because diversification itself is ambiguous.
Indices assign a scale with low to high values, however those values still require
interpretation. As noted in Figure 1 in the introduction, farmers on the subsistenceoriented spectrum are motivated differently than those on the market-oriented spectrum,
though they may have similar “scores” as measured by any of the three diversity indices.
Diversification is a much more nuanced concept than is generally understood by the
development community; in scenarios of extreme poverty and high risk, “specializing” is
more likely to indicate a family striving to reach a minimal caloric threshold than
specialization for the market. As such, an interesting extension of diversification research
could be to assign households to one of the two spectrums (based on historical market
participation data, consumption, or qualitative interviews) and conduct the analysis for
the two groups separately.
This paper contributes to the understanding of smallholder behavior in
Mozambique, specifically what drives diversification on small farms. Policies that
expand information and market access to isolated farms have the potential to change
cropping behaviors . However, we must also understand the impact of changes in land
allocation. The World Bank found no relationship between diversification and income in
Mozambique (2006), and there is very little in the literature that attempts to estimate it.
We also do not know how crop diversification affects the diet diversity of individuals
within a household, although many development policies postulate a linkage. This would
be the logical next step for further research, as the nutritional impacts are likely to differ
based on a family’s motivation for diversification (ie. subsistence or commercial.)

67

APPENDIX

68

Table 14. Average Household Land Allocated to Crops by Province
Nampula
Zambezia
Tete
Manica
2008 2011 2008 2011 2008 2011 2008 2011
Maize
27% 25% 41% 35% 47% 51% 59% 55%
Sorghum
7%
9%
7%
6%
0%
0% 13%
9%
Rice
2%
3%
6%
6%
0%
0%
0%
1%
Sesame
3%
3%
1%
0%
0%
0%
3%
2%
Horticulture
2%
1%
1%
3% 11%
7%
7%
9%
Cowpea
7%
8%
4%
3%
5%
3%
5%
4%
Cassava
32% 29% 17% 17%
1%
1%
2%
6%
Cotton
2%
3%
1%
1%
1%
1%
0%
0%
Millet
0%
0%
0%
0%
0%
0%
1%
0%
Small
Groundnut
9%
7%
2%
1%
3%
4%
2%
2%
Sweet
Potato
0%
0%
0%
1%
2%
1%
1%
3%
Large
Groundnut
3%
3%
2%
1%
5%
5%
1%
1%
Common
Bean
0%
0%
2%
2% 11% 11%
2%
3%
Pigeon Pea
2%
3% 13% 18%
0%
0%
0%
2%
Jugo Bean
2%
2%
0%
0%
0%
1%
1%
1%
Orange
Sweet
Potato
0%
0%
0%
0%
0%
0%
0%
0%
Potato
0%
0%
0%
0%
4%
4%
0%
0%
Sugar
1%
0%
1%
2%
0%
0%
0%
0%
Tobacco
0%
0%
1%
2%
6%
6%
0%
0%
Sunflower
0%
0%
0%
0%
1%
0%
1%
1%
Yam
0%
0%
0%
0%
0%
0%
0%
1%
Mung Bean
1%
3%
0%
0%
0%
0%
0%
0%
Soy
0%
0%
0%
1%
3%
3%
0%
0%
Green Bean
0%
0%
0%
0%
0%
0%
0%
0%
Paprika
0%
0%
0%
0%
0%
0%
0%
0%
Ginger
0%
0%
0%
0%
0%
0%
0%
0%
Sisal
0%
0%
0%
0%
0%
0%
0%
0%
Tea
0%
0%
0%
0%
0%
0%
0%
0%
Number of
Households
198
196 248 244 254 256 210 206
Source: MSU/MINAG Survey 2008, 2011.

69

Sofala
2008 2011
44% 38%
16%
9%
9% 10%
8%
6%
5%
8%
3%
4%
3%
6%
3%
3%
2%
1%
2%

2%

1%

2%

1%

1%

1%
1%
1%

1%
4%
1%

0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%

1%
0%
1%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%

265

264

Table 15. Household Fixed Effects Regression Results with
Infrastructure Index, 2008 - 2011
Crop Count Herfindahl Index
BP Index
0.534
0.137
-0.484
Expected Maize Price
(1.118)
(0.115)
(0.586)
Expected Common
-1.623***
0.087
-0.505*
Bean Price
(0.592)
(0.060)
(0.265)
3.459***
-0.313**
1.530**
Expected Cowpea Price
(1.215)
(0.128)
(0.628)
0.654*
-0.034
0.037
Infrastructure Index
(0.339)
(0.034)
(0.176)
0.144***
-0.013**
0.038
Household Size
(0.052)
(0.006)
(0.026)
0.149
-0.040
-0.119
Proportion Male
(0.458)
(0.061)
(0.360)
0.049
-0.002
0.018*
Land Size
(0.027)*
(0.002)
(0.009)
0.442
-0.022
0.133
Asset Index
(0.252)*
(0.025)
(0.117)
-0.633
-0.023
0.000
Year 2011
(0.513)
(0.054)
(0.263)
N=1,130. Robust standard errors reported in parentheses, clustered at
household level; *** (p<.01), ** (p<.05), * (p<.10). ΨLower value
indicates more even portfolio/ higher diversity. Higher BP index indicates
more domination by one crop.

70

REFERENCES

71

REFERENCES

Aker, J. (2010). Information from Markets Near and Far: Mobile Phones and
Agricultural Markets in Niger. The Economic Journal: Applied Economics, 2,
46-59.
Allen, J. (2012). Determinants of land allocation in a multi-crop farming system: An
application of the fractional multinomial logit model to agricultural households
in Mali. (M.S.), Michigan State University.
Aneani, F., Anchirinah, V. M., Owusu-Ansah, F., & Asamoah, M. (2011). An analysis
of the extent and determinants of crop diversification by cocoa (Theobroma
cacao) farmers in Ghana. African Journal of Agricultural Research, 6(18), 10.
Antle, J. (1983). Infrastructure and Aggregate Agricultural Productivity: International
Evidence. Economic Development and Cultural Change. 31(3): 609-619.
Arndt, C., Chinowsky, P., Strzepek, K., & Thurlow, J. (2012). Climate Change, Growth
and Infrastructure Investment: The Case of Mozambique. Review of Development
Economics, 16(3), 463-475.
Arndt, C., Hussain, M. A., Jones, E. S., Nhate, V., Tarp, F., & Thurlow, J. (2012).
Explaining the Evolution of Poverty: The Case of Mozambique. American
Journal of Agricultural Economics, 94(4), 854–872.
Benfica, R. & Tschirley, D. (2012). The Dynamics of Agricultural Market Participation
and Performance in Central and Northern Mozambique: Evidence from a Panel
Survey. Flash No. 61E. Research Results from the Directorate of Economics,
Ministry of Agriculture.
Benin, S., Smale, M., Pender, J., Gebremedhin, B., & Ehui, S. (2004). The economic
determinants of cereal crop diversity in the Ethiopian highlands. Agricultural
Economics, 31, 197-208.
Besley, T., & Case, A. (1993). Modeling technology adoption in developing
countries. The American Economic Review, 83(2), 296.
Bezabih, M., & Sarr, M. (2012). Risk Preferences and Environmental Uncertainty:
Implications for Crop Diversification Decisions in Ethiopia. Environmental
Resource Economics, 53, 483–505.

72

Binswanger, H., Khandker, S. & Rosenzweig, M. (1993.) How infrastructure and
financial institutions affect agricultural output and investment in India. Journal
of Development Economics. 41(1993): 337-366.
Bittinger, A. K. (2010). Crop Diversification and Technology Adoption: The role of
market isolation in Ethiopia. (M.S.), Montana State University, Bozeman,
Montana.
Bizimana, J.-C., Bessler, D., & Angerer, J. (2013). Impact of Market Information System
(E-Soko) on Beans Markets Integration: Case of Rwanda. Paper presented at the
Southern Agricultural Economics Association Annual Meeting, Orlando.
Brouwer, R., & Brito, L. (2012). Cellular phones in Mozambique: Who has them and
who doesn't? Technological Forecasting & Social Change, 79, 231-243.
Burke, W. (2009). Triple Hurdle Model of Smallholder Production and Market
Participation in Kenya’s Dairy Sector. Master’s Thesis: Michigan State
University. East Lansing, MI.
Cairns, J. (2012). Thesis Re-print: Does Selling Fruits or Vegetables Provide a Strategic
Advantage to Selling Maize for Small-Holders in Mozambique? A DoubleHurdle Correlated Random Effects Approach to Evaluating Market
Decisions Agricultural Research Institute of Mozambique Research Report
Series.
Central Intelligence Agency (CIA). (2013). The World Factbook: Mozambique.
Available: https://www.cia.gov/library/publications/the-worldfactbook/geos/mz.html
Chamberlain, G. (1982). Multivariate regression models for panel data. Journal of
Econometrics. 18(1): 5-46.
Cragg, J.G. 1971. Some Statistical Models for Limited Dependent Variables with
Application to the Demand for Durable Goods. Econometrica 39.5: 829‐44.
Cungara, B., Fagilde, G., Garrett, J., Uaiene, R., & Headey, D. (2011). Growth Without
Change: The Elusiveness of Agricultural and Economic Transformation in
Mozambique. Paper presented at the Dialogue on Promoting Agricultural Growth
in Mozambique, Maputo, Mozambique.
Cunguara, B., Mudema, J., Mather, D., & Tschirley, D. (2012). Changes in smallholder
cropping and input use in Central and Northern Mozambique, 2008/2011 Flash:
Research results from the Directorate of Economics, Ministry of
Agriculture (Vol. 60).

73

de Janvry, A., Fafchamps, M., & Sadoulet, E. (1991). Peasant Household Behaviour
with Missing Markets: Some Paradoxes Explained. The Economic Journal,
101(409).
DeBoer, J., & Chandra, S. (1978). A model of crop selection in semi-subsistence
agriculture and an application to mixed agriculture in Fiji. American Journal of
Agricultural Economics, 60(3), 436-444.
Department of State. (2013). Congressional Budget Justification, Volume 2: Foreign
Operations. Fiscal Year 2013.
Dercon, S., Gilligan, D., Hoddinott, J. and Woldehanna, T. (2009). The Impact of
Agricultural Extension and Roads on Poverty and Consumption Growth in
Fifteen Ethiopian Villages. American Journal of Agricultural Economics. 91(4):
1007-1021.
Dominguez-Torres, C., & Briceño-Garmendia, C. (2011). Mozambique’s Infrastructure:
A Continental Perspective Africa Infrastructure Country Diagnostic.
Washington, DC: The World Bank.
Dorosh, P., Wang, H.G., You, L. & Schmidt, E. (2012). Road connectivity, population,
and crop production in Sub-Saharan Africa. Agricultural Economics. 43(2012):
89-103.
Fafchamps, M. (1992). Cash Crop Production, Food Price Volatility, and Rural Market
Integration in the Third World American Journal of Agricultural Economics,
74(1), 9.
Fafchamps, M. & Minten, B. Impact of SMS-Based Agricultural Information on Indian
Farmers. The World Bank Economic Review. 26(3): 383-414.
Famine Early Warning Systems Network (FEWSNET). (2013). Mozambique Price
Bulletins. From United States Agency for International Development:
http://www.fews.net/pages/country.aspx?gb=mz&l=en.
Food and Agriculture Organization (FAO). (2011). Nutrition Country Profile: Republic
of Mozambique. Food and Agriculture Organization of the United Nations—
Nutrition and Consumer Protection Division.
Goetz, S.J. 1992. A Selectivity Model of Household Food Marketing Behavior in SubSaharan Africa. American Journal of Agricultural Economics 74.2: 444-452.
Greene, W. 2004. Fixed Effects and Bias due to the Incidental Parameters Problem in the
Tobit Model. Econometric Reviews 23.2: 125-47.

74

Gubert, F., & Robilliard, A. S. (2007). Risk and Schooling Decisions in Rural
Madagascar: A Panel Data-Analysis. Journal of African Economies, 17(2), 207238.
Herfindahl, O. C. (1950) Concentration in the U.S. Steel Industry. Unpublished doctoral
dissertation, Columbia University.
Horioka, C. Y., & Wan, J. (2007). The Determinants of Household Saving in China: A
Dynamic Panel Analysis of Provincial Data. Journal of Money, Credit and
Banking, 39(8).
Hua, W., Hite, D., & Sohngen, B. (2005). Assessing the relationship between crop
choice and land use change using a Markov model. Paper presented at the
American Agricultural Economics Association Annual Meeting Providence,
Rhode Island.
Ibrahim, H., Rahman, S. A., Envulus, E. E., & Oyewole, S. O. (2009). Income and Crop
Diversification Among Farming Households in a Rural Area of North Central
Nigeria. Journal of Tropical Agriculture, Food, Environment and Extension,
8(2).
Immink, M.D.C., & Alarcon, J. (1993). Household Income, Food Availability, and
Commercial Crop Production by Smallholder Farmers in the Western Highlands
of Guatemala. Economic Development and Cultural Change. 41(2): 319-342.
Jacoby, H. (2000). Access to Markets and the Benefits of Rural Roads. The Economic
Journal, 110(July), 713-737.
Janvry, A. d., Fafchamps, M., & Sadoulet, E. (1991). Peasant Household Behaviour with
Missing Markets: Some Paradoxes Explained. The Economic Journal, 101(409),
17.
Jensen, R. (2007). The Digital Provide: Information (Technology), Market Performance,
and Welfare in the South Indian Fisheries Sector. The Quarterly Journal of
Economics, CXXII(3).
Marenya, P., & Barrett, C. (2007). Household-level determinants of adoption of
improved natural resources management practices among smallholder farmers in
western Kenya Food Policy, 32, 21.
Mather, D. (2012). Determinants of crop income in rural Mozambique, 20022005 Research Report for the Directorate of Economics, Ministry of Agriculture:
Michigan State University.
Mather, D., Boughton, D., & Jayne, T. S. (2011). Smallholder Heterogeneity and Maize
Market Participation in Southern and Eastern Africa: Implications for Investment
75

Strategies to Increase Marketed Food Staple Supply International Development
Working Paper. East Lansing: Michigan State University.
Mather, D. & Donovan, C. (2007). The Impacts of Prime-age Adult Mortality on Rural
Household Income, Assets, and Poverty in Mozambique. Background paper for
the Mozambique Poverty, Gender, and Social Impact Assessment, Africa Region,
The World Bank, Washington, D.C. and MINAG Directorate of Economics
Research Paper 65E, Maputo, Mozambique.
Mellor, J. (1985). The Changing World Food Situation. Food Policy Statement.
International Food Policy Research Institute. Washington, D.C. January 1985.
Mesfin, W., Fufa, B., & Haji, J. (2011). Pattern, Trend and Determinants of Crop
Diversification: Empirical Evidence from Smallholders in Eastern
Ethiopia. Journal of Economics and Sustainable Development, 2(8).
Mittal, S., Gandhi, S. & Tripathi, G. (2010). Socio-Economic Impact of Mobile Phones
on Indian Agriculture. Indian Council for Research on International Economic
Relations. Working Paper No. 246.
Mosera, C., & Barrett, C. (2006). The complex dynamics of smallholder technology
adoption: the case of SRI in Madagascar. Agricultural Economics, 35, 373–388.
Mundlak, Y. (1978). On the Pooling of Time Series and Cross Section Data.
Econometrica. 46(1): 69-85.
Muto, M., & Yamano, T. (2009). Impact of Mobile Phone Coverage Expansion on
Market Participation: Panel Data Evidence from Uganda. World Development,
37(12), 1887-1896.
Ndhlovu, D. (2010). Determinants of farm households' cropland allocation and crop
diversification decisions: The role of fertilizer subsidies in Malawi. (M.S.),
Norwegian University of Life Sciences.
Parker, F. and Berger, W. (1971). Faunal and solution patterns of planktonic
Foraminifera in surface sediments of the South Pacific. Deep Sea Research and
Oceanographic Abstracts. 18(1): 73–107.
Poverty and Wellbeing in Mozambique: Third National Poverty Assessment (2010):
National Directorate of Studies and Policy Analysis (DNEAP), Ministry of
Planning and Development.
Rahman, S. (2008). The Economic Determinants of Crop Diversity on Farms in Rural
Bangladesh. Journal of International Agricultural Trade and Development, 5(1).

76

Reardon, T., Delgado, C., & Matlon, P. (1992). Determinants and Effects of Income
Diversification Amongst Farm Households in Burkina Faso.The Journal of
Development Studies, 28(2).
Reyes, B. (2012). The Economic Impact of Improved Bean Varieties and Determinants
of Market Participation: Evidence from Latin America and Angola. (Doctor of
Philosophy), Michigan State University, East Lansing.
Singh, I., Squire, L., & Strauss, J. (1986). A Survey of Agricultural Household Models:
Recent Findings and Policy Implications. The World Bank Economic Review,
1(1), 30.
Stifel, D., & Minten, B. (2008). Isolation and agricultural productivity. Agricultural
Economics, 39, 15.
Taylor, E., & Adelman, I. (2003). Agricultural Household Models: Genesis, Evolution,
and Extensions. Review of Economics of the Household, 1(Jan-Apr 2003).
United Nations Commodity Trade Statistics Database (COMTRADE). (2013).
United States Government (USG). (2011). Mozambique FY2011-2015 Multi-Year
Strategy Feed the Future Multi-Year Strategies. Washington, DC.
Van Dusen, E. & Taylor, J.E. (2003). Missing Markets and Crop Diversity: Evidence
from Mexico. University of California at Berkeley Working Paper. February 1,
2003.
Wooldridge, J. (2010). Econometric Analysis of Cross Section and Panel Data. The MIT
Press. Cambridge, MA.
World Bank. (2006). Mozambique Agricultural Development Strategy: Stimulating
Smallholder Agricultural Growth: The World Bank: Agriculture, Environment,
and Social Development Unit.
World Bank. (2009). Project Information Document: Roads and Bridges Management
and Maintenance Program - Phase II Additional Financing. Washington, Dc: The
World Bank.
World Bank. (2013). World Bank Supports Mozambique’s Plans to Boost Agriculture
and Increase Food Security. Press Release: April 25, 2013. Available:
http://www.worldbank.org/en/news/press-release/2013/04/25/world-banksupports-mozambique-plans-to-boost-agriculture-and-increase-food-security
World Food Programme (WFP). (2013). Hunger Statistics. Available:
http://www.wfp.org/hunger/stats.

77

World Telecommunication/ICT Indicators Database. (2012).
Zita, G. (2012). The Quality of Mobile Services (Voice and SMS) in Mozambique: A
Proposed Methodology to Effectively Apply Appropriate Quality of Service
Standards. Maputo: United States Agency for International
Development/Mozambique Support Program for Economic and Enterprise
Development.

78