FOOD CONSUMPTION PATTERNS IN LIGHT OF RISING INCOMES,
URBANIZATION AND FOOD RETAIL MODERNIZATION:
EVIDENCE FROM EASTERN AND SOUTHERN AFRICA
By
Michael James Dolislager

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food, and Resource Economics—Doctor of Philosophy
2017

ABSTRACT
FOOD CONSUMPTION PATTERNS IN LIGHT OF RISING INCOMES,
URBANIZATION AND FOOD RETAIL MODERNIZATION:
EVIDENCE FROM EASTERN AND SOUTHERN AFRICA
By
Michael James Dolislager
Over the past fifteen years, Eastern and Southern Africa (ESA) has experienced
significant changes in its agrifood system. Driven by rapid urbanization and growth in per capita
incomes, changing food consumption patterns have been at the core of this broader agrifood
system change. Responding to the region’s diet transformation, firms – including a limited
presence of supermarkets – have arisen in food marketing and distribution, food processing, and
food services.
Food consumption patterns are changing in three ways: food is becoming more
purchased, more perishable, and more processed. These changes drive a multitude of effects on
the agrifood system. This dissertation provides new insight into the nature of the diet
transformation that is unfolding in the region, and on some of its drivers and effects. The first
essay, “Diversification “Beyond Staple Foods” in the Diets of Poor Rural and Urban
Consumers in Developing Eastern and Southern Africa”, focuses on the effects of income and
urbanization on the commodity makeup and source of foods in household diets. Transitioning
food consumption patterns are not solely a middle-class story, as conventionally assumed – in
fact, poor households are already consuming surprisingly high levels of purchased food and are
consuming greater shares of non-grain foods as their incomes rise. Spatial considerations of
increasing city size and reduced distance to cities also have significant positive effects on

purchased share, affirming the expectation that households will purchase more food (compared
to consuming own production) when they have increased access to markets.
Essay two, “Consumption of Processed Food in the Diets of Poor Rural and Urban
Consumers in Developing Eastern and Southern Africa”, further analyzes the same drivers of
food consumption patterns on the consumption of processed food. Processed foods, specifically
highly processed foods, have penetrated the diets of rural and urban households at all levels
across the income distribution of ESA. The income-induced diet change towards processed food
begins among the poor regardless of household distance from urban areas, albeit at greater shares
among households within larger cities. The patterns of increased processed share with income
growth and urbanicity signal a strong future demand for increased food market infrastructure.
The final essay, “City size, supermarkets, and processed foods: Evidence from Zambia”,
incorporates food retail modernization into the analysis of processed food consumption patterns.
Utilizing a unique dataset that contains disaggregated household consumption data, including
source and distance to source of each consumed food item, this analysis shows the following:
First, conditional on the presence of a supermarket, households in smaller cities consume greater
shares of food, by value, from supermarkets than households in larger cities. Second, although
supermarket use positively affects the consumption of processed food, households that consume
some food from supermarkets continue to purchase the majority of their processed food from
non-supermarket retailers.
Together, these three essays provide detail to how the macro trends of income growth,
rapid urbanization and the beginning of food retail modernization significantly affect the
consumption patterns of ESA, which in turn affect ESA’s entire agrifood system.

Copyright by
MICHAEL JAMES DOLISLAGER
2017

ACKNOWLEDGEMENTS

I wish to express my gratitude for the mentorship and guidance that I have received from
Dr. David Tschirley, who has been my major advisor and has funded my research at Michigan
State University. Thank you for the many hours you spent working with me throughout these
past years. This would not have been possible without your support. I would also like to thank
the other members of my doctoral committee, Drs. Thomas Reardon, Nicole Mason, Steven
Haggblade and Robert Richardson, for their invaluable contributions to this research. Thank you
for the walks around the track, for the meetings and calls working through analysis, and for the
many comments and emails that have occurred to bring this research to where it is today.
I also want to extend a big thanks to my friends and family. Thank you to the many of
you who have been with me through this process. Special thanks to Helder Zavale, for the hours
you spent helping me with empirical analysis early in my graduate experience, and Chewe
Nkonde, for all your encouragement as we made our way through graduate school. I am
especially grateful to you both for your friendship. To my parents, James and Carla Dolislager,
thank you for your prayers, encouragement and support during this time and throughout my
entire life. And last, but not least, a thanks to my wife, Elyse Dolislager: thank you for your love
and patience through these past few years, you mean more to me than words can express.

v

TABLE OF CONTENTS

LIST OF TABLES ................................................................................................................. viii
LIST OF FIGURES ............................................................................................................... xiii
KEY TO ABBREVIATIONS ................................................................................................ xv
INTRODUCTION ................................................................................................................. 1
REFERENCES ............................................................................................................... 5
ESSAY 1: DIVERSIFICATION “BEYOND STAPLE FOODS” IN THE DIETS OF POOR
RURAL AND URBAN CONSUMERS IN DEVELOPING EASTERN AND SOUTHERN
AFRICA ................................................................................................................................. 7
1.1 Introduction ................................................................................................................ 7
1.2 Conceptual Framework .............................................................................................. 10
1.3 Methods...................................................................................................................... 14
1.3.1 Context ........................................................................................................... 14
1.3.2 Data ................................................................................................................ 18
1.3.3 Definitions...................................................................................................... 19
1.3.4 General Model ............................................................................................... 21
1.3.5 Empirical Model ............................................................................................ 22
1.3.6 Estimation Methods ....................................................................................... 28
1.4 Results ........................................................................................................................ 31
1.4.1 Descriptive Results ........................................................................................ 32
1.4.2 Nonparametric analysis of the relationship between expenditure and food
consumption ................................................................................................... 34
1.4.3 Expenditure Threshold Analysis .................................................................... 40
1.4.4 Engel’s Curve Model regression analysis ...................................................... 44
1.5 Conclusions ................................................................................................................ 57
APPENDIX ...................................................................................................................... 60
REFERENCES ................................................................................................................ 73
ESSAY 2: CONSUMPTION OF PROCESSED FOOD IN THE DIETS OF POOR RURAL
AND URBAN CONSUMERS IN DEVELOPING EASTERN AND SOUTHERN AFRICA
................................................................................................................................................ 77
2.1 Introduction ................................................................................................................ 77
2.2 Conceptual Framework .............................................................................................. 78
2.3 Methods...................................................................................................................... 84
2.3.1 Context ........................................................................................................... 84
2.3.2 Data ................................................................................................................ 87
2.3.3 Definitions...................................................................................................... 89
2.3.4 General Model ............................................................................................... 90
2.3.5 Empirical Model ............................................................................................ 91

vi

2.3.6 Estimation Methods ....................................................................................... 96
2.4 Results ........................................................................................................................ 99
2.4.1 Descriptive statistics and nonparametric analysis of the relationship between
processed food consumption and total household expenditure ..................... 99
2.4.2 Expenditure Threshold Analysis .................................................................... 104
2.4.3 Descriptive statistics of processed food consumption by city size ................ 108
2.4.4 Nonparametric analysis of the relationship between distance to cities and
processed food consumption .......................................................................... 110
2.4.5 Engel’s Curve Model regression analysis ...................................................... 111
2.4.6 Mediation model analysis .............................................................................. 118
2.5 Conclusions ................................................................................................................ 120
APPENDIX ...................................................................................................................... 123
REFERENCES ................................................................................................................ 138
ESSAY 3: CITY SIZE, SUPERMARKETS, AND PROCESSED FOODS: EVIDENCE FROM
ZAMBIA ................................................................................................................................ 143
3.1 Introduction ................................................................................................................ 143
3.2 Conceptual Framework .............................................................................................. 146
3.3 Methods...................................................................................................................... 154
3.3.1 Data ................................................................................................................ 154
3.3.2 Definitions...................................................................................................... 157
3.3.3 General Model ............................................................................................... 158
3.3.4 Empirical Model ............................................................................................ 158
3.3.5 Estimation Methods ....................................................................................... 167
3.4 Results ........................................................................................................................ 172
3.4.1 Descriptive Statistics ...................................................................................... 172
3.4.2 Regression Results ......................................................................................... 177
3.5 Conclusions ................................................................................................................ 186
APPENDIX ...................................................................................................................... 189
REFERENCES ................................................................................................................ 205
CONCLUSIONS.................................................................................................................... 209

vii

LIST OF TABLES

Table 1-1: National expenditure and population by expenditure strata and city size ............ 17
Table 1-2: Household descriptive characteristics .................................................................. 19
Table 1-3: Dependent Variables ............................................................................................ 20
Table 1-4: Determinants of demand in the Engel’s Curve Model ......................................... 24
Table 1-5: Food budget shares of key food consumption patterns, aggregated country and
settlement ............................................................................................................................... 32
Table 1-6: Food budget shares of select commodity aggregates, pooled data by settlement
................................................................................................................................................ 34
Table 1-7: Expenditure thresholds and slopes prior to and after expenditure thresholds in the
consumption patterns of purchased share and Beyond Staple Foods items – rural data ....... 40
Table 1-8: Expenditure thresholds and slopes prior to and after expenditure thresholds in the
consumption patterns of purchased share and Beyond Staple Foods items – urban data ...... 41
Table 1-9: Expenditure thresholds and slopes prior to and after expenditure thresholds for the
aggregates of select commodity based aggregates – rural data ............................................. 42
Table 1-10: Expenditure thresholds and slopes prior to and after expenditure thresholds for the
aggregates of select commodity based aggregates – urban data ............................................ 43
Table 1-11: Average partial effects of expenditure for key food consumption patterns – rural data
................................................................................................................................................ 45
Table 1-12: Average partial effects of expenditure and city size for key food consumption
patterns – urban data .............................................................................................................. 46
Table 1-13: Indirect effects of urban household settlement within primary and secondary cities,
estimates by country .............................................................................................................. 50
Table 1-14: Average partial effects of expenditure and distances to various city sizes for key
food consumption patterns – rural Zambia ............................................................................ 55
Table 1-15: Indirect effects of distance from rural household settlement to primary, secondary or
tertiary cities, estimates by country ....................................................................................... 57
Table 1-A-1: Food budget shares of select commodity aggregates, country data by settlement
................................................................................................................................................ 61

viii

Table 1-A-2: Average partial effects of the household determinants for key food consumption
patterns, urban household estimates by country .................................................................... 63
Table 1-A-3: Average partial effects of the household determinants for key food consumption
patterns, rural household estimates by country ...................................................................... 64
Table 1-A-4: Average partial effects of the household determinants for select commodity
aggregates, urban household estimates by country ................................................................ 65
Table 1-A-5: Average partial effects of the household determinants for select commodity
aggregates, rural household estimates by country ................................................................. 69
Table 2-1: Impact of city size on food choice: urban environment, food environment, and
demand factors ....................................................................................................................... 83
Table 2-2: National expenditure and population by expenditure strata and city size ............ 86
Table 2-3: Household descriptive characteristics .................................................................. 87
Table 2-4: Processed food aggregates.................................................................................... 90
Table 2-5: Determinants of demand in the Engel’s Curve Model ......................................... 93
Table 2-6: Food budget shares of processed food, urban households by country and expenditure
................................................................................................................................................ 100
Table 2-7: Food budget shares of processed food, rural households by country and expenditure
................................................................................................................................................ 100
Table 2-8: Expenditure thresholds and slopes prior to and after expenditure thresholds in
processed food consumption patterns .................................................................................... 105
Table 2-9: Expenditure thresholds and slopes prior to and after expenditure thresholds for
processed food aggregates – urban data................................................................................. 106
Table 2-10: Expenditure thresholds and slopes prior to and after expenditure thresholds for
processed food aggregates – rural data .................................................................................. 107
Table 2-11: Food budget shares of processed food, aggregated country and settlement ...... 109
Table 2-12: Food budget shares of processed food aggregates, pooled data by settlement .. 109
Table 2-13: Average partial effects of expenditure and city size on processed share – urban data
................................................................................................................................................ 112
Table 2-14: Average partial effects of expenditure and distances to various city sizes on
processed share – rural data .................................................................................................. 113

ix

Table 2-15: Indirect effects of urban household settlement within primary and secondary cities
on processed share, estimates by country .............................................................................. 119
Table 2-16: Indirect effects of distance from rural household settlement to primary, secondary or
tertiary cities on processed share – rural Zambia................................................................... 120
Table 2-A-1: Relationship between food classification scheme in this paper and that in Monteiro
et al. (2010) ............................................................................................................................ 124
Table 2-A-2: Food budget shares of processed food aggregates, by country and expenditure
................................................................................................................................................ 124
Table 2-A-3: Food budget shares of processed food aggregates, country data by settlement
................................................................................................................................................ 128
Table 2-A-4: Average partial effects of the household determinants on processed share, urban
household estimates by country ............................................................................................. 129
Table 2-A-5: Average partial effects of the household determinants on processed share, rural
household estimates by country ............................................................................................. 130
Table 2-A-6: Average partial effects of the household determinants for processed food
aggregates, urban household estimates by country ................................................................ 131
Table 2-A-7: Average partial effects of the household determinants for processed food
aggregates, rural household estimates by country ................................................................. 135
Table 3-1: Impact of city size on food choices: urban environment, food environment, and
demand factors ....................................................................................................................... 153
Table 3-2: City populations ................................................................................................... 155
Table 3-3: Total daily household expenditure per adult equivalent ...................................... 156
Table 3-4: Kilometers to supermarket relative to closest non-supermarket food retailer to the
household .............................................................................................................................. 156
Table 3-5: Shares of market value by acquisition type .......................................................... 157
Table 3-6: Dependent variables ............................................................................................. 161
Table 3-7: Key for determinants of demand in equation 3.1 ................................................. 161
Table 3-8: Household characteristics of the sample .............................................................. 167
Table 3-9: Household weighted averages of consumption shares ......................................... 173
Table 3-10: Household weighted averages of sources of food aggregates ............................ 175

x

Table 3-11: Household weighted averages of consumption shares by low and high processing
levels ...................................................................................................................................... 176
Table 3-12: Fractional probit estimates of the average partial effects of the determinants of
household share of supermarket purchases in total food consumption.................................. 178
Table 3-13: Second stage of control function regression; estimating the share of processed food
in total food consumption ...................................................................................................... 181
Table 3-14: City indirect effects via specific explanatory variables ...................................... 185
Table 3-A-1: Prices of commonly consumed items ............................................................... 190
Table 3-A-2: First stage of control function regression; OLS estimates of the determinants of the
share of supermarket purchases in total food consumption ................................................... 191
Table 3-A-3: Second stage of control function regression; estimating the share of low processed
food in total food consumption .............................................................................................. 192
Table 3-A-4: Second stage of control function regression; estimating the share of high processed
food in total food consumption .............................................................................................. 193
Table 3-A-5: Estimates of the marginal effects of the determinants of processed food in total
food consumption (OLS) ....................................................................................................... 194
Table 3-A-6: Estimates of the marginal effects of the determinants of low processed food in total
food consumption (OLS) ....................................................................................................... 195
Table 3-A-7: Estimates of the marginal effects of the determinants of high processed food in
total food consumption (2SLS) .............................................................................................. 196
Table 3-A-8: Second stage of two stage least squares; estimating the share of processed food in
total food consumption (2SLS) .............................................................................................. 197
Table 3-A-9: Second stage of two stage least squares; estimating the share of low processed food
in total food consumption (2SLS) .......................................................................................... 198
Table 3-A-10: Second stage of two stage least squares; estimating the share of high processed
food in total food consumption (2SLS) ................................................................................. 199
Table 3-A-11: Fractional probit estimates of the determinants of household share of supermarket
purchases in total food consumption (CRE) .......................................................................... 200
Table 3-A-12: First stage of control function regression; OLS estimates of the determinants of
the share of supermarket purchases in total food consumption (CRE).................................. 201
Table 3-A-13: Second stage of control function regression; estimating the share of processed
food in total food consumption (CRE)................................................................................... 202

xi

Table 3-A-14: Second stage of control function regression; estimating the share of low processed
food in total food consumption (CRE)................................................................................... 203
Table 3-A-15: Second stage of control function regression; estimating the share of high
processed food in total food consumption (CRE) .................................................................. 204

xii

LIST OF FIGURES

Figure 1-1: Conceptual framework ........................................................................................ 12
Figure 1-2: Nonparametric (LOWESS curve) analysis of select food consumption patterns – rural
data ......................................................................................................................................... 35
Figure 1-3: Nonparametric (LOWESS curve) analysis of select food consumption patterns –
urban data ............................................................................................................................... 36
Figure 1-4: Nonparametric (LOWESS curve) analysis of select commodity aggregates – pooled
rural data ................................................................................................................................ 38
Figure 1-5: Nonparametric (LOWESS curve) analysis of select commodity aggregates – pooled
urban data ............................................................................................................................... 39
Figure 1-6: Estimated effects of expenditure on purchased share and the consumption of Beyond
Staple Foods items – rural data .............................................................................................. 46
Figure 1-7: Estimated effects of expenditure on purchased share and the consumption of Beyond
Staple Foods items – urban data ............................................................................................ 47
Figure 1-8: Locally weighted scatterplot smoothing curves representing the typical purchased
share relative to household distance to city ........................................................................... 53
Figure 1-9: Locally weighted scatterplot smoothing curves representing the typical food budget
shares of the consumption Beyond Staple Foods items relative to household distance to city
................................................................................................................................................ 54
Figure 2-1: Conceptual framework ........................................................................................ 80
Figure 2-2: Nonparametric (LOWESS curve) analysis of consumption patterns of processed food
................................................................................................................................................ 101
Figure 2-3: Nonparametric (LOWESS curve) analysis of processed food aggregates – pooled
data ......................................................................................................................................... 104
Figure 2-4: Locally weighted scatterplot smoothing curves representing the typical processed
share relative to household distance to city ........................................................................... 111
Figure 2-5: Estimated effects of expenditure on purchased share and the consumption of
processed food ....................................................................................................................... 114
Figure 3-1: Conceptual approach linking city size to consumer food choices ...................... 147

xiii

Figure 3-2: Estimation of indirect effects .............................................................................. 172
Figure 3-3: Estimated effects of expenditure on supermarket use and processed food
consumption ........................................................................................................................... 179
Figure 3-A-1: Processing level aggregates compared with aggregates from Monteiro et al. 2010
................................................................................................................................................ 190

xiv

KEY TO ABBREVIATIONS

2SLS

Two Stage Least Squares

AE

Adult Equivalent

BSF

Beyond Staple Foods

CRE

Correlated Random Effects

ESA

Eastern & Southern Africa

FDI

Foreign Direct Investment

IV

Instrument Variable

LOWESS

Locally Weighted Scatterplot Smoothing

OLS

Ordinary Least Squares

PPP

Purchasing Power Parity

SF

Staple Foods

UCS

2007/2008 Zambia Urban Consumption Survey

FP

Fractional Probit

xv

INTRODUCTION

Over the past fifteen years Eastern and Southern Africa (ESA)1 has experienced
significant changes in its agrifood system. Driven by rapid urbanization and growth in per capita
incomes, changing food consumption patterns – amounting to a diet transformation - have been
at the core of this broader agrifood system change. Average daily expenditures per adult
equivalent in ESA remain low, near three dollars2, yet World Bank data show that countries in
this region have achieved real growth of per capita income3 averaging near three and a half
percent per annum since 2000. Urban population since this time has risen from 21% of total
population to over one quarter currently4, and the United Nations Population Division anticipates
that forty-four percent of the population will live in urban areas by 2050. Responding to the
region’s diet transformation, 10s of thousands of micro, small, and medium size firms have
arisen in food marketing and distribution, food processing, and food services including prepared
food away from home. Lastly, this region is experiencing food retail modernization as part of
the final wave of the supermarket revolution. Although supermarket presence in Sub-Saharan
Africa remains the lowest of any major region of the world, rapid increases have been seen in
recent years in major urban areas of ESA, driven by rising incomes, urbanization and market
liberalization that drove an influx of domestic investment and foreign direct investment.

Eastern and Southern Africa is referring to the following developing Eastern and Southern African nations:
Ethiopia, Botswana, Kenya, Malawi, Mozambique, Tanzania, Uganda, and Zambia
2 Author’s calculation from LSMS survey data.
3 Author’s calculation using GNI per capita, PPP (constant international dollars).
4 Author’s calculations using United Nations Population Division data. Urban share of population calculated at 21%
in the year 2000.
1

1

Diets are changing in three ways: food is becoming more purchased, more perishable,
and more processed. These changes drive a multitude of effects on the agrifood system.
Consumer demand for food incentivizes the production of food; therefore household food
demand affects the production decisions of farmers and potentially affect the policy actions of
governments that support certain production practices. Increased purchasing of food through
markets drives private investment in those markets, and requires significant public investment in
marketing infrastructure (especially wholesale markets but also retail) and in complementary
infrastructure such as roads, electrical networks, and plumbing. Consumer demand for more
perishable foods – fresh produce, meats, and milk products – drives a need for investment in cold
chains and changing marketing practices to handle such foods. Increased demand for processed
food presents enormous opportunities for investment by local entrepreneurs to satisfy this
demand, at the same time that it drives increased imports of processed foods, directly competing
with local businesses.
This dissertation provides new insight into the nature of the diet transformation that is
unfolding in the region, and on some of its drivers and effects. The first essay, “Diversification
“Beyond Staple Foods” in the Diets of Poor Rural and Urban Consumers in Developing Eastern
and Southern Africa”, focuses on the effects of income and urbanization on the commodity
makeup and source of foods in household diets. Bennett’s Law (1941) states that, as incomes
increase, households will consume greater shares of non-grain foods, establishing a relationship
between income and food consumption patterns. The concept of supernumerary income (Stone,
1954) indicates that households allocate initial income toward required expenditures and that
income above a certain threshold is allocated at the household’s discretion. Combining Bennett’s
Law and the concept of supernumerary income suggests that food consumption patterns will

2

change as incomes rise above a level required to meet basic household needs. Yet simple
calculations on average household food consumption patterns potentially appear to show diet
change happening well below what could be considered supernumerary levels of income. Using
nonparametric and parametric analyses, essay one seeks to understand these patterns by testing
for changes in the direction and magnitude of the relationships between income and food
consumption patterns of specific commodities and channels of food acquisition.
The rapid urbanization of ESA highlights the importance of understanding the effects of
urban residence, city size and proximity to cities on food consumption patterns. Urbanization has
been shown to affect household lifestyle patterns (Huang and Bouis, 2001; FerrĂŠ et al., 2012;
Fafchamps and Shilpi, 2003), and this essay considers such effects on commodity food budget
shares and food purchase shares in ESA.
Essay two, “Consumption of Processed Food in the Diets of Poor Rural and Urban
Consumers in Developing Eastern and Southern Africa”, further analyzes the same drivers of
food consumption patterns on the consumption of processed food. Household consumption
patterns of processed food are expected to vary much as the consumption of non-staple food
varies across the income distribution and with increased urbanicity (de Haen et al., 2003; Pingali,
2007). Given such expectations, this essay analyzes consumption patterns with respect to
processed food across the entire income distribution and explores the impact of city size and
proximity to city on these patterns.
The final essay, “City size, supermarkets, and processed foods: Evidence from Zambia”,
incorporates food retail modernization into the analysis of food consumption patterns.
Supermarkets are a key component in food retail modernization, one that provides households
with increased access to a greater diversity of types of foods, thereby affecting household food

3

consumption patterns (Rischke et al., 2015). To this author’s knowledge, no previous study has
unpacked the impact of urbanization on shopping and consumption patterns by examining how
this impact varies by city size and by distance to cities. Utilizing a unique dataset containing
disaggregated household consumption data that includes source and distance to source of each
consumed food item, this essay tests for the effects of city size on supermarket use and
consumption of processed food, recognizing the affect that increasing supermarket use could
have on processed food consumption.

4

REFERENCES

5

REFERENCES

Bennett, M. K. (1941). Wheat in national diets. Wheat Studies, (02).
De Haen, H., Stamoulis, K., Shetty, P., & Pingali, P. (2003). The world food economy in the
twenty‐first century: challenges for international co‐operation. Development Policy
Review, 21(5‐6), 683-696.
Fafchamps, M., & Shilpi, F. (2003). The spatial division of labour in Nepal. The Journal of
Development Studies, 39(6), 23-66.
FerrĂŠ, C., Ferreira, F. H., & Lanjouw, P. (2012). Is There a Metropolitan Bias? The relationship
between poverty and city size in a selection of developing countries. The World Bank
Economic Review, lhs007.
Huang, J., & Bouis, H. (2001). Structural changes in the demand for food in Asia: empirical
evidence from Taiwan. Agricultural Economics, 26(1), 57-69.
Rischke, R., Kimenju, S. C., Klasen, S., & Qaim, M. (2015). Supermarkets and food
consumption patterns: The case of small towns in Kenya. Food Policy, 52, 9-21.
Pingali, P. (2007). Westernization of Asian diets and the transformation of food systems:
implications for research and policy. Food policy, 32(3), 281-298.
Stone, R. (1954). Linear expenditure systems and demand analysis: an application to the pattern
of British demand. The Economic Journal, 64(255), 511-527.

6

ESSAY 1: DIVERSIFICATION “BEYOND STAPLE FOODS”
IN THE DIETS OF POOR RURAL AND URBAN CONSUMERS
IN DEVELOPING EASTERN AND SOUTHERN AFRICA

1.1 Introduction
The past two decades of rapid per capita income growth and urbanization in many countries
of developing Eastern & Southern Africa (ESA)5 raise a host of questions regarding related impact
on consumer behavior. One question follows: has this rapid income growth had any meaningful
impact on food consumption patterns of households within ESA? The answer matters, because
what people eat and how the food is sourced carries profound implications for the structure of the
agrifood system, the level and type of needed public and private investments, the level and
distribution of employment, in addition to nutrition and health.
Bennett’s Law indicates that as incomes increase household diets will transition away from
grains and towards what we refer to as “Beyond Staple Foods” (BSF), which include meat, dairy,
fish, fresh produce, among other non-grain food products (Bennett, 1941). The observed trend of
rising household incomes is therefore expected to have increased the consumption of BSF and if
rising income continues the consumption of BSF is expected to continue to disproportionately rise.
The extent to which income has and will affect food consumption patterns is conditional on the
theory of supernumerary income, which suggests that households will alter their consumption
patterns only after required expenses have been met as it is the discretionary income households
use to acquire other utility generating goods (Stone, 1954). These theories combined indicate that
households at income levels below a supernumerary level of income would consume primarily

Countries include: Botswana, Burundi, Ethiopia, Kenya, Lesotho, Malawi, Mozambique,
Namibia, Rwanda, South Sudan, Swaziland, Tanzania, Uganda, Zambia, and Zimbabwe.
5

7

cheap grains that enable households to meet their basic food need for sufficient and reliable
calories, and that at incomes above this level, households would devote resources to good-tasting
food, novel food, and instrumental food6 (Satter, 2007). The empirically observed income level
that corresponds with a supernumerary income matters due to the expectation that food
consumption patterns will differ prior to and beyond attainment of this income level. Following
the theory of supernumerary income, conventional wisdom tends to view significant diet change
as a predominantly middle class (and higher) phenomenon, as the presence of discretionary income
is commonly used to denote the point of delineation between poor and middle class households in
ESA.
The food consumption patterns that impact the agrifood system are also affected by nonincome determinants, such as opportunity costs of time. Household opportunity cost of time would
vary by household income as well as participation in activities such as nonfarm employment,
which is also shown to affect food consumption patterns (Senauer et al., 1986; Kennedy &
Reardon, 1996; GĂłmez et al., 2013). Opportunity cost of time is further increased by household
settlement in urban areas that offer additional employment or entrepreneurial opportunities and a
generally faster pace of life (Bettencourt et al., 2007). The work of Huang & Bouis (2001) use data
from Taiwan to highlight these patterns by showing that general changes in “tastes and lifestyles”,
specifically in food consumption patterns, result from structural change such as the aforementioned
engagement in nonfarm employment and settlement in urban areas. Haggblade et al. (2010) noted
that households in Africa increasingly depend on nonfarm employment, additionally the 2014
revision of the World Urbanization Prospects by the United Nations cites significant growth in

6

Instrumental food is defined by Satter 2007 as food that contributes to desired physical, cognitive, or spiritual outcomes.

8

recent and the projected future urbanization rates for ESA, each highlighting growth in the drivers
of food consumption patterns in ESA that were found to be significant in Taiwan.
Huang & Bouis (2001) reason that urban settlement would affect food consumption
patterns by highlighting multiple differences between urban and rural settlements, including the
increased availability of food in markets and the broader exposure to different cultures. These
differences also exist to varying degrees across urban areas of varying population sizes, with cities
of greater populations providing increased availability of food and broader exposure to different
cultures. Thiele & Weiss (2003) have shown that city size (in terms of population) has a significant
effect on the diet diversity of German households, although similar analyses have not been
completed using data of low-income countries. City size has been shown to affect other lifestyle
patterns of households within low-income countries such as income and poverty – where income
is greater in larger cities, and medium and small cities lead to greater quantities of migrating
households escaping from poverty (FerrĂŠ et al., 2012; Christiaensen et al., 2013; BerdeguĂŠ et al.,
2015). The lifestyle patterns of rural households are also affected by urbanization, as many lifestyle
patterns such as employment and agricultural production are affected by a household’s distance to
urban areas (Fafchamps & Shilpi, 2003; Sharma, 2016).
The effects of city size and distance to cities are evident pertaining to multiple lifestyle
patterns, yet to our knowledge, their effects have not been estimated regarding food consumption
patterns. We will estimate the effects of city size and distance to cities on the share of purchased
food and the share of BSF in the total value of consumed food. We will also empirically test for
the level at which these food consumption patterns vary across income strata, according to the joint
application of Bennett’s Law and the theory of supernumerary income.

9

The paper proceeds as follows: Section two presents the conceptual model used to
articulate the decision process and hypotheses. Section three addresses the methodology, which
includes the context of the study, description of data, definitions, general and empirical models,
and estimation procedures. Section four reflects on the results and provides discussion. Lastly,
section five provides conclusions.

1.2 Conceptual Framework
Food consumption patterns are based on the desire of households to maximize their own
utility. Due to the heterogeneity of households, each household possesses a unique utility
function that is based on the consumer demand characteristics of individual and household
preferences for taste, energy density, and convenience; of social and individual norms and beliefs
regarding informal institutions, beauty, and lifestyle; and of health concerns including foods’
nutritional content and perceived safety. The maximization of these utility functions is
conditional on being able to afford these goods within a household’s income (m).
𝑀𝑎𝑥 𝑢(𝒙), 𝑠. 𝑡. 𝒑𝒙 ≤ 𝑚

(1.1)

𝑜𝑝𝑡𝑖𝑚𝑎𝑙 𝑏𝑢𝑛𝑑𝑙𝑒: 𝑥(𝒑, 𝑚)
Food to be consumed can either be purchased from food markets or sourced from
household’s own production. The prices included in the household’s utility maximization include
market prices as well as shadow prices for the production and preparation of all consumed food.
Bennett’s Law provides a relationship between household income and food consumption
patterns. Stone’s theory of supernumerary income further refines the expected relationship
between income and consumption patterns, by suggesting that the bulk of income induced diet
change would result from purchases by households with increasing levels of discretionary

10

income. Given that nearly two thirds of the population within ESA7 has incomes below the
international poverty line8, this theory suggests that relatively little diet change would have
resulted from the recent income growth. The conceptual framework will relax the assumption
that diet change will primarily occur among the non-poor. To facilitate discussion I will coin a
term, the subnumerary kink, which will refer to a change in the direction or speed of diet change
that would occur at income levels below those that would have normally associated with
increasing discretionary income. If I find a subnumerary kink in food consumption patterns, then
I would want to know the specific empirical configuration of the pattern: at what income levels
does the rise start?; does it build-up incrementally or in quantum steps? I would want also to
understand its drivers; what factors create a situation in which poor and rural households are
more likely than they previously were to change their consumption behaviors in this way. Put
another way, what factors might be driving this changes at lower levels of income (whether a
households resides in a rural- or urban area) than in the past? More formally, controlling for
income, what non-income factors might be driving these observed patterns?
To incorporate key non-income drivers, Rischke et al. (2010) conceptualized a
framework that combines consumer demand, as discussed above, with the food environment,
which highlight supply-side factors such as access, availability, price and market-induced
desirability of food. Access, not to be confused with economic access, refers to the cost to access
foods, which is subject to the density of retailers and the household’s time and expense of
reaching markets. Availability identifies the types and variety of foods offered by both formal
and informal retailers. Prices vary across products and across retailers (GĂłmez and Ricketts

Authors’ calculation using the World Bank’s PovcalNet database.
The poverty line is defined as $1.90 daily per adult equivalent expenditure, calculated with
purchasing power parity adjusted 2011 constant international dollars (World Bank, 2015)
7
8

11

2013). Market-induced desirability recognizes that the food environment affects the household
desire for products via shopping atmosphere and marketing.
Our conceptual approach expands the framework used by Rischke et al. to include the
effects that city size (defined by population) has on the urban environment that are exogenous to
the food environment and consumer demand. The effects of city size are relevant to urban and
rural consumption patterns alike. Urban consumption patterns will be directly affected by the
urban environment of the city in which the household is located, and the effect of urban
environment on rural households is expected to attenuate with increasing distance from the urban
area. The anticipated attenuation of these effects is due to increased cost of accessing cities and
the food markets therein as the distance from a household to cities increases (Fafchamps and
Shilpi, 2003; Sharma, 2016).

Figure 1-1: Conceptual framework

12

I suggest the following effects of a rising city size on the factors of urban environment
that affect aspects of the food environment and consumer demand as follows. First, based on
robust empirical patterns across Africa, incomes rise and poverty rates fall (FerrĂŠ et al., 2012,
World Bank, 2009; Christiaensen et al., 2013). Both should increase the purchasing of food and
the consumption of BSF foods.
Second, the size of the market increases more than proportionally to city size: this
follows directly from a positive relationship between city size and mean consumer incomes.
Increased size of market attracts investment in retail food markets, thereby improving physical
access to them, which favors shopping in them. Increased access to food retail markets expands
the variety of foods that is available.
Third, the cost of infrastructural investment rises, given higher costs of land – presuming
market mechanisms operate in the land market. Higher investment cost works asymmetrically
against investment in formal food retail: open air markets often arise spontaneously (informally)
with no explicit cost of land, and street vendors do not typically pay for the space they occupy.
These costs raise food prices in the more formal markets, reducing real incomes of consumers,
and incentivize the use of informal retail that may not have the same variety of food as formal
retailers. The higher land costs within and near larger cities provide a disincentive to urban and
peri-urban farming that could be used for own production of food.
Fourth, consumers are more exposed to modern advertising. Advertising increases the
likelihood that consumer preferences will align with the advertised foods that are commonly
non-staple foods made more available from modern food retailers.
Fifth, access to public and private motorized transport increases. Access to motorized
transport has an asymmetric effect on physical access to food: motorized transport improves

13

physical access for outlets such as supermarkets that are larger and fewer in numbers, but should
have little or no effect on access to other outlet types, which have developed historically to serve
largely foot-bound consumers. Access to transport thus favors the use of supermarket shopping,
which exposes to consumers to greater variety of foods.
Sixth, congestion increases increase the cost of moving a given distance by motorized
transport. However, alternative forms of transportation such as by foot or bicycle – both
negatively associated with household incomes – are relatively insensitive to city size. Congestion
thus asymmetrically affects physical access to food, it diminishes the access advantage conveyed
by motorized transport and so reduces the probability of shopping in a formal retail market.
Additionally, the increased cost of transportation increases the costs of cross-shopping9, reducing
the benefits that multiple markets offer in terms of available variety of food.
This framework does not provide consistent expected effects of city size on the key food
consumption patterns, but the generally anticipated trend leads me to hypothesize positive
cumulative effects of city size and negative cumulative effects of distance to cities on the shares
of purchased food and BSF foods in the household’s total value of consumed food.

1.3 Methods
1.3.1 Context
Developing Eastern and Southern Africa, which I term “ESA” here, has been rapidly
urbanizing. In the year 2000, twenty-one percent of the population in developing Eastern and

The practice of cross-shopping is where households shop at various locations. Cross-shopping
is common in China (Goldman 2000), Israel (Hino 2014) and in South Africa, where households
travel to locations away from their area of residence to acquire food (D’Haese and van
Huylenbroeck 2005, Strydom 2011).
9

14

Southern Africa lived in urban settlements; by 2015, the urbanization level was 26 percent, and
the United Nations Population Division10 projects that by 2050 the urban population share will
increase to forty-four percent. I will use the nations of Malawi, Tanzania, Uganda and Zambia to
represent this urbanizing region of Eastern Southern Africa. I define three city “types” based on
population: primary cities with populations greater than one million, secondary cities with
populations between one million and one hundred thousand, and tertiary cities with populations
below one hundred thousand. These countries include three primary cities, thirty-eight secondary
cities and over three hundred tertiary cities.
Two other characteristics of note regarding this region include the rising average per
capita income and a modernizing food system. Data utilized are from 2010, and in the decade
prior to 2010, the population weighted annual growth in per capita GNI11 of these four nations
was positive each year averaging near four percent since 2004. The relationship between
household income and consumption of food, as indicated by Bennett’s Law, requires income also
to be accounted for in this analysis. Daily per adult equivalent income is rising from low levels,
resulting in an average income of just above three dollars12 per day (Table 1-1). In spite of the
low average incomes, the “supermarket revolution” is occurring in Eastern and Southern Africa
(Reardon and Timmer, 2007). This revolution is modernizing the food system with the expansion
of supermarket and other formal food retail outlets, it is impacting how households buy and sell

World Urbanization Prospects: The 2014 Revision
11 World Bank data (http://data.worldbank.org/): GNI/capita, PPP (constant 2011 international $)
12 Expenditure data are converted to daily expenditure levels valued in purchasing power parity
(PPP) adjusted constant 2011 international dollars using historic exchange rates obtained from
XE.com and the conversion information for constant PPP valuation from worldbank.org.
Household expenditures are evaluated at per adult equivalent (AE) values, which are calculated as
one AE for each household member fifteen years of age or older, 0.75 AE for each child aged five
to fourteen, and 0.50 AE for each child younger than five years old.
10

15

food. Formal retailers typically enter markets by establishing their retail outlets near relatively
affluent areas within the greater market (Battersby and Peyton, 2014), which has resulted in
supermarkets directing significant investment into larger cities and lesser investments into
secondary and tertiary cities, thereby altering the food retail options across city size.

16

Table 1-1: National expenditure and population by expenditure strata and city size
Total Expenditure (per adult equivalent)
Malawi
National
Rural
Urban

Tertiary City

Uganda

Zambia

Pooled

2.48

3.03

3.51

3.13

3.11

(1.67)

(2.26)

(2.54)

(1.74)

(2.19)

1.97

2.38

2.78

1.78

2.37

(1.52)

(1.97)

(2.23)

(1.30)

(1.91)

5.31

4.91

6.10

5.66

5.39

(3.09)

(3.50)

(4.37)

(3.74)

(3.78)

7.85

8.11

7.41

7.82

(5.83)

(6.39)

(4.62)

(5.72)

Primary City
Secondary City

Tanzania

5.25

4.35

8.67

5.20

5.06

(3.14)

(3.34)

(4.56)

(3.68)

(3.43)

5.52

3.52

5.10

4.48

4.31

(2.84)

(2.76)

(3.77)

(2.99)

(3.13)

Represented Population: Total (`000s) and Cumulative Density by Expenditure Strata
$0.50- $1.00- $1.50- $2.00- $3.00- $5.00Total
< $0.50
> $10.00
1.00
1.50
2.00
3.00
5.00
10.00
101,923

2.1

13.8

30.2

45.2

67.4

86.2

96.7

100

77,041

2.6

16.9

36.3

53.2

76.6

93.4

99.1

100

24,882

0.4

4.1

11.2

20.4

38.8

64.0

89.2

100

6,126

0.0

1.5

3.4

6.6

17.0

41.7

77.4

100

7,178

0.2

3.7

11.3

22.1

42.7

67.4

90.8

100

11,578

0.8

5.6

15.2

26.6

48.0

73.6

94.4

100

Source: authors’ calculations using national household level surveys
Notes: Daily average expenditure levels per adult equivalent are presented in purchasing power parity adjusted constant 2011 international dollars. Primary font indicates the population weighted average.
Parentheses indicate the population weighted median.

17

1.3.2 Data
City level population data are obtained from http://www.citypopulation.de/, which assembles the
most recent available census data and uses country definitions of urban settlements. City
population data is paired with household level data to identify the population size where urban
households reside. Distance measurements for this analysis are approximated using distance
from center of rural constituency to the nearest primary, secondary and tertiary cities for the
Zambia data using the distance calculator at https://www.daftlogic.com/.
Household data are acquired from four nationally representative household surveys:
2010/2011 Malawi Integrated Household Survey (sample size of 12,271), 2010/2011 Tanzania
National Panel Survey (3,924), 2009/2010 Uganda National Panel Survey (2,975), and 2010
Zambia Living Conditions Monitoring Survey (19,397). These datasets were selected due to the
nations’ geographic proximity, similarities in national demographics, the timing of the surveys,
and the similarity of the data collection instruments. The enumerators of the surveys conducted
in-person interviews during 2010 and each interview collected household data on consumption,
employment, and other household characteristics.
Table 1-2 shows descriptive statistics of pooled data from the four datasets. The
expenditure data are converted into daily expenditure levels valued in purchasing power parity
(PPP) constant 2011 international dollars, using historic exchange rates obtained from XE.com
and conversion factors for constant PPP valuation from worldbank.org. Expenditures are
converted to per adult equivalent (AE) terms, and the daily expenditures per AE are used to
calculate total household expenditure, total food consumption, and totals for various food
consumption aggregates. Each survey collected data on the value of food consumption with a
seven day recall, save for the Zambia survey, which used a two- or four week recall depending

18

on the food item. The valuation of consumed own production varies by dataset. The Uganda and
Zambia surveys collected stated values of food consumed from own production13, while Malawi
and Tanzania collected quantities; I value these quantities using imputed household and
community prices of marketed food (see below for detail).

Table 1-2: Household descriptive characteristics
National
Total household expenditure
per AE
Nonfarm employment
Dependency ratio
Household adult equivalents
Hectares of farmed land
Own a gas or electric stove
Own a refrigerator

Pooled Data
Primary
Urban
Cities
5.39
7.82

Rural

Secondary
Cities
5.06

Tertiary
Cities
4.31
(3.13)

3.11

2.37

(2.19)

(1.91)

(3.78)

(5.72)

(3.43)

27.1

21.4

44.7

52.6

41.9

42.1

(20.0)

(0.0)

(50.0)

(50.0)

(40.0)

(40.0)

47.1

49.8

38.9

33.4

38.7

41.9

(50.0)

(50.0)

(40.0)

(33.3)

(40.0)

(42.9)

6.7

6.8

6.4

6.5

6.2

6.4

(5.6)

(5.6)

(5.4)

(5.5)

(5.2)

(5.4)

3.5

4.1

1.4

0.5

1.2

2.0

(0.9)

(1.2)

(0.0)

(0.0)

(0.0)

(0.2)

12.0

6.9

1.9

21.5

33.2

23.8

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

8.7

2.3

27.7

48.8

29.3

12.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

13.5

12.3

17.1

20.7

12.9

17.8

(14.0)

(14.0)

(17.0)

(17.0)

(13.0)

(17.0)

45.6

46.2

43.7

43.9

42.8

44.2

(43.0)

(44.0)

(41.0)

(41.0)

(41.0)

(41.0)

Female headed household

22.1

21.2

24.9

26.1

21.2

26.6

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Own a phone

55.4

46.7

82.4

93.3

84.3

75.5

(100.0)

(0.0)

(100.0)

(100.0)

(100.0)

(100.0)

9.5

10.7

5.7

10.3

4.7

3.9

(4.5)

(5.7)

(2.0)

(2.0)

(2.0)

(1.4)

Maximum education attained
within the household
Age of household head

Kilometers to Market
Own a car
Own a motorcycle
Own a bicycle

2.9

1.2

8.2

12.1

10.7

4.6

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

4.7

4.6

5.2

3.9

4.3

6.4

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

47.0

51.9

31.9

13.7

35.4

39.3

(0.0)

(100.0)

(0.0)

(0.0)

(0.0)

(0.0)

Source: authors’ calculations using national household level surveys
Notes: Daily average expenditure levels per adult equivalent are presented in purchasing power parity adjusted constant 2011 international dollars.
Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

1.3.3 Definitions
Shares of aggregated food values are calculated from the household expenditure values to
represent purchased share and share of BSF foods, shares of purchased food value and BSF food

The observations where Uganda reported farm-gate prices, I imputed market purchase prices
to represent the value of own production.
13

19

value, respectively, in total food expenditure. Purchased food value does not differentiate
between formal or informal food retailers, but represents the total value of household purchases
of food. Food items are defined as BSF foods if their primary ingredient is dairy, meat, fruit,
vegetables, nuts, vegetable oil, spices or sugar, and otherwise defined as low-income Bennett’s
Law (SF) food14.
The BSF aggregation is further disaggregated into 27 commodity based food aggregates
based on the food item’s primary input (Table 1-3). An example of a food item that is aggregated
into a commodity aggregate would be bread being aggregated into the wheat commodity
grouping because its primary ingredient is wheat flour.

Table 1-3: Dependent Variables
Share
Purchased Share
Beyond Staple Foods items
27 commodity based food aggregates
Rice
Maize
Wheat
Other Cereals
Cassava
Potatoes
Other Tubers
Pulses (with groundnuts)
Nuts

Definition
Share of purchased food value in total food expenditure
Foods whose primary ingredient is dairy, meat, fruit, vegetables,
nuts, vegetable oil, spices or sugar
Oil Crops
Oils & Fats
Staple Vegetables
Other Vegetables
Fruits
Plantains
Beef
Other Meat
Aquatic Products

Dairy Products
Poultry
Eggs
Sugar and Sweets
Spices
Non-Alcoholic Beverages
Alcoholic Beverages
Food Away From Home
Other Foods

City size classifies cities by population: primary cities have populations greater than one
million, secondary cities have populations below one million and above than 100,000, and
tertiary cities have populations below one hundred thousand.

14

In Uganda, matoke is a low cost staple food and is therefore categorized as a SF food.
20

1.3.4 General model
To estimate the effects of city size and distance to cities I follow a four part analysis.
Parts 1 and 2 establish the empirical configuration of the pattern of diet change across the income
distribution, while parts 3 and 4 explore drivers that include the effects of city size on food
consumption patterns.
First, I complete nonparametric analysis in the form of LOWESS curves. This analysis
imposes no structure on the data, allowing the data to reveal patterns of consumption across the
expenditure distribution. This analysis provides a first assessment of where in the expenditure
distribution (e.g., above or below the poverty line) changes in consumption patterns begin, and
whether the change is sudden and rapid or gradual.
Second, I formally establish breakpoints, and rates of change before and after them, by
testing for expenditure thresholds at which a significant change occurs in (a) the rate of change
or (b) the direction of the relationship between total expenditure and the consumption shares.
This analysis is not as flexible as the nonparametric analysis, but by imposing linear relations on
the food consumption patterns before and after an estimated expenditure breakpoint, it formally
estimates particular expenditure levels at which patterns change, and establishes rates of change
before the breakpoint (among lower income households) and above it (among higher income
households).
Third, I examine the drivers of consumption change by using regression analysis of Engel
curves to estimate the marginal effects of income and non-income determinants on food
consumption shares. The results of the nonparametric and threshold analysis are applied to select
an appropriate functional form of the Engel curve model for regression analysis.

21

Fourth, I use a mediation model to estimate the indirect effects that city size and distance
to city have on food consumption patterns that occur via other determinant variables. When
combined with the regression analysis of part 3, the indirect analysis provides results of the total
effects of city size and distance to cities.
All analyses are completed separately for rural and urban households and by country,
with descriptive analysis additionally highlighting city size differences that are analyzed in the
parametric analysis.

1.3.5 Empirical model
The empirical model for the nonparametric analysis and the expenditure threshold
analysis will estimate the marginal effect of income on the share of the value of a particular food
aggregate in the total value of food consumed, without controlling for any other effects. The first
two stages estimate Engel’s curves for each share of the food aggregates, with adult equivalent
total expenditure as the sole independent variable.
𝐷𝐹𝑜𝑜𝑑 𝐴𝑔𝑔𝑟𝑒𝑔𝑎𝑡𝑒 = 𝑓(𝑌 ∗ )

(1.2)

The models estimate the dependent variables: purchased share and the share of BSF foods, along
with the twenty-seven commodity food aggregates, which add more detail to patterns of
commodity based food consumption.
The third and fourth steps in the analysis use an Engel curve model to estimate the
marginal effects of income and non-income household variables, including city size and distance
to city, on each of the dependent variables in Table 1-3. The conceptual model posits that city
size influences the urban environment, which affects both the food environment and consumer
demand characteristics. To empirically specify the model, I include variables that capture as

22

many of the supply and demand side factors as possible. I include dummy variables for primary
and secondary cities that reflect city size differences, and in the rural analysis I include variables
for the distances to various city sizes. To the extent that variables are missing, or imperfectly
capture the concepts in the conceptual model, I use the conceptual model to interpret the
meaning of the city size variables and generate defensible conclusions.
The national datasets capture consumer demand characteristics relatively well, with
variables for household per capita expenditure along with demographic variables that proxy for
preferences, norms and beliefs, and health concerns. Data are less complete on the food
environment. The national datasets provide four variables that are relevant to accessing food:
own a car, own a motorcycle, own a bicycle, and distance to nearest market (for rural
households). Measures of availability, price and food desirability are not available. The national
datasets do not capture household level prices well due to the high prevalence of non-standard
units. Systematic differences across cities will be captured by city dummy variables in pooled
regressions.
Regarding the six urban environment factors within the conceptual model, I am able to
directly control for two of them: income and access to motorized transport. Both are
hypothesized to have positive effects on purchased share and the share of BSF food. The four
remaining factors are captured by city size dummies – market size, congestion, investment cost,
and advertising. Two of these, congestion and investment cost, are hypothesized to have
negative or mixed effects on purchased share and the share of BSF food. It would thus not be
surprising if I find that increasing city size decreases the shares of purchased or BSF food.
The effects of income, city size, distance to cities and the other selected variables are
estimated by the following Engel’s Curve Model:

23

𝐷𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒 = 𝑓(𝐴𝑐 ∗ , 𝐴𝑣 ∗ , 𝑃∗ , 𝐷 ∗ , 𝑌 ∗ , 𝑍 ∗ , 𝑊 ∗ , 𝐻 ∗ )

(1.3)

The dependent variables are purchased share and the share of BSF food, with commodity
based food consumption further estimated for each of the twenty-seven commodity based
aggregates (Table 1-3). The determinants of equation three represent the factors of food
environment and consumer demand characteristics, each with household variables that proxy for
their effects (Table 1-4). Each of the independent variables are described below.

Table 1-4: Determinants of demand in the Engel’s Curve Model

Ac*

Access

Own a car, Own a
motorcycle, Own a bicycle,
Distance to market (rural),
City size (urban), Distance
to cities (rural)

Disposable
Income

Z*

Individual and
Household
Preferences

Price

City size (urban), Distance
to cities (rural)

W*

Social and
Individual
Norms and
Beliefs

Desirability

City size (urban), Distance
to cities (rural)

H*

Health Concerns

Availability

City size (urban), Distance
to cities (rural)

P*

D*

Av*

Y*

Daily total expenditure per adult
equivalent, City size (urban),
Distance to cities (rural)
Nonfarm employment, Dependency
ratio, Adult equivalents, Farmed
hectares of land, Own a stove, Own a
refrigerator, Maximum level of
education, Age of household head,
Female head of household, City size
(urban), Distance to cities (rural)
Own a stove, Own a refrigerator,
Maximum level of education, Age of
household head, Female head of
household, Own a telephone, City
size (urban), Distance to cities (rural)
Maximum level of education, Female
head of household, Dependency
ratio, City size (urban), Distance to
cities (rural)

The variables that capture city size effects on urban environment are dummy variables for
residence in primary- and secondary cities (tertiary cities are excluded). For rural households,
the variables regarding access to markets are distance to nearest primary city, distance to nearest
secondary city, distance to nearest tertiary city. The distance (kilometer) variables are in
logarithmic values to account for diminishing marginal effects of distance. Greater distance from
cities reduces the effects of urban environment on food environment, consumer demand
characteristics and ultimately food consumption decisions. The attenuation of these effects is

24

caused by increased costs of accessing urban areas and the diminishing spillover of the effects of
the urban environment at greater distances from cities. I expect that the impact of distance will
be greater for larger cities, as the density of commercial activity within larger cities is likely to
result in greater commercial networks immediately surrounding the cities. The expansive
commercial network allows for households near large cities to maintain consumption patterns
more similar to those of urban residents, and as distance increases to large cities household food
consumption patterns will experience greater marginal change.
The dummy variables ownership of a car, ownership of a motorcycle, and ownership of a
bicycle are included to proxy for access to foods. Household ownership of a car would reduce the
time required in transportation to markets and would increase household ability to buy and
transport food purchased in bulk. Ownership of a motorcycle is expected to have a similar effect
as ownership of a car, but offering less benefit on the ability to shop in bulk. Ownership of a
bicycle also reduces the transportation costs associated with small purchases, but would likely
have little effect on household ability to shop in bulk. Distance to market is also included as an
independent variable to proxy for access to foods. The distance is measured in kilometers and
enters the estimation in a logarithmic form to account for diminishing effects of distance. The
greater the distance to the market the greater the shadow price of transportation that is required to
purchase food, which is expected to reduce purchased share and BSF food consumption by
limiting access to a variety of food. Tanzania and Zambia datasets include variables of household
identified distance from household to market, whereas the Malawi and Uganda datasets include
variables of distance from community to market that are collected by enumeration teams. Due to
certain distances being collected only at the community level, the distance to market variables
are only included in the analysis of consumption patterns of rural households.

25

Total daily household expenditure per AE is included in the model serving as a proxy for
income, a key factor of consumer demand. The specific form that expenditure enters the
estimation equations is explained in the estimation section.
Variables that proxy for individual and household preferences are included within the
model. Nonfarm employment is calculated as the percentage of working age15 adults employed
(including self-employment) in nonfarm activities. A household’s dependency ratio is calculated
by dividing the number of dependents by the total members within the household. Nonfarm
employment and dependency ratio both increase households’ opportunity cost of time,
incentivizing households to reduce their time allocated towards producing and preparing food.
The additional opportunity cost of time is expected to increase the consumption of purchased
food in lieu of consuming food from own production. The number of household adult
equivalents is the total number of adult equivalents within a household. Based on economies of
scale within households, the number of adult equivalents reduces the average household
opportunity cost of time. More household adult equivalents are therefore expected to reduce
purchased share.
Independent variables that proxy for the shadow prices of household ability to produce,
prepare and/or preserve food that contribute to individual and household preferences are land
farmed, own a gas/electric stove and own a refrigerator. Land farmed indicates the hectares of
land that a household farmed during the previous harvest season. This variable is included in
logarithmic form to account for diminishing marginal effects of the amount of land farmed on
food consumption. Larger areas of farmed land are more likely to be used for commercial
purposes, reducing the marginal effect on household consumption as farmed area increases.

15

Ages 15-64
26

Farming one’s own land is expected to lead to greater consumption of SF food from own
production, so reducing demand for purchased and BSF food. Ownership of a gas/electric stove
will reduce the preparation cost of food and thus reduce purchased food, which on average
requires less preparation. Ownership of a refrigerator will reduce the cost of food preservation,
enabling households to purchase greater quantities of perishable foods. The dummy variables for
ownership of a gas/electric stove or a refrigerator take values of one if affirmative and zero if
negative.
The final three variables that proxy for individual and household preferences are the
following: The maximum level of education within a household is an ordinal value based on the
level of education acquired that varies by country dataset that represents the household’s
maximum educational level. Increased education informs households on food safety and
cleanliness issues, influencing which food items to consume and from where to acquire their
food (Turrell and Kavanagh, 2006). The age of the head of the household is included in the
empirical model as elderly households are more likely to maintain traditional diets, therefore
older households will consume lower shares of purchased and BSF foods than younger
households. Finally, the gender of the head of the household could influence the consumption
patterns as consumption decisions have been shown to vary by gender (Smith, 2003). One
highlighted characteristic of female heads of households is a willingness to delay gratification,
leading to the expectation of less consumption of the primary food consumption patterns.
Variables included to proxy for social and individual norms include household ownership
of a telephone, in addition to ownership of a gas/electric stove, ownership of a refrigerator,
maximum level of education, age of household head, and female head of household, which have
already been described. A dummy variable for household ownership of a phone has the value of

27

one if the household owns a phone, zero otherwise. Ownership of a phone indicates a step
towards modernization and household potential willingness to modernize their food consumption
patterns.
The variables included to proxy for health concerns are maximum level of education,
female head of household, and dependency ratio, which have already been described

1.3.6 Estimation methods
LOWESS curve regressions (Cleveland, 1979; Cleveland and Devlin, 1988) are first
applied to observe the nonparametric relationship between the food budget share and the
household per adult equivalent total expenditure. The second application of LOWESS curves is
used to map the relationship between distance to various city types and rural food consumption
patterns. The nonparametric estimation is conducted via locally weighted scatterplot smoothing
(LOWESS), which is a robust form of local polynomial smoothing. The LOWESS procedure
does not impose functional form on the relationship between dependent and independent
variables. Rather it uses locally weighted least squares estimation on subsets of the data to form a
curve that represents the relationship between the variables. Compared with other forms of local
polynomial smoothing, this specific form of local polynomial smoothing is more resistant to
statistical outliers because it applies smaller local weights to estimates with large residuals.
Accounting for statistical outliers is valuable given the nature of household survey data. To
appropriately represent the sample population, the LOWESS analysis is conducted on expanded
data where duplicate observations are recognized according to population weights.
In the second step of the analysis, I test for expenditure thresholds at which the direction
or rate of change in the food consumption patterns could change. I apply a non-linear estimation
method to find the best fit of two linear least squared curves that are connected at the point that

28

optimizes the fit of the two curves. This is known as a piecewise regression (Serber and Wild,
1989) and takes the following form:
𝐸[𝑦|𝑥] = {

𝛽10 + 𝛽11 𝑥, 𝑥 ≤ 𝛼
𝛽20 + 𝛽21 𝑥, 𝑥 > 𝛼

(1.4)

𝑠. 𝑡. 𝛽10 + 𝛽11 𝛼 = 𝛽20 + 𝛽21 𝛼
In this analysis, household food budget shares of food item aggregates (y) are regressed
against household total expenditure per adult equivalent (x). This method provides inflection
point values and slopes of each curve around the inflection points with robust standard errors.
Statistically significant changes in the consumption patterns are identified when the piecewise
regression identifies a kink in the consumption curve at the inflection point.
The nonparametric analysis and the threshold analysis lead into the third step in the
analysis by identifying the functional form of the Engel curve that would provide the best fit for
the patterns of food consumption. In the base functional form of an Engel curve, a share of
consumption is the dependent variable and the independent variables include income and other
household variables. Multiple forms of this model have been proposed regarding the specific
form in which income relates to the dependent variable. Four of the options are; the linear
relationship suggested in Engel’s original work (1857), the linear relationship between log
income and the dependent variable as identified by Working (1943), Working’s model with the
addition of the inverse of income as proposed by Leser in 1963, or the functional form proposed
by Banks, Blundell and Lewbel (1996) that includes both log income and squared log income.
The inclusion of non-linear relationships between income and the consumption of goods has
been an improvement to the linear functional form as the non-linear relationships enable the
analysis of consumption patterns that exhibit differing marginal, potentially inverted, effects of
income on the relative shares of consumption of goods over strata of household income.

29

For the selection of the appropriate functional form of the Engel curve I follow a two-step
process. First, the results of the nonparametric and semi-parametric analyses are used to narrow
the options of functional form based on the general shapes of the observed patterns between total
expenditure and the highlighted expenditure shares and the potential for the functional form to fit
the observed patterns. Second, regressions using the reduced list of functional forms are
completed with the food expenditure shares serving as the dependent variables, and the estimated
fit of the regressions with the various functional forms are considered. Comparisons of adjusted
R2 or Pearson’s chi-squared values divided by degrees of freedom are made between like
dependent variables, and the functional form that most consistently provides the best fit is
selected. I will follow the recommendation of Subramanian and Deaton (1996) to treat total
expenditure as exogenous in food demand analysis.
The shares of consumption that serve as the dependent variables in Engel curve
regressions are bound between zero and one. The standard estimation method for such a
dependent variable is to convert it with a log-odds transformation and then to estimate with
ordinary least squares (Wooldridge, 2010). The data used in this analysis are primarily collected
in a one-week survey period, which typically results in many households with particular food
budget shares of zero or one hundred percent – values that cannot be used in the log-odds
transformation. Given the bounded nature of the dependent variable, the likelihood of reported
zero or one hundred percent budget shares, and the assumption of asymptotically normal
standard errors, the appropriate method of estimation in this specific equation is the fractional
probit model (Papke and Wooldridge, 1996 and 2008). This model estimates the marginal effects
of the predictors of a dependent variable that takes values of a closed set of zero to one.

30

Due to the selection of the fractional probit model, the second step of the selection of the
functional form of the Engel curve will be completed by comparing Pearson’s chi-squared values
divided by degrees of freedom from regressions using each functional form. This estimation
method, coupled with the above empirical model, enables the estimation of the marginal effects
of the independent variables that serve as the determinants for the food consumption patterns.
In the fourth step of the analysis a mediation model is used to test for indirect effects of
the variables city size and distance to city on food consumption patterns. City size and distance to
city (predictor variables) could affect other determinant variables (mediator variables) that then
affect food consumption patterns; this indirect effect would not be captured without the use of a
mediation model. The indirect effects of predictor variables are calculated by separately
multiplying the estimated effect of a mediator variable on a dependent variable by an estimated
effect of the predictor variable on the mediator. OLS is used to estimate the city size and distance
to city effects on the mediator variables, and bootstrapping is used to obtain standard errors for
the indirect effects (Wooldridge, 2010). The independent variables used in the regressions that
determine the effect of the predictor variables on the mediator variables are all of the
independent variables in the empirical model, save for the independent variable being treated as
the mediator variable. This estimation of indirect effects follows the methodology described in
Hayes and Preacher (2010).

1.4 Results
I first address descriptive statistics, followed by the analysis of LOWESS curves and
piecewise regressions to observe varying consumption patterns across income strata. Then I will
discuss the results of the full Engel’s Curve Model regression and the indirect effects of city size

31

on urban consumption patterns. I will return to LOWESS curve analysis to show patterns
between distance to cities and food consumption patterns, and I will conclude this section with
the indirect effects of distance to cities on food consumption patterns.

1.4.1 Descriptive statistics
Table 1-5 highlights patterns of purchased share and consumption of BSF foods, showing
average food consumption shares across country by national, rural settlement, and urban
settlement by city size. These descriptive statistics do not show causation, but are valuable in
observing preliminary patterns between food consumption and city size.

Table 1-5: Food budget shares of key food consumption patterns, aggregated country and
settlement
National
Rural
Urban
Primary City
Secondary City
Tertiary City

National
Rural
Urban
Primary City
Secondary City
Tertiary City

Purchased Share
Tanzania
Uganda
58.6
46.5

Pooled
54.8

Malawi
53.9

Zambia
61.8

(53.0)

(52.2)

(59.0)

(40.2)

45.6

47.9

48.8

39.5

46.6

(41.8)

(45.8)

(46.1)

(33.6)

(43.5)

(66.5)

83.4

87.3

86.7

71.2

90.5

(95.7)

(95.6)

(98.3)

(84.3)

(97.6)

92.9

97.6

82.2

95.1

(100.0)

(100.0)

(92.0)

(100.0)

90.0

90.0

91.5

76.8

90.7

(97.6)

(97.2)

(99.2)

(84.3)

(96.8)

74.2

77.6

77.6

66.6

85.1

(86.1)

(86.5)

(87.9)

(79.1)

(93.6)

Pooled
48.8

Beyond Staple Foods items
Malawi
Tanzania
Uganda
50.9
51.6
40.4

Zambia
56.0

(49.8)

(52.4)

(52.3)

(39.8)

45.5

49.2

48.2

37.1

(57.5)

52.4

(46.1)

(50.4)

(48.8)

(35.7)

(52.9)

59.1

60.1

61.4

52.1

62.9

(60.0)

(61.6)

(62.7)

(52.6)

(64.3)

65.5

70.7

56.3

64.5

(66.7)

(71.3)

(54.7)

(65.3)

60.3

60.5

59.9

53.6

62.5

(61.3)

(61.7)

(60.1)

(52.2)

(64.1)

55.0

58.5

56.9

50.4

61.9

(55.8)

(60. 8)

(56.5)

(51.8)

(63.5)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

32

Purchased share is strongly related to settlement type: in every country, the level of
purchased share approaches one hundred percent in primary cities and declines monotonically
with falling city size, being lowest in rural areas. A key distinction between rural and urban
households is that purchased share is almost twice as high among urban households as rural
households. However, purchased share in rural areas averages over 45% in every country other
than Uganda (40%), higher than what is observed in previous studies on purchased share in rural
developing Africa (von Braun et al., 1991; Garrett and Ruel, 1999). The patterns observed for the
shares of BSF food are consistent with the ordinal patterns of purchased share, but at a far
smaller magnitude in variation, specifically, rural consumption shares of BSF foods average
nearly eighty percent of the shares in urban areas.
Table 1-6 further disaggregates commodity based food consumption patterns and displays
pooled national data of eight of the twenty-seven commodity-based food aggregates – three lowincome Bennett’s Law (SF) foods and five BSF foods. Wheat, although a grain, is considered by
many households within ESA to be a luxury good and therefore has a consumption pattern
similar to that of BSF foods, i.e., monotonic decrease with falling city size. The other highlighted
aggregates follow patterns between rural and urban households that are expected based on what
is observed in Table 1-5. Monotonic relationships are less prevalent across city size for the
commodity aggregates using the pooled data, but there is more consistency with the expected
relationships between city size and the commodity aggregates when observing data that remains
disaggregated by country (Table 1-A-1). Food away from home16 represents a significant portion

Food away from is included in the commodity aggregates as food away from home is commonly
reported generally as food away from home without indication of specific base commodity.
16

33

of urban food share (13.4%), and this share is shown to rise monotonically across city size when
considering country specific data (Table 1-A-1).

Table 1-6: Food budget shares of select commodity aggregates, pooled data by settlement
Staple Foods
Pooled Data
National
Rural

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
6.3
3.4
2.8

18.1

2.5

6.1

Oils &
Fats
3.2

(13.1)

(0.0)

(0.0)

(2.5)

(4.9)

(0.0)

(0.0)

19.9

1.6

7.4

2.9

6.2

2.8

2.8

5.1

(15.5)

(0.0)

(0.6)

(2.1)

(4.5)

(0.0)

(0.0)

(0.0)

13.4

Maize

Wheat

Cassava

Food Away
from Home
7.1
(0.0)

Urban

12.5

5.1

2.1

4.2

6.8

5.4

2.8

(8.6)

(3.6)

(0.0)

(3.6)

(5.7)

(3.4)

(0.0)

(0.0)

Primary City

8.4

6.3

1.0

3.6

6.6

6.1

2.9

21.0
(13.3)

Secondary City
Tertiary City

(5.7)

(5.2)

(0.0)

(3.2)

(5.9)

(5.3)

(0.0)

14.6

6.2

1.3

5.3

7.9

5.1

2.4

9.2

(11.5)

(4.3)

(0.0)

(4.5)

(6.5)

(2.6)

(0.0)

(0.0)

13.3

3.8

3.2

3.9

6.2

5.1

3.1

11.9

(9.0)

(1.4)

(0.0)

(3.3)

(5.1)

(0.0)

(0.0)

(0.0)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

1.4.2 Nonparametric analysis of the relationship between expenditure and food consumption
LOWESS results (Figure 1-2, left panel) show that rural purchased share follows a
pattern of initial decline in the share of purchased food followed by a monotonic rise in each of
the four countries. This pattern resembles the functional forms of Engel curves for luxury goods
as suggested by Leser (1963) or Banks, Blundell and Lewbell (1997). These curves indicate that
a typical rural household at any income level does not purchase less than 40% of their consumed
food value in any of the observed countries other than Uganda. The initial high levels of
purchased share in rural households are likely caused by a lack of physical capital (especially
land) required to produce a desired level of own production in addition to need based cash
transfers.
Nonparametric analysis of the relationship between income and hectares of previously
cultivated land17 shows that the poorest rural households have the least land, suggesting that poor

17

Limited sample to households with less than five hectares of land.
34

households face high shadow prices for the production of food, supporting high shares of
purchased share at low income levels. Households with expenditures below one dollar consume
more food received as gifts or food in kind compared with the rest of the population (10% vs
6%), therefore it is logical that need based cash transfers that result in food purchases would be
the most prevalent at the lowest expenditure levels.

Figure 1-2: Nonparametric (LOWESS curve) analysis of select food consumption patterns – rural
data
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

Urban households within the same countries exhibit a similar estimated relationship
between income and purchased share with two exceptions (Figure 1-3, left panel). First, urban
Zambia does not display an initial decline, instead following an approximate log-linear
relationship as suggested by Working (1943) for luxury goods. The lack of an initial decline in
purchased share at low income levels suggests that Zambia has relatively less urban agriculture
than the other represented nations. Second, urban households exhibit higher levels of purchased

35

share than rural households. Considering pooled data within the estimated range of total
expenditure, the highest estimate of purchased share for rural households is less than the lowest
estimate of purchased share for urban households.

Figure 1-3: Nonparametric (LOWESS curve) analysis of select food consumption patterns –
urban data
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

For both rural and urban households the slopes of the fitted estimates of purchased share
are positive at all income levels above the poverty line, with the slopes of the pooled data
beginning to rise at expenditure levels below the poverty line, indicating a transition towards
purchased food at low income levels.
As expected, the LOWESS curves in Figures 1-2 and 1-3 (right panels) are consistent
with Bennett’s Law as BSF foods have a near linear relationship with increasing slopes starting
at low household expenditure levels. The disaggregated commodity based consumption patterns,
shown in Figures 1-4 (rural) and 1-5 (urban), show greater variation in the consumption patterns

36

of individual commodity aggregates than the aggregated pattern of BSF foods in Figures 1-2 and
1-3. Among rural households the food budget shares of the commodity aggregates for maize and
cassava (Figure 1-4, top left), and staple vegetables (top right) fall by one-third between daily
expenditures of $0.50 and $2.00. Over the same income range in urban areas, the estimated food
budget share of cassava declines over two thirds (Figure 1-5, top left). Commodity aggregates
with rising food budget shares among poor households include wheat, beef, dairy and food away
from home. Food away from home exhibits the greatest increase in consumption share across
income, again highlighting household demand for attributes such as time saving as expenditures
rise. These patterns suggest that households follow the food need hierarchy that indicated that
households initially satisfy their need for calories (cheap calories in maize and cassava) and then
address higher order needs such as good tasting and novel food, which are satisfied with the
consumption of higher valued foods, i.e. meat and food away from home.

37

Figure 1-4: Nonparametric (LOWESS curve) analysis of select commodity aggregates – pooled
rural data
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

38

Figure 1-5: Nonparametric (LOWESS curve) analysis of select commodity aggregates – pooled
urban data
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

In sum, LOWESS curves provide fitted estimates that indicate changing food
consumption patterns across expenditure strata, specifically among the poor. The slopes of the
consumption patterns estimated by the LOWESS curves vary across country and settlement, but
the general pattern of the primary highlighted food consumption patterns is one of initial decline
in food budget share followed by a monotonic rise that begins below the poverty line.

39

1.4.3 Expenditure Threshold Analysis
The expenditure threshold analysis tests for the presence of statistically significant
expenditure thresholds that indicate changing directions or rates of change in the consumption
patterns of purchased share or the share of BSF foods. Expenditure threshold analysis indicates
that patterns of rural and urban purchased share vary across income levels, and that the steady
rise of purchased share begins among poor households.
The results for rural purchased share in Table 1-7 show that in three of four countries
there is a statistically significant pattern of declining food budget shares of purchased food
followed by a rise in food budget shares, and that three of four of the countries observe the
change in direction below or near the poverty line. The pooled data from the four countries show
the expenditure threshold of purchased share to occur at $1.56.

Table 1-7: Expenditure thresholds and slopes prior to and after expenditure thresholds in the
consumption patterns of purchased share and Beyond Staple Foods items – rural data
Countries
Malawi
Tanzania
Uganda

Purchased Share
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$1.31
-0.084
0.046
(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

$1.14

-0.188

0.042

1.312

0.004

0.055

(0.000)

(0.011)

(0.000)

(0.001)

(0.904)

(0.000)

$2.09

-0.046

0.056

0.041

(0.000)

(0.046)

(0.000)

(0.000)

Zambia
Pooled

Beyond Staple Foods items
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
1.950
0.103
0.029
(0.000)

0.059

1.568

0.026

0.015

(0.000)

(0.103)

(0.035)

(0.026)

$1.56

-0.069

0.043

1.297

-0.005

0.037

(0.000)

(0.000)

(0.000)

(0.000)

(0.789)

(0.000)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

In urban households the slopes of the pattern of purchased share are positive both before
and after the expenditure breakpoints when they are statistically significant (Table 1-8). This
indicates that urban households prefer greater shares of purchased food at all income levels,

40

which is expected because of high shadow prices of agricultural production for households living
in urban settlement due the limited access to farmland.

Table 1-8: Expenditure thresholds and slopes prior to and after expenditure thresholds in the
consumption patterns of purchased share and Beyond Staple Foods items – urban data
Countries
Malawi
Tanzania

Purchased Share
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$2.64
0.081
-0.001
(0.000)

(0.000)

(0.929)

(0.000)

(0.001)

$1.58

-0.075

0.063

$1.53

-0.022

0.045

(0.001)

(0.461)

(0.000)

(0.008)

(0.740)

(0.000)

0.064

$0.63

-1.050

0.055

(0.003)

(0.000)

(0.028)

(0.000)

Uganda
Zambia
Pooled

Beyond Staple Foods items
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$2.01
0.069
0.025
(0.000)

$2.02

0.087

0.017

$1.85

0.047

0.015

(0.000)

(0.000)

(0.000)

(0.000)

(0.002)

(0.000)

$1.67

-0.019

0.048

$1.42

-0.010

0.035

(0.010)

(0.737)

(0.000)

(0.036)

(0.845)

(0.000)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

The expenditure threshold analysis for the share of BSF foods leads to two results of
note. First, the estimates affirm Bennett’s Law and indicate that low incomes households are
rapidly moving towards BSF foods, which is shown with greater positive slopes prior to the
expenditure breakpoints than beyond them in all but one case where both slopes were statistically
significant – the exception has an expenditure breakpoint of $0.63 (Table 1-8). These results
suggest that a supernumerary income is not a necessary condition for household consumption
patterns to exhibit Bennett’s Law.
Second, the patterns of the disaggregated commodity aggregates further support the nonnecessary condition of supernumerary income for a shifting consumption towards BSF foods
(Tables 1-9 and 1-10). The select commodity aggregates continue to show that the bulk of the
transition within BSF food consumption occurs among poor households. Fifteen of the 18 nonpooled statistically significant pairs of slopes representing commodity food consumption patterns
have the same direction and are of greater absolute value before the expenditure breakpoints than

41

beyond the breakpoints: negative slopes for maize, cassava and staple vegetables, positive for
wheat, oil and fats, beef, dairy and food away from home.

Table 1-9: Expenditure thresholds and slopes prior to and after expenditure thresholds for the
aggregates of select commodity based aggregates – rural data
Countries
Malawi
Tanzania

Expenditure
Breakpoint
$1.95

Pooled

Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Countries
Malawi
Tanzania

Pooled

Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Expenditure
Breakpoint
$2.35

Rice
Slope Before
Breakpoint
0.016

Slope After
Breakpoint
0.008

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

$1.38

-0.026

-0.043

(0.184)

(0.404)

(0.000)

-0.007

-0.008

$4.49

0.004

0.061

(0.284)

(0.088)

(0.000)

(0.000)

(0.091)

$1.32

-0.051

-0.022

$1.46

0.025

0.013

(0.001)

(0.013)

(0.000)

(0.000)

(0.000)

(0.000)

$2.13

-0.074

-0.026

$1.25

0.008

0.006

(0.000)

(0.000)

(0.000)

(0.026)

(0.000)

(0.000)

Expenditure
Breakpoint
$1.40

Cassava
Slope Before
Breakpoint
0.008

Slope After
Breakpoint
-0.005

Expenditure
Breakpoint
$2.31

Oils & Fats
Slope Before
Breakpoint
0.013

Slope After
Breakpoint
0.001

0.007
(0.000)

(0.000)

(0.084)

(0.000)

(0.000)

(0.000)

(0.339)

$1.64

-0.073

-0.006

$1.96

0.002

-0.001

(0.000)

(0.000)

(0.016)

(0.044)

(0.407)

(0.670)

$2.38

-0.057

-0.013

$2.69

0.006

-0.003

(0.000)

(0.000)

(0.121)

(0.000)

(0.060)

(0.071)

$1.94

-0.029

-0.008

$3.30

0.002

0.005

(0.000)

(0.000)

(0.385)

(0.008)

(0.205)

(0.290)

$1.96

-0.031

-0.009

$1.98

0.004

0.000

(0.000)

(0.000)

(0.002)

(0.000)

(0.014)

(0.561)

Expenditure
Breakpoint

Beef
Slope Before
Breakpoint
0.007

Slope After
Breakpoint

Staple Vegetables
Expenditure Slope Before Slope After
Breakpoint
Breakpoint
Breakpoint
$1.29
-0.061
-0.012
(0.000)

(0.000)

(0.000)

$1.80

-0.029

-0.009

$1.70

0.014

0.003

(0.000)

(0.000)

(0.000)

(0.002)

(0.016)

(0.242)

0.001

0.001

$2.42

0.016

0.007

(0.486)

(0.140)

(0.003)

(0.000)

(0.193)

$1.55

0.006

0.001

$0.86

0.013

0.005

(0.003)

(0.034)

(0.391)

(0.011)

(0.073)

(0.004)

$1.79

-0.024

-0.006

$2.41

0.013

0.005

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.091)

Expenditure
Breakpoint
$1.53

Dairy
Slope Before
Breakpoint
0.004

Slope After
Breakpoint
0.007

(0.001)

(0.000)

(0.000)

(0.000)

(0.000)

(0.010)

$1.96

0.010

0.001

$2.01

0.012

0.037

(0.028)

(0.129)

(0.787)

(0.000)

(0.197)

(0.000)

$2.27

0.020

0.003

(0.000)

(0.000)

(0.461)

$2.00

0.005

0.004

$3.09

0.000

0.002

(0.414)

(0.000)

(0.116)

(0.000)

(0.791)

(0.050)

$2.20

0.013

0.003

$2.86

0.014

0.025

(0.000)

(0.000)

(0.170)

(0.001)

(0.000)

(0.003)

Uganda
Zambia

Slope After
Breakpoint
-0.038

(0.000)

Uganda
Zambia

Maize
Slope Before
Breakpoint
-0.140

(0.000)

Food Away From Home
Expenditure Slope Before Slope After
Breakpoint
Breakpoint
Breakpoint
$1.78
0.007
0.002

0.006
(0.004)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

42

Table 1-10: Expenditure thresholds and slopes prior to and after expenditure thresholds for the
aggregates of select commodity based aggregates – urban data
Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Countries

Expenditure
Breakpoint
$1.38

Maize
Slope Before
Breakpoint
-0.246

Slope After
Breakpoint
-0.033

Expenditure
Breakpoint
$2.57

Rice
Slope Before
Breakpoint
0.030

Slope After
Breakpoint
0.000

(0.000)

$1.40

(0.000)

(0.000)

(0.000)

(0.000)

(0.956)

0.031

-0.036

$2.39

0.014

0.000

(0.004)

(0.607)

(0.000)

(0.000)

(0.011)

(0.947)

$3.12

-0.051

0.029

$0.90

-0.019

0.010

(0.000)

(0.001)

(0.048)

(0.003)

(0.322)

(0.000)

$2.06

-0.066

-0.028

$2.49

0.030

0.009

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.014)

$1.99

-0.048

-0.031

$2.51

0.015

0.003

(0.062)

(0.020)

(0.000)

(0.000)

(0.000)

(0.328)

Expenditure
Breakpoint

Cassava
Slope Before
Breakpoint
-0.002

Slope After
Breakpoint
-0.002

Expenditure
Breakpoint
$1.61

Oils & Fats
Slope Before
Breakpoint
0.040

Slope After
Breakpoint
0.003

(0.034)

(0.024)

(0.000)

(0.000)

(0.251)

$1.77

-0.040

-0.002

$1.76

-0.008

0.000

(0.000)

(0.246)

(0.222)

(0.015)

(0.353)

(0.979)

$1.96

-0.049

0.006

$2.31

0.009

0.000

(0.000)

(0.099)

(0.491)

(0.002)

(0.102)

(0.934)

$1.59

-0.034

-0.003

$2.03

-0.007

-0.001

(0.000)

(0.000)

(0.000)

(0.000)

(0.046)

(0.518)

$1.81

-0.034

0.001

-0.001

-0.001

(0.000)

(0.055)

(0.636)

(0.359)

(0.306)

Beef
Slope Before
Breakpoint
0.004

Slope After
Breakpoint
0.014

Malawi
Tanzania
Uganda
Zambia
Pooled

Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Countries
Malawi
Tanzania

Staple Vegetables
Expenditure Slope Before
Slope After
Breakpoint
Breakpoint
Breakpoint
$1.97
-0.041
-0.011
(0.000)

(0.006)

(0.000)

(0.007)

(0.661)

(0.000)

$0.74

0.068

-0.006

$1.47

-0.026

0.009

(0.007)

(0.346)

(0.007)

(0.002)

(0.348)

(0.014)

$1.80

-0.006

0.005

$1.99

0.026

0.007

(0.037)

(0.679)

(0.051)

(0.057)

(0.231)

(0.304)

$1.30

0.003

-0.006

$2.00

0.016

0.004

(0.028)

(0.763)

(0.000)

(0.000)

(0.000)

(0.000)

$0.71

0.040

-0.007

$1.38

-0.007

0.009

(0.018)

(0.469)

(0.000)

(0.036)

(0.728)

(0.000)

Expenditure
Breakpoint
$1.62

Dairy
Slope Before
Breakpoint
0.001

(0.000)

(0.521)

(0.000)

(0.638)

(0.519)

$2.54

0.006

-0.002

$1.48

-0.008

0.046

(0.052)

(0.440)

(0.557)

(0.018)

(0.863)

(0.000)

0.010

0.012

$0.43

-3.746

0.006

(0.171)

(0.003)

(0.000)

(0.000)

(0.548)

$2.90

0.007

0.004

$1.35

0.001

0.002

(0.001)

(0.000)

(0.025)

(0.158)

(0.312)

(0.032)

$1.22

0.017

0.005

$1.41

-0.012

0.024

(0.003)

(0.004)

(0.002)

(0.018)

(0.745)

(0.000)

Uganda
Zambia
Pooled

Expenditure
Breakpoint
$1.47

Food Away From Home
Expenditure Slope Before Slope After
Breakpoint
Breakpoint
Breakpoint
0.001
0.001

Slope After
Breakpoint
0.010

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

Overall, the semi-parametric expenditure threshold analysis highlights three key results.
First, there is evidence of statistically significant changes in the direction and/or slope of the
shares of food aggregates relative to total expenditure levels. Second, if the expenditure

43

breakpoint signifies a change in the direction of the food consumption pattern, the breakpoint is
commonly below the poverty line. Third, if the expenditure breakpoint signifies a change in the
slope of the food consumption pattern, the slope is commonly of greater absolute value at
expenditures below the expenditure breakpoint than above. These findings indicate that there is
significant change in food consumption patterns among poor households in ESA, and that change
is often more rapid for the poor than it is for the non-poor.

1.4.4 Engel’s Curve Model regression analysis
The fractional probit regressions of the Engel curves build upon the previous analyses by
better isolating the impact of income on food consumption patterns and the effects of the various
non-income determinants. Following the two-step process for the selection of the functional form
of the Engel curve, the functional form suggested by Banks, Blundell and Lewbel (1997) was
selected. The nonparametric and semi-parametric analyses highlight the need for the functional
form to allow for changing slopes of the food consumption patterns, narrowing the functional
form options to those suggested by Banks et al. (1997) and Leser (1963). Fractional probit
regressions of the highlighted food consumption patterns with dependent variables specific to
these two functional forms resulted in thirteen of the twenty-four regressions having a superior
goodness of fit with the Banks et al. suggested functional form. The selected functional form
includes non-income independent variables and two independent variables for expenditure: log
total household expenditure and the squared log of total household expenditure. As expenditure
is included in the model as two separate variables, the average partial effects of expenditure are
calculated using the following equation:

44

𝜕
𝜕(𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒)

1

= [𝑎 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒 + 2𝑏

𝑙𝑛(𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒)
𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒

] ∗ Ε𝜙[𝑋𝛽]

(1.5)

where 𝑎 equals the marginal effects of log expenditure and 𝑏 equals the marginal effects of
squared log expenditure, and the scaling factor in the right of the equation is equal to the average
of the normal densities of the predicted values of the dependent variable.
The regression analysis portion of the results section will first consider the effects of
expenditure. Then I will observe non-expenditure drivers of consumption, including the direct
and the indirect effects of city size on urban consumption patterns. Then I will briefly return to
LOWESS analysis to graphically represent the relationship between distance to cities and rural
food consumption patterns, and conclude by showing the direct and indirect effects of distance to
cities on food consumption patterns.
The estimated effects of expenditure provide three results of note. First, the average
partial effects of expenditure on purchase share are positive in seven of eight of rural or urban
estimates, but in only one of two significant marginal effects (Tables 1-11 and 1-12). It is
surprising that this was not more statistically significant given the clear patterns observed in the
nonparametric and threshold analyses, but it speaks to the importance of non-income
determinants on the share of purchased food. The estimated effects of expenditures on purchased
share graphed in Figures 1-6 and 1-7 resemble the patterns of the previous analyses, consistent
with the correlation of expenditure and purchase share.

Table 1-11: Average partial effects of expenditure for key food consumption patterns – rural data
Rural
Household total
expenditure per
AE

Malawi
-0.015
(0.000)

Purchased Share
Tanzania
Uganda
0.004
0.007
(0.525)

(0.214)

Zambia
0.007

Malawi
0.050

(0.150)

(0.000)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

45

Beyond Staple Foods items
Tanzania
Uganda
Zambia
0.040
0.023
0.004
(0.000)

(0.000)

(0.269)

Figure 1-6: Estimated effects of expenditure on purchased share and the consumption of Beyond
Staple Foods items – rural data

Table 1-12: Average partial effects of expenditure and city size for key food consumption
patterns – urban data
Urban
Household total
expenditure per AE
Household located in a
primary city
Household located in a
secondary city

Malawi
0.001
(0.756)

Purchased Share
Tanzania
Uganda
0.013
0.004

Zambia
0.005

Malawi
0.008
(0.000)

(0.001)

(0.296)

(0.296)

0.090

0.063

0.037

Beyond Staple Foods items
Tanzania
Uganda
Zambia
0.025
0.015
0.009
(0.000)

(0.000)

(0.000)

0.017

-0.029

-0.010

(0.000)

(0.048)

(0.000)

(0.172)

(0.130)

(0.058)

0.090

0.051

0.036

0.022

0.007

-0.017

-0.035

-0.008

(0.000)

(0.000)

(0.349)

(0.000)

(0.445)

(0.208)

(0.229)

(0.087)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

46

Figure 1-7: Estimated effects of expenditure on purchased share and the consumption of Beyond
Staple Foods items – urban data

Second, the average partial effects of expenditure on BSF food are positive and
statistically significant in seven of eight estimations, consistent with Bennett’s Law (Tables 1-11
and 1-12). When these estimated effects are represented across the expenditure strata their
marginal effects are positive at all expenditure levels above one dollar (Figures 1-6 and 1-7).
These findings further support that supernumerary income is not a condition for households to
increase their share of BSF food.
Third, the statistically significant average partial effects of expenditure on the selected
commodity aggregates are consistently negative for maize, cassava and staple vegetables;
consistently positive for wheat, beef, dairy and food away from home; and mixed for oils and
fats. These consistent patterns highlight the heterogeneity of effects within the greater Bennett’s
Law aggregates.

47

Three findings regarding the effects of city size on urban consumption patterns are of
note. First, larger cities positively influence households to purchase a larger share of the food
they consume. The marginal effects of households living in primary cities and households living
in secondary cities on purchased share were estimated and six of the seven estimates18 from the
four countries were positive and statistically significant. As shown with the data from Tanzania
and Zambia, the marginal effect of a household living in primary city on purchased share was
twice the magnitude of a household living in a secondary city, indicating a positive relationship
between city size and purchased food consumption. These findings are consistent with the
conceptual model, where larger cities improve the access to markets do to the increased presence
of markets ad alter desirability to shop in a formal market due to the increased advertising in
larger cities.
Second, city size does not significantly affect the consumption share of BSF food. Stage
et al. (2010) found that although urban households consumed greater shares of BSF food than
rural households, this difference is not driven by urbanization but rather by other household
variables such as income. The findings here echo Stage et al. (2010). The primary effect of city
size on the consumption of BSF foods is via increased income, and as this effect is controlled for
with household expenditure the remaining effect of city size is insignificant.
Third, further disaggregation of BSF foods does reveal multiple city size effects. Based
on the consumption patterns of Table 1-6, it was expected that large cities would positively
impact the consumption of wheat, staple vegetables, beef, dairy and food away from home, and
negatively affect the consumption of maize and cassava. Table 1-12 shows that when controlling

Note that no city in Malawi is large enough to qualify as “primary”, resulting in seven, rather
than eight, estimated coefficients.
18

48

for various consumption determinants, the impacts of city size on these food consumption
patterns follow expectations with the exceptions of beef and dairy, with limited and contradicting
statistically significant marginal effects. These estimated effects indicate that although income
and city size do not uniformly affect all Bennett’s Law foods, income and city size do affect the
consumption of certain commodities.
Five additional effects of non-expenditure determinant variables are of note, with their
marginal effects shown in Tables 1-A-2 and 1-A-3 of the appendix. First, consistent with
literature that indicates that nonfarm employment increases the opportunity cost of time and
thereby influences consumption patterns (Senauer et al., 1986; Kennedy and Reardon, 1996), the
marginal effects of nonfarm employment are positive and significant in urban Tanzania and
urban Zambia for both purchased share and share of BSF foods. This finding highlights the
significant effects on nonfarm labor on rural households’ decisions to engage with markets.
Second, dependency ratio, another variable related to opportunity cost of time that would
contribute to the consumer demand factor of individual and household preferences, generally has
a negative effect on purchased share and BSF food. Third, the amount of land farmed negatively
affects the shares of purchased food as it directly provides the opportunity to consume food from
own production.
Fourth, ownership of a telephone has a strongly significant positive effect on purchase
share and share of BSF foods in more than half of the national estimates. This finding is in
agreement with the hypothesis that ownership of a phone reveals that individual and household
preferences are more open to modern lifestyle patterns. Fifth, the effect of the age of the
household head is negative and significant on purchased share, affirming that elderly households
are less open to modern lifestyle patterns.

49

The fractional probit model estimates the direct effects of city size on household
consumption patterns, but as discussed earlier, city size also affects other determinant variables
of consumption such as household income. The mediation model is used to estimate the indirect
effects of city size on consumption patterns through household total expenditure, nonfarm labor,
the amount of land farmed by the household and the maximum education attained within the
household. These three variables are selected as mediator variables for the following reasons:
difference in city size could impact household expenditure, it could expose household members
to various nonfarm employment opportunities, it could vary household access to farmland, and
larger cities could offer greater opportunities for education. There were no statistically
significant indirect marginal effects of city size found using the Malawi data, nor were there
significant indirect effects via education, but the remaining countries produced the following
three findings of note (Table 1-13).

Table 1-13: Indirect effects of urban household settlement within primary and secondary cities,
estimates by country
Mediator
Variable
Household total
expenditure per
AE
Nonfarm
employment

Food
Consumption
Pattern
Purchased Share
Beyond Staple
Foods items
Purchased Share
Beyond Staple
Foods items

Hectares of
farmed land
(log)

Purchased Share

Maximum
education
attained within
the household

Purchased Share

Beyond Staple
Foods items

Beyond Staple
Foods items

Malawi
Secondary
Cities
-0.004

Tanzania
Primary Secondary
Cities
Cities
0.056
-0.003

Uganda
Primary Secondary
Cities
Cities
0.043
0.029

Zambia
Primary Secondary
Cities
Cities
0.018
0.001

(0.657)

(0.014)

(0.689)

(0.085)

(0.480)

(0.000)

-0.006

0.106

-0.005

0.050

0.034

0.018

(0.568)

0.001

(0.546)

(0.000)

(0.644)

(0.031)

(0.462)

(0.000)

(0.562)

0.000

0.008

0.006

0.004

0.000

0.005

0.001

(0.650)

(0.008)

(0.086)

(0.497)

(0.857)

(0.000)

(0.010)

0.000

0.006

0.005

0.003

0.000

0.002

0.000

(0.696)

(0.004)

(0.081)

(0.363)

(0.822)

(0.001)

(0.050)

0.001

0.010

0.009

0.012

0.019

0.014

0.008

(0.321)

(0.001)

(0.016)

(0.011)

(0.007)

(0.000)

(0.000)

0.000

0.002

0.002

0.003

0.005

0.001

0.001

(0.740)

(0.354)

(0.385)

(0.301)

(0.240)

(0.161)

(0.150)

0.000

0.001

0.001

0.009

0.012

0.000

0.000

(0.599)

(0.517)

(0.507)

(0.158)

(0.279)

(0.844)

(0.747)

-0.001

-0.001

-0.001

0.000

0.000

0.000

0.000

(0.218)

(0.418)

(0.421)

(0.847)

(0.866)

(0.854)

(0.776)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

50

First, primary cities have a positive indirect effect on the consumption of purchased and
BSF food via total household expenditure. Household expenditure is positively correlated with
these food consumption patterns and the mediation model confirms that primary cities positively
impact household expenditure. The positive indirect effect of city size on BSF food is important,
given that city size did not have a significant direct effect on BSF food, by indicating that the
type of urbanization does, albeit indirectly, influence the consumption patterns of these foods as
indicated by Satterthwaite et al. (2010).
Second, nonfarm employment is also a significant mediator variable in the estimation of
indirect effects. As observed in Tanzania and Zambia, larger cities have a positive effect on
nonfarm employment, which, via increased opportunity cost of time, increases household
demand for purchased share and the consumption of BSF food. These indirect effects amplify the
direct effects of city size on purchased food, while indicating a second avenue in which city size
has an indirect effect on the consumption of BSF food.
The final indirect effect of urban household city size on food consumption patterns
relates to urban farming – both farming in urban areas and farming by urban households on rural
land. Rural households engage in farming more often than urban households, but urban
households continue to farm. The percentage of households that engage in farming has an
inverse relationship with city size (primary city 9%, secondary city 29%, tertiary city 55%) 19
such that city size is a statistically significant determinant of the amount of land farmed by
households. Primary and secondary cities have positive indirect effects on purchased share

Average land farmed by settlement type: Primary 0.5ha, Secondary 1.2ha, Tertiary 2.0ha,
Rural 4.1ha.
19

51

through their negative effect on the amount of land farmed. The use of the mediator model
highlights these three indirect effects that are not accounted for when testing for direct effects.
Both non-parametric and parametric analysis show that distance to city affects the
consumption patterns of rural households. Figures 1-8 and 1-9 display the LOWESS curve
estimates of the relationships between distance to city and the primary food consumption
patterns. Each figure exhibits inverse relationships with declining marginal effects between
distance to cities and the food consumption aggregates20. Along with these negative
relationships, it is shown that proximity to larger cities has a larger effect on the consumption
patterns than distances to smaller cities. Both of these observations are consistent with previous
findings of new economic geography in that increased distance from urban centers reduces the
interaction of rural households with urban markets, and larger markets, which would be more
prevalent in larger cities, offer a greater variety of marketed goods that increase the capabilities
for households to diversify their consumption patterns (Renkow 2007).

The anomalous relationship between distance to tertiary cities and the consumption of highincome Bennett’s Law food is driven by a small sample of households. Seventy-five percent of
households are within seventy-five kilometers of a tertiary city, and it is the furthest ten percent
of households that are driving the positive relationship between BSF food and distance to city.
20

52

Figure 1-8: Locally weighted scatterplot smoothing curves representing the typical purchased
share relative to household distance to city
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

53

Figure 1-9: Locally weighted scatterplot smoothing curves representing the typical food budget
shares of the consumption Beyond Staple Foods items relative to household distance to city

While the LOWESS analysis reveals correlations between the spatial effects and food
consumption patterns, it does not control for the effects of other drivers of demand. Engel curve
analysis provides the ability to estimate the effects of distance to city while controlling for
multiple non-spatial determinant variables, leading to the observation of the following three
patterns.
First, both of the key food consumption patterns are negatively impacted by distance to
city (Table 1-14). Consistent with the von ThĂźnen model, distance to markets increase
transaction costs of acquiring the various foods from urban markets, limiting household ability to
optimize their preferred bundle of foods to consume (Nelson, 2002). Owning a car reduces the

54

impact of the additional transaction costs, but with only one percent of rural households owning
a car, the distance to urban areas is a significant limiting factor of consumption.

Table 1-14: Average partial effects of expenditure and distances to various city sizes for key food
consumption patterns – rural Zambia
Rural Zambia
Household total expenditure per AE

Purchased
Share
0.007

Beyond Staple
Foods items
0.004

(0.150)

(0.269)

Distance to primary city (log)

-0.058

-0.024

(0.000)

(0.000)

Distance to secondary city (log)

-0.007

-0.012

Distance to tertiary city (log)

(0.118)

(0.002)

-0.008

-0.004

(0.000)

(0.020)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value

Second, the distance to a primary city has a far stronger impact on food consumption
patterns than the distances to smaller cities (Table 1-14). Table 1-6 showed that households
within primary cities consume greater shares of purchased food and BSF foods, so it follows that
households near this city size would also have higher levels of the particular food consumption
pattern. These findings are consistent with the hypothesized relationship between distance to
primary cities and food consumption patterns that was based on the reach of commercial
networks outside of primary cities.
Third, the pattern of average partial effects of distance on commodity consumption
generally follows the expectations of Bennett’s Law (Table 1-A-5). In particular, distance
decreases the consumption of wheat, oils and fats, fresh vegetables, and beef, while increasing
the consumption of cassava. Perhaps surprisingly, increased distance to cities decreases the
consumption of maize. Cassava is commonly consumed in rural Zambia and could partially
explain the negative effect of distance to cities on maize consumption. Results on dairy are also
somewhat anomalous, with significant but opposite effects in primary and secondary cities.

55

Overall, however, the patterns of the relationships between distance to city and the consumption
of commodities are as expected.
Indirect effects of distances to cities on rural food consumption patterns are estimated
using total household expenditure, nonfarm employment, and the maximum attained education
within the household as mediator variables. These variables are mediator variables as distance to
cities could affect household expenditure, it could limit nonfarm employment opportunities that
commonly occur in and around cities, and distance to cities could limit access to higher levels of
education. The marginal effects of distances to cities on the three mediator variables generate
indirect effects of the same sign for the distances to cities on the key measures of food
consumption, as the marginal effects of these three mediator variables on the primary measures
of food consumption are positive.
The overall finding from the mediation analysis is that increasing distance to cities
decreases the primary food consumption patterns through indirect negative effects on household
income, engagement in non-farm employment, and educational attainment (Table 1-15). This
finding holds most clearly for the indirect effects of primary cities, where three of the six
coefficients are negative and statistically significant.

56

Table 1-15: Indirect effects of distance from rural household settlement to primary, secondary or
tertiary cities, estimates by country
Mediator Variable
Household total expenditure
per AE

Food Consumption Pattern
Purchased Share

Distance to
Secondary
Cities
-0.001

Tertiary
Cities
0.000

(0.043)

(0.169)

(0.476)

Beyond Staple Foods items

-0.001

0.000

0.000

(0.045)

(0.152)

(0.493)

Purchased Share

-0.008

0.001

-0.001

(0.000)

(0.447)

(0.001)

Beyond Staple Foods items

-0.001

0.000

0.000

(0.102)

(0.600)

(0.101)

Purchased Share

0.000

0.000

0.000

(0.531)

(0.976)

(0.076)

Beyond Staple Foods items

0.000

0.000

0.000

(0.518)

(0.977)

(0.117)

Nonfarm employment

Maximum education attained
within the household

Primary
Cities
-0.001

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

1.5 Conclusions
I considered how the consumption patterns of purchased food and high-income Bennett’s
Law foods vary across income, by city size and by distances to cities of differing size. To my
knowledge, this is the first paper to analyze the spatial effects of city size and distance to city on
food consumption patterns in Africa. The analysis of the effects of household income and city
size on the food consumption patterns of ESA are highlighted by the following results.
I find that the surprisingly high levels of purchased shares of food in total household
consumed food value across ESA are a result of households transitioning food consumption
towards purchased at subnumerary levels of income. This finding highlights that these transitions
in consumption patterns are not solely a middle-class story, as conventionally assumed, but poor
households are purchasing greater shares of their food and are consuming greater shares BSF
foods as their incomes rise.
Nonparametric analysis shows purchased share is present in rural households at all levels
of expenditure, to the extent that the fitted regressions of the share of purchased food do not fall
below forty percent at any expenditure level in three of the four countries in this analysis.

57

Expenditure threshold analysis identifies that the consumption patterns of purchased share and
the share of BSF foods are increasing below the poverty line. These significant initial levels and
growth at low expenditure levels of purchased share indicate that poor households are currently
able to access markets. Given the large portion of the population in ESA that currently has low
income levels, coupled with the pattern of increased purchased share with income growth, this
finding signals a strong future demand for increased food market infrastructure.
The Engel curve analysis highlighted many significant non-income determinants such as
nonfarm employment that commonly occur at low income levels and in increasing magnitudes
with rising incomes. These non-income determinants of consumption accentuate the incomeinduced patterns and contribute to the rise in the consumption patterns of purchased share and
BSF foods at subnumerary levels of income.
Recognizing the level of market interaction and the extent of income-induced change in
food consumption patterns across the entire income distribution, specifically among poor
households, considerable pressure on the food systems in developing countries should be
expected with future increases in household income.
The location of the future agrifood system investments is of due to the varying food
demand across city size. Increasing city size has significant positive effects on purchased share
in urban households, affirming the expectation that households will purchase more food when
they have increased access to markets.
When controlling for other determinants of consumption, the consumption of BSF food is
not shown to be positively affected by city size. Although, when considering the positive effects
of city size on household expenditure and engagement in nonfarm employment, along with the
positive effects that these determinant variables have on the consumption of BSF food, the

58

analysis shows that large cities indirectly have positive effects on the consumption of BSF food
via these mediator variables that capture the effects of city size.
The distance to cities of any size has a diminishing negative effect on purchased share
and the consumption of BSF foods by rural households. This relationship is consistent with the
expectations that distance to urban centers will increase the cost of accessing urban markets,
reducing the use of urban markets and resulting in lower shares of purchased or BSF food. The
magnitude of the effect of distance to city varies by city size, with large cities having the
strongest effects. Large cities are likely to have significant commercial networks that extend
market reach further into rural areas, resulting in increased engagement with urban markets by
rural households that live near large cities.
The spatial effects of city size and distance to city are shown in this analysis to be
significant drivers of the food consumption patterns in Eastern and Southern Africa. As it is
anticipated that urbanization will continue within this region and elsewhere in developing
nations, the recognition of how urbanization occurs and where investments should be made
should be considered when effort is put forth to meet the changing food demands of households
within and near urban areas.

59

APPENDIX

60

APPENDIX

Table 1-A-1: Food budget shares of select commodity aggregates, country data by settlement
Staple Foods items
Malawi Data
National
Rural
Urban

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
10.2
1.3
1.0

28.6

2.7

2.7

Oils &
Fats
3.2

(25.3)

(0.0)

(0.0)

(2.1)

(8.4)

(0.0)

(0.0)

30.6

2.1

2.9

2.6

10.2

1.0

0.8

1.1

(27.7)

(0.0)

(0.0)

(1.5)

(8.3)

(0.0)

(0.0)

(0.0)

Maize

Wheat

Cassava

Food Away
from Home
1.3
(0.0)

17.8

6.2

1.4

6.6

10.6

3.2

2.5

2.1

(14.7)

(5.5)

(0.0)

(6.0)

(9.1)

(0.0)

(0.0)

(0.0)

Primary City
Secondary City
Tertiary City

17.5

6.3

1.2

6.9

10.9

3.3

2.5

2.1

(14.7)

(5.7)

(0.0)

(6.3)

(9.3)

(0.0)

(0.0)

(0.0)

19.0

5.9

1.8

5.7

9.3

2.9

2.3

1.7

(14.7)

(5.1)

(0.0)

(4.8)

(8.5)

(0.0)

(0.0)

(0.0)

Staple Foods items
Tanzania Data
National
Rural
Urban
Primary City
Secondary City
Tertiary City

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
7.6
3.8
3.3

19.3

2.3

4.2

Oils &
Fats
3.6

(15.2)

(0.0)

(0.0)

(3.0)

(6.2)

(0.0)

(0.0)

21.4

1.6

5.1

3.4

7.6

3.0

3.7

9.0

(18.2)

(0.0)

(0.0)

(2.7)

(6.0)

(0.0)

(0.0)

(0.0)

Maize

Wheat

Cassava

Food Away
from Home
12.2
(2.9)

13.2

4.4

1.7

4.1

7.7

5.8

2.2

21.3

(8.8)

(3.4)

(0.0)

(3.6)

(6.5)

(4.7)

(0.0)

(15.2)

6.8

5.6

0.8

3.6

7.4

6.4

1.7

33.0

(5.2)

(4.7)

(0.0)

(3.2)

(6.8)

(5.6)

(0.0)

(32.2)

12.9

4.7

1.5

4.7

7.6

6.4

2.0

18.1

(9.3)

(3.4)

(0.0)

(4.1)

(6.1)

(5.4)

(0.0)

(10.3)

17.1

3.6

2.4

4.1

7.8

5.2

2.7

16.4

(12.6)

(2.0)

(0.0)

(3.6)

(6.6)

(2.4)

(0.0)

(7.4)

61

Table 1-A-1: (continued)
Staple Foods items
Uganda Data
National
Rural
Urban
Primary City
Secondary City
Tertiary City

9.3

1.3

10.5

(3.8)

(0.0)

(4.2)

(1.4)

(2.4)

(0.0)

(0.0)

10.0

0.8

12.3

2.0

2.8

3.7

3.3

3.3

(4.2)

(0.0)

(6.0)

(1.1)

(2.1)

(0.0)

(0.0)

(0.0)

11.8

Maize

Wheat

Cassava

National
Rural
Urban
Primary City
Secondary City
Tertiary City

Food Away
from Home
5.2
(0.0)

6.6

3.2

4.1

2.8

4.2

6.2

4.3

(3.4)

(0.0)

(0.6)

(2.5)

(3.6)

(2.9)

(0.0)

(0.0)

6.3

4.8

2.2

2.9

4.9

6.7

5.0

15.0

(3.2)

(3.5)

(0.0)

(2.8)

(4.3)

(5.3)

(3.1)

(0.0)

4.7

3.0

5.4

2.8

4.7

8.9

4.4

11.3

(2.7)

(0.0)

(0.0)

(2.9)

(4.2)

(8.6)

(3.1)

(0.0)

7.0

2.6

4.7

2.8

3.8

5.7

4.0

10.7

(3.4)

(0.0)

(0.6)

(2.4)

(3.2)

(0.0)

(0.0)

(0.0)

Staple Foods items
Zambia Data

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
3.1
4.3
3.5

Oils &
Fats
2.1

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
4.7
2.4
1.6

22.6

5.3

6.0

Oils &
Fats
4.5

(17.9)

(1.7)

(0.0)

(3.6)

(3.6)

(0.0)

(0.0)

25.9

3.3

8.8

4.0

3.8

1.5

1.2

0.1

(21.8)

(0.0)

(0.0)

(2.9)

(2.5)

(0.0)

(0.0)

(0.0)

Maize

Wheat

Cassava

Food Away
from Home
0.2
(0.0)

16.4

9.2

0.7

5.3

6.4

4.1

2.4

0.5

(13.5)

(7.3)

(0.0)

(4.5)

(5.3)

(0.0)

(0.7)

(0.0)

14.6

9.9

0.1

4.6

6.6

4.9

3.0

0.8

(12.3)

(8.6)

(0.0)

(3.9)

(5.4)

(4.2)

(1.8)

(0.0)

17.3

9.4

0.3

5.5

6.4

3.8

2.4

0.4

(14.4)

(7.5)

(0.0)

(4.8)

(5.3)

(0.0)

(0.6)

(0.0)

17.1

8.0

2.0

5.8

6.3

3.8

1.8

0.2

(13.7)

(5.4)

(0.0)

(4.8)

(5.1)

(0.0)

(0.0)

(0.0)

62

Table 1-A-2: Average partial effects of the household determinants for key food consumption patterns, urban household estimates by
country
Urban
Household total expenditure
per AE
Household located in a
primary city
Household located in a
secondary city

Malawi
0.001
(0.756)

Purchased Share
Tanzania
Uganda
0.013
0.004

Zambia
0.005

(0.001)

(0.296)

(0.296)

0.090

0.063

0.037

(0.000)

(0.048)

(0.000)

0.090

0.051

0.036

0.022

High-Income Bennett's Law Food
Malawi
Tanzania
Uganda
Zambia
0.008
0.025
0.015
0.009
(0.000)

(0.000)

(0.000)

(0.000)

0.017

-0.029

-0.010

(0.172)

(0.130)

(0.058)

0.007

-0.017

-0.035

-0.008

(0.000)

(0.000)

(0.349)

(0.000)

(0.445)

(0.208)

(0.229)

(0.087)

Nonfarm employment

-0.012

0.110

0.036

0.077

-0.007

0.089

0.025

0.030

(0.639)

(0.000)

(0.419)

(0.000)

(0.665)

(0.000)

(0.397)

(0.000)

Dependency ratio

-0.023

-0.026

0.132

-0.023

-0.043

-0.128

-0.217

-0.026

(0.374)

(0.293)

(0.062)

(0.009)

(0.028)

(0.000)

(0.000)

(0.006)

Household adult equivalents

0.000

0.003

-0.008

0.004

-0.002

0.009

-0.009

0.001

(0.919)

(0.099)

(0.215)

(0.000)

(0.426)

(0.000)

(0.014)

(0.375)

Farmed hectares of land (log)

-0.078

-0.037

-0.099

-0.028

0.012

-0.009

-0.026

-0.003

(0.004)

(0.000)

(0.000)

(0.000)

(0.617)

(0.348)

(0.151)

(0.205)

Own a gas or electric stove

0.049

-0.007

0.006

0.032

0.001

(0.000)

(0.740)

(0.201)

(0.052)

(0.977)

(0.587)

Own a refrigerator

0.038

0.040

0.007

0.027

0.015

-0.006

(0.005)

(0.008)

(0.152)

(0.054)

(0.178)

Maximum education attained
within the household

0.003

0.000

0.004

0.002

0.007

-0.001

0.000

0.001

(0.503)

(0.565)

(0.000)

(0.011)

(0.028)

(0.423)

(0.823)

(0.082)

Age of household head

-0.001

-0.001

-0.001

-0.001

0.000

-0.001

-0.001

0.000

(0.001)

(0.002)

(0.286)

(0.000)

(0.918)

(0.001)

(0.244)

(0.010)

Female head of household

-0.022

0.009

0.049

-0.019

-0.003

0.002

-0.037

-0.005

(0.184)

(0.592)

(0.107)

(0.000)

(0.763)

(0.864)

(0.072)

(0.278)

Own a telephone

0.037

0.113

0.070

0.013

0.003

-0.002

0.010

0.000

(0.018)

(0.000)

(0.088)

(0.011)

(0.749)

(0.928)

(0.689)

(0.940)

Own a car

0.059

-0.027

0.057

0.000

0.051

-0.043

0.010

0.025

(0.002)

(0.334)

(0.218)

(0.952)

(0.004)

(0.024)

(0.737)

(0.001)

Own a motorcycle

0.027

0.018

0.045

-0.015

-0.003

-0.013

-0.032

-0.035

Own a bicycle

0.003
(0.380)

(0.565)

(0.352)

(0.337)

(0.428)

(0.916)

(0.521)

(0.166)

(0.145)

-0.082

-0.038

-0.106

-0.030

-0.003

0.009

-0.053

-0.012

(0.000)

(0.014)

(0.002)

(0.000)

(0.748)

(0.492)

(0.018)

(0.018)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

63

Table 1-A-3: Average partial effects of the household determinants for key food consumption patterns, rural household estimates by
country
Rural
Household total expenditure
per AE

Malawi
-0.015
(0.000)

Purchased Share
Tanzania
Uganda
0.004
0.007
(0.525)

(0.214)

Zambia
0.007
(0.150)

High-Income Bennett's Law Food
Malawi
Tanzania
Uganda
Zambia
0.050
0.040
0.023
0.004
(0.000)

(0.000)

(0.000)

(0.269)

Distance to primary city (log)

-0.058

-0.024

(0.000)

(0.000)

Distance to secondary city
(log)

-0.007

-0.012

(0.118)

(0.002)

Distance to tertiary city (log)

-0.008

-0.004

(0.000)

(0.020)

Nonfarm employment

0.138

0.175

0.130

0.314

0.013

0.053

0.097

0.035

(0.000)

(0.000)

(0.000)

(0.000)

(0.387)

(0.000)

(0.000)

(0.080)

Dependency ratio

-0.070

-0.091

-0.046

-0.052

-0.017

-0.083

-0.130

0.014

(0.000)

(0.003)

(0.195)

(0.009)

(0.191)

(0.000)

(0.000)

(0.372)

Household adult equivalents

-0.011

-0.004

-0.001

0.001

0.004

0.005

-0.010

-0.004

(0.000)

(0.005)

(0.827)

(0.730)

(0.113)

(0.000)

(0.000)

(0.029)

Farmed hectares of land (log)

-0.042

-0.060

-0.057

-0.031

0.009

-0.018

0.018

0.008

(0.008)

(0.000)

(0.000)

(0.000)

(0.443)

(0.003)

(0.008)

(0.017)

Own a gas or electric stove

0.051

0.107

0.116

0.054

-0.025

0.029

(0.392)

(0.086)

(0.000)

(0.085)

(0.357)

(0.258)

Own a refrigerator

0.263

0.118

0.014

-0.032

-0.017

0.001

(0.000)

(0.004)

(0.722)

(0.255)

(0.469)

Maximum education attained
within the household

0.017

0.002

0.001

0.009

0.001

0.000

0.000

0.004

(0.000)

(0.061)

(0.470)

(0.000)

(0.721)

(0.593)

(0.648)

(0.000)

Age of household head

-0.002

0.000

-0.002

-0.002

0.000

0.000

0.000

0.000

(0.000)

(0.929)

(0.003)

(0.000)

(0.517)

(0.301)

(0.519)

(0.559)

Female head of household

-0.026

0.003

-0.014

-0.013

-0.020

-0.016

-0.007

-0.011

(0.020)

(0.888)

(0.423)

(0.249)

(0.007)

(0.167)

(0.527)

(0.159)

Own a telephone

0.045

0.087

0.061

0.065

0.010

0.039

0.031

0.011

(0.000)

(0.000)

(0.000)

(0.000)

(0.140)

(0.000)

(0.006)

(0.159)

Distance to market (log)

-0.017

0.004

-0.003

-0.018

-0.002

-0.002

0.004

0.000

(0.000)

(0.517)

(0.656)

(0.000)

(0.411)

(0.527)

(0.360)

(0.966)

Own a car

0.125

0.070

0.128

0.132

0.033

-0.022

0.010

0.103

(0.043)

(0.262)

(0.060)

(0.006)

(0.193)

(0.447)

(0.849)

(0.015)

Own a motorcycle

0.010

0.031

0.030

-0.022

0.063

0.003

0.023

0.044

(0.752)

(0.426)

(0.295)

(0.745)

(0.049)

(0.903)

(0.203)

(0.225)

0.007

-0.020

-0.055

0.008

0.014

-0.011

0.011

0.008

(0.420)

(0.183)

(0.000)

(0.389)

(0.024)

(0.235)

(0.280)

(0.266)

Own a bicycle

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

64

(0.976)

Table 1-A-4: Average partial effects of the household determinants for select commodity aggregates, urban household estimates by
country
Staple Foods items
Urban Malawi

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
-0.007
0.008
0.004

Maize

Wheat

Cassava

Household total expenditure per
AE
Household located in a
secondary city

-0.006

0.004

-0.001

Oils &
Fats
0.001

(0.008)

(0.000)

(0.170)

(0.134)

(0.000)

(0.002)

(0.006)

(0.200)

0.003

-0.003

-0.004

0.010

0.020

-0.001

-0.001

0.003

(0.730)

(0.380)

(0.019)

(0.001)

(0.000)

(0.840)

(0.588)

(0.167)

Nonfarm employment

-0.027

0.015

-0.002

-0.006

-0.009

0.000

0.001

0.000

(0.040)

(0.020)

(0.525)

(0.324)

(0.161)

(0.968)

(0.690)

(0.917)

Dependency ratio

0.031

-0.010

-0.004

-0.002

0.011

-0.003

0.006

-0.014

(0.074)

(0.240)

(0.229)

(0.732)

(0.159)

(0.679)

(0.306)

(0.015)

Household adult equivalents

-0.001

0.002

0.000

-0.001

-0.007

0.003

0.000

0.000

(0.787)

(0.208)

(0.938)

(0.134)

(0.000)

(0.001)

(0.717)

(0.921)

Farmed hectares of land (log)

-0.028

-0.008

-0.003

-0.010

0.015

-0.001

-0.003

-0.013

(0.177)

(0.257)

(0.514)

(0.282)

(0.214)

(0.866)

(0.508)

(0.113)

Own a gas or electric stove

-0.039

0.004

-0.007

0.002

0.014

0.009

0.010

-0.002

(0.000)

(0.409)

(0.000)

(0.660)

(0.011)

(0.073)

(0.078)

(0.504)

Own a refrigerator

-0.020

-0.001

0.003

0.013

0.004

0.002

0.005

-0.009

(0.049)

(0.799)

(0.454)

(0.010)

(0.517)

(0.717)

(0.220)

(0.001)

Maximum education attained
within the household

-0.011

0.001

0.000

0.002

-0.001

0.003

0.001

-0.002

(0.000)

(0.462)

(0.547)

(0.058)

(0.582)

(0.025)

(0.173)

(0.014)

Age of household head

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.143)

(0.275)

(0.965)

(0.335)

(0.420)

(0.388)

(0.535)

(0.251)

Female head of household

-0.019

-0.009

-0.003

0.010

-0.002

-0.005

-0.001

-0.001

(0.061)

(0.069)

(0.138)

(0.096)

(0.749)

(0.284)

(0.731)

(0.819)

Own a telephone

-0.010

0.009

-0.006

0.005

0.016

-0.001

0.010

0.004

(0.253)

(0.069)

(0.013)

(0.292)

(0.000)

(0.845)

(0.001)

(0.172)

Own a car

-0.047

0.008

-0.003

0.002

0.019

0.003

-0.001

0.008

(0.006)

(0.261)

(0.179)

(0.614)

(0.021)

(0.582)

(0.698)

(0. 249)

Own a motorcycle

-0.037

0.037

0.000

0.005

0.021

-0.010

-0.006

-0.005

(0.280)

(0.079)

(0.990)

(0.653)

(0.276)

(0.134)

(0.023)

(0.512)

0.006

-0.015

0.000

-0.002

0.005

0.000

0.004

-0.002

(0.423)

(0.000)

(0.837)

(0.669)

(0.205)

(0.920)

(0.203)

(0.438)

Own a bicycle

65

Food Away
from Home
0.000

Table 1-A-4: (continued)
Staple Foods items
Urban Tanzania

Oils &
Fats
-0.002

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
-0.007
0.002
0.001

Food Away
from Home
0.020

Maize

Wheat

Cassava

Household total expenditure per
AE
Household located in a primary
city
Household located in a
secondary city

-0.023

0.000

-0.003

(0.000)

(0.850)

(0.009)

(0.000)

(0.000)

(0.004)

(0.044)

-0.026

0.017

-0.003

-0.002

0.019

-0.001

-0.013

0.037

(0.001)

(0.000)

(0.383)

(0.502)

(0.000)

(0.867)

(0.000)

(0.071)

0.002

0.008

-0.003

0.004

0.003

0.005

-0.010

-0.011

(0.820)

(0.114)

(0.557)

(0.152)

(0.537)

(0.531)

(0.004)

(0.584)

Nonfarm employment

-0.070

-0.003

-0.009

0.000

-0.013

-0.008

-0.019

0.125

(0.000)

(0.662)

(0.055)

(0.941)

(0.045)

(0.378)

(0.001)

(0.000)

Dependency ratio

0.023

0.017

0.013

0.008

0.011

0.012

0.006

-0.232

(0.175)

(0.030)

(0.069)

(0.135)

(0.134)

(0.312)

(0.426)

(0.000)

Household adult equivalents

-0.007

0.000

0.000

-0.001

-0.003

0.002

0.000

0.010

(0.000)

(0.836)

(0.156)

(0.001)

(0.000)

(0.000)

(0.273)

(0.000)

Farmed hectares of land (log)

0.014

-0.006

-0.001

0.000

-0.003

0.000

-0.001

-0.001

(0.033)

(0.100)

(0.648)

(0.992)

(0.421)

(0.989)

(0.685)

(0.960)

Own a gas or electric stove

0.006

0.000

0.001

-0.001

0.005

-0.002

0.015

-0.018

(0.682)

(0.989)

(0.885)

(0.823)

(0.501)

(0.800)

(0.027)

(0.474)

Own a refrigerator

-0.023

-0.005

-0.006

0.000

0.004

0.010

0.009

-0.025

(0.028)

(0.232)

(0.041)

(0.907)

(0.448)

(0.160)

(0.025)

(0.144)

Maximum education attained
within the household

0.000

0.000

0.000

0.000

0.000

0.000

0.000

-0.002

(0.456)

(0.260)

(0.048)

(0.791)

(0.869)

(0.539)

(0.348)

(0.078)

Age of household head

0.001

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.008)

(0.492)

(0.018)

(0.026)

(0.117)

(0.043)

(0.052)

(0.472)

Female head of household

-0.022

0.002

0.000

0.005

0.005

0.004

-0.002

-0.033

(0.020)

(0.670)

(0.987)

(0.085)

(0.212)

(0.506)

(0.526)

(0.048)

Own a telephone

-0.005

0.016

-0.003

0.012

0.000

0.012

0.000

-0.025

(0.688)

(0.002)

(0.572)

(0.001)

(0.941)

(0.179)

(0.974)

(0.464)

Own a car

0.012

0.007

-0.009

0.015

0.025

-0.008

0.009

-0.062

(0.384)

(0.304)

(0.014)

(0.063)

(0.007)

(0.318)

(0.174)

(0.018)

Own a motorcycle

-0.018

0.002

0.004

0.001

0.012

-0.003

-0.001

-0.062

Own a bicycle

(0.000)

(0.148)

(0.788)

(0.447)

(0.844)

(0.133)

(0.771)

(0.813)

(0.035)

-0.012

-0.006

0.005

-0.003

-0.003

0.008

0.008

-0.016

(0.197)

(0.147)

(0.181)

(0.169)

(0.498)

(0.199)

(0.039)

(0.348)

66

Table 1-A-4: (continued)
Staple Foods items
Urban Uganda

Oils &
Fats
0.000

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
-0.001
0.003
0.002

Food Away
from Home
0.004

Maize

Wheat

Cassava

Household total expenditure per
AE
Household located in a primary
city
Household located in a
secondary city

-0.010

0.001

-0.004

(0.000)

(0.012)

(0.013)

(0.526)

(0.129)

(0.001)

(0.011)

(0.004)

0.014

0.011

-0.010

-0.001

0.010

-0.003

0.007

-0.002

(0.154)

(0.067)

(0.129)

(0.765)

(0.010)

(0.721)

(0.320)

(0.928)

-0.006

0.001

0.019

0.000

0.008

0.021

0.002

-0.034

(0.618)

(0.945)

(0.228)

(0.949)

(0.089)

(0.150)

(0.831)

(0.139)

Nonfarm employment

-0.002

0.007

0.001

0.000

-0.002

0.003

-0.007

0.039

(0.853)

(0.408)

(0.961)

(0.913)

(0.678)

(0.783)

(0.450)

(0.208)

Dependency ratio

0.024

0.015

0.033

0.005

-0.005

0.029

0.002

-0.223

(0.275)

(0.140)

(0.080)

(0.367)

(0.503)

(0.107)

(0.877)

(0.000)

Household adult equivalents

0.001

0.001

-0.001

-0.001

-0.002

0.000

-0.002

-0.003

(0.704)

(0.572)

(0.682)

(0.110)

(0.001)

(0.864)

(0.089)

(0.323)

Farmed hectares of land (log)

0.015

-0.006

0.016

-0.006

-0.007

0.008

0.010

-0.022

(0.028)

(0.172)

(0.002)

(0.054)

(0.044)

(0.277)

(0.121)

(0.409)

Maximum education attained
within the household

0.000

0.000

0.000

0.000

0.000

0.001

0.000

-0.001

(0.049)

(0.114)

(0.552)

(0.011)

(0.353)

(0.007)

(0.943)

(0.020)

Age of household head

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.355)

(0.283)

(0.435)

(0.844)

(0.674)

(0.923)

(0.437)

(0.830)

Female head of household

0.015

0.005

0.002

0.005

0.010

-0.003

-0.008

-0.059

(0.139)

(0.455)

(0.756)

(0.050)

(0.006)

(0.753)

(0.225)

(0.003)

Own a telephone

-0.014

0.012

0.001

0.000

-0.001

-0.004

-0.001

0.057

(0.258)

(0.041)

(0.947)

(0.979)

(0.810)

(0.731)

(0.960)

(0.011)

Own a car

-0.009

-0.001

-0.017

-0.005

0.009

0.007

0.005

-0.024

(0.431)

(0.909)

(0.035)

(0.301)

(0.125)

(0.578)

(0.678)

(0.434)

Own a motorcycle

-0.001

0.005

-0.016

0.004

0.002

-0.016

0.013

-0.050

Own a bicycle

(0.909)

(0.459)

(0.027)

(0.335)

(0.550)

(0.065)

(0.127)

(0.006)

-0.013

-0.014

0.018

0.004

0.004

-0.009

-0.003

-0.044

(0.158)

(0.002)

(0.051)

(0.184)

(0.269)

(0.247)

(0.681)

(0.030)

67

Table 1-A-4: (continued)
Staple Foods items
Urban Zambia

Oils &
Fats
-0.002

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
-0.003
0.003
0.002

Food Away
from Home
0.000

Maize

Wheat

Cassava

Household total expenditure per
AE
Household located in a primary
city
Household located in a
secondary city

-0.013

0.003

-0.003

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

0.011

0.012

-0.007

-0.011

0.007

-0.002

0.006

0.002

(0.004)

(0.001)

(0.000)

(0.000)

(0.000)

(0.414)

(0.001)

(0.017)

0.015

0.010

-0.009

-0.003

0.003

-0.006

0.004

0.002

(0.000)

(0.002)

(0.000)

(0.028)

(0.068)

(0.006)

(0.016)

(0.056)

Nonfarm employment

-0.025

0.008

-0.006

-0.002

0.004

0.006

0.002

0.004

(0.000)

(0.142)

(0.000)

(0.456)

(0.180)

(0.048)

(0.595)

(0.014)

Dependency ratio

0.012

0.005

0.001

0.000

-0.010

0.000

0.005

-0.008

(0.102)

(0.435)

(0.586)

(0.872)

(0.003)

(0.929)

(0.077)

(0.000)

Household adult equivalents

-0.003

0.004

-0.001

-0.001

-0.002

0.002

0.001

0.001

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.033)

(0.002)

(0.004)

Farmed hectares of land (log)

-0.002

-0.002

0.002

-0.001

-0.002

-0.002

0.000

-0.001

(0.136)

(0.160)

(0.000)

(0.363)

(0.003)

(0.014)

(0.434)

(0.106)

Own a gas or electric stove

-0.006

-0.001

-0.002

0.007

0.001

-0.001

0.002

-0.001

(0.199)

(0.749)

(0.007)

(0.000)

(0.623)

(0.741)

(0.373)

(0.505)

Own a refrigerator

-0.005

0.000

0.001

0.004

-0.001

0.000

0.001

-0.004

(0.267)

(0.895)

(0.223)

(0.052)

(0.700)

(0.888)

(0.471)

(0.040)

Maximum education attained
within the household

-0.001

0.000

-0.001

0.000

0.000

0.000

0.000

0.000

(0.008)

(0.599)

(0.000)

(0.213)

(0.908)

(0.110)

(0.513)

(0.174)

Age of household head

0.001

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.000)

(0.082)

(0.173)

(0.479)

(0.622)

(0.786)

(0.036)

(0.000)

Female head of household

0.003

0.000

0.000

0.001

0.001

0.001

0.000

-0.001

(0.457)

(0.919)

(0.664)

(0.402)

(0.754)

(0.576)

(0.794)

(0.083)

Own a telephone

-0.007

0.013

-0.003

0.001

-0.004

0.005

0.004

0.001

(0.111)

(0.001)

(0.000)

(0.678)

(0.097)

(0.050)

(0.074)

(0.475)

Own a car

-0.011

-0.006

0.003

0.006

0.003

0.008

0.003

0.001

(0.047)

(0.234)

(0.251)

(0.003)

(0.210)

(0.004)

(0.172)

(0.347)

Own a motorcycle

0.027

0.022

-0.002

0.010

-0.002

-0.012

-0.005

-0.002

(0.139)

(0.173)

(0.208)

(0.195)

(0.710)

(0.046)

(0.260)

(0.061)

0.000

0.008

0.002

0.000

-0.001

-0.004

-0.002

-0.001

(0.915)

(0.022)

(0.008)

(0.805)

(0.518)

(0.050)

(0.103)

(0.397)

Own a bicycle

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

68

Table 1-A-5: Average partial effects of the household determinants for select commodity aggregates, rural household estimates by
country
Staple Foods items
Rural Malawi

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
-0.021
0.040
0.048

Maize

Wheat

Cassava

Household total expenditure per
AE

-0.056

0.029

0.000

Oils &
Fats
0.015

(0.000)

(0.000)

(0.685)

(0.000)

(0.000)

(0.001)

(0.000)

Nonfarm employment

-0.042

0.009

0.001

0.010

-0.009

0.001

0.000

0.003

(0.003)

(0.001)

(0.921)

(0.000)

(0.104)

(0.685)

(0.997)

(0.137)

Dependency ratio

0.011

0.001

-0.008

-0.005

-0.009

-0.001

0.000

-0.002

(0.365)

(0.773)

(0.078)

(0.019)

(0.082)

(0.783)

(0.893)

(0.319)

Household adult equivalents

-0.007

0.000

0.002

0.000

-0.009

0.001

0.001

0.000

(0.001)

(0.174)

(0.067)

(0.840)

(0.000)

(0.014)

(0.001)

(0.416)

Farmed hectares of land (log)

0.002

0.002

-0.026

-0.002

0.004

0.003

0.003

0.000

(0.891)

(0.324)

(0.000)

(0.407)

(0.428)

(0.114)

(0.096)

(0.753)

Own a gas or electric stove

0.010

-0.004

0.004

0.014

0.038

0.007

0.005

0.001

(0.666)

(0.148)

(0.705)

(0.067)

(0.119)

(0.483)

(0.408)

(0.853)

Own a refrigerator

-0.006

0.003

-0.006

0.013

-0.006

0.015

-0.002

-0.001

(0.777)

(0.407)

(0.383)

(0.015)

(0.702)

(0.186)

(0.429)

(0.905)

Maximum education attained
within the household

-0.003

0.001

0.001

0.003

0.002

0.001

0.002

0.000

(0.289)

(0.015)

(0.434)

(0.000)

(0.033)

(0.024)

(0.004)

(0.427)

Age of household head

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.430)

(0.001)

(0.781)

(0.018)

(0.490)

(0.526)

(0.003)

(0.000)

Female head of household

0.013

-0.001

0.004

-0.001

0.004

0.001

-0.001

-0.001

(0.069)

(0.447)

(0.138)

(0.538)

(0.210)

(0.730)

(0.727)

(0.174)

Own a telephone

-0.007

0.006

-0.002

0.007

0.002

0.001

-0.001

0.002

(0.296)

(0.000)

(0.463)

(0.000)

(0.464)

(0.421)

(0.452)

(0.046)

Distance to market (log)

-0.004

0.000

0.003

-0.001

-0.002

0.001

0.000

0.000

(0.087)

(0.346)

(0.015)

(0.080)

(0.011)

(0.103)

(0.365)

(0.368)

Own a car

-0.002

0.010

-0.026

0.003

0.003

-0.006

0.014

0.001

(0.965)

(0.200)

(0.000)

(0.457)

(0.906)

(0.000)

(0.165)

(0.892)

Own a motorcycle

-0.012

0.001

-0.013

0.000

-0.012

-0.002

0.005

-0.002

(0.486)

(0.842)

(0.025)

(0.974)

(0.255)

(0.702)

(0.397)

(0.414)

-0.011

0.004

-0.003

0.004

0.006

0.001

0.002

-0.001

(0.052)

(0.004)

(0.186)

(0.000)

(0.028)

(0.333)

(0.119)

(0.308)

Own a bicycle

69

Food Away
from Home
0.003
(0.008)

Table 1-A-5: (continued)
Staple Foods items
Rural Tanzania

Oils &
Fats
-0.002

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
-0.015
0.005
0.005

Food Away
from Home
0.024

Maize

Wheat

Cassava

Household total expenditure per
AE

-0.037

0.004

-0.017

(0.000)

(0.000)

(0.000)

(0.010)

(0.000)

(0.001)

(0.003)

Nonfarm employment

-0.034

0.010

-0.008

-0.001

-0.002

-0.002

-0.039

0.085

(0.017)

(0.000)

(0.321)

(0.774)

(0.620)

(0.619)

(0.000)

(0.000)

Dependency ratio

0.012

-0.006

0.022

-0.006

0.001

-0.007

0.010

-0.077

(0.557)

(0.119)

(0.095)

(0.147)

(0.879)

(0.329)

(0.407)

(0.000)

Household adult equivalents

-0.003

0.000

-0.001

-0.001

-0.003

0.000

0.001

0.003

(0.000)

(0.406)

(0.035)

(0.001)

(0.000)

(0.588)

(0.002)

(0.000)

Farmed hectares of land (log)

0.035

-0.002

-0.005

-0.001

0.000

-0.002

0.006

0.003

(0.000)

(0.073)

(0.185)

(0.214)

(0.962)

(0.258)

(0.019)

(0.534)

Own a gas or electric stove

-0.023

-0.001

-0.004

0.003

0.020

-0.002

-0.005

-0.019

(0.351)

(0.809)

(0.752)

(0.341)

(0.072)

(0.804)

(0.743)

(0.252)

Own a refrigerator

-0.068

0.023

-0.010

-0.003

-0.002

0.011

0.025

-0.034

(0.003)

(0.004)

(0.264)

(0.319)

(0.694)

(0.223)

(0.258)

(0.001)

Maximum education attained
within the household

0.000

0.000

0.000

0.000

0.000

0.000

-0.001

0.000

(0.958)

(0.331)

(0.247)

(0.025)

(0.113)

(0.288)

(0.002)

(0.821)

Age of household head

-0.001

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.012)

(0.736)

(0.155)

(0.026)

(0.036)

(0.104)

(0.187)

(0.340)

Female head of household

0.005

0.003

-0.002

0.003

0.003

0.004

0.013

-0.018

(0.649)

(0.235)

(0.705)

(0.128)

(0.395)

(0.348)

(0.040)

(0.061)

Own a telephone

-0.002

0.004

-0.011

0.011

0.001

0.008

0.014

0.003

(0.863)

(0.028)

(0.049)

(0.000)

(0.684)

(0.016)

(0.002)

(0.718)

Distance to market (log)

0.004

-0.001

-0.007

0.001

0.003

0.000

-0.003

-0.004

(0.293)

(0.337)

(0.005)

(0.150)

(0.030)

(0.725)

(0.118)

(0.141)

Own a car

0.009

-0.006

-0.021

0.003

0.041

0.007

-0.016

-0.005

(0.730)

(0.016)

(0.010)

(0.551)

(0.023)

(0.562)

(0.137)

(0.802)

Own a motorcycle

-0.037

0.001

-0.014

-0.003

0.003

0.010

0.002

0.007

(0.070)

(0.705)

(0.146)

(0.415)

(0.628)

(0.220)

(0.809)

(0.686)

0.010

0.001

0.000

-0.003

-0.003

-0.004

0.001

-0.015

(0.310)

(0.428)

(0.989)

(0.147)

(0.287)

(0.301)

(0.803)

(0.060)

Own a bicycle

70

(0.000)

Table 1-A-5: (continued)
Staple Foods items
Rural Uganda

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
0.000
0.008
0.005

Maize

Wheat

Cassava

Household total expenditure per
AE

-0.009

0.004

-0.024

Oils &
Fats
0.001

(0.001)

(0.000)

(0.000)

(0.026)

(0.885)

(0.000)

(0.000)

Nonfarm employment

0.027

0.001

-0.047

0.000

0.004

0.016

-0.003

0.024

(0.016)

(0.618)

(0.001)

(0.948)

(0.119)

(0.003)

(0.580)

(0.000)

Dependency ratio

0.059

0.000

0.048

-0.003

-0.014

0.008

-0.013

-0.048

(0.003)

(0.975)

(0.026)

(0.408)

(0.004)

(0.331)

(0.179)

(0.000)

Household adult equivalents

-0.002

0.000

0.003

0.000

0.000

0.000

-0.001

-0.003

(0.379)

(0.338)

(0.106)

(0.260)

(0.277)

(0.700)

(0.026)

(0.034)

Farmed hectares of land (log)

-0.002

-0.001

0.010

0.001

-0.006

-0.003

0.010

0.000

(0.796)

(0.283)

(0.067)

(0.653)

(0.000)

(0.283)

(0.000)

(0.927)

Maximum education attained
within the household

-0.001

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.012)

(0.000)

(0.308)

(0.315)

(0.758)

(0.231)

(0.909)

(0.462)

Age of household head

0.000

0.000

-0.001

0.000

0.000

0.000

0.000

0.000

(0.214)

(0.742)

(0.062)

(0.009)

(0.228)

(0.456)

(0.275)

(0.331)

Female head of household

-0.012

0.003

-0.002

0.002

0.004

-0.009

0.000

-0.013

(0.169)

(0.158)

(0.845)

(0.325)

(0.073)

(0.042)

(0.980)

(0.065)

Own a telephone

0.008

0.003

-0.012

0.004

0.005

0.003

0.019

0.002

(0.349)

(0.020)

(0.217)

(0.031)

(0.010)

(0.467)

(0.000)

(0.649)

Distance to market (log)

-0.002

-0.001

0.001

0.000

-0.002

0.000

0.003

-0.005

(0.466)

(0.041)

(0.779)

(0.384)

(0.018)

(0.847)

(0.020)

(0.048)

Own a car

-0.016

0.000

-0.044

0.005

0.006

0.000

0.035

-0.024

(0.518)

(0.919)

(0.005)

(0.445)

(0.511)

(0.972)

(0.081)

(0.000)

Own a motorcycle

-0.011

0.000

-0.026

0.005

0.008

0.005

0.010

0.001

(0.404)

(0.796)

(0.038)

(0.150)

(0.009)

(0.369)

(0.143)

(0.933)

-0.012

-0.002

0.027

0.003

0.003

-0.002

-0.003

-0.007

(0.127)

(0.047)

(0.003)

(0.038)

(0.104)

(0.554)

(0.422)

(0.186)

Own a bicycle

71

Food Away
from Home
0.007
(0.000)

Table 1-A-5: (continued)
Staple Foods items
Rural Zambia

Oils &
Fats
-0.004

Beyond Staple Foods items
Staple
Beef
Dairy
Vegetables
0.000
0.004
0.003

Food Away
from Home
0.000

Maize

Wheat

Cassava

Household total expenditure per
AE

-0.028

0.012

0.000

(0.000)

(0.000)

(0.938)

(0.000)

(0.798)

(0.006)

(0.000)

(0.569)

Distance to primary city (log)

-0.045

-0.008

0.137

-0.003

-0.006

-0.004

-0.006

0.000

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.001)

(0.000)

(0.458)

Distance to secondary city (log)

-0.007

-0.003

0.034

-0.001

-0.002

-0.001

0.002

0.000

(0.052)

(0.001)

(0.000)

(0.105)

(0.005)

(0.240)

(0.003)

(0.668)

Distance to tertiary city (log)

0.001

0.000

0.007

0.000

-0.002

-0.001

0.000

0.000

(0.746)

(0.934)

(0.000)

(0.698)

(0.000)

(0.134)

(0.219)

(0.086)

Nonfarm employment

-0.042

0.018

-0.031

0.018

0.013

0.002

0.001

0.000

(0.030)

(0.001)

(0.084)

(0.000)

(0.011)

(0.635)

(0.750)

(0.286)

Dependency ratio

0.017

-0.012

-0.003)

-0.002

-0.002

0.003

0.002

-0.001

(0.307)

(0.009)

(0.805)

(0.567)

(0.634)

(0.455)

(0.474)

(0.063)

Household adult equivalents

-0.003

0.002

0.005

-0.001

0.000

0.000

0.001

0.000

(0.064)

(0.001)

(0.001)

(0.000)

(0.993)

(0.857)

(0.005)

(0.179)

Farmed hectares of land (log)

0.007

0.001

-0.017

0.001

0.000

0.002

0.002

0.000

(0.034)

(0.290)

(0.000)

(0.202)

(0.629)

(0.001)

(0.018)

(0.351)

Own a gas or electric stove

-0.067

-0.007

-0.048

-0.007

0.006

-0.002

-0.004

0.000

(0.001)

(0.136)

(0.035)

(0.347)

(0.191)

(0.539)

(0.055)

(0.041)

Own a refrigerator

0.014

0.007

-0.014

0.012

0.002

-0.001

0.003

0.000

(0.637)

(0.397)

(0.612)

(0.293)

(0.613)

(0.834)

(0.540)

(0.718)

Maximum education attained
within the household

-0.003

0.001

-0.003

0.002

0.001

0.000

0.000

0.000

(0.045)

(0.105)

(0.001)

(0.000)

(0.010)

(0.077)

(0.049)

(0.762)

Age of household head

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

(0.671)

(0.028)

(0.017)

(0.651)

(0.330)

(0.236)

(0.688)

(0.907)

Female head of household

0.023

0.000

-0.003

-0.002

-0.002

0.001

0.000

0.000

(0.015)

(0.856)

(0.632)

(0.325)

(0.369)

(0.751)

(0.938)

(0.741)

Own a telephone

0.013

0.009

-0.034

0.011

0.003

0.003

0.003

0.000

(0.144)

(0.001)

(0.000)

(0.000)

(0.089)

(0.243)

(0.055)

(0.072)

Distance to market (log)

-0.002

-0.006

0.007

0.000

-0.004

0.000

0.001

0.000

(0.350)

(0.000)

(0.000)

(0.985)

(0.000)

(0.417)

(0.114)

(0.142)

Own a car

-0.079

-0.001

-0.080

0.017

0.006

0.008

0.005

0.000

(0.000)

(0.886)

(0.000)

(0.181)

(0.608)

(0.194)

(0.511)

(0.966)

Own a motorcycle

0.007

-0.009

-0.079

0.012

0.019

-0.006

0.001

0.000

(0.854)

(0.215)

(0.000)

(0.115)

(0.287)

(0.044)

(0.767)

(0.673)

-0.031

0.006

0.009

0.004

0.001

-0.001

-0.004

0.000

(0.000)

(0.009)

(0.169)

(0.007)

(0.603)

(0.626)

(0.020)

(0.162)

Own a bicycle

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

72

REFERENCES

73

REFERENCES

Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel curves and consumer demand.
Review of Economics and statistics, 79(4), 527-539.
Battersby, J., & Peyton, S. (2014, June). The geography of supermarkets in Cape Town:
Supermarket expansion and food access. In Urban Forum (Vol. 25, No. 2, pp. 153-164).
Springer Netherlands.
Bennett, M. K. (1941). Wheat in national diets. Wheat Studies, (02).
BerdeguĂŠ, J. A., Carriazo, F., Jara, B., Modrego, F., & Soloaga, I. (2015). Cities, Territories, and
Inclusive Growth: Unraveling Urban–Rural Linkages in Chile, Colombia, and Mexico.
World Development.
Bettencourt, L. M., Lobo, J., Helbing, D., KĂźhnert, C., & West, G. B. (2007). Growth, innovation,
scaling, and the pace of life in cities. Proceedings of the national academy of sciences,
104(17), 7301-7306.
Christiaensen, L., Weerdt, J., & Todo, Y. (2013). Urbanization and poverty reduction: the role of
rural diversification and secondary towns1. Agricultural Economics, 44(4-5), 435-447.
Cleveland, W. S. (1979). Robust locally weighted regression and smoothing scatterplots. Journal
of the American statistical association, 74(368), 829-836.
Cleveland, W. S., & Devlin, S. J. (1988). Locally weighted regression: an approach to regression
analysis by local fitting. Journal of the American Statistical Association, 83(403), 596-610.
D’Haese, M., & Van Huylenbroeck, G. (2005). The rise of supermarkets and changing expenditure
patterns of poor rural households case study in the Transkei area, South Africa. Food
Policy, 30(1), 97-113.
Fafchamps, M., & Shilpi, F. (2003). The spatial division of labour in Nepal. The Journal of
Development Studies, 39(6), 23-66.
FerrĂŠ, C., Ferreira, F. H., & Lanjouw, P. (2012). Is There a Metropolitan Bias? The relationship
between poverty and city size in a selection of developing countries. The World Bank
Economic Review, lhs007.
Garrett, J. L., & Ruel, M. T. (1999). Are determinants of rural and urban food security and
nutritional status different? Some insights from Mozambique. World Development, 27(11),
1955-1975.

74

GĂłmez, M. I., Barrett, C. B., Raney, T., Pinstrup-Andersen, P., Meerman, J., Croppenstedt, A.,
Carishma, B., & Thompson, B. (2013). Post-green revolution food systems and the triple
burden of malnutrition. Food Policy, 42, 129-138.
GĂłmez, M. I., & Ricketts, K. D. (2013). Food value chain transformations in developing
countries: Selected hypotheses on nutritional implications. Food Policy, 42, 139-150.
Haggblade, S., Hazell, P., & Reardon, T. (2010). The rural non-farm economy: Prospects for
growth and poverty reduction. World Development, 38(10), 1429-1441.
Hayes, A. F., & Preacher, K. J. (2010). Quantifying and testing indirect effects in simple mediation
models when the constituent paths are nonlinear. Multivariate behavioral research, 45(4),
627-660.
Hino, H. (2014). Shopping at different food retail formats: Understanding cross-shopping behavior
through retail format selective use patterns. European Journal of Marketing, 48(3/4), 674698.
Huang, J., & Bouis, H. (2001). Structural changes in the demand for food in Asia: empirical
evidence from Taiwan. Agricultural Economics, 26(1), 57-69.
Leser, C. E. V. (1963). Forms of Engel functions. Econometrica: Journal of the Econometric
Society, 694-703.
Lewis, W. A. (1954). Economic development with unlimited supplies of labour. The manchester
school, 22(2), 139-191.
Nelson, G. C. (2002). Introduction to the special issue on spatial analysis for agricultural
economists. Agricultural Economics, 27(3), 197-200.
Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response variables
with an application to 401(K) plan participation rates. Journal of Applied Econometrics,
volume 11, 619-632.
Papke, L. E., & Wooldridge, J. M. (2008). Panel data methods for fractional response variables
with an application to test pass rates. Journal of Econometrics, 145(1), 121-133.
Popkin, B. M., Adair, L. S., & Ng, S. W. (2012). Global nutrition transition and the pandemic of
obesity in developing countries. Nutrition reviews, 70(1), 3-21.
Reardon, T., & Timmer, C. P. (2007). Transformation of markets for agricultural output in
developing countries since 1950: How has thinking changed?. Handbook of agricultural
economics, 3, 2807-2855.
Renkow, M. (2007). Cities, Towns, and the Rural Nonfarm Economy in S. Haggblade, Hazell, P.
B., & Reardon, T. (Eds.). (pp. 183-198). International Food Policy Res Institute.

75

Rischke, R., Kimenju, S. C., Klasen, S., & Qaim, M. (2015). Supermarkets and food consumption
patterns: The case of small towns in Kenya. Food Policy, 52, 9-21.
Satterthwaite, D., McGranahan, G., & Tacoli, C. (2010). Urbanization and its implications for food
and farming. Philosophical Transactions of the Royal Society B: Biological Sciences,
365(1554), 2809-2820.
Serber, G. A. F., & Wild, C. J. (1989). Nonlinear Regression. John Wiley & Sons, 435.
Sharma, A. (2016). Urban Proximity and Spatial Pattern of Land Use and Development in Rural
India. The Journal of Development Studies, 1-19.
Stage, J., Stage, J., & McGranahan, G. (2010). Is urbanization contributing to higher food prices?
Environment and Urbanization, 22(1), 199-215.
Stone, R. (1954). Linear expenditure systems and demand analysis: an application to the pattern
of British demand. The Economic Journal, 64(255), 511-527.
Strydom, J. W. (2011). Retailing in disadvantaged communities: the outshopping phenomenon
revisited. Journal of Contemporary Management, 8, 150-172.
Subramanian, S., & Deaton, A. (1996). The demand for food and calories. Journal of political
economy, 104(1), 133-162.
Thiele, S., & Weiss, C. (2003). Consumer demand for food diversity: evidence for Germany.
Food Policy, 28(2), 99-115.
Turrell, G., & Kavanagh, A. M. (2006). Socio-economic pathways to diet: modelling the
association between socio-economic position and food purchasing behaviour. Public
health nutrition, 9(03), 375-383.
Von Braun, J., De Haen, H., & Blanken, J. (1991). Commercialization of agriculture under
population pressure: Effects on production, consumption, and nutrition in Rwanda (Vol.
85). Intl Food Policy Res Inst.
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press.
Working, H. (1943). Statistical laws of family expenditure. Journal of the American Statistical
Association, 38(221), 43-56.
World Bank. (2015, October 4). The international poverty line has just been raised to $1.90 a day,
but global poverty is basically unchanged. How is that even possible? [Blog post].
Retrieved from http://blogs.worldbank.org/developmenttalk/international-poverty-linehas-just-been-raised-190-day-global-poverty-basically-unchanged-how-even

76

ESSAY 2: CONSUMPTION OF PROCESSED FOOD
IN THE DIETS OF POOR RURAL AND URBAN CONSUMERS
IN DEVELOPING EASTERN AND SOUTHERN AFRICA

2.1 Introduction
I explore how the consumption patterns of processed food in developing Eastern and
Southern Africa (ESA)21 vary across the household income distribution and spatial
considerations. Having a clear understanding of processed food consumption patterns has value,
because the food people consume – specifically the share of processed food – generates profound
implications on the structure of the agrifood system, the level and types of public and private
investments, the level and distribution of employment, and nutrition and health.
Household consumption patterns of processed food are expected to vary much as the
consumption of staple food varies across the income distribution (Bennett, 1941; Pingali, 2007);
an expectation that households would consume more processed food with rising incomes and
increased urbanicity (de Haen et al., 2003). Furthermore, household consumption patterns are
expected to change among urban middle class households that attain the supernumerary income
(Stone 1954) and sufficient market access to indulge their desire to luxury goods, such as
processed food.
By contrast, I find that both urban and rural households are rapidly transitioning towards
diets with higher shares of processed food, beginning at income levels below the international
poverty line22. Given the surprisingly low income level at which I observe diet change, I seek to

Countries include Botswana, Burundi, Ethiopia, Kenya, Lesotho, Malawi, Mozambique,
Namibia, Rwanda, South Sudan, Swaziland, Tanzania, Uganda, Zambia, and Zimbabwe.
22 The World Bank (2015) defines the poverty line $1.90 daily per adult equivalent expenditure,
calculated with purchasing power parity adjusted 2011 constant international dollars.
21

77

understand the pattern in diet change – its speed and direction – across the entire income
distribution.
I recognize that non-income factors such as rural nonfarm employment (Senauer et al.,
1986), education (Turrell & Kavanagh, 2006), preferences from urbanization and proximity to
urban areas (Thiele and Weiss, 2003), and even “lifestyle changes” (Huang and Bouis, 2001) can
affect the food consumption patterns, and I include these in the analysis. Beyond the urban/rural
distinction, city size has been shown to affect other lifestyle patterns of households within lowincome countries such as income and poverty – where income is greater in larger cities, and
medium and small cities lead to greater quantities of migrating households escaping from
poverty (FerrĂŠ et al., 2012; Christiaensen et al., 2013; BerdeguĂŠ et al., 2015). Therefore, I choose
to test for the effects of city size on the consumption pattern of processed food.
The paper proceeds as follows. Section two presents the conceptual model. Section three
addresses the context, data, definitions, model, and estimation procedures. Section four presents
results, and section five concludes.

2.2 Conceptual Framework
Households select their food consumption patterns to maximize their utility from the food
that they consume. The heterogeneity of households’ results in each household possessing a
unique utility function based on the consumer demand characteristics regarding preferences for
taste, energy density, and convenience; on social and individual norms and beliefs regarding
informal institutions, beauty, and lifestyle; and on health concerns including foods’ nutritional
content and perceived safety. Households maximize utility functions conditional the affordability
of these goods subject to their income (m).

78

𝑀𝑎𝑥 𝑢(𝒙), 𝑠. 𝑡. 𝒑𝒙 ≤ 𝑚

(2.1)

𝑜𝑝𝑡𝑖𝑚𝑎𝑙 𝑏𝑢𝑛𝑑𝑙𝑒: 𝑥(𝒑, 𝑚)
Prices include market prices as well as shadow prices for the production and preparation
as households can purchase food from markets or source food from own production.
I begin the conceptual framework with Bennett’s Law providing a relationship between
household income and food consumption patterns. Stone’s theory of supernumerary income adds
to this relationship by suggesting that income induced diet change results from purchases by
households with increasing levels of discretionary income. I relax the assumption of required
supernumerary income as nearly two thirds of the population within ESA23 lives below the
international poverty line24, which otherwise would have consolidated much of the recent
income-induced diet change among the non-poor. If I find a change in the direction or speed of
diet change that occurs at income levels below those that would have normally associated with
increasing discretionary income, then I would want to know the specific empirical configuration
of the pattern; specifically, at what income levels does the pattern begin and what is the shape of
the pattern? I would want also to understand its drivers of consumption patterns that affect the
consumption patterns of the poor. More formally, controlling for income, what non-income
factors drive consumption patterns of poor households?
Rischke et al. (2010) conceptualized a framework that incorporates non-income drivers
by combining consumer demand, as discussed above, with food environment – highlight supplyside factors such as access, availability, price and market-induced desirability of food. Access
refers to the cost to access foods, which is subject to the density of retailers and households’

Authors’ calculation using the World Bank’s PovcalNet database.
The poverty line is defined as $1.90 daily per adult equivalent expenditure, calculated with
purchasing power parity adjusted 2011 constant international dollars (World Bank, 2015)
23
24

79

costs of reaching them. Availability refers to the variety of foods offered by formal and informal
markets. Prices vary across products and across retailers (GĂłmez and Ricketts 2013). Desirability
is the market-induced effects of shopping atmosphere and marketing.
Our conceptual framework expands that of Rischke et al. to include the effects of city
size (defined by population) on the urban environment that are exogenous drivers of food
environment and consumer demand (Figure 2-1). The urban environment of the city in which an
urban household is located would directly affect the household consumption patterns. Urban
environment affects rural household consumption patterns in the same way as they would affect
urban household consumption patterns, although the increasing costs of accessing cities and the
food markets therein as the distance from a household to cities increases would attenuate these
affects.

Figure 2-1: Conceptual framework

80

I suggest six effects of rising city size on the factors of urban environment that are
pertinent to food environment and consumer demand.
First, incomes rise and poverty rates fall as shown in recent empirical work across Africa
(FerrĂŠ et al., 2012, World Bank, 2009; Christiaensen et al., 2013). Both should increase the
consumption of processed food.
Second, the market size increases more than proportionally to city size due to the positive
relationship between city size and mean consumer incomes. Increased size of market attracts
investment in retail food markets; thereby improving physical access to them, which markets
expands the variety of foods that are available.
Third, infrastructure investment costs rises, given higher costs of land associated with
larger cities. Higher investment cost asymmetrically affects investment: greater impact on formal
food retailers that buy land compared to informal retailers that often do not pay for the space
they occupy. These costs are transferred to food prices, incentivizing consumers the use of
informal retail that may not have the same variety of food as formal retailers. The higher land
costs within and near larger cities increase the costs of urban and peri-urban farming, reducing
the likelihood of food production for own consumption.
Fourth, increased consumer exposure to modern advertising. Advertising affects
consumer preferences, encouraging the consumption of advertised food from modern food
retailers.
Third, increases access to public and private motorized transport. Motorized transport
improves physical access for outlets such as supermarkets that are larger and fewer in numbers,
but should have little or no effect on access to other outlet types, which have developed
historically to serve largely foot-bound consumers. Access to transport thus favors the use of

81

supermarket shopping, which provide consumers with lower prices of processed food (Neven et
al., 2006).
Fourth, congestion increases, diminishing the access advantage conveyed by motorized
transport and so increasing the cost of shopping in a formal retail market. The increased cost of
transportation also increases the costs of cross shopping25, reducing the benefits that multiple
markets offer in terms of available variety of food, increasing household benefits of shopping
from formal retail market, which offer a variety of food from one location.
Table 2-1 summarizes my expectations of the effects of city size on processed food
consumption via urban environment’s effect on the food environment and consumer demand
factors. Of the six urban environment factors, four drive strictly positive relationships between
city size and processed food consumption: mean incomes, market size, exposure to advertising,
and access to motorized transport. The cost of infrastructure investment and congestion have
mixed effects.

The practice of cross shopping is where households shop at various locations. Cross shopping
is common in China (Goldman 2000), Israel (Hino 2014) and in South Africa, where households
travel to locations away from their area of residence to acquire food (D’Haese and van
Huylenbroeck 2005, Strydom 2011).
25

82

Table 2-1: Impact of city size on food choice: urban environment, food environment, and
demand factors
Change in exogenous
urban environment
factor as city size rises

Impact on supply (food environment) and demand factors
Food environment

Demand factors
Higher incomes; rising
health concerns; more
“modern” norms &
beliefs

Increased incomes,
lower poverty

Disproportionate
increase in market size

Increased access to formal retail;
increased availability of processed foods

Higher investment cost

Reduced physical access to formal
markets; increased cost of owning
farmland

Greater exposure to
advertising

Increased desirability of processed food

Greater access to
motorized transport

Asymmetric impact on physical access of
markets – increases it for supermarkets
but little or no effect for other outlet types

Higher congestion

Reduces advantage conveyed by
motorized transport to purchase all food
from formal markets

Impact on
processed
food demand

Positive

Positive

Positive or
Negative
More “modern” norms
& beliefs around food

Positive

Positive

Higher opportunity cost
of time

Positive or
Negative

The net effect of city size on processed food consumption depends on the size and effect
of each factor. This framework does not provide a singular hypothesis for the effect of city size;
therefore, I considered the cumulative nature of the suggested effects and now hypothesize a
positive net effect of city size on processed food consumption.

83

2.3 Methods
2.3.1 Context
Developing Eastern and Southern Africa (ESA) is rapidly urbanizing. United Nations
Population Division26 estimates that twenty-one percent of the population in ESA lived in urban
settlements in 2000; by 2015, the urbanization level was 26 percent, and it projects forty-four
percent of the population will live in urban settlements by 2050. To represent this urbanizing
region of ESA I will use the nations of Malawi, Tanzania, Uganda and Zambia. These nations
include a variety of population based city “types”, including three primary cities with
populations greater than one million, thirty-eight secondary cities with populations between one
million and one hundred thousand, and over three hundred tertiary cities with populations below
one hundred thousand.
In addition to rapid urbanization, ESA is characterized by rising average per capita
income and a modernizing food system. Between 2000 and 2010, the population weighted
annual growth in per capita GNI27 of these four nations was positive each year and averaged
nearly four percent since 2004. In spite of the rapid growth of incomes, household data shows
that daily per adult equivalent incomes remain low, averaging incomes of just above three
dollars28 per day (Table 2-2). In the midst of the low average incomes, the “supermarket
revolution” is occurring in ESA, evidenced by the expansion of supermarket and other formal
food retail outlets (Reardon and Timmer, 2007). Formal retailers typically enter markets by
establishing their retail outlets near relatively affluent areas within the greater market (Battersby

World Urbanization Prospects: The 2014 Revision
World Bank data (http://data.worldbank.org/): GNI/capita, PPP (constant 2011 international $)
28 Daily expenditure levels valued in purchasing power parity (PPP) adjusted constant 2011
international dollars.
26
27

84

and Peyton, 2014), which has resulted in supermarkets directing significant investment into
larger cities and lesser investments into secondary and tertiary cities, thereby altering the food
retail options across city size.

85

Table 2-2: National expenditure and population by expenditure strata and city size
Total Expenditure (per adult equivalent)
Malawi
National
Rural
Urban

Tertiary City

Uganda

Zambia

Pooled

2.48

3.03

3.51

3.13

3.11

(1.67)

(2.26)

(2.54)

(1.74)

(2.19)

1.97

2.38

2.78

1.78

2.37

(1.52)

(1.97)

(2.23)

(1.30)

(1.91)

5.31

4.91

6.10

5.66

5.39

(3.09)

(3.50)

(4.37)

(3.74)

(3.78)

7.85

8.11

7.41

7.82

(5.83)

(6.39)

(4.62)

(5.72)

5.25

4.35

8.67

5.20

5.06

(3.14)

(3.34)

(4.56)

(3.68)

(3.43)

Primary City
Secondary City

Tanzania

5.52

3.52

5.10

4.48

4.31

(2.84)

(2.76)

(3.77)

(2.99)

(3.13)

Represented Population: Total (`000s) and Cumulative Density by Expenditure Strata
$0.50- $1.00- $1.50- $2.00- $3.00- $5.00Total
< $0.50
> $10.00
1.00
1.50
2.00
3.00
5.00
10.00
101,923

2.1

13.8

30.2

45.2

67.4

86.2

96.7

100

77,041

2.6

16.9

36.3

53.2

76.6

93.4

99.1

100

24,882

0.4

4.1

11.2

20.4

38.8

64.0

89.2

100

6,126

0.0

1.5

3.4

6.6

17.0

41.7

77.4

100

7,178

0.2

3.7

11.3

22.1

42.7

67.4

90.8

100

11,578

0.8

5.6

15.2

26.6

48.0

73.6

94.4

100

Source: authors’ calculations using national household level surveys
Notes: Daily average expenditure levels per adult equivalent are presented in purchasing power parity adjusted constant 2011 international dollars. Primary font indicates the population weighted average.
Parentheses indicate the population weighted median.

86

2.3.2 Data
Four nationally representative household surveys provide the household data for this
analysis: 2010/2011 Malawi Integrated Household Survey (sample size of 12,271), 2010/2011
Tanzania National Panel Survey (3,924), 2009/2010 Uganda National Panel Survey (2,975), and
2010 Zambia Living Conditions Monitoring Survey (19,397). Due to the nations’ geographic
proximity, similarities in national demographics, the timing of the surveys, and the similarity of
the data collection instruments I selected these datasets. In-person interviews performed by
enumerators of these surveys collected household data on consumption, employment, and other
household characteristics (Table 2-3).

Table 2-3: Household descriptive characteristics
National
Total household expenditure
per AE
Nonfarm employment
Dependency ratio
Household adult equivalents
Hectares of farmed land
Own a gas or electric stove
Own a refrigerator

Pooled Data
Primary
Urban
Cities
5.39
7.82

Rural

Secondary
Cities
5.06

Tertiary
Cities
4.31
(3.13)

3.11

2.37

(2.19)

(1.91)

(3.78)

(5.72)

(3.43)

27.1

21.4

44.7

52.6

41.9

42.1

(20.0)

(0.0)

(50.0)

(50.0)

(40.0)

(40.0)

47.1

49.8

38.9

33.4

38.7

41.9

(50.0)

(50.0)

(40.0)

(33.3)

(40.0)

(42.9)

6.7

6.8

6.4

6.5

6.2

6.4

(5.6)

(5.6)

(5.4)

(5.5)

(5.2)

(5.4)

3.5

4.1

1.4

0.5

1.2

2.0

(0.9)

(1.2)

(0.0)

(0.0)

(0.0)

(0.2)

12.0

6.9

1.9

21.5

33.2

23.8

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

8.7

2.3

27.7

48.8

29.3

12.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

13.5

12.3

17.1

20.7

12.9

17.8

(14.0)

(14.0)

(17.0)

(17.0)

(13.0)

(17.0)

45.6

46.2

43.7

43.9

42.8

44.2

(43.0)

(44.0)

(41.0)

(41.0)

(41.0)

(41.0)

Female headed household

22.1

21.2

24.9

26.1

21.2

26.6

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Own a phone

55.4

46.7

82.4

93.3

84.3

75.5

(100.0)

(0.0)

(100.0)

(100.0)

(100.0)

(100.0)

9.5

10.7

5.7

10.3

4.7

3.9

(4.5)

(5.7)

(2.0)

(2.0)

(2.0)

(1.4)

Maximum education attained
within the household
Age of household head

Kilometers to Market
Own a car
Own a motorcycle
Own a bicycle

2.9

1.2

8.2

12.1

10.7

4.6

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

4.7

4.6

5.2

3.9

4.3

6.4

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

47.0

51.9

31.9

13.7

35.4

39.3

(0.0)

(100.0)

(0.0)

(0.0)

(0.0)

(0.0)

Source: authors’ calculations using national household level surveys
Notes: Daily average expenditure levels per adult equivalent are presented in purchasing power parity adjusted constant 2011 international dollars.
Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

87

The expenditure data are valued in purchasing power parity (PPP) constant 2011
international dollars, using historic exchange rates obtained from XE.com and conversion factors
for constant PPP valuation from worldbank.org. I use per adult equivalent (AE)29 levels of
expenditure to calculate total household expenditure, total food consumption, and totals for
various food consumption aggregates. Each survey collected data on the value of food
consumption with a seven day recall, with the exception of Zambia, which used a two or four
week recall depending on the food item. The valuation of consumed own production varies by
dataset. Stated values of food consumed from own production are collected by the Uganda and
Zambia surveys30, while the Malawi and Tanzania datasets include quantities of consumed own
production that I value using imputed household and community prices of marketed food. To
estimate these values, I calculated prices of purchased food items that match items from
consumed own production. The medians of these purchased food prices are then calculated based
on locality and multiplied by the quantities of consumed own production to estimate the value of
consumption.
City population data from http://www.citypopulation.de/, a website that assembles recent
census data, paired with household level data identify the population size of the cities where
urban households reside. Distance measurements for rural Zambia are approximated using
distance from center of rural constituency to the nearest primary, secondary and tertiary cities
with the distance calculator at https://www.daftlogic.com/.

Per adult equivalent values are calculated as one AE for each household member fifteen years of age or older, 0.75
AE for each child aged five to fourteen and 0.50 AE for each child younger than five years old.
30 The observations where Uganda reported farm-gate prices, I imputed market purchase prices to represent the
value of own production.
29

88

2.3.3 Definitions
Processed share is the term I use to represent the share of processed food value in total
food expenditure. Processed food is defined as food that prior to household acquisition has been
transformed from its original state, beyond removal from the plant and (for non-perishables)
drying. Processed food therefore identifies food embedded with the labor of humans or
machines.
To analyze the consumption of processed food with more granularity I create five
processing sub-aggregates through the following three steps. First, I categorize food items into
maize and non-maize food aggregates; as maize is the primary staple food in the region and
could dominate the consumption patterns. Second, I disaggregate both the maize and non-maize
food aggregates into processed and unprocessed aggregates as defined above. Third, I further
disaggregate the processed non-maize aggregate into processed low and processed high food
aggregates. I assign to the high category to food items if they satisfy at least two of the following
three conditions: multiple ingredients; physical change induced by heating, freezing, extrusion,
or chemical processes (i.e., more than simple physical transformation); and packaging more
complex than simple paper or plastic. The resulting sub-aggregates are unprocessed maize,
processed maize, unprocessed non-maize, low processed non-maize, and high processed nonmaize (Table 2-4). Although this aggregation delineates between unprocessed, processed low
and processed high based on the structure of the food system and the value-added that they
represent, this aggregation closely mirror the processed food aggregates considered in nutritionbased literature31 (Monteiro et al., 2010).

Table 2-A-1 shows a sample of the food items and how the items map into both the Monteiro et al. (2010)
aggregates and the processing aggregates.
31

89

Table 2-4: Processed food aggregates
Share
Processed consumption
Unprocessed Maize
Processed Maize
Unprocessed
Low Processed
High Processed

Definition
Share of any food acquired in processed form
Maize acquired in unprocessed form, e.g. maize grain
Maize acquired in processed form, e.g. maize meal
Non-maize food items acquired as unprocessed, e.g. common bean
Non-maize food items acquired as low processed, e.g. wheat flour
Non-maize food items acquired as high processed, e.g. bread

City size is the term used to classify cities by population. For analysis, I categorize cities
by population thresholds: primary cities have populations greater than one million, secondary
cities have populations below one million and above than 100,000, and tertiary cities have
populations below one hundred thousand.

2.3.4 General model
I follow a four-part analysis to estimate the effects of city size and distance to cities on
food consumption patterns. I establish the empirical configuration of the pattern of diet change
across the income distribution in the first two parts, and explore drivers of the consumption
patterns, that include the effects of city size, in the last two parts.
Part 1 is nonparametric analysis in the form of LOWESS curves, which impose no
structure on the data, allowing the data to reveal patterns of consumption across the expenditure
distribution. This analysis provides a first assessment of identifying at what expenditure level
changes in consumption patterns begin, and whether changes are sudden and rapid or gradual.
Part 2 formally establishes expenditure breakpoints, and rates of change before and after
them, by testing for thresholds where significant change occurs in (a) the rate of change or (b)
the direction of the relationship between food consumption patterns and total expenditure. By
imposing linear relations on the food consumption patterns before and after an estimated
expenditure breakpoint, part 2 formally estimates particular expenditure levels at which patterns

90

change, and estimates patterns of change at expenditures below the breakpoint (among lower
income households) and above it (among higher income households).
Part 3 examines drivers of consumption change with the use of Engel curve based
regressions to estimate the marginal effects of income and non-income determinants on food
consumption shares. Parts 1 & 2 direct the selection of an appropriate functional form of the
Engel curve model that I use in the regression analysis.
Part 4 uses a mediation model to estimate the indirect effects of city size and distance to
city on food consumption patterns occurring via other determinant variables. The total effects of
city size and distance to cities incorporate the estimated effects from the regression analysis in
part 3 with the indirect effects from part 4.
I complete each part of the analysis separately for rural and urban households and by
country, with descriptive and non-parametric analysis highlighting city size differences that I
further analyze in the parametric analysis.

2.3.5 Empirical model
The nonparametric analysis and the expenditure threshold analysis will estimate the
marginal effect of income (𝑌 ∗ ), without controlling for any other determinants of consumption,
on processed share and five sub-aggregates that depict processed food consumption.
𝐷𝐹𝑜𝑜𝑑 𝐴𝑔𝑔𝑟𝑒𝑔𝑎𝑡𝑒 = 𝑓(𝑌 ∗ )

(2.2)

In parts 3 & 4 of the analysis, I use an Engel curve model to estimate the marginal effects
of income and non-income household variables, including city size and distance to city, on each
of the dependent variables shown in Table 2-4. The conceptual model posits that city size
influences the urban environment, which affects both the food environment and consumer

91

demand characteristics. To specify the model empirically, I include variables that capture as
many of the supply and demand side factors as possible. In the urban analysis, I include dummy
variables for primary and secondary cities that reflect city size differences, and in the rural
analysis, I include variables for the distances to various city sizes. I use the conceptual model to
interpret the meaning of the city size variables and generate defensible conclusions in instances
where data to represent factors of city size are unavailable or imperfectly capture the concepts in
the conceptual model.
Data from the national datasets sufficiently capture consumer demand characteristics,
with variables that proxy for incomes, preferences, norms and beliefs, and health concerns. The
national datasets provide less comprehensive coverage of the food environment. Four variables
from the datasets are relevant to accessing food: own a car, own a motorcycle, own a bicycle,
and distance to nearest market (for rural households), but measures of availability, price and food
desirability are not available. The national datasets do not capture household level prices well
due to the high prevalence of non-standard units, limiting the ability to apply prices to food
aggregates. I capture systematic differences across cities with the inclusion of city dummy
variables in the pooled regressions.
I am able to directly control for two of the six urban environment factors within the
conceptual model: income and access to motorized transport. City size dummies capture the four
remaining factors of city size: market size, congestion, investment cost, and advertising. I
estimate the marginal effects of income, city size, distance to cities and the other determinant
variables with the following Engel’s Curve Model:
𝐷𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒 = 𝑓(𝐴𝑐 ∗ , 𝐴𝑣 ∗ , 𝑃∗ , 𝐷 ∗ , 𝑌 ∗ , 𝑍 ∗ , 𝑊 ∗ , 𝐻 ∗ )

92

(2.3)

The dependent variables are processed share and the processing sub-aggregates (Table 24). The determinants identified in equation three are the factors of food environment and
consumer demand characteristics, each with household variables that proxy for their effects as
shown in Table 2-5 and described below.

Table 2-5: Determinants of demand in the Engel’s Curve Model
Ac*

Access

Own a car, Own a
motorcycle, Own a bicycle,
Distance to market (rural),
City size (urban), Distance
to cities (rural)

Y*

Disposable
Income

Z*

Individual
and
Household
Preferences

Availability

City size (urban), Distance
to cities (rural)

P*

Price

City size (urban), Distance
to cities (rural)

W*

Social and
Individual
Norms and
Beliefs

D*

Desirability

City size (urban), Distance
to cities (rural)

H*

Health
Concerns

Av*

Daily total expenditure per adult
equivalent, City size (urban), Distance to
cities (rural)
Nonfarm employment, Dependency ratio,
Adult equivalents, Farmed hectares of
land, Own a stove, Own a refrigerator,
Maximum level of education, Age of
household head, Female head of
household, City size (urban), Distance to
cities (rural)
Own a stove, Own a refrigerator,
Maximum level of education, Age of
household head, Female head of
household, Own a telephone, City size
(urban), Distance to cities (rural)
Maximum level of education, Female
head of household, Dependency ratio,
City size (urban), Distance to cities (rural)

City size variables that affect the urban environment in the urban analyses are dummy
variables for residence in primary- and secondary cities (tertiary cities are excluded), and in the
rural analyses are distance to nearest primary city, distance to nearest secondary city, distance to
nearest tertiary city. In the urban analyses, these variables simply identify the general city size
where the household resides, in turn identifying the urban environment. In the rural analyses,
these variables account for accessing the urban environment and are included in logarithmic
form to account for diminishing marginal effects of distance. As distances increase, there is a
reduced effect of urban environment on food environment, consumer demand characteristics and
ultimately food consumption decisions. The attenuation of these effects are a result of increased
costs of accessing urban areas and the diminished spillover effects of the urban environment at

93

greater distances. I expect that households near larger cities will have consumption patterns
similar to those of urban residents due to the density of commercial activity within larger cities
that is likely to lead to greater commercial networks immediately surrounding the cities.
I use the dummy variables ownership of a car, ownership of a motorcycle, and ownership
of a bicycle to proxy for access to foods. Ownership of any of these transportation vehicles
reduce the cost of access, while increasing the ability to shop in bulk with decreasing magnitude
in the following rank: car, motorcycle, and bicycle. Distance to market also acts as a proxy for
access to foods as greater distances increase the shadow price of transportation to markets,
reducing household ability to purchase processed food. Distance is measured in kilometers and it
enters the estimation in a logarithmic form to account for diminishing effects of distance.
Variables of household distance from household to market are available for Tanzania and
Zambia, and distance from community to market are available for Malawi and Uganda. Distance
to market variables are only included in rural household analysis, as certain distances are only
available at the community level.
Total daily household expenditure per AE proxies for income, a key factor of consumer
demand. The estimation methods section explains how expenditure enters the estimation
equations.
I include nine variables, in addition to city size and distance to cities variables, to proxy
for individual and household preferences. I calculate nonfarm employment as the percentage of
working age32 adults employed (including self-employment) in nonfarm activities. I calculate the
household’s dependency ratio by dividing the number of dependents by the total members within
the household. Both nonfarm employment and dependency ratio increase household opportunity

32

Ages 15-64
94

cost of time, thereby incentivizing households to reduce their time allocated towards preparing
food, which incentivizes the consumption of processed foods. I calculate household adult
equivalents as the total number of adult equivalents within a household because more household
equivalents reduce the opportunity cost of time based on economies of scale within households,
therefore reducing the benefit of processed food consumption.
Household shadow prices regarding household ability to produce, prepare and/or preserve
food vary based on household characteristics such as the amount of land farmed, ownership of a
gas/electric stove and ownership of a refrigerator. Land farmed is the variable representing the
hectares of land that a household farmed during the previous harvest season, which enters the
analysis in logarithmic form to account for diminishing marginal effects of the amount of land
farmed on food consumption patterns. Farming one’s own land increases the likelihood of
consuming food from own production, which would directly reduce the share of processed food
consumption that by definition households acquired in the processed form. Ownership of a
gas/electric stove will reduce the preparation cost of food, diminishing the benefits of processed
food. Ownership of a refrigerator reduces the cost of food preservation, reducing the benefit of
low perishability that is commonly associated with processed food.
The maximum level of education within a household is an ordinal value based on the level
of education acquired by any member of the household. Education is positively associated with
healthier consumption patterns, which could affect the consumption of processed foods (Turrell
and Kavanagh, 2006). Elderly households are more likely to maintain traditional diets, with low
levels of processed food; therefore, I include age of the head of the household in the empirical
model. Finally, as females have a greater propensity to delay gratification and consume healthier

95

food options (Smith, 2003), I include the gender of the head of the household in the empirical
model.
Ownership of a telephone is a proxy for social and individual norms in addition to
ownership of a gas/electric stove, ownership of a refrigerator, maximum level of education, age
of household head, and female head of household, which I already described. Ownership of a
phone is a zero one dummy variable that indicates household tendency towards modernization
and willingness to modernize food consumption patterns.
The variables that proxy for health concerns are maximum level of education, female
head of household, and dependency ratio, which I have previously described.

2.3.6 Estimation methods
I use LOWESS curve regressions (Cleveland, 1979; Cleveland and Devlin, 1988) to
observe the nonparametric relationship between the food budget share and the household per
adult equivalent total expenditure. My second application of LOWESS curves is to map the
relationship between distance to various city types and rural food consumption patterns. The
LOWESS procedure is a robust form of local polynomial smoothing that does not impose
functional form on the relationship between dependent and independent variables. Compared
with other forms of local polynomial smoothing, this specific form of local polynomial
smoothing is more resistant to statistical outliers because it applies smaller local weights to
estimates with large residuals, which is valuable given the nature of household survey data. I
completed the LOWESS analysis on expanded data where I generated duplicate observations
according to population weights to represent the sample population.

96

I test for expenditure thresholds in the second part of the analysis, to observe changes in
the direction or rate of change in the food consumption patterns. I use a non-linear estimation to
find the best fit of two linear curves that join at a singular point that optimizes the fit of the two
curves. This estimation method is as a piecewise regression (Serber and Wild, 1989) that takes
the following form:
𝐸[𝑦|𝑥] = {

𝛽10 + 𝛽11 𝑥, 𝑥 ≤ 𝛼
𝛽20 + 𝛽21 𝑥, 𝑥 > 𝛼

(2.4)

𝑠. 𝑡. 𝛽10 + 𝛽11 𝛼 = 𝛽20 + 𝛽21 𝛼
In the piecewise regressions, the household food budget shares of food item aggregates
(y) are regressed against household total expenditure per adult equivalent (x), generating
inflection point values and slopes of each curve around the inflection points with robust standard
errors. This estimation method shows statistically significant changes in the consumption
patterns when the piecewise regression identifies a kink in the consumption curve at the
inflection point.
The third part of the analysis incorporates the nonparametric analysis and the threshold
analysis by contributing to the identification of the functional form of the Engel curve. Four
common forms of Engel’s curves are: the linear relationship suggested in Engel’s original work
(1857), the linear relationship between log income and the dependent variable as identified by
Working (1943), Working’s model with the addition of the inverse of income as proposed by
Leser in 1963, and the functional form proposed by Banks, Blundell and Lewbel (1996) that
includes both log income and squared log income. Each of these forms follow the base form of a
share of consumption that is dependent on income and other household variables. The inclusion
of non-linear relationships between income and the consumption of goods has been an
improvement to the linear functional form as the non-linear relationships enable the analysis of

97

consumption patterns that exhibit differing marginal, potentially inverted, effects of income on
the relative shares of consumption of goods over strata of household income.
I follow a two-step process to select the appropriate functional form of the Engel curve.
First, I use the results of the nonparametric and semi-parametric analyses to narrow the options
of the functional form based on the general shapes of the patterns between total expenditure and
processed share. Second, I complete regressions using the reduced list of functional forms and
compare the fit of the regressions with adjusted R2 or chi-squared values to identify the best fit. I
treat total expenditure as exogenous in food demand analysis as recommended by Subramanian
and Deaton (1996).
Consumption shares bound the dependent variables in the Engel’s Curve Model between
zero and one. The standard estimation method for such a dependent variable is to convert it with
a log-odds transformation and then to estimate with ordinary least squares (Wooldridge, 2010).
Food consumption data from in one-week survey results in many households with food budget
shares of zero or one hundred percent, disallowing the use of the log-odds transformation. Given
the characteristics of the dependent variable and the assumption of asymptotically normal
standard errors, the appropriate method of estimation is the fractional probit model (Papke and
Wooldridge, 1996 and 2008). This model estimates the marginal effects of the predictors of a
dependent variable that takes values of a closed set of zero to one.
In the fourth part of the analysis, I use a mediation model to test for indirect effects of the
variables city size and distance to city on food consumption patterns. City size and distance to
city are predictor variables that could affect other determinant variables (mediator variables),
which also affect food consumption patterns. I would not capture the indirect effects of city size
and distance to city variables that occur through the mediator variables without the use of a

98

mediation model. I calculate the indirect effects of predictor variables by multiplying the
estimated effect of a mediator variable on a dependent variable with the estimated effect of the
predictor variable on the mediator variable. I use OLS to estimate the city size and distance to
city effects on the mediator variables, and bootstrap the standard errors of the indirect effects
(Wooldridge, 2010). The estimation of indirect effects follows the methodology described in
Hayes and Preacher (2010).

2.4 Results
I use descriptive statistics, LOWESS curves analysis and piecewise regressions to show
the relationship of processed food consumption with income and then across city size in urban
areas and distance to cities in rural areas. Then I will discuss the results of the full Engel’s Curve
Model regression and the results of the mediation model analysis.

2.4.1 Descriptive statistics and nonparametric analysis of the relationship between processed
food consumption and total household expenditure
Tables 2-6 & 2-7 show the consumption pattern of processed food across the expenditure
strata of urban and rural households, while the LOWESS curves (Figure 2-2) visually represent
these same relationships. These descriptive statistics and estimated curves do not show causation,
but are valuable in observing preliminary patterns between household total expenditure and
processed share.

99

Table 2-6: Food budget shares of processed food, urban households by country and expenditure
Average Processed Share by
Expenditure Terciles

Average Processed Share by Expenditure Strata
Urban

Total
62.5

54.5

$0.501.00
56.4

(66.3)

(60.0)

(58.5)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
51.4

$1.502.00
56.2

$2.003.00
62.0

$3.005.00
65.7

$5.0010.00
68.9

(55.6)

(57.0)

(65.4)

(68.2)

(71.4)

> $10.00

Low

Middle

High

66.4

50.0

57.8

65.7

(71.4)

(52.7)

(59.6)

(68.7)

66.2

44.4

63.1

56.2

54.9

59.3

67.9

74.1

78.8

56.8

58.7

72.2

(69.8)

(56.7)

(66.8)

(59.3)

(59.2)

(65.5)

(70.3)

(76.2)

(80.3)

(60.4)

(63.4)

(75.0)

54.7

73.3

44.7

37.9

46.4

42.5

52.8

60.9

69.1

44.8

44.9

59.9

(56.2)

(78.6)

(52.5)

(42.1)

(47.1)

(38.9)

(55.6)

(61.0)

(67.7)

(47.1)

(43.8)

(60.5)

65.4

46.0

50.3

56.4

59.3

63.7

67.3

69.8

72.5

51.7

60.4

69.0

(68.4)

(46.7)

(53.5)

(58.2)

(62.2)

(66.7)

(68.8)

(71.1)

(73.8)

(54.3)

(63.5)

(70.8)

62.7

52.9

56.8

52.9

54.2

56.7

63.3

68.7

73.6

52.8

55.7

67.6

(66.8)

(56.7)

(60.8)

(57.0)

(57. 6)

(61.3)

(66.7)

(70.7)

(75.0)

(56.0)

(60.1)

(69.7)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

Table 2-7: Food budget shares of processed food, rural households by country and expenditure
Average Processed Share by
Expenditure Terciles

Average Processed Share by Expenditure Strata
Rural

Total
37.0

38.5

$0.501.00
34.7

(35.1)

(32.9)

(29.0)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
34.0

$1.502.00
35.3

$2.003.00
38.3

$3.005.00
43.0

$5.0010.00
52.0

(30.3)

(32.9)

(37.4)

(42.9)

(52.3)

> $10.00

Low

Middle

High

56.3

34.7

35.2

42.6

(69. 1)

(29.8)

(32.3)

(42.3)

39.0

50.7

36.1

34.6

35.0

38.1

43.0

54.4

65.7

36.0

37.4

46.3

(36.5)

(51.5)

(31.1)

(30.4)

(31.8)

(36.1)

(41.5)

(53.7)

(65.5)

(32.0)

(35.1)

(46.9)

30.7

25.9

26.7

26.5

27.4

28.5

33.9

43.6

64.7

26.4

28.6

40.6

(26.6)

(9.9)

(26.8)

(20.2)

(22.2)

(25.9)

(29.0)

(45.9)

(71.4)

(22.0)

(24.4)

(38.9)

33.2

29.1

27.3

30.8

34.3

37.6

43.6

56.1

59.1

28.3

34.3

45.2

(28.9)

(24.2)

(23.1)

(26.5)

(31.1)

(34.5)

(43.9)

(59.0)

(67.3)

(24.1)

(30.8)

(45.7)

35.6

41.2

32.1

31.9

32.8

34.7

39.6

49.9

63.5

32.0

34.1

43.9

(32.3)

(36.6)

(26.8)

(27.4)

(28.5)

(32.8)

(37.1)

(51.2)

(67.7)

(26.8)

(31.2)

(43.7)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

100

Figure 2-2: Nonparametric (LOWESS curve) analysis of consumption patterns of processed food
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

Figure 2-2 shows a general pattern in both urban and rural households of initial decline in
the share of processed food followed by a monotonic rise of processed share as total
expenditures rise. The only exception to this pattern is in urban Zambia, where there is no initial
decline. The overall positive correlation between processed food consumption and expenditure is
consistent with the patterns observed in low-income countries in Asia, where households
consume greater shares of processed food at higher levels of household expenditure (Pingali
2007). The pattern shown here resembles the functional forms of Engel curves for luxury goods
as suggested by Leser (1963) or Banks, Blundell and Lewbell (1997). Beyond the general
pattern, there are three results of note.
First, the consumption of processed food is greater for urban households (62.7%) than for
rural households (35.6%). The larger average shares of purchased food33 by urban households

33

Author’s calculation from pooled data

101

(83.4%) than by rural households (45.6%) are likely to drive this relationship, as processed food
by definition cannot be acquired by own production.
Second, the positive pattern of processed share begins below the poverty line in all
countries for both urban and rural households. The LOWESS estimates of the pooled data show
that lowest estimated share of processed food in urban households (53.1%) occurs at total
expenditure of $1.22 and the lowest share of processed food in rural households (31.8%) occurs
at total expenditure of $1.16. These estimates, which are consistent with the average processed
shares by expenditure strata, show two findings: One, the demand for processed food occurs at
all levels of income; it is not merely a middle-class story. Two, processed food are a significant
portion of households’ diets at any income level, as evidenced by the lowest share of processed
food consumption being more than half of the food value consumed by urban households, and
nearly a third of the food value consumed by rural households.
Third, the poorest households do not consume the lowest share of processed food value
across the expenditure strata. I suggest two explanations of this. One, the poorest households
have the least amount of land that could be used to produce food for own consumption. This lack
of capital results in poor households not being able to produce sufficient food for own
consumption necessitating their purchase of food, including processed food. Two, the poorest
households are likely to receive the greatest share of their income in the form of remittances,
which leads to these households purchasing a significant share of their food value. The data do
not consistently show the share of income from remittances, but the data show that the share of
food value received in the forms of food in kind or gifts34, which I assume is correlated with
remittance income, is greatest among the poorest households.

Ten percent of food value among households with total daily expenditure below $1.00, six percent of food value
among households with total daily expenditure above $1.00.
34

102

Figure 2-3 and Table 2-A-2 of the appendix show additional depth regarding patterns of
processed food consumption in the form of processing level aggregates. Highly processed food,
one of the processing level aggregates, has garnered much attention in nutrition literature focused
on low-income countries (Popkin et al., 2012; GĂłmez et al., 2013; Monteiro et al., 2013). Highly
processed food contains many utility enhancing attributes such as convenience and taste in the
form of increased salt and fat. My findings show that highly processed foods exhibit the largest
increase in food budget share among both rural (9% - 21%) and urban (17% - 31%) households
across the expenditure stratum35, with the transition towards greater shares of highly processed
food beginning at total expenditures of under one dollar per day. Unprocessed maize shows the
greatest decline in food budget share, where three of the four countries consume their maximum
shares of unprocessed maize at expenditures levels below $1.00. These patterns reinforce and
expound upon the patterns identified in nutrition literature.

35

These results are for pooled households with daily total expenditure below $5.00.

103

Figure 2-3: Nonparametric (LOWESS curve) analysis of processed food aggregates – pooled
data
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

2.4.2 Expenditure threshold analysis
The expenditure threshold analysis shows that patterns of rural and urban processed share
vary across income levels, and that the rise of processed share commonly begins among poor
households (Table 2-8). Expenditure breakpoints for rural Malawi and Tanzania are $1.10 and
$0.87 respectively, breakpoints that reveal negative slopes in the consumption patterns below the
breakpoint and positive beyond the breakpoint. Positive relationships between expenditure and
processed share bracket the expenditure breakpoints of rural Uganda and Zambia. These results
suggest that supernumerary income is not a necessary condition for households to increasing
processed food consumption.

104

Table 2-8: Expenditure thresholds and slopes prior to and after expenditure thresholds in
processed food consumption patterns
Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Rural
Slope Before
Breakpoint
-0.055

Slope After
Breakpoint
0.038

Expenditure
Breakpoint
$3.55

Urban
Slope Before
Breakpoint
0.058

Slope After
Breakpoint
-0.019

(0.000)

(0.070)

(0.000)

$0.87

-0.325

0.038

(0.000)

(0.000)

(0.544)

$1.53

-0.089

(0.000)

(0.004)

0.057

(0.000)

(0.000)

(0.309)

$3.29

(0.000)

0.014

0.095

$2.12

-0.025

0.053

(0.000)

(0.159)

(0.000)

(0.041)

(0.714)

(0.062)

$2.14

0.061

0.041

$2.08

0.091

0.024

(0.003)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

$0.86

-0.200

0.034

$1.40

-0.053

0.042

(0.000)

(0.000)

(0.000)

(0.000)

(0.438)

(0.000)

Expenditure
Breakpoint
$1.10

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

The urban expenditure threshold analysis is less statistically significant than the rural
analysis, but the statistically significant results indicate a positive relationship between
household expenditure and processed food consumption, regardless of the expenditure level.
Urban Zambia provides the only instance where the slopes of this relationship are significant
above and below the breakpoint, and it is of note that the slope prior to the expenditure
breakpoint is greater than beyond the expenditure breakpoint.
Tables 2-9 & 2-10 display the results of the expenditure breakpoint analysis for the
processing sub-aggregates, with five findings noted here. First, each36 of 13 statistically
significant pairs of slopes representing rural or urban country level processing aggregate
consumption patterns have the same direction and are of greater absolute value before the
expenditure breakpoints than beyond the breakpoints: negative slopes for maize and unprocessed
non-maize, positive for non-maize processed low and processed high foods. This pattern of the
relative magnitude of slopes highlights the growth of processed food consumption among the
poor.

This excludes the slopes that bracket the expenditure breakpoint for highly processed non-maize in Uganda,
which has a statistically significant expenditure breakpoint at $0.45. Half of one percent of the observations in
Uganda have expenditures less than $0.45.
36

105

Table 2-9: Expenditure thresholds and slopes prior to and after expenditure thresholds for processed food aggregates – urban data
Countries
Malawi
Tanzania

Maize Unprocessed
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$2.50
-0.048
-0.001
(0.000)

(0.007)

(0.745)

(0.000)

(0.000)

(0.001)

$2.62

0.012

-0.038

$2.68

-0.026

-0.007

(0.615)

(0.001)

(0.004)

(0.051)

(0.412)

-0.011

-0.006

$1.95

-0.054

-0.004

(0.423)

(0.285)

(0.001)

(0.202)

(0.682)

-0.051

-0.009

$0.84

0.118

-0.020

(0.000)

Uganda
Zambia
Pooled

Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Maize Processed
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$0.97
-0.492
-0.018

$2.00
(0.000)

(0.000)

(0.000)

(0.001)

(0.375)

(0.000)

$2.59

-0.010

-0.020

$2.69

-0.034

-0.006

(0.164)

(0.437)

(0.008)

(0.000)

(0.000)

(0.219)

Unprocessed
Slope
Expenditure
Before
Breakpoint
Breakpoint
$1.19
0.047

Slope
After
Breakpoint
-0.023

Processed Low
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$2.66
0.052
0.004

Processed High
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$2.57
0.063
0.024

(0.000)

(0.401)

(0.000)

(0.000)

(0.000)

(0.628)

(0.000)

(0.000)

$1.47

0.061

-0.035

$1.42

-0.094

0.021

$1.47

0.001

(0.006)

0.051

(0.001)

(0.362)

(0.000)

(0.000)

(0.191)

(0.001)

(0.064)

(0.988)

(0.000)

$2.28

0.033

-0.053

$0.90

-0.129

0.022

$0.44

-3.256

0.032

(0.010)

(0.613)

(0.049)

(0.067)

(0.619)

(0.042)

(0.000)

(0.000)

(0.005)

$2.09

-0.041

-0.015

$1.88

0.070

0.020

$2.81

0.039

0.019

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

(0.001)

$1.46

0.053

-0.026

$1.24

-0.038

0.021

0.037

0.036

(0.000)

(0.194)

(0.000)

(0.001)

(0.488)

(0.000)

(0.000)

(0.000)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

106

Table 2-10: Expenditure thresholds and slopes prior to and after expenditure thresholds for processed food aggregates – rural data
Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Countries
Malawi
Tanzania
Uganda
Zambia
Pooled

Maize Unprocessed
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
-0.045
-0.040

Maize Processed
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$1.48
-0.140
-0.009

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

$1.93

0.022

-0.046

$1.79

-0.062

0.000

(0.000)

(0.178)

(0.000)

(0.000)

(0.000)

(0.936)

$3.20

0.002

-0.024

$2.44

-0.017

0.010

(0.000)

(0.784)

(0.036)

(0.000)

(0.032)

(0.241)

$2.29

-0.036

-0.024

$0.55

-0.168

0.005

(0.043)

(0.000)

(0.002)

(0.000)

(0.090)

(0.009)

$1.56

-0.024

-0.033

$1.81

-0.055

0.001

(0.173)

(0.071)

(0.000)

(0.000)

(0.000)

(0.663)

Unprocessed
Slope
Expenditure
Before
Breakpoint
Breakpoint
$2.00
0.035

Slope
After
Breakpoint
-0.006

Processed Low
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$2.18
0.050
0.011
(0.000)

(0.000)

Processed High
Slope
Slope
Expenditure
Before
After
Breakpoint
Breakpoint Breakpoint
$1.90
0.051
0.029

(0.000)

(0.000)

(0.269)

(0.016)

(0.000)

(0.000)

(0.000)

$0.77

0.232

-0.009

0.008

$2.00

0.016

0.045

(0.000)

(0.083)

(0.170)

(0.015)

(0.000)

(0.097)

(0.000)

$2.56

-0.007

-0.051

$1.73

0.006

0.025

$3.20

0.008

0.034

(0.000)

(0.616)

(0.000)

(0.036)

(0.827)

(0.000)

(0.000)

(0.053)

(0.010)

$3.72

-0.024

0.025

$1.98

0.034

0.020

$2.11

0.027

0.017

(0.000)

(0.000)

(0.564)

(0.005)

(0.000)

(0.004)

(0.001)

(0.000)

(0.000)

$2.10

0.036

-0.018

$0.75

-0.044

0.015

$3.08

0.024

0.040

(0.000)

(0.000)

(0.011)

(0.009)

(0.489)

(0.000)

(0.000)

(0.000)

(0.000)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

107

Second, among rural households the slopes of processed maize consumption patterns are
negative below the expenditure breakpoint ($1.46 pooled), and positive above the breakpoint.
This result is consistent with poor rural households increasing their income with increased
production of maize, a substitute for processed maize, yet choose to acquire processed grain at
low levels of total expenditure.
Third, non-maize processed food drives the income-induced growth of processed share in
urban areas as shown by positive slopes of processed share at expenditures above the breakpoint
and negative slopes of processed maize at expenditures.
Overall, the non-parametric and semi-parametric expenditure analysis suggest that there
is significant change in processed food consumption patterns among poor households in ESA,
and that the changing consumption patterns are often more rapid for the poor than for the nonpoor.

2.4.3 Descriptive statistics of processed food consumption by city size
Table 2-11 shows the food budget shares of processed food across settlement types that
are consistent with the hypothesis of increased consumption of processed food by households
within larger cities, which the results show with each country having a positive monotonic
relationship between processed share and city size.

108

Table 2-11: Food budget shares of processed food, aggregated country and settlement
National
Rural
Urban
Primary City
Secondary City
Tertiary City

Processed Share
Tanzania
Uganda
46.1
35.9

Pooled
42.2

Malawi
40.9

Zambia
44.4

(40.6)

(40.0)

(45.7)

(32.8)

35.6

37.0

39.0

30.7

33.2

(32.3)

(35.1)

(36.5)

(26.6)

(28.9)

(44.7)

62.7

62.5

66.2

54.7

65.4

(66.8)

(66.3)

(69.8)

(56.2)

(68.4)

71.0

76.8

61.3

69.2

(73.1)

(77.8)

(59.8)

(70.6)

66.1

64.2

68.8

53.2

66.5

(68.4)

(67.7)

(70.8)

(52.7)

(68.8)

56.2

56.6

58.7

52.4

59.4

(60.1)

(60.1)

(62.3)

(55.5)

(63.5)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

The food budget shares of the processed food aggregates across city size in Table 2-12
from which I highlight two consumption patterns. First, three of the five aggregates display a
monotonic relationship between processed share and increasing city size; negative for
unprocessed maize and unprocessed non-maize, positive for processed low non-maize. Table 2A-3, the country-specific form of Table 2-12 found in the appendix shows these same monotonic
patterns37, strengthening the pattern observed in the pooled data. These patterns are consistent
with the patterns of Table 2-11, as they show greater consumption of processed food within
larger cities.

Table 2-12: Food budget shares of processed food aggregates, pooled data by settlement
Pooled Data
National

Maize
Unprocessed
Processed
11.6
6.5

Unprocessed
46.2

Non-Maize
Processed Low
18.6

Processed High
17.1

(1.4)

(0.0)

(43. 1)

(16.1)

(11.4)

Rural

14.0

5.9

50.4

16.5

13.2

(6.5)

(0.0)

(48.9)

(13.3)

(8.6)

Urban

4.0

8.4

33.2

25.2

29.1

(0.0)

(5.1)

(30.6)

(24.9)

(24.5)

0.4

8.0

28.7

25.0

37.9

(0.0)

(5.4)

(26.6)

(24.5)

(34.1)

3.1

11.6

30.8

26.9

27.7

(0.0)

(8.5)

(28.8)

(26.8)

(24.5)

Primary City
Secondary City
Tertiary City

6.6

6.7

37.2

24.2

25.4

(0.0)

(2.8)

(33.6)

(24.1)

(20.3)

Source: authors’ calculations using national household level surveys

Unprocessed maize and unprocessed non-maize food budget shares in Uganda minimally deviate from the
monotonicity of the pattern.
37

109

Second, low processed foods are likely the “entry point” for processed foods among
rural- and low-income urban households, as they show no clear pattern with respect to city size
gradations (Table 2-12). Rising incomes and denser urban settlement likely drive a transition
towards more highly processed foods, suggesting that highly processed foods are the driving
force behind the consumption patterns of processed food.

2.4.4 Nonparametric analysis of the relationship between distance to cities and processed food
consumption
Figure 2-4 displays the LOWESS curve estimates of the relationship between distance to
city and processed share. Each curve, representing the relationships between processed share and
distances to cities of difference size, displays an inverse relationship between distance to city and
processed share, which is consistent with previous studies that have studied the effects of
distance to urban centers on lifestyle patterns (Fafchamps and Shilpi, 2003; Sharma, 2016). In
addition to these negative relationships, the proximity to larger cities has a larger effect on
processed share than distances to smaller cities. The difference in the effects of distance by city
size is consistent with the conceptual model that includes the attenuation of the effects of urban
environment that are sensitive to city size.

110

Food Budget Share of Processed Consumption

.3

.35

.4

.45

.5

.55

Rural Zambia

0

100

200
Distance (km)

Distance to Primary City
Distance to Tertiary City

300

400

Distance to Secondary City

Figure 2-4: Locally weighted scatterplot smoothing curves representing the typical processed
share relative to household distance to city
Note: LOWESS curve estimates use a bandwidth equal to 0.4.

2.4.5 Engel’s Curve Model regression analysis
The fractional probit regressions analysis builds upon the previous analyses by isolating
the impact of determinant variables using the full Engel’s Curve Model. I selected the functional
form of the Engel curve suggested by Banks, Blundell and Lewbel (1997). The descriptive
analyses highlighted the need for the functional form to allow for changing slopes of the food
consumption patterns, which narrowed the functional form options to those suggested by Banks
et al. (1997) and Leser (1963). Due to superior goodness of fit of the fractional probit regressions
I selected the Banks et al. suggested functional form. This functional form includes non-income
independent variables and two independent variables for expenditure: log total household
expenditure and the squared log of total household expenditure. As expenditure is included in the

111

model as two separate variables, I calculate the average partial effects of expenditure with the
following equation:
𝜕

1

= [𝑎 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒 + 2𝑏
𝜕(𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒)

𝑙𝑛(𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒)
𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒

] ∗ Ε𝜙[𝑋𝛽]

(2.5)

where 𝑎 equals the marginal effects of log expenditure, 𝑏 equals the marginal effects of squared
log expenditure, and the scaling factor in the right of the equation is equal to the average of the
normal densities of the predicted values of the dependent variable.
Here I first consider the effects of expenditure on processed food consumption and then
the effects non-expenditure variables, including the effects of city size and distance to cities, on
processed food consumption.
Households with higher total expenditures consume larger shares of processed food, as
shown by the positive average partial effects of total expenditure on processed share in Tables 213 & 2-14. This finding is consistent with previous analyses on the consumption of processed
food (Veeck & Veeck, 2000; Pingali, 2007; Popkin et al., 2012; Tschirley et al., 2015), and
provides a base relationship for the following three findings.

Table 2-13: Average partial effects of expenditure and city size on processed share – urban data
Urban
Household total
expenditure per AE
Household located in a
primary city
Household located in a
secondary city

Malawi
0.007
(0.004)

Processed Share
Tanzania
Uganda
0.017
0.009

Zambia
0.010

(0.000)

(0.001)

0.068

0.000

(0.000)

0.037

(0.000)

(1.000)

(0.000)

0.053

0.047

-0.053

0.043

(0.000)

(0.003)

(0.049)

(0.000)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

112

Table 2-14: Average partial effects of expenditure and distances to various city sizes on
processed share – rural data
(marginal effects of distance only for rural Zambia)
Processed Share
Rural
Malawi
Tanzania
Uganda
Household total
0.004
0.009
0.022
expenditure per AE
(0.287)
(0.062)
(0.000)
Distance to primary
city (log)
Distance to secondary
city (log)
Distance to tertiary
city (log)

Zambia
0.010
(0.009)

-0.045
(0.000)

-0.002
(0.554)

-0.005
(0.001)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

First, the marginal effects of expenditure on processed share are positive at all levels of
income (Figure 2-5) 38, and among rural households, the marginal effects are positive in all
countries at expenditure levels above $1.55. These effects show no supernumerary level of
expenditure in urban areas and, if one exists, in rural areas this level of expenditure would be
below the poverty line. The differences from the slopes of processed share represented by
LOWESS curves (Figure 2-2) necessitates the later analysis of additional determinants affecting
processed food consumption.

The graphed negative marginal effect of expenditure on processed share in Uganda is theoretical, as the dataset
does not include households at this level of total expenditure.
38

113

Figure 2-5: Estimated effects of expenditure on purchased share and the consumption of
processed food

Second, processed maize is not driving the transition towards processed food within
urban areas, as expenditure positively affects the consumption of processed non-maize
aggregates and negatively affects maize and unprocessed non-maize aggregates (Table 2-A-6).
Third, there is evidence of the adherence to Bennett’s Law within rural areas, and that
household desire of non-staple food at higher incomes contributes to the increased consumption
of processed food. Total expenditure has a positive effect on all non-maize aggregates other than
in rural Uganda, where households commonly consider matoke to be the primary staple food and
matoke is primarily included in the non-maize unprocessed sub-aggregate (Table 2-A-7).
The fractional probit regression analysis reveals three key results specific to nonexpenditure determinants of urban processed food consumption. First, larger city size positively
affects the consumption of processed food (Table 2-13). Three of the four countries show
significant marginal effects of the city size dummy variables that indicate a positive relation,
with only secondary cities in Uganda consuming a smaller share of processed food than do

114

households in tertiary cities. There is not a pattern of relative marginal effects between primary
and secondary cities, merely the significant difference between cities with populations above one
hundred thousand and those below. This result supports the hypothesis that larger cities influence
households to consume greater shares of potentially timesaving processed food, relative to
households within smaller cities.
Second, under further disaggregation of the consumption of processed food, I show that
household settlement in larger cites continues to affect processed food consumption (Table 2-A6). In each of the four countries, the marginal effects of city size on unprocessed and processed
maize shares show that households in larger cities favor the purchase of the latter. In three of the
four countries, the marginal effects of primary and secondary cites on the consumption of
unprocessed food is negative when significant, indicating that increased city size is associated
with less consumption of unprocessed food. Lastly, the significant marginal effects of city size
on low processed food are mixed and on high processed food are positive. Given the positive
marginal effect of large cities on processed consumption, this result supports the earlier
descriptive results that indicate that highly processed food drives overall growth of processed
food consumption in urban areas.
Third, five additional effects of non-expenditure determinant variables for the urban
consumption of processed food are of note (Tables 2-A-4 & 2-A-6). First, households with
greater participation in nonfarm employment consume greater shares of processed food,
specifically greater shares of highly processed food and lesser shares of unprocessed maize. This
finding is consistent with literature that indicates that nonfarm employment increases the
opportunity cost of time, which would adjust household preferences to favor timesaving food
items (Senauer et al., 1986; Kennedy and Reardon, 1996). Second, dependency ratio, another

115

consumer demand factor of individual and household preferences, generally has a negative effect
on processed food consumption, specifically highly processed food that show negative effects of
dependency ratio in each country. Third, the amount of land farmed negatively affects processed
share in each country as the production of food enabled by the farming of land directly provides
the opportunity to consume food from own production, which by definition is unprocessed.
Fourth, ownership of a telephone has a positive effect on processed share, which is in agreement
with the hypothesis that ownership of a phone suggests individual and household preferences
that are consistent with modern lifestyle patterns. Fifth, the effect of the age of the household
head, although minimal, is negative and significant on processed share, affirming that elderly
households are less open to modern lifestyle patterns.
I will highlight four results pertaining to the marginal effects of non-expenditure
determinants on processed food consumption by rural households.
First, distance to cities has a negative effect on processed share in rural Zambia (Table 214). This is consistent with the von ThĂźnen model that says that distance to cities increase
transaction costs of acquiring foods from urban markets (Nelson, 2002), thereby increasing the
effective price and reducing household access to processed foods.
Second, the effect of distance to a city has a monotonically increasing negative effect on
processed share across city sizes. Table 2-6 shows greater consumption of processed food in
primary cities, therefore it is consistent that households near primary cities would also consume
greater shares of processed food.
Third, further examining processed foods (Table 2-A-7), I find that, distance to primary
and secondary cities are significant determinants for all processing sub-aggregates, other than
low processed food. Distance to tertiary cities is only statistically significant for high processed

116

foods, and this determinant has a negative marginal effect that is in agreement with the marginal
effects of distance to primary or secondary cities on high processed food. The negative marginal
effects of distance to all cities, including tertiary cities, reveal household preference for high
processed food and the ability of cities of any size to provide it. Unexpected results are the
negative marginal effects of distances to larger cities on unprocessed maize. The positive
marginal effect of distance to cities on unprocessed non-maize partially explains the unexpected
result as unprocessed non-maize is an aggregate that includes cassava, which households in rural
Zambia consider as a substitute for unprocessed maize.
Fourth, determinant variables other than proximity to city have significant effects on
processed food consumption providing four observations of note. First, increased opportunity
cost of time from household participation in nonfarm employment is significant to the processed
food consumption of rural households as indicated by positive marginal effects on processed
food in each country (Tables 2-A-5). Nonparametric analysis of the relationship between
nonfarm employment and total household expenditure shows a positive relationship between
these variables beginning at incomes as low as one dollar per day; this relationship contributes to
the rise in processed share beginning among the poor. Haggblade et al. (2010) noted that
households in rural Africa increasingly depend on nonfarm employment; therefore, this finding
leads to an expectation of future increased processed food consumption at low expenditures.
Second, the average partial effects of rural farmed hectares of land are negative for processed
share and all statistically significant sub-aggregates for processed food. The amount of farmed
hectares of land positively relates to household total expenditure; therefore, the amount of
farmed hectares of land partially offsets a portion of the positive effects of expenditure on
processed share. Third, high levels of education positively influence the consumption of

117

processed share in Malawi and Zambia, and when statistically significant it reduces the
unprocessed sub-aggregates and increases the processed sub-aggregates (Table 2-A-7). Fourth,
female head of household has a negative effect on highly processed food, which, assuming that
highly processed foods are less healthy aligns with female-headed households selecting healthier
eating patterns (Smith, 2003).

2.4.6. Mediation model analysis
City size affects multiple determinant variables within the Engel’s Curve Model,
therefore I use mediation model analysis to estimate the indirect effects of city size on processed
share that occur through household total expenditure, nonfarm labor, the amount of land farmed
by the household and the maximum education attained within the household. I select these four
variables as mediator variables for the following reasons: difference in city size could affect
household expenditure (FerrĂŠ et al., 2012), it could expose household members to various
nonfarm employment opportunities, it could vary household access to farmland, and larger cities
could offer greater opportunities for education. I do not find statistically significant indirect
effects of city size with the Malawi data, nor via education, but the remaining countries and
mediator variables led to the following results.
Table 2-15 shows the indirect effects that primary and secondary cities have on processed
share with three results of note. First, primary cities in Tanzania and Zambia have positive
indirect effects on processed share, suggesting a stronger positive effect of primary cities on
processed share than shown in Tables 2-13 & 2-A-4. Second, primary cities in Tanzania along
with primary and secondary cities in Zambia significantly increase household participation in
nonfarm employment, which positively effects processed share, resulting in positive and

118

significant indirect effects of city size. Third, city size indirectly increases the consumption of
processed food via the amount of household farmed land. Larger cities provide households with
less opportunity for farming larger areas of land39, and land can be used for the production of
processed food, therefore city size increases processed share by reducing the consumption of an
alternative food aggregate. The positive indirect effects of city size via hectares of farmed land
on processed share show that city size reduces farmed land, which would have reduced
processed share, therefore city size increases processed food consumption beyond what direct
effects would show.

Table 2-15: Indirect effects of urban household settlement within primary and secondary cities
on processed share, estimates by country
Mediator Variable
Household total
expenditure per AE
Nonfarm
employment
Hectares of farmed
land (log)
Maximum
education

Malawi
Primary Secondary
Cities
Cities
-0.005

Tanzania
Primary
Secondary
Cities
Cities
0.073
-0.004

Uganda
Primary
Secondary
Cities
Cities
0.025
0.017

Zambia
Primary
Secondary
Cities
Cities
0.029
0.002

(0.574)

(0.000)

(0.658)

(0.153)

(0.522)

(0.000)

0.000

0.007

0.005

0.002

0.000

0.002

(0.565)

0.001

(0.732)

(0.011)

(0.090)

(0.646)

(0.897)

(0.000)

(0.015)

0.001

0.009

0.008

0.007

0.011

0.015

0.008

(0.382)

(0.019)

(0.047)

(0.073)

(0.023)

(0.000)

(0.000)

-0.001

0.000

0.000

0.004

0.005

0.000

0.000

(0.286)

(0.831)

(0.835)

(0.182)

(0.290)

(0.948)

(0.933)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

I estimate the indirect effects of distances to cities on rural processed food consumption
patterns using the same mediator variables that I use while estimating urban indirect effects
(Table 2-16). The overall finding from the rural mediation analysis is that increasing distance to
cities decreases processed share through indirect negative effects on household income, nonfarm
employment, and hectares of farmed land. The one anomalous result is the positive indirect for
secondary cities on processed share via hectares of farmed land. Increasing rural household

The percentage of urban households that engage in farming also has an inverse relationship with city size
(primary city 9%, secondary city 29%, tertiary city 55%).
39

119

distance from secondary cities results in statistically significant reduction in farmed land, causing
this positive effect. The use of the mediator model highlights the indirect effects that testing for
direct effects in typical regression analysis do not capture.

Table 2-16: Indirect effects of distance from rural household settlement to primary, secondary or
tertiary cities on processed share – rural Zambia
Mediator Variable
Household total expenditure
per AE

Primary
Cities
-0.001

Distance to
Secondary
Cities
0.000

Tertiary
Cities
0.000

(0.075)

(0.203)

(0.337)

Nonfarm employment

-0.005

0.000

-0.001

(0.000)

(0.455)

(0.002)

Hectares of farmed land (log)

-0.002

0.002

0.000

(0.006)

(0.000)

(0.538)

Maximum education attained
within the household

0.000

0.000

0.000

(0.534)

(0.975)

(0.079)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

2.5 Conclusions
I considered how the consumption patterns of processed food varies across income, by
city size and by distances to cities of differing size. To my knowledge, this is the first paper to
analyze the spatial effects of city size and distance to city on processed food consumption in
Africa.
Processed foods have penetrated the diets of rural (36%) and urban (63%) households at
all levels across the income distribution of ESA, which is both a result of increased market
access because of public and private investments and an indication of demand for future
processing investments within the agrifood system. I find that income-induced diet change that
begins among the poor contribute to these high levels of processed food consumption. The
pattern of a rising share of processed food value with increasing total expenditure in the diets of
poor households consistently appears in the analysis of descriptive statistics, nonparametric

120

analysis, expenditure threshold analysis and Engel’s curve regression analysis. These findings
highlight that the rise in processed food consumption is not solely a middle-class story, as
conventionally assumed, but poor households are purchasing significant shares of processed
food.
Our analysis of sub-aggregates of processed food suggests that the consumption of highly
processed food is the driving force behind the consumption pattern of processed food in general,
where low processed food serves as a transition from unprocessed to highly processed food. This
contributes to discussions regarding nutrition and the sophistication of the processing sector.
Given the large portion of the population in ESA that currently has low income, coupled
with the pattern of increased processed share with income growth, these findings signal a strong
future demand for increased food market infrastructure.
The future growth of investments into the agrifood system should consider proximity to
urban areas as I found that the consumption patterns of processed food significantly vary by
households’ settlement in or near cities of various size. I consistently observe these patterns
across the four-step analysis that includes descriptive, nonparametric, fractional probit
regression, and mediation model analyses.
City size has significant positive effects on processed food consumption in urban
households, affirming the expectation that city size will alter the urban environment such that
households will have greater access to processed foods and the consumer demand characteristics
will exhibit an increased desire of processed food.
The distance to cities of any size has a diminishing negative effect on processed food
consumption by rural households. The distance to city effects with the greatest magnitudes were
for primary cities, which are likely to have significant commercial networks that extend market

121

reach further into rural areas that result in increased engagement with urban markets by rural
households.
I show that the spatial effects of city size and distance to city are significant drivers of
processed food consumption in ESA. As urbanization is expected to continue within this region
and elsewhere in developing nations, the recognition of how urbanization occurs should be
considered when effort is put forth to meet the changing food demands of households near and
within urban areas.

122

APPENDIX

123

APPENDIX

Table 2-A-1: Relationship between food classification scheme in this paper and that in Monteiro et al. (2010)
Processing
Aggregates
Monteiro
Aggregates

Unprocessed

Low Processed

Unprocessed or Minimally Processed
Whole grains
Beans & pulses
Fresh fruits
Fresh vegetables
Roots & tubers
Fresh fish
Eggs

Rice
Butchered meat
Dried fish

High Processed

Processed Culinary or Ingredients
Sugar
Spices
Wheat flour
Maize meal
Other flours

Vegetable oils
Vegetable fats
Pasta

Pasteurized milk*

Ultra-Processed
Breads
Biscuits
Soft drinks
Cheeses
Processed meats
Prepared food away
from home

Note: Pasteurized milk is part of Monteiro et al.’s Unprocessed or Minimally Processed aggregate

Table 2-A-2: Food budget shares of processed food aggregates, by country and expenditure
Average Processed Share by
Expenditure Terciles

Average Maize Unprocessed Share by Expenditure Strata
Rural

Total
20.7

20.5

$0.501.00
25.5

(17.5)

(0.0)

(25.3)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
23.7

$1.502.00
21.2

$2.003.00
17.6

$3.005.00
13.0

$5.0010.00
9.3

(23.7)

(20.2)

(16.4)

(11.1)

(6.5)

> $10.00

Low

Middle

High

12.7

24.7

21.2

14.4

(3.9)

(23.5)

(20.3)

(12.2)

15.6

14.4

16.8

18.4

19.7

15.7

11.7

5.9

2.0

18.2

16.6

10.1

(10.3)

(0.0)

(5.3)

(11.0)

(17.4)

(14.2)

(8.6)

(1.8)

(0.0)

(11.5)

(14.3)

(6.9)

5.6

1.9

7.1

5.4

6.5

5.7

5.3

3.0

2.2

6.2

6.0

3.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

21.2

25.7

24.2

22.4

20.8

18.4

15.7

8.5

8.6

24.2

20.3

14.7

(16.5)

(18.5)

(20.3)

(19.0)

(17.7)

(13.9)

(9.3)

(5.1)

(2.4)

(19.7)

(16.8)

(8.4)

14.0

17.7

18.7

16.6

16.2

12.6

9.6

5.2

3.9

16.4

14.5

9.3

(6.5)

(0.0)

(9.1)

(9.1)

(10.8)

(7.1)

(5.3)

(1.1)

(0.0)

(5.3)

(9.5)

(5.1)

124

Table 2-A-2: (continued)
Average Processed Share by
Expenditure Terciles

Average Maize Unprocessed Share by Expenditure Strata
Urban

Total
5.3

11.3

$0.501.00
8.8

(0.0)

(0.0)

(0.0)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
11.5

$1.502.00
7.1

$2.003.00
4.4

$3.005.00
3.4

$5.0010.00
1.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

> $10.00

Low

7.6

14.2

6.0

3.8

(0.0)

(0.0)

(0.0)

(0.0)

5.4

14.5

8.2

6.5

9.6

10.1

4.4

1.6

0.7

8.2

9.8

2.7

(24.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

1.6

1.3

2.6

7.1

3.1

2.2

1.6

0.9

0.2

4.5

1.8

1.0

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

3.4

16.2

10.5

7.4

6.0

3.8

2.5

1.4

0.8

10.3

5.2

1.8

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

4.0

11.3

7.9

7.4

7.3

6.8

3.2

1.4

1.1

7.9

6.7

2.2

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Total

Average Processed Share by
Expenditure Terciles

9.9

24.8

$0.501.00
16.2

(0.0)

(0.0)

(0.0)

5.8

13.6

10.1

7.1

4.7

4.1

4.0

4.7

3.2

8.1

4.3

4.1

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(2.5)

(0.0)

(0.0)

(0.0)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
10.3

$1.502.00
7.3

$2.003.00
5.4

$3.005.00
5.1

$5.0010.00
4.5

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

> $10.00

Low

3.8

15.6

7.6

5.0

(1.3)

(0.0)

(0.0)

(0.0)

Tanzania
Uganda
Zambia
Pooled

High

4.5

8.4

6.6

4.7

5.5

3.7

4.5

2.5

2.3

5.6

3.8

3.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

4.7

7.2

3.9

4.6

4.5

5.1

5.4

5.3

5.0

4.5

4.6

5.3

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(2.1)

(2.8)

(0.0)

(0.0)

(0.0)

5.9

14.2

9.5

6.7

5.3

4.2

4.3

3.8

3.0

8.0

4.7

4.3

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Total

Average Processed Share by
Expenditure Terciles

12.5

41.1

$0.501.00
25.7

6.3

18.8

16.8

10.2

(10.1)

(51.1)

(27.1)

(15.8)

(14.9)

(14.3)

(9. 9)

(7.3)

(4.3)

(14.7)

(16.5)

(8.7)

7.8

3.2

12.8

14.3

9.2

8.8

7.7

5.6

3.0

12.8

8.7

6.1

(5.1)

(0.0)

(9.0)

(9.2)

(6.3)

(5.9)

(6.2)

(4.9)

(2.4)

(9.2)

(5.7)

(4.6)

< $0.50
Malawi

Middle

(0.0)

Average Maize Processed Share by Expenditure Strata
Urban

High

(0.0)

Average Maize Processed Share by Expenditure Strata
Rural

Middle

$1.001.50
15.1

$1.502.00
15.8

$2.003.00
13.9

$3.005.00
10.8

$5.0010.00
9.0

> $10.00

Low

Middle

High

5.0

7.8

13.5

9.0

7.6

6.2

5.3

3.8

1.6

9.5

5.3

4.1

(2.0)

(0.0)

(9.3)

(0.0)

(0.0)

(2.7)

(3.0)

(2.0)

(0.8)

(0.0)

(2.0)

(2.2)

13.0

17.9

19.0

18.1

16.8

15.7

13.0

10.0

6.3

18.2

16.9

10.8

(10.7)

(6.3)

(15.0)

(17.8)

(16.3)

(15.3)

(12.3)

(9.0)

(5.3)

(15.5)

(16.5)

(9.4)

8.4

12.0

15.4

14.4

11.1

9.9

8.2

5.9

3.6

12.9

10.1

6.8

(5.1)

(0.0)

(10.4)

(10.9)

(7.3)

(6.6)

(6.3)

(4.6)

(2.5)

(8.1)

(5.8)

(4.8)

125

Table 2-A-2: (continued)
Average Processed Share by
Expenditure Terciles

Average Non-Maize Unprocessed Share by Expenditure Strata
Rural

Total
42.3

41.1

$0.501.00
39.8

(41.0)

(39.2)

(38.5)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
42.2

$1.502.00
43.5

$2.003.00
44.2

$3.005.00
44.0

$5.0010.00
38.7

(40.9)

(43.0)

(43.7)

(41.4)

(35.9)

> $10.00

Tanzania
Uganda
Zambia
Pooled

31.0

40.5

43.6

42.9

(38.9)

(43.7)

(41.2)

45.3

35.0

47.1

47.0

45.3

46.2

45.3

39.7

32.3

45.9

46.0

43.6

(29.7)

(43.4)

(43.1)

(41.9)

(44.8)

(43.0)

(38.8)

(32.2)

(42.6)

(44.4)

(41.7)

63.8

72.1

66.3

68.1

66.1

65.7

60.8

53.4

33.1

67.4

65.4

55.5

(65.6)

(82.6)

(65.0)

(70.1)

(71.3)

(67.6)

(63.8)

(51.0)

(25.7)

(70.1)

(67.7)

(55.5)

45.6

45.2

48.5

46.9

44.9

44.0

40.8

35.4

32.3

47.5

45.4

40.0

(43.1)

(42.9)

(45.8)

(45.4)

(43.4)

(42.0)

(37.7)

(31.7)

(28.1)

(45.5)

(43.2)

(37.7)

50.4

41.1

49.2

51.5

51.0

52.7

50.8

44.9

32.5

51.7

51.5

46.8

(48.9)

(38.0)

(46.7)

(49.5)

(49.4)

(51.2)

(50.8)

(45.3)

(28.7)

(50.1)

(50.1)

(45.7)

Average Processed Share by
Expenditure Terciles

Total
32.2

34.1

$0.501.00
34.9

(30.9)

(31.8)

(37.2)

$1.001.50
37.1

$1.502.00
36.6

$2.003.00
33.7

$3.005.00
30.8

$5.0010.00
29.2

(36.7)

(34.3)

(32.4)

(29.8)

(27.9)

> $10.00

Tanzania
Uganda
Zambia
Pooled

Middle

High

26.0

35.8

36.2

30.4

(35.5)

(34.6)

(28.9)

28.4

41.1

28.7

37.3

35.5

30.5

27.7

24.2

20.5

35.0

31.5

25.1

(26.1)

(31.1)

(26.6)

(37.4)

(31.8)

(29.1)

(25.1)

(22.3)

(19.7)

(34.1)

(30.0)

(23.5)

43.7

25.4

52.7

55.0

50.5

55.3

45.6

38.2

30.7

50.8

53.3

39.1

(42.2)

(21.4)

(46.4)

(52.3)

(52.9)

(55.9)

(41.9)

(38.3)

(32.2)

(49.8)

(54.4)

(38.3)

31.2

37.8

39.2

36.2

34.7

32.5

30.3

28.8

26.7

38.0

34.3

29.1

(29.4)

(35.7)

(35.9)

(34.5)

(33.2)

(30.3)

(29.2)

(27.7)

(25.5)

(35.3)

(32.8)

(28.0)

33.2

35.8

35.2

39.7

38.5

36.5

33.5

29.9

25.3

39.3

37.5

30.2

(30.6)

(31.1)

(34.1)

(37.4)

(35.0)

(32.1)

(30.6)

(28.1)

(24.2)

(37.2)

(32.7)

(28.3)

Total
18.9

12.3

$0.501.00
14.8

(16.3)

(9.2)

(12.4)

(15.0)

14.9

21.1

12.6

12.7

14.4

15.9

15.3

(11.4)

(12.5)

(5.8)

(7.9)

(10.6)

(13.2)

(13.3)

16.6

9.9

14.0

14.5

14.3

16.6

19.2

(13.8)

(2.3)

(9.3)

(8.8)

(10.8)

(13.4)

(17.7)

< $0.50
Malawi

Low

(25.6)

Average Processed Share by
Expenditure Terciles

Average Non-Maize Low Processed Share by Expenditure Strata
Rural

High

(43.0)

< $0.50
Malawi

Middle

(26.5)

Average Non-Maize Unprocessed Share by Expenditure Strata
Urban

Low

$1.001.50
17.6

$1.502.00
19.5

$2.003.00
22.1

$3.005.00
23.3

$5.0010.00
27.0

(17.2)

(20.5)

(22.0)

(26.1)

> $10.00

Low

Middle

High

25.1

15.1

19.5

23.4

(24.7)

(12.5)

(17.2)

(21.9)

19.3

25.5

13.4

15.5

16.6

(17.8)

(24.7)

(8.2)

(12.0)

(14.4)

22.0

20.0

13.9

16.7

20.9

(22.0)

(15.6)

(9.2)

(13.8)

(19.7)

18.9

15.3

16.2

17.5

19.9

21.1

24.2

30.0

30.5

16.3

19.7

24.7

(15.2)

(10.4)

(12.3)

(13.6)

(16.4)

(18.0)

(22.7)

(29.1)

(27.3)

(12.3)

(16.1)

(22.7)

16.5

17.0

14.1

14.8

15.7

17.2

18.1

21.7

23.2

14.2

17.0

19.7

(13.3)

(10.0)

(9.9)

(10.5)

(12.6)

(14.3)

(16.0)

(19.4)

(23.1)

(9.8)

(13.8)

(17.8)

126

Table 2-A-2: (continued)
Average Processed Share by
Expenditure Terciles

Average Non-Maize Low Processed Share by Expenditure Strata
Urban

Total
27.7

11.6

$0.501.00
22.3

(27.3)

(17.0)

(19.2)

< $0.50
Malawi
Tanzania
Uganda
Zambia
Pooled

$1.001.50
22.8

$1.502.00
25.0

$2.003.00
28.4

$3.005.00
31.0

$5.0010.00
30.3

(23.1)

(25.0)

(28.9)

(30.2)

(31.0)

> $10.00

Tanzania
Uganda
Zambia
Pooled

25.1

20.8

25.5

29.5

(19. 0)

(25.4)

(29.6)

22.4

25.1

25.1

20.6

20.5

21.2

24.4

23.2

19.7

21.8

21.1

23.2

(22.9)

(24.4)

(18.8)

(21.1)

(21.7)

(25.0)

(23.3)

(18.1)

(21.4)

(21.7)

(23.0)

25.6

29.9

20.2

21.3

23.2

19.6

27.8

28.7

23.6

21.7

23.4

27.1

(25.1)

(30.9)

(13.7)

(14.7)

(18.8)

(16.5)

(25.8)

(27.6)

(25.1)

(14.7)

(20.2)

(26.2)

30.4

19.0

18.0

23.9

25.9

28.6

31.3

34.3

36.8

20.2

26.4

33.3

(29.9)

(16.2)

(16.2)

(22.3)

(24.5)

(27.7)

(30.5)

(33.8)

(36.0)

(18.3)

(25.5)

(32.5)

25.2

23.3

22.5

21.7

22.6

22.9

27.1

27.3

25.1

21.5

23.0

26.8

(24.9)

(22.4)

(21.0)

(20.5)

(22.6)

(22.5)

(26.6)

(26.7)

(24.4)

(19.6)

(22.7)

(26.3)

Average Processed Share by
Expenditure Terciles

Total
8.2

1.4

$0.501.00
3.7

$1.001.50
6.1

$1.502.00
8.5

$2.003.00
10.8

$3.005.00
14.6

$5.0010.00
20.5

(5.1)

(0.0)

(1.1)

(3.8)

(6.0)

(8.8)

(12.1)

(19.2)

18.3

16.0

13.5

14.8

15.9

18.1

23.7

30.4

37.0

14.5

17.6

25.6

(14.0)

(11.7)

(9.8)

(11.0)

(12.9)

(13.9)

(18.5)

(27.8)

(36.3)

(10.7)

(13.7)

(21.7)

15.9

> $10.00

Tanzania
Uganda
Zambia
Pooled

Middle

High

27.4

4.1

8.0

14.2

(1.4)

(5.6)

(11.4)

9.5

7.6

6.1

7.3

7.5

8.2

10.2

19.0

42.4

7.0

8.1

(4.7)

(0.0)

(3.7)

(2.8)

(3.9)

(4.6)

(5.9)

(15.6)

(27.0)

(3.3)

(4.6)

(9.0)

9.6

6.6

7.2

8.7

9.9

11.4

13.9

20.8

23.6

7.4

10.0

15.2
(12.0)

(6.9)

(3.1)

(4.8)

(6.3)

(7.7)

(8.7)

(11.4)

(19.2)

(19.8)

(4.9)

(7.7)

13.2

9.9

8.5

10.4

11.7

13.3

17.1

24.4

37.4

9.8

12.4

19.9

(8.6)

(5.6)

(4.7)

(6.9)

(8.3)

(9.6)

(12.0)

(21.2)

(32.1)

(5.8)

(8.8)

(15.2)

Total
22.2

1.8

$0.501.00
8.4

(21.1)

(0.0)

(4.6)

< $0.50
Malawi

Low

(26.0)

Average Processed Share by
Expenditure Terciles

Average Non-Maize High Processed Share by Expenditure Strata
Urban

High

(22.4)

< $0.50
Malawi

Middle

(25.3)

Average Non-Maize High Processed Share by Expenditure Strata
Rural

Low

$1.001.50
13.5

$1.502.00
15.5

$2.003.00
19.6

$3.005.00
23.9

$5.0010.00
29.6

(12.2)

(14.4)

(19.3)

(22.5)

(27.9)

> $10.00

Low

Middle

High

35.0

10.4

15.6

26.1

(35.8)

(9.3)

(14.4)

(24.6)

36.0

16.2

25.2

21.3

25.2

29.3

35.8

45.3

56.1

22.2

29.0

43.0

(33.5)

(14.7)

(21.5)

(16.5)

(20.2)

(25.8)

(33.1)

(45.9)

(56.3)

(19.1)

(25.6)

(42.3)

24.1

35.6

11.0

7.6

15.6

16.7

19.7

28.4

43.9

13.6

16.3

28.7

(17.7)

(10.7)

(6.4)

(3.0)

(6.4)

(10.1)

(16.4)

(22.6)

(33.5)

(5.8)

(10.1)

(22.8)

21.9

9.1

13.3

14.3

16.7

19.4

22.9

25.6

29.4

13.3

17.1

24.9

(20.3)

(5.2)

(11.4)

(12.0)

(14.1)

(18.3)

(21.6)

(24.1)

(27.6)

(11.4)

(15.0)

(23.4)

29.1

17.6

18.9

16.8

20.5

23.9

28.0

35.5

44.9

18.3

22.6

34.0

(24.5)

(10.7)

(13.8)

(13.0)

(16.5)

(19.5)

(24.3)

(30.9)

(40.8)

(13.4)

(17.7)

(29.3)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

127

Table 2-A-3: Food budget shares of processed food aggregates, country data by settlement
Malawi Data
National
Rural
Urban
Primary City
Secondary City
Tertiary City

Tanzania Data
National
Rural
Urban
Primary City
Secondary City
Tertiary City

Uganda Data
National
Rural
Urban
Primary City
Secondary City
Tertiary City

Zambia Data
National
Rural
Urban
Primary City
Secondary City
Tertiary City

Maize
Unprocessed
Processed
18.3
10.3

Unprocessed
40.7

Non-Maize
Processed Low
20.3

Processed High
10.4
(6.6)

(14.0)

(0.0)

(38.8)

(18.3)

20.7

9.9

42.3

18.9

8.2

(17.5)

(0.0)

(41.0)

(16.3)

(5.1)

5.3

12.5

32.2

27.7

22.2

(0.0)

(10.1)

(30.9)

(27.3)

(21.1)

0.0

0.0

0.0

0.0

0.0

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

4.2

13.2

31.6

28.1

22.8

(0.0)

(11.1)

(30.4)

(27.7)

(21.7)

9.0

10.0

34.4

26.4

20.3

(1.6)

(3.9)

(32.9)

(25.4)

(17.8)

Unprocessed
41.0

Non-Maize
Processed Low
16.9

Processed High
22.9
(17.8)

Maize
Unprocessed
Processed
13.0
6.3
(4.9)

(0.0)

(37.4)

(14.3)

15.6

5.8

45.3

14.9

18.3

(10.3)

(0.0)

(43.0)

(11.4)

(14.0)

5.4

7.8

28.4

22.4

36.0

(0.0)

(5.1)

(26.1)

(22.4)

(33.5)

0.2

6.6

23.1

22.1

48.1

(0.0)

(5.2)

(22.2)

(21.4)

(48.8)

3.0

10.0

28.2

24.5

34.3

(0.0)

(7.0)

(26.1)

(24.5)

(31.4)

9.8

7.3

31.5

21.4

30.0

(1.5)

(3.6)

(30.0)

(22.1)

(26.3)

Maize
Unprocessed
Processed
4.7
4.6

Unprocessed
59.4

Non-Maize
Processed Low
18.6

Processed High
12.7
(6.4)

(0.0)

(0.0)

(61.0)

(16.0)

5.6

4.5

63.8

16.6

9.5

(0.0)

(0.0)

(65.6)

(13.8)

(4.7)

1.6

5.0

43.7

25.6

24.1

(0.0)

(2.0)

(42.2)

(25.1)

(17.7)

0.5

5.7

38.2

25.8

29.7

(0.0)

(2.6)

(39.9)

(26.1)

(23.0)

0.1

4.6

46.6

24.2

24.5

(0.0)

(2.6)

(47.3)

(23.0)

(17.5)

2.2

4.8

45.5

25.7

21.9

(0.0)

(1.8)

(42.4)

(24.8)

(15.5)

Unprocessed
40.6

Non-Maize
Processed Low
22.9

Processed High
13.9
(10.5)

Maize
Unprocessed
Processed
15.0
7.6
(7.1)

(0.0)

(36.7)

(20.9)

21.2

4.7

45.6

18.9

9.6

(16.5)

(0.0)

(43.1)

(15.2)

(6.9)

3.4

13.0

31.2

30.4

21.9

(0.0)

(10.7)

(29.4)

(29.9)

(20.3)

0.6

14.0

30.2

30.9

24.3

(0.0)

(12.0)

(29.0)

(30.4)

(23.2)

3.0

14.3

30.5

30.4

21.9

(0.0)

(11.7)

(29.0)

(29.6)

(20.1)

7.2

9.9

33.5

30.0

19.4

(0.0)

(7.5)

(31.1)

(29.4)

(17.3)

Source: authors’ calculations using national household level surveys
Notes: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

128

Table 2-A-4: Average partial effects of the household determinants on processed share, urban
household estimates by country
Urban
Household total expenditure
per AE
Household located in a
primary city
Household located in a
secondary city

Malawi
0.007

Zambia
0.010

(0.000)

(0.001)

(0.000)

0.068

0.000

0.037

(0.000)

(1.000)

(0.000)

0.053

0.047

-0.053

0.043

(0.000)

(0.003)

(0.049)

(0.000)

Nonfarm employment

0.008

0.094

0.015

0.037

(0.726)

(0.000)

(0.619)

(0.000)

Dependency ratio

-0.043

-0.075

-0.146

-0.001

(0.091)

(0.005)

(0.001)

(0.918)

Household adult equivalents

0.001

0.007

-0.006

0.006

(0.684)

(0.000)

(0.098)

(0.000)

Farmed hectares of land (log)

-0.098

-0.032

-0.056

-0.028

(0.005)

(0.010)

(0.009)

(0.000)

Own a gas or electric stove

0.035

-0.009

0.001

(0.007)

(0.702)

(0.902)

Own a refrigerator

0.021

-0.002

0.005

(0.138)

(0.851)

Maximum education attained
within the household

0.007

0.000

0.001

(0.087)

(0.837)

(0.008)

(0.787)

Age of household head

-0.001

-0.001

-0.002

-0.001

(0.139)

(0.010)

(0.014)

(0.001)

Female head of household

-0.025

-0.008

-0.018

-0.008

(0.066)

(0.638)

(0.381)

(0.096)

Own a telephone

-0.009

0.073

0.067

0.029

(0.514)

(0.005)

(0.026)

(0.000)

Own a car

0.014

-0.033

-0.031

0.000

(0.518)

(0.159)

(0.329)

(0.954)

Own a motorcycle

-0.006

-0.008

-0.038

0.020

(0.856)

(0.687)

(0.207)

(0.292)

-0.065

-0.025

-0.100

-0.023

(0.000)

(0.102)

(0.000)

(0.000)

Own a bicycle

(0.004)

Processed Share
Tanzania
Uganda
0.017
0.009

(0.404)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

129

0.000

Table 2-A-5: Average partial effects of the household determinants on processed share, rural
household estimates by country
Rural
Household total expenditure
per AE

Malawi
0.004
(0.287)

Processed Share
Tanzania
Uganda
0.009
0.022
(0.062)

(0.000)

Zambia
0.010
(0.009)

Distance to primary city (log)

-0.045

Distance to secondary city (log)

-0.002

Distance to tertiary city (log)

-0.005

(0.000)
(0.554)
(0.001)

0.080

0.117

0.107

0.183

(0.000)

(0.000)

(0.000)

(0.000)

Dependency ratio

-0.049

-0.075

-0.047

-0.032

(0.004)

(0.006)

(0.071)

(0.045)

Household adult equivalents

-0.009

-0.002

-0.003

-0.003

(0.001)

(0.073)

(0.249)

(0.073)

Farmed hectares of land (log)

-0.029

-0.044

-0.034

-0.023

(0.040)

(0.000)

(0.000)

(0.000)

Own a gas or electric stove

-0.029

0.000

0.039

(0.517)

(0.992)

(0.065)

Nonfarm employment

Own a refrigerator
Maximum education attained
within the household

0.128

0.007

0.037

(0.003)

(0.794)

(0.117)

0.014

0.001

0.000

0.007

(0.001)

(0.164)

(0.914)

(0.000)

Age of household head

-0.001

0.000

0.000

0.000

(0.008)

(0.794)

(0.539)

(0.039)

Female head of household

-0.008

0.018

-0.010

-0.004

(0.421)

(0.267)

(0.491)

(0.633)

Own a telephone

0.028

0.063

0.053

0.049

(0.002)

(0.000)

(0.000)

(0.000)

Distance to market (log)

-0.013

-0.004

-0.008

-0.008

(0.000)

(0.418)

(0.095)

(0.001)

Own a car

0.121

0.077

-0.057

0.118

(0.008)

(0.201)

(0.250)

(0.000)

Own a motorcycle
Own a bicycle

0.022

0.027

0.023

0.041

(0.514)

(0.421)

(0.237)

(0.447)

-0.021

-0.023

-0.034

0.009

(0.008)

(0.086)

(0.003)

(0.193)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

130

Table 2-A-6: Average partial effects of the household determinants for processed food
aggregates, urban household estimates by country
Urban Malawi
Household total expenditure
per AE
Household located in a
secondary city
Nonfarm employment
Dependency ratio
Household adult equivalents
Farmed hectares of land (log)

Maize
Unprocessed
Processed
-0.001
-0.004

Unprocessed
-0.007

Non-Maize
Processed Low
0.005

Processed High
0.012

(0.680)

(0.002)

(0.000)

(0.001)

(0.000)

-0.034

0.037

-0.018

0.008

0.008

(0.000)

(0.000)

(0.031)

(0.287)

(0.249)

0.017

-0.037

-0.020

0.022

0.022

(0.362)

(0.026)

(0.210)

(0.095)

(0.081)

0.013

0.019

0.034

-0.012

-0.046

(0.461)

(0.225)

(0.068)

(0.464)

(0.005)

0.002

-0.004

-0.004

0.005

0.000

(0.212)

(0.060)

(0.034)

(0.013)

(0.983)

0.044

-0.180

0.045

-0.008

-0.016

(0.000)

(0.001)

(0.089)

(0.730)

(0.330)

Own a gas or electric stove

-0.011

-0.025

-0.023

0.014

0.032

(0.228)

(0.006)

(0.028)

(0.175)

(0.004)

Own a refrigerator

-0.020

0.002

0.001

0.008

0.001

(0.023)

(0.840)

(0.918)

(0.459)

(0.906)

Maximum education attained
within the household

-0.006

-0.005

-0.001

0.005

0.004

(0.045)

(0.083)

(0.760)

(0.056)

(0.091)

Age of household head

0.000

-0.001

0.000

0.000

0.000

(0.088)

(0.008)

(0.450)

(0.673)

(0.888)

0.005

-0.020

0.022

-0.004

-0.002

(0.691)

(0.066)

(0.086)

(0.695)

(0.832)

0.006

-0.016

0.002

-0.007

0.020

(0.535)

(0.173)

(0.839)

(0.485)

(0.033)

Own a car

-0.021

-0.019

0.020

-0.006

0.026

(0.141)

(0.160)

(0.230)

(0.706)

(0.058)

Own a motorcycle

-0.032

0.005

0.043

-0.039

0.025

(0.120)

(0.834)

(0.135)

(0.059)

(0.279)

0.047

-0.039

0.019

-0.005

-0.019

(0.000)

(0.000)

(0.047)

(0.542)

(0.012)

Female head of household
Own a telephone

Own a bicycle

131

Table 2-A-6: (continued)
Urban Tanzania
Household total expenditure
per AE
Household located in a
primary city
Household located in a
secondary city

Maize
Unprocessed
Processed
-0.017
-0.010

Unprocessed
-0.012

Non-Maize
Processed Low
-0.002

Processed High
0.023

(0.000)

(0.000)

(0.000)

(0.289)

(0.000)

-0.057

0.025

-0.022

-0.009

0.051

(0.000)

(0.003)

(0.080)

(0.440)

(0.006)

-0.028

0.037

-0.011

0.007

0.006

(0.001)

(0.001)

(0.423)

(0.549)

(0.723)

Nonfarm employment

-0.053

-0.010

-0.028

0.004

0.109

(0.000)

(0.420)

(0.108)

(0.776)

(0.000)

Dependency ratio

-0.003

0.022

0.070

0.077

-0.176

(0.869)

(0.149)

(0.002)

(0.000)

(0.000)

Household adult equivalents

-0.004

-0.004

-0.004

0.003

0.008

(0.004)

(0.000)

(0.002)

(0.001)

(0.000)

Farmed hectares of land (log)

0.023

-0.028

0.002

-0.019

0.000

(0.000)

(0.000)

(0.808)

(0.035)

(0.988)

0.015

-0.005

-0.007

-0.004

-0.001

Own a gas or electric stove

(0.513)

(0.688)

(0.634)

(0.770)

(0.964)

Own a refrigerator

-0.004

-0.015

0.007

0.031

-0.018

(0.773)

(0.090)

(0.546)

(0.013)

(0.259)

Maximum education attained
within the household

0.000

0.000

0.001

0.001

-0.001

(0.378)

(0.930)

(0.376)

(0.124)

(0.216)

0.000

0.000

0.001

-0.001

-0.001

Age of household head

(0.228)

(0.077)

(0.052)

(0.053)

(0.033)

Female head of household

-0.015

-0.006

0.024

0.014

-0.020

(0.112)

(0.479)

(0.074)

(0.198)

(0.187)

Own a telephone

-0.012

0.018

-0.036

0.060

0.008

(0.300)

(0.075)

(0.097)

(0.000)

(0.767)

Own a car

-0.027

0.022

0.038

0.008

-0.057

(0.012)

(0.060)

(0.056)

(0.635)

(0.028)

Own a motorcycle

-0.018

-0.006

0.028

0.039

-0.049

(0.079)

(0.578)

(0.098)

(0.026)

(0.056)

0.007

-0.020

0.017

0.008

-0.012

(0.405)

(0.014)

(0.192)

(0.471)

(0.446)

Own a bicycle

132

Table 2-A-6: (continued)
Urban Uganda
Household total expenditure
per AE
Household located in a
primary city
Household located in a
secondary city

Maize
Unprocessed
Processed
-0.004
-0.006

Unprocessed
-0.008

Non-Maize
Processed Low
0.001

Processed High
0.009

(0.075)

(0.000)

(0.003)

(0.726)

(0.000)

-0.008

0.020

0.007

-0.017

0.002

(0.056)

(0.024)

(0.714)

(0.272)

(0.914)

-0.015

0.012

0.068

-0.023

-0.037

(0.000)

(0.292)

(0.012)

(0.204)

(0.128)

Nonfarm employment

-0.003

0.002

-0.009

-0.012

0.033

(0.669)

(0.849)

(0.762)

(0.619)

(0.287)

Dependency ratio

0.017

0.005

0.126

0.096

-0.221

(0.112)

(0.797)

(0.004)

(0.009)

(0.000)

0.000

0.001

0.007

0.000

-0.007

(0.758)

(0.270)

(0.080)

(0.897)

(0.036)

0.010

-0.002

0.038

-0.011

-0.053

(0.014)

(0.762)

(0.042)

(0.514)

(0.040)

0.000

0.001

-0.001

0.002

-0.001

(0.397)

(0.005)

(0.013)

(0.000)

(0.264)

0.000

0.000

0.001

-0.001

-0.001

(0.062)

(0.630)

(0.053)

(0.103)

(0.378)

0.004

0.009

0.012

0.033

-0.060

Household adult equivalents
Farmed hectares of land (log)
Maximum education attained
within the household
Age of household head
Female head of household

(0.481)

(0.266)

(0.545)

(0.045)

(0.002)

Own a telephone

-0.001

-0.015

-0.067

0.033

0.060

(0.936)

(0.167)

(0.023)

(0.147)

(0.011)

Own a car

-0.002

-0.009

0.031

0.011

-0.013

(0.883)

(0.260)

(0.341)

(0.681)

(0.705)

Own a motorcycle

0.003

-0.006

0.037

0.002

-0.033

(0.720)

(0.539)

(0.224)

(0.942)

(0.101)

0.008

-0.021

0.093

-0.021

-0.058

(0.242)

(0.001)

(0.000)

(0.241)

(0.006)

Own a bicycle

133

Table 2-A-6: (continued)
Urban Zambia
Household total expenditure
per AE
Household located in a
primary city
Household located in a
secondary city

Maize
Unprocessed
Processed
-0.007
-0.008

Unprocessed
-0.006

Non-Maize
Processed Low
0.009

Processed High
0.008

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

-0.033

0.050

-0.002

-0.025

0.015

(0.000)

(0.000)

(0.713)

(0.000)

(0.001)

-0.019

0.045

-0.017

-0.011

0.013

(0.000)

(0.000)

(0.000)

(0.018)

(0.001)

Nonfarm employment

-0.017

-0.007

-0.019

0.016

0.031

(0.000)

(0.311)

(0.010)

(0.053)

(0.000)

Dependency ratio

0.001

0.009

-0.002

0.015

-0.029

(0.824)

(0.202)

(0.820)

(0.127)

(0.001)

Household adult equivalents

-0.002

-0.001

-0.004

0.004

0.004

(0.002)

(0.193)

(0.000)

(0.000)

(0.000)

Farmed hectares of land (log)

0.011

-0.024

0.011

-0.004

-0.005

(0.000)

(0.000)

(0.000)

(0.092)

(0.013)

Own a gas or electric stove

-0.003

-0.004

0.000

0.004

-0.001

(0.235)

(0.327)

(0.965)

(0.436)

(0.905)

Own a refrigerator

-0.005

-0.002

-0.003

0.007

-0.002

(0.061)

(0.637)

(0.631)

(0.210)

(0.730)

Maximum education attained
within the household

-0.001

-0.001

0.000

0.001

0.000

(0.051)

(0.114)

(0.772)

(0.042)

(0.629)

Age of household head

0.000

0.001

0.000

0.000

-0.001

(0.002)

(0.001)

(0.025)

(0.015)

(0.000)

0.001

0.002

0.008

0.003

-0.013

Female head of household

(0.583)

(0.642)

(0.092)

(0.556)

(0.002)

Own a telephone

-0.002

-0.005

-0.026

0.021

0.019

(0.501)

(0.321)

(0.000)

(0.001)

(0.000)

Own a car

0.001

-0.011

0.000

-0.005

0.009

(0.914)

(0.034)

(0.992)

(0.492)

(0.151)

0.008

0.019

-0.024

-0.031

0.032

(0.377)

(0.328)

(0.153)

(0.104)

(0.210)

0.011

-0.013

0.011

-0.004

-0.006

(0.000)

(0.002)

(0.016)

(0.409)

(0.159)

Own a motorcycle
Own a bicycle

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

134

Table 2-A-7: Average partial effects of the household determinants for processed food
aggregates, rural household estimates by country
Rural Malawi
Household total expenditure
per AE

Maize
Unprocessed
Processed
-0.025
-0.023

Unprocessed
0.019

Non-Maize
Processed Low
0.026

Processed High
0.054

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

Nonfarm employment

-0.068

0.022

-0.032

0.022

0.031

(0.000)

(0.121)

(0.051)

(0.069)

(0.000)

Dependency ratio

0.045

-0.032

0.006

-0.007

-0.013

(0.002)

(0.009)

(0.660)

(0.487)

(0.025)

0.003

-0.010

0.006

-0.002

0.003

(0.262)

(0.000)

(0.002)

(0.218)

(0.002)

0.027

-0.045

0.002

-0.008

0.003

(0.015)

(0.004)

(0.892)

(0.360)

(0.569)

0.032

-0.020

-0.002

-0.043

0.013

(0.443)

(0.487)

(0.970)

(0.173)

(0.274)

Own a refrigerator

-0.089

0.101

-0.076

0.046

0.010

(0.007)

(0.035)

(0.035)

(0.090)

(0.331)

Maximum education attained
within the household

-0.002

-0.001

-0.013

0.008

0.005

(0.630)

(0.671)

(0.000)

(0.001)

(0.000)

Age of household head

0.000

0.000

0.000

0.000

0.000

(0.064)

(0.169)

(0.217)

(0.591)

(0.000)

0.006

0.006

0.002

-0.009

-0.009

Household adult equivalents
Farmed hectares of land (log)
Own a gas or electric stove

Female head of household

(0.531)

(0.386)

(0.807)

(0.119)

(0.005)

Own a telephone

-0.007

-0.001

-0.020

0.010

0.014

(0.381)

(0.891)

(0.006)

(0.051)

(0.000)

Distance to market (log)

0.007

-0.009

0.006

-0.001

-0.002

(0.029)

(0.000)

(0.007)

(0.672)

(0.100)

Own a car

-0.022

0.014

-0.110

0.027

0.050

(0.556)

(0.706)

(0.020)

(0.386)

(0.002)

Own a motorcycle

0.007

-0.048

-0.023

0.040

-0.009

(0.709)

(0.039)

(0.501)

(0.177)

(0.430)

0.016

-0.026

0.006

0.004

0.002

(0.022)

(0.000)

(0.383)

(0.461)

(0.463)

Own a bicycle

135

Table 2-A-7: (continued)
Rural Tanzania
Household total expenditure
per AE

Maize
Unprocessed
Processed
-0.023
-0.015

Unprocessed
0.009

Non-Maize
Processed Low
-0.001

Processed High
0.025

(0.000)

(0.000)

(0.068)

0.823)

0.000)

Nonfarm employment

-0.044

0.004

-0.080

0.005

0.104

(0.003)

(0.690)

(0.000)

(0.635)

(0.000)

Dependency ratio

-0.008

0.017

0.083

-0.004

-0.084

(0.719)

(0.222)

(0.002)

(0.795)

(0.000)

Household adult equivalents

-0.002

-0.002

0.004

-0.002

0.002

(0.118)

(0.002)

(0.000)

(0.002)

(0.028)

Farmed hectares of land (log)

0.044

-0.014

-0.003

-0.029

-0.006

(0.000)

(0.006)

(0.712)

(0.000)

(0.269)

Own a gas or electric stove

-0.007

-0.017

-0.005

0.034

-0.021

(0.829)

(0.307)

(0.920)

(0.132)

(0.307)

Own a refrigerator

-0.057

-0.019

0.014

0.045

-0.029

(0.020)

(0.065)

(0.571)

(0.007)

(0.034)

Maximum education attained
within the household

0.000

0.000

-0.001

0.001

0.000

(0.696)

(0.637)

(0.069)

(0.004)

(0.765)

Age of household head

-0.001

0.000

0.000

0.000

0.000

(0.050)

(0.449)

(0.349)

(0.719)

(0.436)

Female head of household

-0.009

0.011

-0.009

0.015

-0.011

(0.421)

(0.243)

(0.548)

(0.119)

(0.263)

Own a telephone

-0.019

0.019

-0.046

0.028

0.018

(0.056)

(0.013)

(0.001)

(0.000)

(0.038)

Distance to market (log)

0.009

-0.005

-0.005

0.002

-0.001

(0.022)

(0.090)

(0.342)

(0.534)

(0.677)

Own a car

-0.047

0.064

-0.055

0.034

-0.002

(0.360)

(0.034)

(0.090)

(0.149)

(0.934)

Own a motorcycle

-0.022

-0.013

-0.004

0.033

0.003

(0.331)

(0.194)

(0.879)

(0.081)

(0.870)

0.015

-0.003

0.010

-0.004

-0.013

(0.135)

(0.644)

(0.405)

(0.603)

(0.107)

Own a bicycle
Rural Uganda
Household total expenditure
per AE

Maize
Unprocessed
Processed
-0.005
-0.004

Unprocessed
-0.019

Non-Maize
Processed Low
0.010

Processed High
0.013

(0.062)

(0.035)

(0.000)

(0.000)

(0.000)

Nonfarm employment

-0.009

0.033

-0.100

0.025

0.044

(0.333)

(0.000)

(0.000)

(0.043)

(0.000)

Dependency ratio

0.038

0.021

0.013

0.024

-0.084

(0.009)

(0.157)

(0.635)

(0.204)

(0.000)

0.000

-0.002

0.003

0.003

-0.003

(0.992)

(0.107)

(0.119)

(0.232)

(0.029)

0.004

-0.005

0.031

-0.022

-0.006

(0.393)

(0.445)

(0.000)

(0.001)

(0.171)

0.000

-0.001

0.000

0.001

0.000

(0.379)

(0.016)

(0.675)

(0.024)

(0.546)

0.000

0.000

0.000

0.000

0.000

(0.453)

(0.006)

(0.899)

(0.828)

(0.231)

Female head of household

-0.014

0.003

0.024

0.007

-0.020

(0.019)

(0.711)

(0.092)

(0.494)

(0.013)

Own a telephone

-0.010

0.019

-0.042

0.023

0.013

(0.124)

(0.002)

(0.001)

(0.014)

(0.054)

Distance to market (log)

-0.002

0.000

0.010

-0.002

-0.006

(0.351)

(0.975)

(0.036)

(0.594)

(0.055)

Own a car

0.002

-0.016

0.057

-0.015

-0.020

Household adult equivalents
Farmed hectares of land (log)
Maximum education attained
within the household
Age of household head

Own a motorcycle
Own a bicycle

(0.927)

(0.240)

(0.274)

(0.551)

(0.421)

-0.012

0.002

-0.010

0.014

0.010

(0.169)

(0.854)

(0.622)

(0.283)

(0.368)

0.009

-0.020

0.026

-0.001

-0.011

(0.157)

(0.000)

(0.030)

(0.923)

(0.111)

136

Table 2-A-7: (continued)
Rural Zambia
Household total expenditure
per AE

Maize
Unprocessed
Processed
-0.022
-0.007

Unprocessed
0.010

Non-Maize
Processed Low
0.010

Processed High
0.007

(0.000)

(0.000)

(0.012)

(0.000)

(0.000)

Distance to primary city (log)

-0.022

-0.019

0.070

-0.005

-0.016

(0.000)

(0.000)

(0.000)

(0.156)

(0.000)

Distance to secondary city (log)

-0.012

0.006

0.013

0.004

-0.009

(0.002)

(0.005)

(0.001)

(0.190)

(0.000)

Distance to tertiary city (log)

0.002

-0.001

0.004

-0.002

-0.002

(0.334)

(0.211)

(0.037)

(0.049)

(0.008)

Nonfarm employment

-0.114

0.037

-0.111

0.064

0.057

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

Dependency ratio

0.036

-0.017

-0.005

-0.003

-0.012

(0.033)

(0.022)

(0.788)

(0.823)

(0.066)

Household adult equivalents

-0.001

-0.002

0.004

0.000

0.000

(0.567)

(0.006)

(0.021)

(0.910)

(0.653)

Farmed hectares of land (log)

0.024

-0.013

0.002

-0.006

-0.001

(0.000)

(0.000)

(0.590)

(0.011)

(0.517)

Own a gas or electric stove

-0.110

0.003

0.029

0.037

-0.016

(0.000)

(0.778)

(0.215)

(0.023)

(0.079)

Own a refrigerator

0.020

0.003

-0.061

-0.014

0.041

(0.509)

(0.846)

(0.011)

(0.399)

(0.009)

Maximum education attained
within the household

-0.003

0.001

-0.005

0.004

0.003

(0.012)

(0.254)

(0.000)

(0.000)

(0.000)

Age of household head

0.000

0.000

0.001

0.000

0.000

(0.150)

(0.009)

(0.001)

(0.054)

(0.000)

0.018

0.006

-0.013

0.001

-0.010

(0.052)

(0.266)

(0.146)

(0.900)

(0.005)

0.001

0.011

-0.052

0.026

0.012

(0.953)

(0.037)

(0.000)

(0.000)

(0.001)

0.001

-0.004

0.007

0.001

-0.005

Female head of household
Own a telephone
Distance to market (log)

(0.679)

(0.004)

(0.012)

(0.725)

(0.000)

Own a car

-0.096

0.009

-0.048

0.064

0.030

(0.000)

(0.547)

(0.096)

(0.004)

(0.058)

Own a motorcycle

-0.018

0.024

-0.027

0.019

-0.003

Own a bicycle

(0.682)

(0.258)

(0.487)

(0.635)

(0.767)

-0.024

-0.006

0.016

0.007

0.010

(0.002)

(0.100)

(0.041)

(0.210)

(0.002)

Source: Authors’ calculations
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value.

137

REFERENCES

138

REFERENCES

Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel curves and consumer demand.
Review of Economics and statistics, 79(4), 527-539.
Battersby, J., & Peyton, S. (2014, June). The geography of supermarkets in Cape Town:
Supermarket expansion and food access. In Urban Forum (Vol. 25, No. 2, pp. 153-164).
Springer Netherlands.
Bennett, M. K. (1941). Wheat in national diets. Wheat Studies, (02).
BerdeguĂŠ, J. A., Carriazo, F., Jara, B., Modrego, F., & Soloaga, I. (2015). Cities, Territories, and
Inclusive Growth: Unraveling Urban–Rural Linkages in Chile, Colombia, and Mexico.
Christiaensen, L., Weerdt, J., & Todo, Y. (2013). Urbanization and poverty reduction: the role of
rural diversification and secondary towns1. Agricultural Economics, 44(4-5), 435-447.
Cleveland, W. S. (1979). Robust locally weighted regression and smoothing scatterplots. Journal
of the American statistical association, 74(368), 829-836.
Cleveland, W. S., & Devlin, S. J. (1988). Locally weighted regression: an approach to regression
analysis by local fitting. Journal of the American Statistical Association, 83(403), 596-610.
De Haen, H., Stamoulis, K., Shetty, P., & Pingali, P. (2003). The world food economy in the
twenty‐first century: challenges for international co‐operation. Development Policy
Review, 21(5‐6), 683-696.
D’Haese, M., & Van Huylenbroeck, G. (2005). The rise of supermarkets and changing expenditure
patterns of poor rural households case study in the Transkei area, South Africa. Food
Policy, 30(1), 97-113.
Fafchamps, M., & Shilpi, F. (2003). The spatial division of labour in Nepal. The Journal of
Development Studies, 39(6), 23-66.
FerrĂŠ, C., Ferreira, F. H., & Lanjouw, P. (2012). Is There a Metropolitan Bias? The relationship
between poverty and city size in a selection of developing countries. The World Bank
Economic Review, lhs007.
Goldman, A. (2000). Supermarkets in China: the case of Shanghai. The international review of
retail, distribution and consumer research, 10(1), 1-21.
GĂłmez, M. I., Barrett, C. B., Raney, T., Pinstrup-Andersen, P., Meerman, J., Croppenstedt, A.,
Carishma, B., & Thompson, B. (2013). Post-green revolution food systems and the triple
burden of malnutrition. Food Policy, 42, 129-138.

139

GĂłmez, M. I., & Ricketts, K. D. (2013). Food value chain transformations in developing
countries: Selected hypotheses on nutritional implications. Food Policy, 42, 139-150.
Haggblade, S., Hazell, P., & Reardon, T. (2010). The rural non-farm economy: Prospects for
growth and poverty reduction. World Development, 38(10), 1429-1441.
Hayes, A. F., & Preacher, K. J. (2010). Quantifying and testing indirect effects in simple mediation
models when the constituent paths are nonlinear. Multivariate behavioral research, 45(4),
627-660.
Hino, H. (2014). Shopping at different food retail formats: Understanding cross-shopping behavior
through retail format selective use patterns. European Journal of Marketing, 48(3/4), 674698.
Huang, J., & Bouis, H. (2001). Structural changes in the demand for food in Asia: empirical
evidence from Taiwan. Agricultural Economics, 26(1), 57-69.
Leser, C. E. V. (1963). Forms of Engel functions. Econometrica: Journal of the Econometric
Society, 694-703.
Lewis, W. A. (1954). Economic development with unlimited supplies of labour. The manchester
school, 22(2), 139-191.
Monteiro, C. A., Levy, R. B., Claro, R. M., Castro, I. R. R. D., & Cannon, G. (2010). A new
classification of foods based on the extent and purpose of their processing. Cadernos de
saude publica, 26(11), 2039-2049.
Monteiro, C. A., Moubarac, J. C., Cannon, G., Ng, S. W., & Popkin, B. (2013). Ultra‐processed
products are becoming dominant in the global food system. Obesity reviews, 14(S2), 2128.
Nelson, G. C. (2002). Introduction to the special issue on spatial analysis for agricultural
economists. Agricultural Economics, 27(3), 197-200.
Neven, D., Reardon, T., Chege, J., & Wang, H. (2006). Supermarkets and consumers in Africa:
the case of Nairobi, Kenya. Journal of International Food & Agribusiness Marketing, 18(12), 103-123.
Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response variables
with an application to 401(K) plan participation rates. Journal of Applied Econometrics,
volume 11, 619-632.
Papke, L. E., & Wooldridge, J. M. (2008). Panel data methods for fractional response variables
with an application to test pass rates. Journal of Econometrics, 145(1), 121-133.
Pingali, P. (2007). Westernization of Asian diets and the transformation of food systems:
implications for research and policy. Food policy, 32(3), 281-298.

140

Popkin, B. M., Adair, L. S., & Ng, S. W. (2012). Global nutrition transition and the pandemic of
obesity in developing countries. Nutrition reviews, 70(1), 3-21.
Reardon, T., & Timmer, C. P. (2007). Transformation of markets for agricultural output in
developing countries since 1950: How has thinking changed?. Handbook of agricultural
economics, 3, 2807-2855.
Rischke, R., Kimenju, S. C., Klasen, S., & Qaim, M. (2015). Supermarkets and food consumption
patterns: The case of small towns in Kenya. Food Policy, 52, 9-21.
Senauer, B., Sahn, D., & Alderman, H. (1986). The effect of the value of time on food consumption
patterns in developing countries: evidence from Sri Lanka. American Journal of
Agricultural Economics, 68(4), 920-927.
Serber, G. A. F., & Wild, C. J. (1989). Nonlinear Regression. John Wiley & Sons, 435.
Sharma, A. (2016). Urban Proximity and Spatial Pattern of Land Use and Development in Rural
India. The Journal of Development Studies, 1-19.
Smith, L. C. (Ed.). (2003). The importance of women's status for child nutrition in developing
countries (Vol. 131). Intl Food Policy Res Inst.
Stone, R. (1954). Linear expenditure systems and demand analysis: an application to the pattern
of British demand. The Economic Journal, 64(255), 511-527.
Strydom, J. W. (2011). Retailing in disadvantaged communities: the outshopping phenomenon
revisited. Journal of Contemporary Management, 8, 150-172.
Subramanian, S., & Deaton, A. (1996). The demand for food and calories. Journal of political
economy, 104(1), 133-162.
Thiele, S., & Weiss, C. (2003). Consumer demand for food diversity: evidence for Germany.
Food Policy, 28(2), 99-115.
Tschirley, D., Reardon, T., Dolislager, M., & Snyder, J. (2015). The rise of a middle class in East
and Southern Africa: Implications for food system transformation. Journal of International
Development, 27(5), 628-646.
Turrell, G., & Kavanagh, A. M. (2006). Socio-economic pathways to diet: modelling the
association between socio-economic position and food purchasing behaviour. Public
health nutrition, 9(03), 375-383.
Veeck, A., & Veeck, G. (2000). Consumer segmentation and changing food purchase patterns in
Nanjing, PRC. World Development, 28(3), 457-471.
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press.

141

Working, H. (1943). Statistical laws of family expenditure. Journal of the American Statistical
Association, 38(221), 43-56.
World Bank. (2015, October 4). The international poverty line has just been raised to $1.90 a day,
but global poverty is basically unchanged. How is that even possible? [Blog post].
Retrieved from http://blogs.worldbank.org/developmenttalk/international-poverty-linehas-just-been-raised-190-day-global-poverty-basically-unchanged-how-even

142

ESSAY 3: CITY SIZE, SUPERMARKETS, AND PROCESSED FOODS:
EVIDENCE FROM ZAMBIA

3.1 Introduction
The past two decades in developing Sub-Saharan Africa can be categorized as a time of
significant lifestyle change driven by increasing incomes and rapid urbanization. Two of these
lifestyle changes are the increased utilization of supermarkets and the increased consumption of
processed food.
Retail modernization in developing countries dates back to the 1960s, when supermarkets
in Latin America were gaining popularity in response to growth of household incomes (Moyer &
Hollander, 1968; Harrison et al., 1974). However, it was not until after market liberalization,
which lead to the influx of foreign direct investment (FDI), that supermarkets’ share of retail
food in this part of the world rose rapidly during the 1990s (Reardon et al., 2004). Diffusion of
supermarkets in developing regions is the product of the following supply and demand
conditions (Reardon & Timmer, 2007): On the supply side; market liberalization facilitates the
inflow of FDI, complementary infrastructure provides needed logistics, and large-scale food
manufacturers provide food products to be sold via supermarkets. On the demand side; the
combination of rising incomes and urbanization contribute to market size that can support the
presence of supermarkets and an increase in household opportunity cost of time that strengthen
the benefits these supermarkets offer. These conditions were present in Latin America in the
1990s and are becoming ever more present in developing Africa. In the 1990s there was little
supermarket presence in Sub-Saharan Africa outside of South Africa, but currently the market

143

share of supermarkets in developing Africa, albeit small, is growing40 (Neven & Reardon, 2004;
Traill, 2006).
Consumer access to supermarkets provides at least three potential benefits. First,
supermarkets offer households time-savings with the ability to purchase a variety of foods from a
single location. Second, the cleanliness of supermarkets compared with traditional open-air
markets is a benefit for households with higher health awareness that accompanies increased
education (Turrell & Kavanagh, 2006). Third, supermarkets provide households with cost
savings as they typically offer goods such as processed foods at cheaper prices than other
retailers due to supermarkets’ ability to procure goods in bulk (Neven et al., 2006; Minten et al.,
2010).
Supermarkets offering processed foods at cheaper prices than other outlets could
contribute to the increased consumption of processed food that is primarily attributed to
economic development, characterized by rising incomes and rapid urbanization (Veeck & Veeck,
2000; Pingali, 2007; Popkin et al., 2012; Tschirley et al., 2015). The increased consumption of
processed food that occurs with rising income and urbanization is similar to the relationship
shown by Bennett’s Law, which identifies income as a driver of changing patterns of commodity
based food consumption. Increasing opportunity cost of time accentuates this pattern as it is
related to income growth and has also been found to increase consumption of processed food,
particularly when the opportunity cost of time is driven by the increased participation of females
in nonfarm employment (Senauer et al., 1986; Kennedy & Reardon, 1994).
The concept of “urbanization” embodies multiple characteristics, each of which could
have differing impacts on the pattern of processed food consumption. Urbanization increases

Neven & Reardon (2004) estimate that supermarket share of food retail in urban Kenya was 20%. Traill (2006) projected that
in 2015 supermarkets would capture 15% of Kenya’s national food retail market, up from 10% in 2002.
40

144

population density, providing retailers with a concentrated demand for food that incentivizes the
construction of supermarkets in urban areas (Battersby & Peyton, 2014). The rise of formal retail
outlets in urban areas introduces a greater variety of foods and promotes the consumption of
processed foods (GĂłmez & Ricketts, 2013). Urban areas also offer additional employment- or
entrepreneurial opportunities and a generally faster pace of life (Bettencourt et al., 2007), both of
which increase household opportunity cost of time, further affecting the demand for processed
food.
Urbanization is the growth of the share of population in urban areas relative to rural
areas. This growth occurs via migration from rural to urban areas and the expansion of urban
areas. Recently the poverty and migration literature has begun to take a more disaggregated
view of urbanization and explore the impacts of city size (determined by population) on income
and poverty. Households in larger cities generally have higher incomes than households in
smaller cities, while smaller cities provide greater opportunity for migrating rural households to
escape from poverty (FerrĂŠ et al., 2012; Christiaensen et al., 2013; BerdeguĂŠ et al., 2015). As city
size affects income and the escape from poverty, it is reasonable to expect that city size also
affects food consumption decisions of households. When examining consumer demand for food
variety in Germany, Theile & Weiss (2003) find that households living in larger cities demand
greater variety of food. This paper further considers the effects of city size on food consumption
patterns.
Jointly considering increasing supermarket utilization, processed food consumption and
the impact of city size exposes a gap in the literature. Literature exists regarding supermarket
adoption, processed food consumption, and, more recently, the impact of supermarket adoption
on processed food consumption (Asfaw, 2008; Rischke et al., 2015). Previous studies have also

145

examined how food consumption patterns vary across rural and urban areas (Popkin, 1999; Ecker
& Qaim, 2011; Tschirley et al., 2015). To my knowledge, no previous study has unpacked the
impact of urbanization on the food households in developing countries consume and from where
they acquire it by examining how this impact varies by city size.
This paper will contribute to the literature by examining the effects of city size on
supermarket use and processed food consumption when controlling for the effects of income and
other household variables. It will do this using a unique household dataset covering four cities in
Zambia. The direct and indirect effects of city size on these consumption decisions are tested
using previously applied empirical models that have been adapted to this study.
The paper proceeds as follows: Section two presents the conceptual model used to
articulate the decision process and hypotheses. Section three addresses the methodology, which
includes description of data, definitions, general and empirical models, and estimation
procedures. Section four reflects on the results and provides discussion. Lastly, section five
provides conclusions.

3.2 Conceptual Framework
Rischke et al. (2015) use the concept of the food environment, describing the supply side
facing consumers, combined with demand side factors, to motivate a discussion and generate
hypotheses about where people source their food, what and how much they source, and how they
eat. They do not consider city size. I adapt their framework to generate hypotheses regarding the
effect of city size on the likelihood of consumers shopping in supermarkets and of purchasing
processed foods when they do so.

146

Figure 3-1 summarizes my conceptual approach. City size affects characteristics of the
urban environment that are exogenous to consumers’ food choices; this urban environment
influences the food environment that consumers face (the supply side) along with the factors that
drive consumer demand. Together, the food environment and consumer demand factors
influence the food choices that consumers make: where to shop (supermarkets or other outlet
types) and what to buy (processed foods or others). The food consumed by households can be
influenced directly by the food environment and consumer demand factors, and indirectly
through the impact of these factors on supermarket shopping.

City size
Urban environment
-

Market size
Incomes & poverty
Congestion
Motorized transport
Investment cost
Advertising

Food environment
(supply side)
-

-

Access
Availability
Desirability (shopping
atmosphere, marketing)
Price

Consumer Demand
Characteristics
-

Incomes
Preferences
Norms & beliefs
Health concerns

What to buy

Where to shop

Food Choices
Figure 3-1: Conceptual approach linking city size to consumer food choices

147

The concept of food environment applies a multi-dimensional lens to the factors that
describe the supply side of retail food systems. Given that supermarkets in ESA compete with a
wide range of other outlet types (large and small supermarkets, traditional shops, open air
markets, and often thousands of dispersed street vendors), the factors need to be thought of with
respect to supermarkets per se, and to supermarkets relative to other outlet types. In Rischke et
al. (2015), food environment considers access, availability, desirability, and prices. Access refers
to the cost of gaining physical access to foods, such as the density of outlets, and the time and
money cost of reaching them. This concept is not to be confused with that of economic access.
Availability identifies the types and variety of foods (including processed foods) that
supermarkets and other outlets offer. Desirability recognizes that the food environment affects
household desire for products beyond household desire for product characteristics via shopping
atmosphere and marketing. Prices vary across products and across retailers (GĂłmez & Ricketts
2013). Shopping atmosphere is created by the characteristics of the physical shopping space
(size, design, internal arrangement), and by promotions and product placements. In urban ESA,
supermarkets are unique in the extent to which they purposively invest in shopping atmosphere.
Marketing affects how product characteristics relate to credence attributes created through
branding, labeling, and advertising (e.g. health claims). Note that these marketing investments
are always made first in packaged processed foods.
The food environment households experience is both a result of the retail food system
available to households and the retailers that households select to use. The variation between
retailers regarding availability, desirability and prices of food are accentuated when households
select to shop at specific retailers – such as, the desirability of certain foods being highlighted by
the promotions and product placements that are only experienced when shopping at a particular

148

retailer. Therefore, the decision of where to buy contributes to household decisions of what to
buy.
The demand side in this approach considers, in addition to consumer incomes, their
individual and household preferences (e.g. for taste, energy density, and convenience); their
social and individual norms and beliefs regarding the shopping environment, beauty and body
shapes, and lifestyle; and their health concerns including foods’ nutritional content and perceived
safety.
It is clear factors that influence the created food environment (the supply side), such as
branding, labeling, advertising, packaging, and store layout, also affect the demand side through
their impact on the credence that consumers lend to overt and implicit claims about products.
As consumers make their food choices, learning drives feedback loops to both the supply
side and the demand side: retailers learn by observing consumer decisions, and adjust their
approaches to stocking, marketing, and store layout (and over a longer time horizon, they affect
access through decisions on store location), and consumers learn about products and change their
assessment of their own tastes and preferences over time.
In rough order of certainty, I posit how city size affects aspects of the exogenous urban
environment that drives both the supply side (the food environment) and consumer demand
factors. I suggest the following effects as city size rises:
1. The size of the market increases more than proportionally to city size: this follows directly
from a positive relationship between city size and mean consumer incomes (see next point).
Increased size of market attracts investment in supermarkets, improves physical access to
them, and thus favors shopping in them. Increased market size should also promote the

149

availability of a wider array of foods, including processed foods; all else equal, this should
spur the consumption of processed foods.
2. Mean incomes rise and poverty rates fall (FerrĂŠ et al., 2012). These are robust empirical
patterns across Africa (World Bank, 2009; Christiaensen et al., 2013). Both should increase
shopping in supermarkets and also have direct positive impacts on the purchase of processed
foods, since these foods tend to have higher demand elasticities than do unprocessed foods
(Tschirley et al., 2015).
3. Congestion rises. This means that the cost of moving a given distance by motorized transport
rises. However, the cost of moving by foot or bicycle – both common forms of transport that
are negatively associated with household incomes and with shopping in supermarkets – is
probably insensitive to city size. Congestion thus asymmetrically affects physical access to
food, reducing the access advantage conveyed by motorized transport thus reducing the
probability of shopping in a supermarket in larger cities. Alternatively, due to the increased
cost of transportation, the costs of cross-shopping41 increases and the supermarket benefit of
offering a wide variety of food items in one location is of greater value to households, thus
increasing the probability of shopping in a supermarket in a larger city. Since congestion
increases the time cost of a given activity and thus increases the value of time, it should
directly drive demand for processed food and the convenience it offers.
4. Access to motorized transport – both public and private – increases. This effect follows from
higher consumer incomes and absolute amount of purchasing power in large cities (the
income elasticity of demand for private motorized transport is certainly positive, and public

The practice of cross-shopping is where households shop at various locations. Cross-shopping is common in China (Goldman
2000), Israel (Hino 2014) and in South Africa, where households travel to locations away from their area of residence to acquire
food (D’Haese & Van Huylenbroeck 2005, Strydom 2011).
41

150

transport availability should respond positively to total purchasing power). Access to
motorized transport has an asymmetric effect on physical access to food: motorized transport
improves physical access for outlets such as supermarkets that are larger and fewer in
numbers, but should have little or no effect on access to other outlet types, which have
developed historically to serve largely foot-bound consumers. Access to transport thus trends
in favor of supermarket shopping in large cities, in part because a key advantage of shopping
in such outlets is the ability to buy in bulk. Its effect on consumption of processed foods,
independent of shopping in supermarkets, is not clear.
5. The cost of infrastructural investment rises. This follows from higher congestion, higher
wages, and (if market mechanisms operate in any way in the land market) higher costs of
land. Higher investment cost probably works asymmetrically against investment in
supermarkets as city size rises: open air markets often arise spontaneously (informally) with
no explicit cost of land, and street vendors do not typically pay for the space they occupy.
These costs also raise food prices, reducing real incomes of consumers; any effect on
processed foods would be only through the supermarket effect.
6. Consumers are more exposed to modern advertising. This is a straightforward supply side
response of retailers and food processing companies to the larger market size in large cities,
combined with economies of scale (more eyes see a given billboard; tv and radio advertising
reach more people; the marginal return to advertising should be higher because incomes are
higher). By design, such advertising increases the likelihood that consumers will pursue a
more “modern” shopping experience, thus favoring use of supermarkets and (even
independent of such shopping) consumption of processed foods.

151

Table 3-1 summarizes my expectations regarding the impact of city size on food choices,
working through the urban environment, the food environment, and consumer demand factors.
Of the six urban environment factors, four drive a positive relationship between city size and the
use of supermarkets: market size, mean incomes, access to motorized transport, and exposure to
advertising. The cost of infrastructural investment drives a negative relationship and congestion
has mixed effects. Four factors also drive a positive relationship between city size and processed
food demand: market size, mean incomes, congestion, and advertising. Notably, I don’t expect
negative direct effects on processed food demand from any of the six factors.

152

Table 3-1: Impact of city size on food choices: urban environment, food environment, and
demand factors
Change in
exogenous urban
environment factor
as city size rises
Disproportionate
increase in market
size

Impact on supply (food environment) and demand
factors

Impact on food choices

Food environment

Choice of
supermarkets

Processed
food demand

Positive

Positive

Higher incomes;
operation of Bennet’s
Law; rising health
concerns; more
“modern” norms &
beliefs

Positive

Positive

Higher opportunity
cost of time

Positive or
Negative

Positive

Positive

Only through
supermarkets

Negative

Only through
supermarkets

Positive

Positive

Increased access to
supermarkets; increased
availability of processed
(and other) foods

Increased incomes,
lower poverty

Higher congestion

Greater access to
motorized transport

Higher investment
cost
Greater exposure to
advertising

Demand factors

Asymmetric impact on
physical access – reduces
advantage conveyed by
motorized transport
while also increasing the
benefit of being able to
purchase all food from
one location
Asymmetric impact on
physical access –
increases it for
supermarkets but little or
no effect for other outlet
types
Reduced physical access
to supermarkets
Increased desirability of
food, especially of
packaged, processed
foods

More “modern” norms
& beliefs around food

This framework thus predicts a strongly positive effect of city size on processed food
consumption: all direct factors are positive; any negative effects are indirect through a negative
effect on use of supermarkets and a hypothesized positive effect of supermarkets on processed
food demand; and the possible indirect negative effects of congestion are offset by the positive
direct effects.

153

The net effect of city size on use of supermarkets depends on the size effect of each
factor, which varies on a host of factors specific to any given locale. Note first that there is
clearly a threshold effect: supermarkets do not exist in rural areas or small towns in Africa; urban
settlements have to reach a certain size before they will attract a supermarket. Once that
threshold is reached, however, I see no way, a priori, to predict the relationship between city size
and supermarket use42. Notably, the framework does suggest the possibility of city size
negatively affecting the use of supermarkets, depending on the impact on investment costs and
household response to the level and cost of congestion.

3.3 Methods
3.3.1 Data
To estimate the effects of city size on supermarket use and processed food consumption I
use data from the 2007/2008 Zambia Urban Consumption Survey (UCS), which were gathered
via the joint efforts of the Zambian Central Statistics Office and the Zambia Food Security
Research Project. These data include a sample of 2,160 urban households across four cities:
Lusaka, Kitwe, Kasama and Mansa (Table 3-2). To account for potential seasonal variation in
consumption, data were collected in two rounds: August 2007 and February 2008. August is
soon after primary harvest while February is traditionally the height of the lean season. During
the second round of data collection, 1,865 households were successfully re-interviewed. The
standard enumeration areas were selected in such a way to collect data from households that
represent the neighborhoods across multiple socioeconomic strata. See Hichaambwa et al. (2009)
for more sampling details of this survey.

42

Note that all the cities in this sample have a supermarket.

154

Table 3-2: City populations
Census Year
Annual Growth %
2000
2010
Lusaka
1,084,703
1,747,152
4.88%
Kitwe
363,734
501,360
3.26%
Kasama
74,243
101,845
3.21%
Mansa
41,059
78,153
6.65%
Zambia National
10,100,981 13,216,985
2.73%
Zambia Urban
3,515,393
5,118,243
3.83%
Urban Share
34.8%
38.7%
1.07%
Sources: city data obtained from http://www.citypopulation.de/Zambia-Cities.html, national data obtained from the United
Nations Population Division, Department of Economic and Social Affairs, 2014 Revision of the World Urbanization Prospects.
City

Lusaka, Kitwe, Kasama and Mansa were selected to represent multiple city types in
Zambia. Lusaka is the nation’s capital and is three times larger than the next largest city, Kitwe.
Kitwe is the largest city in the Copperbelt, where the economic activity is driven by copper
mining. Kasama and Mansa are smaller, ranking as the tenth and twelfth largest cities in Zambia
per 2010 census data43. Like Lusaka and Kitwe, Kasama and Mansa are district capitals, but
these cities have a higher share of informal employment than their larger counterparts. These
four cities also vary in per adult equivalent average expenditure44, with average Lusaka
expenditures about double those in the tertiary cities (Table 3-3). I recognize that four cities
comprise a limited set of cities to formulate strong assertions regarding patterns driven by city
size, but given the variation in the city populations I feel that this analysis will provide a basis for
understanding the effects of city size on food consumption patterns.

http://www.citypopulation.de/Zambia-Cities.html
Expenditure data were converted into USD using historic exchange rates obtained from XE.com, and conversion information
for constant 2011 international dollars in purchasing power parity from worldbank.org. Data includes values for food from own
production, gifts of food and imputed rent. Adult equivalents (AE) are calculated as one AE for each household member fifteen
years of age or older, three quarters of an AE for each child aged five to fourteen, and a half an AE for each child younger than
five years old.
43
44

155

Table 3-3: Total daily household expenditure per adult equivalent
Lusaka
Kitwe
Kasama
Mansa

All
8.94
6.98
4.29
4.51

Quintile 1
2.69
2.06
1.28
1.22

Quintile 2
4.86
3.64
2.17
2.29

Quintile 3
7.30
5.19
3.18
3.53

Quintile 4
11.14
7.93
4.58
5.40

Quintile 5
25.63
20.45
11.51
12.03

Source: Zambia Urban Consumption Survey 2007/2008
Note: Quintiles represent total expenditure quintiles, by city.
Values presented in constant 2011 USD, PPP adjusted.

Zambia is part of the final wave of the “supermarket revolution”, with a growing
presence of supermarkets across the country (Reardon et al. 2009). Each of these cities has at
least one large chain supermarket and smaller non-chain supermarkets. Data collected in this
survey include distance from the household to each of the retailers from which it acquired food
(Table 3-4). Average distances to supermarkets vary across cities and expenditure quintiles.
Supermarkets are located closer to wealthy neighborhoods than to less affluent neighborhoods,
consistent with patterns observed in South Africa (Battersby & Peyton, 2014).

Table 3-4: Kilometers to supermarket relative to closest non-supermarket food retailer to the
household
Lusaka
Kitwe
Kasama
Mansa

All
3.2
7.1
3.3
1.8

Quintile 1
3.5
8.4
4.6
2.5

Quintile 2
3.4
7.6
3.9
2.0

Quintile 3
3.7
7.2
3.4
1.9

Quintile 4
3.1
6.8
2.5
1.2

Quintile 5
1.8
4.3
1.8
1.0

Source: Zambia Urban Consumption Survey 2007/2008

Data collected by this survey include household demographics (health, income and
education), asset ownership, general expenditures from the previous six months and detailed
information on food consumption. The food consumption module collected information on 68
commonly consumed food items from the last 30 days that were either purchased, received as
gifts, or from own production. Data on food consumed away from home were collected in a
separate module using 24 hour recall. Table 3-5 identifies the share of market value by

156

acquisition type. Data regarding the source and distance to household were recorded for all
purchased food items. Household level prices of food purchases were not able to be computed
due to non-standard units.

Table 3-5: Shares of market value by acquisition type
Weighted Average (all cities)
Lusaka
Kitwe
Kasama & Mansa Average
Kasama
Mansa

Own
Production
2.5
1.6
4.0
9.7
10.4
8.2

Supermarket
Purchases
7.6
7.9
5.8
9.6
8.7
11.4

Gifts
1.9
1.8
1.6
3.8
4.4
2.4

Non-Supermarket
Purchases
73.4
72.7
77.7
66.7
67.7
64.7

Food Away
From Home
14.7
16.0
10.8
10.2
8.7
13.3

Source: Zambia Urban Consumption Survey 2007/2008
Note: Food away from home is not considered to be non-supermarket purchase, but is segregated out in this table to represent its market share.

3.3.2 Definitions
Key definitions regarding supermarkets, processed food and city size are specified here.
First, supermarket is defined as a self-service retail outlet with a variety of goods that is typically
characterized by multiple cashiers. The definition used for supermarkets both large and small
supermarkets that are both independent or chains. Supermarket share refers to the household of
total household food consumption that was obtained in a supermarket. Second, processed food is
defined as food that prior to household acquisition has been transformed from its original state,
beyond removal from the plant and (for non-perishables) drying. Processed food therefore
identifies food embedded with the labor provided by humans or machines prior to acquisition,
reducing household time and energy required for preparation. Processed share refers to the
household share of processed food in total household food consumption. The processed food
aggregate is further observed using sub-aggregates of low- and high processed foods. High
processed foods are processed foods that satisfy at least two of the three conditions: multiple
ingredients; physical change induced by heating, freezing, extrusion, or chemical processes; and

157

packaging more complex than simple paper or plastic. Although the processed sub-aggregates
are constructed based on implications on the structure of the food system they closely mirror the
processed food aggregates considered nutrition based literature45 (Monteiro et al., 2010). Third,
city size classifies cities by population: primary cities have populations greater than one million
(Lusaka), secondary cities have populations below one million and above than one hundred
thousand (Kitwe), and tertiary cities have populations below one hundred thousand (Kasama and
Mansa).

3.3.3 General Model
Our model estimates how consumption decisions are affected by household variables,
including variables that proxy for household income. An Engel’s Curve Model is an appropriate
general model to use given that it estimates the relationship between shares of consumption
against household income, allowing for estimation of the effects of additional determinants of
consumption such as city size. The functional form of Engel Curves proposed by Banks et al.
(1997) is an appropriate form for this analysis as this functional form allows for the effect of
total expenditure on the shares of food aggregates to vary at different levels of expenditure.

3.3.4 Empirical Model
Our conceptual model posits that city size influences the urban environment, which
affects both the supply side (the food environment) and the demand side of the consumer’s food
consumption decisions. In empirically specifying the model, I search for variables that capture
as many of the supply- and demand side factors as possible. I also include dummy variables for

Figure 3-A-1 shows a sample of the food items included in the Zambia data and how they map into both the
Monteiro et al. 2010 aggregates and the processing sub-aggregates used here.
45

158

each city which, due to the large size differences, can be argued to primarily reflect city size. To
the extent that variables are missing, or imperfectly capture the concepts in the conceptual
model, I use the conceptual model to interpret the meaning of the city size variables and generate
defensible conclusions.
The dataset captures consumer demand characteristics relatively well (Figure 3-1), with
variables for household per capita expenditure along with demographic variables that proxy for
preferences, norms & beliefs, and health concerns.
Data are less complete on the food environment. The most representative variable that I
have regarding food environment is relative distance to supermarket, which I compute as the
distance from household to nearest supermarket minus the distance from household to nearest
non-supermarket food retailer. This variable relates to physical access yet does not fully capture
it. In particular, because it is based only on distance it is unable to capture the effects of
congestion in the urban environment. Logically, greater congestion should increase the real
costs related to distance by increasing the time needed to travel a given distance when using
motorized transport. I thus expect that relative distance to supermarket will have a larger
(negative) effect on supermarket share in the larger cities (which I expect to be more congested).
Measures of availability and food desirability are not available. Systematic differences
across cities will be captured by city dummy variables in pooled regressions. One such
systematic difference is how city size affects the desirability of food items via heterogeneous
prices. The dataset does not capture household level prices due to the high prevalence of nonstandard units. I do know that general price levels are higher in larger cities than in smaller cities,
as indicated by a price index46 calculated from select community level prices captured within the

Expenditure weighted price index comprised of the 10 most consumed items during each round of data collection: commercial
maize meal, bread, chicken, beef, cooking oil, rape, tomato, and sugar in both rounds with rice and dried fish in round 1 and
46

159

data (Table 3-A-1). However, certain processed foods (i.e. bread and commercial maize meal)
are found to have lower average prices in primary cities. The lower prices of processed foods are
expected to contribute to greater consumption of processed foods by households in primary
cities, an effect expected to be captured by the city dummy variables.
Examining the six urban environment factors from Table 3-1 of the conceptual model, I
note that I am able to directly control for two of them: income and access to motorized transport.
Both are hypothesized to have positive effects on use of supermarkets. I am left with four factors
that need to be captured by city size dummies – market size, congestion, investment cost, and
advertising. Two of these, congestion and investment cost, are hypothesized to have negative
effects on the use of supermarkets. It would thus not be surprising if I find that increasing city
size decreases the use of supermarkets.
The effects of city size and the other selected variables on supermarket share and on
processed share are estimated by the following Engel’s Curve Model:
𝐷𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒 = 𝑓(𝐴𝑐 ∗ , 𝐴𝑣 ∗ , 𝐷∗ , 𝑃 ∗ , 𝑌 ∗ , 𝑍 ∗ , 𝑊 ∗ , 𝐻 ∗ )

(3.1)

The dependent variables are supermarket share and processed share (Table 3-6).
Processed share is further disaggregated into shares of low processed and high processed food.
Foods are high processed if they satisfy at least two of the following three conditions: multiple
ingredients; physical change induced by heating, freezing, extrusion, or chemical processes (i.e.,
more than simple physical transformation); and packaging more complex than simple paper or
plastic.

kapenta and beans in round 2. Round 1 Index: Lusaka = 100, Kitwe = 108.7, Kasama = 92.0, Mansa = 99.1. Round 2 Index:
Lusaka = 100, Kitwe = 89.5, Kasama = 76.1, Mansa = 81.5. Overall mean by city Lusaka = 100, Kitwe = 100.9, Kasama = 85.6,
Mansa = 94.1..

160

Table 3-6: Dependent variables
Supermarket Shares
Food acquired from supermarkets such as Shoprite or Melissa
Processed Food Shares
Any Processed Food
Own Production
Unprocessed
Low Processed
High Processed

Any food that is acquired in a processed form
Food that was produced by the household
Food acquired as unprocessed, e.g. maize grain
Food acquired as low processed, e.g. maize flour
Food acquired as high processed, e.g. bread

Independent variables that are identified in Table 3-7 are categorized into three groups:
variables that are only determinants for supermarket share, variables that are only determinants
of processed share, and variables that are shared determinants for both supermarket share and
processed share.

Table 3-7: Key for determinants of demand in equation 3.1
Ac*

Access

Relative distance to
supermarket, Own a car,
Own a motorcycle, Own a
bicycle, Use public
transportation, City size

Y*

Availability

Supermarket share, Lean
season, City size

D*

Desirability

Supermarket share, Lean
season, City size

W*

P*

Prices

City size

H*

Av*

Z*

Disposable
Income

Daily total expenditure per adult
equivalent, City size

Individual and
Household
Preferences

Opportunity cost of time (Percentage
working age females in nonfarm labor,
Dependency ratio), Adult equivalents,
Own farmland, Own a stove, Own a
hotplate, Maximum level of education,
Age of household head, Female head
of household, City size

Social and
Individual
Norms and
Beliefs
Health
Concerns

Maximum level of education, Own a
phone, City size
Maximum level of education, Female
head of household, City size

Note: Supermarket share is only included for processing share dependent variables

Relative distance to supermarket is only included in the supermarket share model.
Distance to supermarket is shown to be negatively correlated with the use of supermarkets
(GĂłmez et al., 2013), and relative distance is a refinement of this measure in an effort to
highlight the opportunity costs of transport to a specific retail location (Goldman, 2002 & 2005).
Any effect of relative distance to supermarket on processed share would only occur via the use of

161

a supermarket. This variable is not intended to capture the effect of distance to household per se,
rather it is to capture the effect of the marginal cost of transportation required to access a
supermarket versus a non-supermarket retailer. It is calculated as the logarithmic value of the
difference in kilometers from household residence to the nearest supermarket where the
household purchased food and the distance to the nearest non-supermarket where the household
purchased food. If a household did not acquire food from a supermarket, the median relative
distance from household to supermarket among all households in the enumeration area is used. If
the relative distance is negative47, the distance measure is replaced with 0.001 to allow for the
calculation of a logarithmic value.
Interaction variables of relative distance to supermarket multiplied by city dummy
variables are included to allow for varying effects of relative distance to supermarket on
supermarket share across the cities. Interaction variables for Lusaka, Kitwe and Kasama are
included in the model, with the relative distance to supermarket variable capturing that effect in
Mansa. I expect stronger negative effects of relative distance to supermarkets in larger cities due
to increased congestion that would increase the cost of motorized transport.
Supermarket share is included as an independent variable in the processed share model.
Supermarket share is likely to be endogenous to processed share due to self-selection. Potential
endogeneity is not expected to be the result of simultaneity as processed food is also available
from non-supermarkets and unprocessed food is available from supermarkets. Treatment of the
potential endogeneity is addressed in the estimation section of the paper. Supermarket share is
included as a determinant of the share of processed food consumed, as supermarkets are
expected to increase the share of processed food consumption (Asfaw, 2008; Rischke et al.,

47

Negative relative distance to supermarket occurs in less than half of one percent of observations.

162

2015). Shopping at a supermarket is expected to increase household access to a greater variety of
food, including processed foods, to expose households to greater marketing efforts that increase
the desirability of processed food, and to offer processed foods at reduced prices.
Interaction variables of supermarket share multiplied by city dummy variables are
included to allow for varying effects of supermarket share on processed food consumption across
the cities. Interaction variables for Lusaka, Kitwe and Kasama are included in the model, with
the supermarket share variable capturing the effects of supermarket use in Mansa.
The following variables are included in both supermarket share and processed share
models.
City dummy variables are included for Lusaka, Kitwe and Kasama, while the effects of a
household residing in Mansa are captured in the intercept. These variables capture systematic
differences across cities in the pooled regressions, identifying the partial direct effects of cities
on the households’ consumption decisions. A mediation model as described by Hayes &
Preacher (2010) will be described in the estimation method of this paper and is used to identify
indirect effects of city size on the consumption decisions.
Total daily household expenditure per AE is included in the model serving as a proxy for
income, a key factor of consumer demand. It enters the estimation in both the logarithmic and
squared logarithmic forms to remain consistent with the Banks et al. (1997) form of the Engel
Curve model. The marginal effects of total expenditure are expected to vary across income, but
generally increase with rising total expenditure for both supermarket and processed shares.
In addition to city size, the following variables are included to capture the effects of
individual and household preferences: percentage working age females in nonfarm labor,

163

dependency ratio, adult equivalents, own farmland, own a gas or electric stove, own a hotplate,
maximum level of education, age of household head, and female head of household.
A rising percentage of working age females employed in nonfarm labor is expected to
increase the households’ opportunity cost of time, resulting in increased demand for processed
food (Senauer et al., 1986; Kennedy & Reardon, 1994); this increased opportunity cost of time is
also expected to impact supermarket retail demand due to the time saving motivation of
acquiring all foods from a single location. The calculation for this variable includes the number
of working age48 women who are employed in nonfarm labor divided by the number of working
age women in the household. The dependency ratio, defined as a percentage of non-working age
members of the household, represents potential increasing opportunity cost of time for those
providing the support to these household dependents as well as a motivation to provide more
healthy food options. Household adult equivalents account for the effect of household size on
consumption decisions. Household ownership of a farm increases the likelihood that the
household would consume food from own production, reducing the household demand to
purchase food from supermarkets and, if consuming food from its own production, directly
reducing the share of acquired processed food. Household ownership of a stove (gas or electric)
and household ownership of a hotplate both reduce the preparation costs of unprocessed food,
which leads to an expectation of reduced processed share. At the same time, ownership of these
appliances indicates a willingness of the household to invest in time saving capital, which
implies a higher opportunity cost of time that would lead to increased supermarket and processed
shares. The maximum years of completed education by a member of the household represents the
household capacity for more educated decisions. Higher education is associated with healthier

48

Ages 15-64

164

food consumption decisions (Turrell & Kavanagh, 2006), potentially influencing the shopping
decisions of households. Age of the head of household is included as elderly households are less
likely to transition away from the more traditional diet and food acquisition patterns. Female
head of households are likely to consume less food from supermarkets and consume less
processed food, as females commonly place a greater emphasis on health and deferred utility
(Smith, 2003).
Social and Individual Norms and Beliefs are captured with the variable of ownership of a
mobile phone and the aforementioned city size and maximum years of completed education
variables. Ownership of a mobile phone could indicate a household’s openness to modernization
and therefore willingness to purchase processed food and shop at supermarkets.
Variables that capture level of household health concerns are maximum level of
education, female head of household and dependency ratio; each are described above.
In addition to city size and supermarket share, five variables are included to capture the
effects of food environment. Ownership of a car, ownership of a motorcycle, ownership of a
bicycle and use of public transportation directly affect access to markets. Household ownership
of a car would reduce the time required in transportation to the supermarkets and
correspondingly, the opportunity costs involved is purchasing food from a supermarket as well as
increase household ability to buy and transport food purchased in bulk. Ownership of a
motorcycle is expected to have a similar effect on transportation costs as ownership of a car, but
offering less benefit on the ability to shop in bulk. Ownership of a bicycle also reduces the
transportation costs associated with small purchases, but would likely have little effect on
household ability to shop in bulk. The use of public transportation allows households to have
improved access to markets without car, motorcycle or bicycle ownership. Each of these

165

transportation variables additionally indicate a reduced transportation cost of households
bringing unprocessed food to processing centers, thereby reducing household cost of processing
unprocessed food. The net effect of these variables is expected to be positive for supermarket and
processed food shares.
The final variable that captures the effects of the food environment factors, availability
and desirability, is the lean season dummy variable, where a value of one indicates the data were
collected during the lean season. This variable captures differences in consumption decisions
driven by the season when the consumption occurs. The lean season is typically defined by
reduced stocks of food and higher food prices directly contributing to the availability and
desirability of the food environment.

166

Table 3-8: Household characteristics of the sample
Independent Variable
Total Daily Household Expenditure per
Adult Equivalent
Percentage Working Age Females
Employed in Nonfarm Labor
Dependency Ratio (%)
Adult Equivalents
Relative Distance to Supermarket (km)
Percentage Own a Car
Percentage Own a Motorcycle
Percentage Who Use Public
Transportation

Pooled
8.1

Lusaka
8.9

Kitwe
7.0

Kasama
4.3

Mansa
4.5

(5.4)

(6.2)

(4.6)

(3.1)

(3.2)

34.2

34.7

32.5

34.9

34.9

(25.0)

(25.0)

(16.7)

(0.0)

(25.0)

39.8

39.3

39.7

44.9

42.6

(40.0)

(40.0)

(40.0)

(50.0)

(42.9)

5.7

5.5

6.1

5.7

6.0

(5.4)

(5.1)

(5.9)

(5.4)

(5.9)

4.1

3.2

7.1

3.3

1.8

(3.0)

(2.8)

(7.5)

(2.5)

(1.5)

11.4

12.9

9.1

5.6

7.0

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

0.5

0.5

0.6

0.6

1.1

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

36.9

71.3

73.3

77.1

40.9

(100.0)

(100.0)

(100.0)

(0.0)

(0.0)

Percentage Own a Bicycle

20.6

15.9

20.9

57.3

54.7

(0.0)

(0.0)

(0.0)

(100.0)

(100.0)

Percentage Own a Farm

14.6

8.7

18.7

50.8

46.3

(0.0)

(0.0)

(0.0)

(100.0)

(0.0)

Percentage Own a Stove (Gas or
Electric)

34.4

34.0

40.5

20.9

21.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Percentage Own a Hotplate

24.9

30.7

13.6

8.2

15.2

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Percentage Own a Mobile Phone

75.1

77.2

74.9

56.4

63.4

(100.0)

(100.0)

(100.0)

(100.0)

(100.0)

11.3

11.4

11.1

10.4

10.8

(12.0)

(12.0)

(12.0)

(10.0)

(12.0)

Highest Completed Grade of Education
in Household

43.7

42.9

45.6

44.3

45.3

(42.0)

(41.0)

(45.0)

(42.0)

(44.0)

Female Head of Household

19.0

20.0

16.6

17.2

20.9

(0.0)

(0.0)

(0.0)

(0.0)

(0.0)

Weighted Percentage of Sample
Number of Household Observations

100
4025

67.8
1330

23.6
1352

5.8
661

2.9
682

Age of Household Head

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

3.3.5 Estimation Method
The estimation of the empirical models is completed in two steps. The first step estimates
the partial effects of all determinant variables on consumption decisions, including the direct
effects of city size. The second step estimates the indirect effects that city size has on
consumption through the other independent variables.
Estimating direct effects: A nested set of econometric issues needs to be dealt with in the
estimation: the likely endogeneity of supermarket share, the proper estimator for share variables

167

bounded between zero and one, and methods for dealing with endogeneity depending on the
estimator chosen49. I discuss each in sequence in the next paragraphs.
Supermarket share is likely endogenous as an explanatory variable of processed share due
to unobserved systematic differences between households that frequently shop at supermarkets
and those that do not – differences that potentially affect household decisions regarding
processed food consumption. To take one example, supermarket shoppers may have higher
average incomes, thus higher opportunity costs of time, and thus a higher likelihood of
consuming processed foods; supermarket shoppers could also be more favorably disposed
towards “modern” consumption habits and for this reason choose processed foods.
Endogeneity would typically be dealt with by use of two stage least squares (2SLS).
Relative distance to supermarket would be used as an instrumental variable (IV) as relative
distance to supermarket has a clear impact on the costs of shopping at a supermarket, but should
not independently affect the share of processed food consumption except through the purchasing
of food from a supermarket. This approach has been applied previously by Rischke et al. (2015).
The first stage regression would estimate supermarket share with equation 3.2.
𝑆𝑖 = 𝜃𝐷𝑖 + 𝜑𝑪𝑖 + 𝜏𝒀𝑖 + 𝜌𝑿𝑖 + 𝜔𝑖

(3.2)

where 𝑆𝑖 is the share of food value purchased from supermarkets in total food value consumed
by households; 𝐷𝑖 is the log of the household net distance to supermarket that captures much of
the effect of the access to market factor (Ac*) of food environment; 𝑪𝑖 are city dummy variables
for Lusaka, Kitwe and Kasama that capture direct effects of city size that affect multiple factors
of food environment and consumer demand; 𝒀𝑖 are the logarithmic and squared logarithmic
forms of daily household total expenditure per adult equivalent, which proxies for the effect of
OLS regression results are included in the appendix as a robustness check to the estimations that control for
potential endogeneity.
49

168

income (Y*) on consumption decisions; 𝑿𝑖 are other demographic and seasonal dummy variables
identified in the general equation that affect multiple factors of food environment and consumer
demand; 𝜃, 𝜑, 𝜏 and 𝜌 are estimated coefficients; and 𝜔𝑖 is the error term. An F-test of the
significance of distance to supermarket on SM share results in a value greater than ten;
suggesting that distance to supermarket is a strong IV (Staiger & Stock 1994).
The second stage in 2SLS estimates processed share continues to be based on equation 1
while incorporating household estimates of supermarket share from the first stage regression, as
shown in the following equation:
𝐹𝑖 = 𝛾𝑆̂𝑖 + 𝛼𝑪𝑖 + 𝜗(𝑪𝑖 ∗ 𝑆̂𝑖 ) + 𝛽𝒀𝑖 + 𝛿𝑿𝑖 + 𝜀𝑖

(3.3)

where 𝐹𝑖 is the share of processed food value in total household food consumption; 𝑆̂𝑖 is the
estimated share of food value purchased from supermarkets in total food value consumed by
households (from equation 2); 𝑪𝑖 ∗ 𝑆̂𝑖 are dummy variables of Lusaka, Kitwe and Kasama
interacted with the estimated supermarket share; 𝑪𝑖 , 𝒀𝑖 , and 𝑿𝑖 represent the same variables as
indicated in equation 2; 𝛾, 𝛼, 𝜗, 𝛽 and 𝛿 are estimated coefficients; and 𝜀𝑖 is the error term.
2SLS presents a weakness in this case, however: the bounded nature of the variables
makes the fractional probit (FP) the best estimator, and the non-linearity of the FP model
requires the use of a control function to account for endogeneity, not 2SLS. The shares of
processed food and of supermarket purchases in total value of consumed food are never greater
than one and never lower than zero. Given the characteristics of the dependent variables and the
assumption of asymptotically normal standard errors, the appropriate method of estimation is the
FP model (Papke & Wooldridge 1996 & 2008).
Due to the nonlinearity of the FP a control function is used to control for endogeneity
(Smith & Blundell, 1986). A control function uses the residuals of the first stage regression as an

169

explanatory variable in the second stage, together with actual values of the endogenous variable.
Insufficient methodology pertaining to predicting residuals from the FP model in the first stage
regression requires the estimation of the first stage with OLS (eq. 4) and the second stage using
FP (eq. 5) to apply the control function that accounts for potential endogeneity (Wooldridge,
2010).
𝑆𝑖 = 𝜃𝐷𝑖 + 𝜑𝑪𝑖 + 𝜏𝒀𝑖 + 𝜌𝑿𝑖 + 𝜔𝑖

(3.4)

𝐹𝑖 = 𝛾𝑆𝑖 + 𝛼𝑪𝑖 + 𝜗(𝑪𝑖 ∗ 𝑆𝑖 ) + 𝛽𝒀𝑖 + 𝛿𝑿𝑖 + 𝜋𝜔
̂𝑖 + 𝜀𝑖

(3.5)

Variables included in equations 4 and 5 mirror those of equations 2 and 3 with the
following exceptions: 𝑆𝑖 is included in the processed share regression as observed at the
household level and the predicted errors from 4 are included in 5. Regarding discussion specific
to supermarket use, I maintain the use of the FP model in the estimation of equation 4 as over
two thirds of the population consumes zero percent of their food from supermarkets. The OLS
estimates used in the first stage of processed share estimation are reported in the appendix (Table
3-A-2). For an initial robustness check the model is also estimated using 2SLS (Tables 3-A-8, 3A-9, & 3-A-10), and these estimates can be found in the appendix with only minor differences
from the FP estimates.
As household expenditure per adult equivalent enters the estimation in both logarithmic
and squared logarithmic forms (𝒀𝑖 ) the estimates of the average partial effects of expenditure are
calculated as follows:
𝜕
𝜕(𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒)

1

= [𝑎 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒 + 2𝑏

𝑙𝑛(𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒)
𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒

] ∗ Ε𝜙[𝑋𝛽]

where 𝑎 equals the marginal effects of log expenditure and 𝑏 equals the marginal effects of
squared log expenditure from equations 4 and 5 for supermarket share or processed share

170

(3.6)

respectively, and is equal to the average of the normal densities of the predicted values of the
dependent variable.
A second robustness check is to take advantage of the two rounds of panel data by using
a correlated random effects (CRE) model to control for unobserved household heterogeneity. As
in 4 and 5, the first stage is estimated with OLS and the second stage with FP and a control
function to account for potential endogeneity:
𝑆𝑖 = 𝜃𝐷𝑖 + 𝜑𝑪𝑖 + 𝜏𝒀𝑖 + 𝜌𝑿𝑖 + 𝜎𝒁̅𝑖 + 𝜔𝑖
𝐹𝑖 = 𝛾𝑆𝑖 + 𝛼𝑪𝑖 + 𝜗(𝑪𝑖 ∗ 𝑆𝑖 ) + 𝛽𝒀𝑖 + 𝛿𝑿𝑖 + 𝜇𝒁̅𝑖 + 𝜋𝜔
̂𝑖 + 𝜀𝑖

(3.7)
(3.8)

Variables included in equations 7 and 8 mirror those of equations 4 and 5 with the
addition of 𝒁̅𝑖 , which represents the household averages of the time variant determinant
variables. Due to low variance in the 𝑍𝑖 between the two rounds, results of the CRE are very
close to those from 4 and 5; the CRE output can be found in the appendix (Tables 3-A-11 – 3-A15).
Finally, in addition to estimating the models with city dummy variables, I drop the city
dummy variables and repeat all of the estimations separately for each city.
Estimating indirect effects: The above estimation methods calculate the direct effects of
cities on the consumption decisions of supermarket share and processed share. The estimation of
indirect effects is completed by treating the non-city determinant variables as mediator variables.
A mediator variable acts as a conduit for a partial effect of a predictor variable (city variables).
The indirect effect of the predictor variable is calculated by multiplying the estimated effect of
the mediator50 on the dependent variable by an estimated effect of the predictor variable on the
mediator:

When total expenditure is the mediator variable, it should be noted that the marginal effect of total expenditure on the
dependent variable follows from equation 6.
50

171

𝑖𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑒𝑓𝑓𝑒𝑐𝑡 =

𝜕(𝑚𝑒𝑑𝑖𝑎𝑡𝑜𝑟 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒)
𝜕(𝑐𝑖𝑡𝑦 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒)

𝜕(𝑑𝑒𝑝𝑒𝑑𝑒𝑛𝑡 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒)

∗ 𝜕(𝑚𝑒𝑑𝑖𝑎𝑡𝑜𝑟 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒)

(3.9)

Total effects of the predictor variable on the dependent variable are the summation of
significant direct and indirect effects. OLS is used to estimate city indirect effects on the
independent variables from the model. Bootstrapping is used to obtain asymptotic
approximations of the statistical significance of the indirect effects (Wooldridge, 2010). This
estimation of indirect effects follows the methodology described in Hayes & Preacher (2010).

Figure 3-2: Estimation of indirect effects

3.4 Results
3.4.1 Descriptive Statistics
Supermarket share and processed share vary across city size (Table 3-9). Households
from tertiary cities choose to purchase at least some of their food from supermarkets at a greater
rate than households from larger cities51, and also obtain a greater share of their food from
supermarkets52. These initial relationships between supermarket utilization and city size follow
the potential negative relationship highlighted in the conceptual model. The primary and
secondary cities have average incomes fifty to one hundred percent higher than the tertiary cities,
which is expected to increase supermarket use, but this alone does not result in a greater share of
supermarket adoption. Higher congestion and increased opportunity costs are effects of city size
on urban environment that could lead to this relationship.

51
52

46% versus 32% and 35%
7% versus 5% and 4%

172

Table 3.9: Household weighted averages of consumption shares
Full Sample Population
Share of food
from
supermarkets
in total food
consumption
Weighted Average
(all cities)
Lusaka
Kitwe
Kasama & Mansa
Average
Kasama
Mansa

Share of
processed food
in total food

Share of food
from
supermarkets
that is processed

Percentage of
households that
shop at
supermarkets
33.7

Subset of Households that
Bought Some Food from a Supermarket
Share of
Share of food
processed food
from
Share of
from
supermarkets
processed food
supermarkets
in total food
in total food
in total
consumption
processed food
15.1
69.4
19.3

5.1

65.2

86.8

(0.0)

(67.1)

(100)

(0.0)

(8.9)

(70.6)

5.4

67.0

87.3

31.9

17.0

70.4

21.5

(0.0)

(68.3)

(100)

(0.0)

(11.0)

(71.3)

(14.2)

12.7

(12.3)

3.5

64.8

85.8

34.5

10.2

70.2

(0.0)

(67.1)

(100)

(0.0)

(6.4)

(70.8)

(8.3)

6.7

52.4

86.5

46.1

14.6

62.0

20.7

(0.0)

(53.9)

(100)

(0.0)

(11.2)

(64.3)

(16.6)

5.6

50.1

85.7

41.1

13.7

60.0

19.7

(0.0)

(50.7)

(100)

(0.0)

(9.8)

(61.4)

(14.6)

8.9

57.2

87.6

56.2

15.9

64.8

22.1

(3.1)

(61.9)

(100)

(100)

(13.0)

(67.0)

(18.0)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

173

The hypothesis linking city size to greater consumption of processed food is consistent
with the results in Table 3-9, with average processed shares more than ten percent greater in
larger cities. Over eighty-five percent of food purchased from supermarkets is processed, which
is greater than the average processed share in any city, so it could reasonably be inferred that
purchasing food from supermarkets will increase processed share. Households that purchase a
portion of their food from supermarkets have average processed shares greater than the entire
sample of households; this is consistent with theory based on supermarkets increasing the
desirability of processed food by pricing it at lower prices than non-supermarket retailers.
However, given that households from small cities have larger supermarket shares and households
from larger cities have larger processed shares, regression analysis is required to determine if
supermarkets increase the share of consumed processed food.
Table 3-9 also shows that households that purchase a portion of their food from
supermarkets continue to purchase approximately eighty percent of their processed food, by
value, from retailers other than supermarkets, highlighting that processed food remains primarily
sourced from non-supermarkets retailers even to households that shop at supermarkets. Nonsupermarket retailers are the dominant source of food in urban Zambia (over eighty percent),
with supermarkets providing a smaller yet nontrivial source of all food – primarily driven by
high processed food and low processed maize (Table 3-10).

174

Table 3-10: Household weighted averages of sources of food aggregates
Supermarkets

5.1

All
Processed
6.5

Non-supermarkets

81.4

81.3

76.0

91.3

88.4

92.5

Food Away From Home

7.8

10.4

16.0

0.0

0.0

0.0

Gifts

2.1

1.7

1.0

2.2

3.1

1.6

All Food

Processed
High
7.0

Processed
Low
6.5

Maize - Low
Processed
8.4

Non-maize
Low Processed
5.9

Own Production

3.6

0.0

0.0

0.0

0.0

0.0

Total

100

100

100

100

100

100

Source: Zambia Urban Consumption Survey 2007/2008

Differences exist in the consumption patterns of low and high processed food across the
cities within this sample (Table 3-11). Lusaka, the largest city in Zambia, is the only city where
the average value of consumed high processed is greater than low processed food. The shares of
processed food aggregates from supermarkets also vary such that the supermarkets in larger
cities source considerable more high processed food than low processed food, relative to
supermarkets in tertiary cities. The ratio of consumed value of high processed food to low
processed food is greater for households that purchase some food from supermarkets relative to
the total population, in all cities other than Mansa, suggesting that supermarkets have stronger
positive effects on the consumption of high processed food than low processed food. The affect
that supermarket use has on processed food consumption will be further addressed within the
regression results.

175

Table 3-11: Household weighted averages of consumption shares by low and high processing levels
Full Sample Population
Share of
low
processed
food in
total food
Weighted Average
(all cities)
Lusaka
Kitwe
Kasama & Mansa
Average
Kasama
Mansa

Share of
high
processed
food in
total food

Share of
food from
supermarkets
that is low
processed

Share of
food from
supermarkets
that is high
processed

Subset of Households that
Bought Some Food from a Supermarket
Share of low
Share of high
Share of low Share of high
processed food processed food
processed
processed
from
from
food in total
food in total
supermarkets
supermarkets
food
food
in total low
in total high
processed food processed food
33.7
35.7
19.3
20.8

33.5

31.8

36.6

50.3

(33.6)

(29.3)

(28.3)

(47.6)

(34.1)

(33.2)

(9.4)

33.4

33.6

38.3

48.9

33.0

37.4

22.1

22.7

(33.4)

(31.7)

(29.9)

(46.2)

(32.8)

(34.1)

(11.2)

(12.5)

15.9

(11.5)

35.2

29.6

28.4

57.4

35.5

34.7

11.0

(35.9)

(27.4)

(6.1)

(63.7)

(36.0)

(32.2)

(1.3)

(9.5)

29.3

23.1

43.6

42.8

33.6

28.3

20.9

20.8

(29.3)

(19.1)

(44.4)

(37.7)

(33.8)

(25.6)

(16.7)

(12.6)

28.7

21.4

41.1

44.5

32.8

27.3

19.2

20.4

(28.7)

(17.8)

(41.2)

(41.4)

(32.9)

(24.3)

(15.5)

(11.2)

30.5

26.7

47.3

40.3

34.9

29.9

23.5

21.3

(31.3)

(23.1)

(48.4)

(34.5)

(35.6)

(26.5)

(17.6)

(14.0)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the population weighted average. Parentheses indicate the population weighted median.

176

3.4.4 Regression Results
Direct effects of determinant variables on supermarket share: Table 3-12 presents the
regression results that address how city size affects supermarket share. Five observations are
noted here. First, city size, independent of household expenditure and other drivers of
consumption, has a negative direct effect on supermarket share, as evidenced by the negative
average partial effects of Lusaka and Kitwe dummy variables on supermarket share in the pooled
regression. This finding is consistent with descriptive statistics that identify tertiary cities as
having greater shares of consumption from supermarkets. The conceptual framework highlighted
this potential result, indicating that increased city size could contribute to reduced use
supermarkets due to household response to congestion and the asymmetric effects of investment
costs.

177

Table 3-12: Fractional probit estimates of the average partial effects of the determinants of
household share of supermarket purchases in total food consumption
Variable
Total Daily Household Expenditure
per Adult Equivalent

Pooled
0.004**
(0.000)

Lusaka dummy variable

-0.075**

Kitwe dummy variable

-0.049**

Kasama dummy variable

-0.017**

Share Supermarket in Total Food
Lusaka
Kitwe
Kasama
0.005**
0.003**
0.012**

Mansa
0.011**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)
(0.000)
(0.000)

Percentage Working Age Females
Employed in Nonfarm Labor

-0.007

-0.008

-0.005

-0.004

-0.014

(0.201)

(0.290)

(0.308)

(0.673)

(0.314)

Dependency Ratio

0.020

0.030

-0.007

0.013

0.008

(0.108)

(0.077)

(0.499)

(0.440)

(0.752)

Adult Equivalents
Relative Distance to Supermarket
in Lusaka (km) (log)
Relative Distance to Supermarket
in Kitwe (km) (log)
Relative Distance to Supermarket
in Kasama (km) (log)
Relative Distance to Supermarket
in Mansa (km) (log)

0.002

0.004

-0.001

-0.003

0.005

(0.262)

(0.164)

(0.446)

(0.266)

(0.057)

-0.008**

-0.011**

(0.000)

(0.000)

-0.005**

-0.005**

(0.000)

(0.000)

-0.006**

-0.002

(0.005)

(0.347)

-0.008*

-0.007**

(0.024)

(0.007)

Own a Car

0.019*
(0.016)

(0.030)

(0.670)

(0.033)

(0.486)

Own a Motorcycle

-0.017

-0.025

0.005

0.015

-0.021

(0.154)

(0.100)

(0.730)

(0.409)

(0.428)

Use Public Transit

-0.005

-0.007

0.001

0.001

-0.010

(0.347)

(0.333)

(0.723)

(0.855)

(0.230)

Own a Bicycle

-0.002

-0.003

-0.001

-0.009

0.010

(0.678)

(0.692)

(0.792)

(0.236)

(0.400)

Own Farmland

-0.002

-0.001**

0.001

-0.004

-0.026*

(0.680)

(0.882)

(0.845)

(0.563)

(0.011)

Own a Stove (Gas or Electric)

0.015*

0.008**

0.025**

0.018

0.044**

(0.040)

(0.454)

(0.000)

(0.091)

(0.000)

Own a Hotplate

0.007

0.004

0.019**

0.003

0.029*

(0.290)

(0.652)

(0.006)

(0.742)

(0.020)

0.009

0.010

0.001

0.009

0.040**
(0.004)

Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household
Lean Season dummy variable

0.023*

0.002

0.041*

-0.009

(0.194)

(0.343)

(0.814)

(0.392)

0.007**

0.007**

0.005**

0.009**

0.003

(0.000)

(0.000)

(0.000)

(0.000)

(0.154)

-0.001*

0.000

0.000

0.000

0.000

(0.228)

(0.265)

(0.856)

(0.284)

(0.024)

0.014

0.015

0.016

-0.006

0(.001

(0.074)

(0.148)

(0.055)

(0.583)

(0.961)

-0.007*

-0.007

-0.008**

-0.002

-0.009

(0.024)

(0.123)

(0.007)

(0.737)

(0.189)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

Second, income has a positive average partial effect on supermarket use at all expenditure
levels in all cities with magnitudes twice as large in tertiary cities than in larger cities. The direct
effects of total expenditure are shown across the income spectrum and by city in the left panel of
Figure 3-3, which are calculated from the same estimation used to generate Table 3-12. The

178

difference in average partial effects of expenditure along with the direct effect of city size on
supermarket share provide justification for households within tertiary cities having larger
supermarket shares than households within primary and secondary cities in spite of higher53 total
expenditure in primary and secondary cities.

Figure 3-3: Estimated effects of expenditure on supermarket use and processed food
consumption

Third, relative distance to supermarkets has a larger – nearly two times larger – negative
effect on supermarket share in the primary city compared to all other cities. This identifies
greater sensitivity to the distance that is required to travel to supermarkets in larger cities,
consistent with the hypothesis that larger cities have greater congestion that increases
transportation costs. Additionally, due to a greater presence of non-supermarket retailers, the
sensitivity to distance could result from nearby non-supermarket retailers that would reduce
household willingness to expend the additional transportation costs to shop at a supermarket.

The average expenditures in the primary city are two times those of the tertiary cities, and the average expenditures in
secondary cities are more than fifty percent greater than those in tertiary cities.
53

179

Fourth, education has a strong positive effect on supermarket share with no evidence that
the effect varies systematically with city size. The direction of the average partial effect is
consistent with the expectation that more educated households will place a premium on the
health benefits associated with the cleanliness of supermarkets. The consistency of the effect of
education across city size shows the transferability of education across demographics.
Fifth, although expected to affect supermarket share, the percentage of working age
females employed in nonfarm labor does not have a statistically significant effect. This variable
is a proxy for opportunity cost of time, which is also controlled for by total household
expenditure. The effect of the additional income earned by female nonfarm employment would
be captured by the expenditure variables, but this finding shows that the time that females
allocate towards nonfarm employment does not statistically affect supermarket share.
Direct effects of determinant variables on processed share: Table 3-13 addresses city size
effects on processed food consumption with seven observations out of note. First, city size
positively affects the consumption of processed food when controlling for other determinant
variables, as indicated by the positive and significant average partial effects of primary and
secondary city dummy variables. This is consistent with the primary hypothesis regarding
processed share, which following the conceptual framework highlighted only positive expected
effects of city size on urban environment and ultimately processed food consumption.

180

Table 3-13: Second stage of control function regression; estimating the share of processed food
in total food consumption
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
0.010**

Lusaka
0.009**

(0.000)

(0.000)

0.007

0.021

(0.945)

(0.855)

(0.000)

-0.091

0.337

(0.425)

(0.150)

(0.983)

0.168

2.867**

(0.196)

(0.006)

Mansa
0.019
(0.770)

0.411**

0.843*

(0.001)

(0.021)

Lusaka dummy variable

0.035**

Kitwe dummy variable

0.037**

Kasama dummy variable

-0.049**

Percentage Working Age Females
Employed in Nonfarm Labor

Share of Processed Food
Kitwe
Kasama
0.011**
0.003

(0.010)
(0.003)
(0.000)

0.006

-0.002

0.021

0.018

(0.375)

(0.849)

(0.051)

(0.266)

(0.013)

Dependency Ratio

-0.006

-0.005

0.015

-0.048

-0.044

(0.683)

(0.801)

(0.454)

(0.212)

(0.242)

Adult Equivalents

0.006**

0.006*

0.005*

0.015*

0.007

(0.001)

(0.019)

(0.023)

(0.013)

(0.074)

Own a Car

-0.016

-0.015

-0.004

-0.210**

-0.020

(0.116)

(0.257)

(0.786)

(0.004)

(0.327)

Own a Motorcycle

0.002

0.030

-0.067*

-0.100

-0.033

(0.930)

(0.416)

(0.019)

(0.055)

(0.266)

0.007

0.013

0.003

-0.013

-0.006

(0.317)

(0.158)

(0.758)

(0.357)

(0.700)

Own a Bicycle

-0.004

-0.009

0.013

0.015

-0.046**

(0.522)

(0.372)

(0.235)

(0.344)

(0.002)

Own Farmland

-0.037**

-0.042**

-0.035**

-0.023

-0.033

Use Public Transit

0.046*

(0.000)

(0.004)

(0.002)

(0.142)

(0.052)

Own a Stove (Gas or Electric)

-0.005

-0.014

0.018

-0.117*

-0.028

(0.591)

(0.257)

(0.141)

(0.016)

(0.386)

Own a Hotplate

-0.014

-0.018

-0.003

-0.051

0.013

(0.073)

(0.055)

(0.838)

(0.053)

(0.585)

Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head

0.014

0.012

0.017

0.038

0.010

(0.108)

(0.313)

(0.199)

(0.054)

(0.651)

-0.005

0.001

0.002

-0.002

-0.020*

(0.692)

(0.467)

(0.461)

(0.022)

(0.127)

-0.002**

-0.002**

-0.001

-0.002**

-0.002*

(0.000)

(0.000)

(0.071)

(0.000)

(0.014)

Female Head of Household

-0.012

-0.007

-0.022

-0.009

-0.032

(0.169)

(0.506)

(0.080)

(0.716)

(0.092)

Lean Season dummy variable

-0.006

-0.006

-0.011

0.000

0.036**

(0.253)

(0.445)

(0.136)

(0.989)

(0.006)

Endogenous Error from 1st Stage
Regression

-0.183

-0.132

-0.410

-2.639*

-0.712*

(0.092)

(0.272)

(0.085)

(0.011)

(0.046)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

Second, shopping in supermarkets significantly increases processed share within tertiary
cities, but not within primary or secondary cities. Relative to tertiary cities, this finding is
consistent with the Rischke et al. (2015) study that found a positive and significant effect of

181

supermarket share on processed food consumption with their data sample of three tertiary cities
within Kenya. The positive, though of low statistical significance, average partial effects of
supermarket share on processed share in the primary and secondary cities limit the ability to
expand this relationship beyond smaller cities (Table 3-13).
Third, total household expenditure has a positive effect on processed share in primary and
secondary cities, but not within tertiary cities. The right panel of Figure 3-3 shows the average
partial effects of income on processed food by city, regardless of the statistical significance, and
this shows positive effects at all levels of expenditure for all cities other than Kasama. The
overall pattern of expenditure on processed food consumption is as hypothesized.
The descriptive statistics show that processed food is available to be acquired from
supermarkets and non-supermarkets alike in each city in the survey. However, these results
show that the consumption of processed food in tertiary cities is access driven by the presence
and utilization of supermarket retailers, whereas the consumption of processed food in primary
and secondary cities is income driven – responding to changes in total expenditure.
Fourth, ownership of farmland negatively affects the consumption of processed food,
with little evidence of systematic variance in the effect across city size. Although the average
partial effects do not systematically vary across city size, the impact of these effects are more
pertinent to households within tertiary cities as more than forty percent of tertiary households
own some farmland, relative to nine percent of households within primary cities.
Fifth, age of household head negatively affects the consumption of processed food,
without evidence of systematic variance in the affect across city size. This effect is expected as
elderly households are less likely to adopt the less traditional consumption patterns of consuming
processed food.

182

Sixth, education is found to have a statistically significant negative effect in only one of
the cities. This result is minimally consistent with the expectation of increased education leading
to healthier consumption decisions given that processed foods are generally considered less
healthy than unprocessed foods.
Seventh, the percentage of working age females employed in nonfarm employment has
positive and significant effects on processed share for households in Kitwe and Mansa, in accord
with the findings of Senauer et al. (1986) and Kennedy & Reardon (1994). There is inadequate
evidence to identify a pattern regarding city size specific effects of participation in nonfarm
employment.
The consumption patterns of processed food are further analyzed by separately
considering the drivers of the consumption for low and high processed food. Five observations
are highlighted from Tables 3-A-3 and 3-A-4 that are located within the appendix.
First, similar to the pattern observed regarding the processed share of households within
tertiary cities, supermarket share positively effects the consumption of low processed food by
households within tertiary cities. The positive average partial effects of supermarket share on
processed food and on low processed food shows that the increased processed food consumption
in tertiary cities due to supermarkets is primarily driven by increased consumption of low
processed food such as maize meal.
Second, the consumption of highly processed food is not directly driven by the use of
supermarket as supermarket share was not statistically significant for high processed food. This
finding was not expected following the patterns observed in the descriptive statistics, which
showed households that use supermarkets have greater consumption of high processed food
relative to low processed food than those who do not use supermarkets (Table 3-11).

183

Third, households within larger cities transition from low processed food to high
processed food as household total expenditures rise. This is shown in Tables 3-A-3 and 3-A-4 by
the negative average partial effects of total expenditure on low processed food and positive
average partial effects of total expenditure on high processed food. The larger absolute value of
the average partial effects on high processed food shows that demand for highly processed food
is driving the rise in consumption of processed food within primary and secondary cities.
Fourth, the average partial effect of expenditure is not significant on low or high
processed food for households within tertiary cities, further reinforcing the idea that consumption
of processed food within tertiary cities is access driven more than income driven.
Fifth, seasonal variability exists regarding the level of processed food consumed by
households. During the lean season there is an increase in the consumption of low processed
food and relative decrease in the consumption of high processed food. This is likely driven by
increased consumption of purchased maize meal during the lean season, as maize grain becomes
scarce on the market at this time and households therefore must acquire it as maize meal.
Indirect effects of determinant variables on supermarket share and processed share: The
effects of city size on supermarket and processed share also include indirect effects that are
driven by the effects that city size has on other determinant variables. Table 3-14 highlights the
indirect effects of city size on supermarket share and processed share via the mediator variables
of relative distance to supermarkets, total daily expenditure per adult equivalent and percentage
of working age females employed in nonfarm labor. The indirect effects of city size are not
statistically significant via the percentage of working age females employed in nonfarm labor,
while the indirect effects via relative distance to supermarkets and total daily expenditure per
adult equivalent are statistically significant and are noted below.

184

Table 3-14: City indirect effects via specific explanatory variables
Supermarket share
Total Daily
Household
Expenditure per
Adult Equivalent
0.012**

Lusaka

(0.025)

(0.000)

(0.808)

(0.000)

Kitwe

-0.011**

0.009**

0.000

0.019**

0.000

(0.001)

(0.001)

(0.912)

(0.000)

(0.922)

-0.005

0.000

0.000

0.001

0.000

(0.064)

(0.886)

(0.940)

(0.885)

(0.955)

Kasama

Percentage
Working Age
Females Employed
in Nonfarm Labor
0.000

Processed Share
Total Daily
Percentage
Household
Working Age
Expenditure per
Females Employed
Adult Equivalent
in Nonfarm Labor
0.027**
0.000

Distance to
Supermarket minus
distance to nearest
retailer (log)
-0.004*

(0.851)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

The indirect effect of city size on supermarket share via relative distance to supermarket
is in agreement with the direct effect: negative and statistically significant. The conceptual model
posits that as city size increases the market size will also increase, incentivizing investment in
supermarkets and non-supermarkets alike. The increased investment in non-supermarkets
increases the likelihood that a non-supermarket would be located near the household, therefore
increasing the relative distance to a supermarket. This combined with the previously noted
negative impact of relative distance to supermarkets drive a negative indirect effect of city size
on supermarket share via relative distance to supermarket; strengthening the total negative effect
of city size on supermarket share.
𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ 𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↑→ % 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↓
𝐷𝑖𝑟𝑒𝑐𝑡 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ % 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↓
𝑇𝑜𝑡𝑎𝑙 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ % 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↓↓
The indirect effect of city size on supermarket share via total income is positive and
significant for primary and secondary cities when calculated at the average city level of total
income. Increasing city size provides increased earning opportunities for households, positively
increasing consumer demand for normal and luxury goods. This indirect effect is in agreement

185

with the common expectation that increased city size would positively affect supermarket share,
a luxury in developing countries. The direct effects captured by the primary and secondary city
dummy variables are negative and of greater absolute value than the indirect effects, thereby
dominating the indirect effects and resulting in negative total effects of primary and secondary
cities on supermarket share.
𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ ℎℎ 𝑖𝑛𝑐𝑜𝑚𝑒 ↑→ % 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↑
𝐷𝑖𝑟𝑒𝑐𝑡 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ % 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↓↓
𝑇𝑜𝑡𝑎𝑙 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ % 𝑠𝑢𝑝𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑡 ↓
The indirect effect of city size on processed share via total income has a positive effect
when calculated at the average city level of total income in primary and secondary cities. These
positive indirect effects strengthen the effects that city size has on processed share beyond that
observed by the direct effects.
𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ ℎℎ 𝑖𝑛𝑐𝑜𝑚𝑒 ↑→ % 𝑝𝑟𝑜𝑐𝑒𝑠𝑠𝑒𝑑 ↑
𝐷𝑖𝑟𝑒𝑐𝑡 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ % 𝑝𝑟𝑜𝑐𝑒𝑠𝑠𝑒𝑑 ↑
𝑇𝑜𝑡𝑎𝑙 𝐸𝑓𝑓𝑒𝑐𝑡: 𝐶𝑖𝑡𝑦 𝑆𝑖𝑧𝑒 ↑→ % 𝑝𝑟𝑜𝑐𝑒𝑠𝑠𝑒𝑑 ↑↑

3.5 Conclusions
Analyzing the urban household data from Zambia, one observes that household
consumption patterns pertaining to the share of food purchased from supermarkets in total food
consumption and the share of processed food in total food consumption do vary across city size
and that city size is a significant determinant of these consumption patterns.
Conventional wisdom indicates that the share of food from supermarkets in total food
consumption would be greater for households in larger cities than for households in smaller

186

cities. This analysis contradicts the conventional wisdom in finding that households within
primary and secondary cities shop at supermarkets less often and have lower average
supermarket shares than households from tertiary cities. The conceptual framework recognizes
that beyond the threshold of city size required to support a supermarket, a greater city size could
in fact reduce the use of supermarkets driven by the urban environment affecting the food
environment such that the absolute or relative access to supermarkets is reduced. Large cities do
have a positive indirect effect on supermarket share through household total expenditure, a
statistically significant mediator variable, but again the negative direct effect of city size
continues to dominate the total effect.
Processed food consumption patterns follow conventional wisdom across city size, where
households from larger cities consume a greater average share of processed food in their diet
than households in smaller cities.
This analysis affirms previous studies that indicate that shopping at supermarkets has a
positive effect on the share of processed food consumption in tertiary cities. This study observes
the same directional effect of supermarket share on processed share in primary and secondary
cities, but these average partial effects lack the statistical significance to definitively confirm the
pattern.
Finally, consistent with the general finding that supermarkets increase processed food
consumption, a subsample of households from all city sizes that purchase at least some of their
food from supermarkets consume greater shares of processed food than the entire data sample.
This same subsample of households continues to purchase approximately eighty percent of their
processed food value from non-supermarket retailers. This later finding is a reminder that

187

although supermarket use is shown to have a positive effect on the consumption of processed
food, the majority of processed food continues to be sourced from non-supermarket retailers.

188

APPENDIX

189

APPENDIX

Current Aggregates:
Monteiro Aggregates:

Unprocessed
Low Processed
High Processed
Unprocessed or Minimally Processed
Processed Culinary or Ingredients
Ultra-Processed
Whole grains
Rice
Sugar
Vegetable oils
Breads
Fresh fruits
Butchered meat
Spices
Vegetable fats
Biscuits
Roots & tubers
Dried fish
Wheat flour
Pasta
Soft drinks
Fresh vegetables
Maize meal
Pasteurized milk*
Cheeses
Fresh fish
Other flours
Prepared food away
from home
Eggs

Figure 3-A-1: Processing level aggregates compared with aggregates from Monteiro et al. 2010
Note: Pasteurized milk is considered to be unprocessed or minimally processed by Monteiro et al. 2010 and high processed in this paper.

Table 3-A-1: Prices of commonly consumed items
Products
Maize meal
Bread
Chicken and other poultry
Beef and other red meat
Cooking oil
Rape
Tomato
Sugar
Rice
Dried fish
Kapenta
Beans
Price Index

Lusaka
1,633
3,801
12,010
15,402
8,214
1,911
1,883
4,340
4,079
20,824
34,220
6,406
100

Prices - Round 1
Kitwe Kasama
1,840
1,728
4,664
4,117
11,233
10,260
16,308
12,560
8,827
7,730
1,730
1,563
1,895
1,598
4,372
4,427
6,088
4,644
27,068
21,745
28,112
18,132
4,967
3,938
108.7
92.0

Mansa
1,812
3,964
9,639
15,983
9,211
1,013
1,458
4,314
4,403
23,547
14,274
3,050
99.1

Lusaka
1,497
4,509
14,041
16,766
9,995
3,236
5,628
4,809
4,421
29,376
37,068
10,846
100

Prices - Round 2
Kitwe Kasama
1,349
2,002
6,117
5,895
14,236 11,792
16,551 12,931
9,556
8,340
2,268
1,271
5,511
2,529
4,138
4,529
4,881
4,053
17,852 16,605
19,630 18,509
7,624
5,464
89.5
76.1

Mansa
2,499
5,232
10,490
16,317
9,648
1,569
3,706
4,945
4,091
23,722
17,916
5,953
81.5

Note: Price indices are weighted by the consumption shares of the top ten items consumed specific to the round in which it was calculated.
Prices are presented in Zambian Kwacha.

190

Lusaka
1,565
4,155
13,025
16,084
9,105
2,574
3,756
4,575
4,250
25,100
35,644
8,626
100

Average Prices
Kitwe Kasama
1,594
1,865
5,391
5,006
12,735 11,026
16,429 12,746
9,192
8,035
1,999
1,417
3,703
2,064
4,255
4,478
5,485
4,349
22,460 19,175
23,871 18,320
6,295
4,701
100.9
85.6

Mansa
2,156
4,598
10,065
16,150
9,430
1,291
2,582
4,629
4,247
23,634
16,095
4,502
94.1

Table 3-A-2: First stage of control function regression; OLS estimates of the determinants of the
share of supermarket purchases in total food consumption
Variable
Total Daily Household Expenditure
per Adult Equivalent

Pooled
0.004**
(0.000)

Lusaka dummy variable

-0.051**

Kitwe dummy variable

-0.052**

Kasama dummy variable

-0.019**

Share Supermarket in Total Food
Lusaka
Kitwe
Kasama
0.004**
0.003**
0.008**
(0.000)

(0.000)

Mansa
0.007**

(0.000)

(0.000)

(0.000)
(0.000)
(0.005)

Percentage Working Age Females
Employed in Nonfarm Labor

-0.004

-0.006

-0.003

0.002

-0.015

(0.422)

(0.406)

(0.516)

(0.800)

(0.232)

Dependency Ratio

0.025

0.038*

-0.006

0.014

0.015

(0.053)

(0.034)

(0.599)

(0.443)

(0.565)

0.001

0.002

-0.001

-0.004

0.003

(0.639)

(0.530)

(0.235)

(0.075)

(0.285)

Relative Distance to Supermarket
in Lusaka (km) (log)
Relative Distance to Supermarket
in Kitwe (km) (log)
Relative Distance to Supermarket
in Kasama (km) (log)
Relative Distance to Supermarket
in Mansa (km) (log)

-0.020**

-0.021**

(0.000)

(0.000)

Own a Car

0.047**

Adult Equivalents

-0.011*

-0.008**

(0.011)

(0.000)

-0.014

-0.004

(0.045)

(0.437)

-0.014**

-0.008*

(0.002)

(0.030)

0.051**

0.019

0.082*

-0.007

(0.001)

(0.007)

(0.158)

(0.020)

(0.786)

Own a Motorcycle

-0.032

-0.059

0.006

0.040

-0.026

(0.321)

(0.209)

(0.759)

(0.436)

(0.642)

Use Public Transit

-0.007

-0.011

0.002

0.002

-0.007

(0.174)

(0.156)

(0.705)

(0.838)

(0.454)

Own a Bicycle

0.002

0.003

-0.002

-0.009

0.006

(0.705)

(0.801)

(0.754)

(0.237)

(0.626)

-0.022*

0.001

0.006

0.002

-0.005

(0.935)

(0.676)

(0.838)

(0.469)

(0.028)

Own a Stove (Gas or Electric)

0.018*

0.013

0.020**

0.042*

0.079**

(0.018)

(0.287)

(0.000)

(0.019)

(0.000)

Own a Hotplate

0.003

-0.001

0.007

0.015

0.047**

Own Farmland

Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household
Lean Season dummy variable

(0.690)

(0.886)

(0.229)

(0.354)

(0.006)

-0.002

-0.002

-0.004

-0.001

0.027*
(0.031)

(0.637)

(0.769)

(0.198)

(0.894)

0.006**

0.006**

0.004**

0.009**

0.002

(0.000)

(0.000)

(0.000)

(0.000)

(0.241)

-0.001*

0.000

0.000

0.000

0.000

(0.388)

(0.367)

(0.909)

(0.828)

(0.018)

0.013

0.012

0.015

-0.008

-0.001

(0.104)

(0.260)

(0.076)

(0.533)

(0.939)

-0.007*

-0.007

-0.008**

-0.001

-0.009

(0.033)

(0.149)

(0.010)

(0.893)

(0.229)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

191

Table 3-A-3: Second stage of control function regression; estimating the share of low processed
food in total food consumption
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
-0.004**
(0.000)

(0.000)

-0.091

-0.062

(0.425)

(0.609)

0.127

(0.124)

(0.634)

2.816**

(0.656)

(0.002)

Mansa
-0.005
(0.916)

0.282**

0.834**

(0.010)

(0.010)

Kitwe dummy variable

0.057**

Dependency Ratio

(0.848)

-0.055

Lusaka dummy variable

Percentage Working Age Females
Employed in Nonfarm Labor

(0.212)

-0.195

0.042**

Kasama dummy variable

Share of Low Processed Food
Lusaka
Kitwe
Kasama
-0.005**
-0.002
-0.020

(0.001)
(0.000)

-0.008
(0.426)

0.000

-0.002

-0.003

0.022

(0.999)

(0.828)

(0.798)

(0.098)

0.042**
(0.006)

0.016

0.024

0.009

-0.048

-0.049
(0.090)

(0.264)

(0.222)

(0.662)

(0.166)

-0.002

-0.004*

-0.001

0.013**

0.005

(0.090)

(0.046)

(0.506)

(0.008)

(0.163)

0.038**

0.045**

0.024

-0.167**

0.037*

(0.000)

(0.001)

(0.136)

(0.000)

(0.031)

Own a Motorcycle

-0.018

-0.017

-0.019

-0.071

-0.050

(0.431)

(0.521)

(0.684)

(0.243)

(0.252)

Use Public Transit

0.010

0.020*

0.003

-0.032**

0.017

(0.142)

(0.028)

(0.734)

(0.003)

(0.132)

Adult Equivalents
Own a Car

0.013

0.015

0.013

0.030

0.001

(0.070)

(0.174)

(0.183)

(0.053)

(0.909)

Own Farmland

-0.025**

-0.012

-0.046**

0.002

-0.026*

(0.000)

(0.305)

(0.000)

(0.878)

(0.048)

Own a Stove (Gas or Electric)

0.023**

0.017

0.039**

-0.086*

-0.043

(0.011)

(0.168)

(0.001)

(0.019)

(0.100)

0.001

0.000

0.010

-0.038*

-0.022

(0.884)

(0.974)

(0.441)

(0.037)

(0.238)

0.009

0.009

-0.001

0.030

0.026

(0.270)

(0.458)

(0.941)

(0.078)

(0.127)

Own a Bicycle

Own a Hotplate
Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household
Lean Season dummy variable
Endogenous Error from 1st Stage
Regression

0.002

0.001

0.004

-0.020*

0.000

(0.240)

(0.496)

(0.057)

(0.016)

(0.926)

0.000

0.000

0.001**

-0.001

0.000

(0.179)

(0.449)

(0.000)

(0.057)

(0.554)

-0.002

-0.001

-0.009

0.008

0.006

(0.843)

(0.961)

(0.504)

(0.698)

(0.654)

0.037**

0.037**

0.033**

0.037**

0.071**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

0.102

0.161

-0.053

-2.639**

-0.545

(0.376)

(0.203)

(0.847)

(0.003)

(0.089)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

192

Table 3-A-4: Second stage of control function regression; estimating the share of high processed
food in total food consumption
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
0.014**
(0.000)

(0.000)

0.098

0.082

(0.442)

(0.565)

0.205

(0.486)

(0.484)

(0.454)

0.260

-0.148

(0.047)

(0.832)

Mansa
0.027
(0.249)

0.090

-0.145

(0.518)

(0.768)

Lusaka dummy variable
Kitwe dummy variable

-0.014

Percentage Working Age Females
Employed in Nonfarm Labor

(0.000)

0.096

-0.005

Kasama dummy variable

Share of High Processed Food
Lusaka
Kitwe
Kasama
0.013**
0.014**
0.025

(0.726)
(0.342)

-0.050**
(0.000)

0.008

0.002

0.024*

-0.002

0.003

(0.310)

(0.867)

(0.024)

(0.845)

(0.882)

Dependency Ratio

-0.022

-0.029

0.003

0.002

0.000

(0.167)

(0.193)

(0.884)

(0.959)

(0.996)

Adult Equivalents

0.008**

0.010**

0.007**

0.002

0.003

(0.000)

(0.000)

(0.005)

(0.621)

(0.446)

Own a Car

-0.046**

-0.052**

-0.023

-0.007

-0.050**

(0.000)

(0.000)

(0.173)

(0.898)

(0.005)

0.020

0.046

-0.039

-0.016

0.017

(0.536)

(0.363)

(0.262)

(0.800)

(0.701)

Use Public Transit

-0.003

-0.006

0.000

0.018

-0.022

(0.695)

(0.552)

(0.970)

(0.091)

(0.143)

Own a Bicycle

-0.018*

-0.025*

0.000

-0.016

-0.046**

(0.017)

(0.018)

(0.968)

(0.215)

(0.005)

Own Farmland

-0.011

-0.026

0.011

-0.028

-0.011
(0.549)

Own a Motorcycle

Own a Stove (Gas or Electric)
Own a Hotplate
Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household

(0.190)

(0.093)

(0.371)

(0.019)

-0.028**

-0.031*

-0.022

-0.017

0.027

(0.002)

(0.017)

(0.123)

(0.609)

(0.542)

-0.016

-0.019

-0.013

-0.010

0.036

(0.052)

(0.070)

(0.321)

(0.605)

(0.234)

-0.015

0.006

0.005

0.019

0.010

(0.486)

(0.710)

(0.143)

(0.486)

(0.510)

-0.001

0.001

-0.005*

0.001

-0.004

(0.598)

(0.744)

(0.043)

(0.846)

(0.194)

-0.002**

-0.002**

-0.002**

-0.001**

-0.002**

(0.000)

(0.000)

(0.000)

(0.004)

(0.003)

-0.009

-0.006

-0.012

-0.020

-0.039*

(0.310)

(0.611)

(0.386)

(0.138)

(0.031)

Lean Season dummy variable

-0.042**

-0.041**

-0.043**

-0.037**

-0.036**

(0.000)

(0.000)

(0.000)

(0.000)

(0.005)

Endogenous Error from 1st Stage
Regression

-0.262*

-0.271

-0.326

0.174

-0.031

(0.047)

(0.072)

(0.272)

(0.805)

(0.950)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

193

Table 3-A-5: Estimates of the marginal effects of the determinants of processed food in total
food consumption (OLS)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa
Lusaka dummy variable
Kitwe dummy variable
Kasama dummy variable
Percentage Working Age Females
Employed in Nonfarm Labor

Pooled
0.011**

Lusaka
0.009**

(0.000)

(0.000)

-0.129**

-0.090**

(0.000)

(0.000)

Share of Processed Food
Kitwe
Kasama
0.013**
0.024**
(0.000)

-0.214**

-0.035

(0.000)

(0.316)

(0.000)

0.058

0.151*

(0.314)

(0.011)

Mansa
0.025**
(0.000)

0.132**

0.059

(0.000)

(0.272)

0.008
(0.273)

0.011
(0.116)

-0.057**
(0.000)

0.009

-0.006

0.007

0.011

(0.072)

(0.458)

(0.385)

(0.399)

0.030**
(0.049)

Dependency Ratio

-0.010

-0.005

0.017

-0.024

-0.044

(0.303)

(0.756)

(0.286)

(0.386)

(0.138)

Adult Equivalents

0.008**

0.006**

0.006**

0.006

0.013**

(0.000)

(0.001)

(0.001)

(0.070)

(0.000)

Own a Car

-0.009

-0.001

-0.008

-0.016

-0.024

(0.238)

(0.958)

(0.523)

(0.582)

(0.326)

Own a Motorcycle

-0.018

0.019

-0.044

0.029

-0.053

(0.445)

(0.672)

(0.224)

(0.662)

(0.360)

Use Public Transit

0.000

0.018*

0.001

-0.003

-0.021

(0.927)

(0.014)

(0.939)

(0.792)

(0.100)

Own a Bicycle

-0.014**

-0.014

0.000

-0.008

-0.045**

(0.007)

(0.133)

(0.980)

(0.505)

(0.001)

Own Farmland

-0.039**

-0.035**

-0.029**

-0.048**

-0.046**
(0.000)

(0.000)

(0.004)

(0.001)

(0.000)

0.004

-0.010

0.022*

0.008

0.020

(0.530)

(0.300)

(0.013)

(0.641)

(0.264)

Own a Hotplate

-0.002

-0.009

-0.007

-0.007

0.038*

(0.664)

(0.262)

(0.435)

(0.690)

(0.024)

Own a Mobile Phone

0.013*

0.002

0.011

0.029

0.025

(0.026)

(0.858)

(0.215)

(0.051)

(0.120)

-0.001

Own a Stove (Gas or Electric)

Highest Completed Grade of
Education in Household

0.001

0.002

-0.001

0.003

(0.364)

(0.083)

(0.450)

(0.255)

(0.843)

Age of Household Head

-0.001**

-0.001**

-0.001*

-0.002**

-0.002**

(0.000)

(0.000)

(0.047)

(0.000)

(0.000)

Female Head of Household

-0.018**

-0.012

-0.018*

-0.035*

-0.021

Lean Season dummy variable

(0.000)

(0.160)

(0.040)

(0.015)

(0.157)

-0.006

-0.014*

-0.012

-0.003

0.024*

(0.115)

(0.027)

(0.057)

(0.803)

(0.042)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

194

Table 3-A-6: Estimates of the marginal effects of the determinants of low processed food in total
food consumption (OLS)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
-0.004**
(0.000)

(0.000)

0.014

0.090**

(0.612)

(0.000)

0.096**

(0.102)

(0.006)

(0.667)

0.024

0.170**

(0.648)

(0.001)

Mansa
-0.002
(0.568)

0.319**

0.298**

(0.000)

(0.000)

Lusaka dummy variable
Kitwe dummy variable

0.056**

Percentage Working Age Females
Employed in Nonfarm Labor

(0.117)

-0.072

0.038**

Kasama dummy variable

Share of Low Processed Food
Lusaka
Kitwe
Kasama
-0.005**
-0.002
0.001

(0.000)
(0.000)

-0.004
(0.579)

0.006

0.000

-0.002

0.015

(0.177)

(0.953)

(0.770)

(0.193)

(0.017)

Dependency Ratio

-0.007

0.006

0.009

-0.024

-0.057*

(0.436)

(0.673)

(0.597)

(0.310)

(0.016)

Adult Equivalents

0.001

-0.004*

0.000

0.004

0.007*

(0.217)

(0.024)

(0.874)

(0.186)

(0.014)

0.024**

0.030**

0.019

0.012

0.046*

Own a Car

0.029*

(0.001)

(0.006)

(0.108)

(0.625)

(0.018)

Own a Motorcycle

-0.003

-0.001

0.019

0.038

-0.064

(0.886)

(0.985)

(0.608)

(0.511)

(0.162)

Use Public Transit

-0.001

0.009

0.002

-0.023*

0.008

(0.858)

(0.246)

(0.776)

(0.024)

(0.427)

0.006

0.009

0.003

-0.004

0.011

(0.184)

(0.335)

(0.742)

(0.661)

(0.285)

Own Farmland

-0.028**

-0.007

-0.043**

-0.017

-0.033**

(0.000)

(0.571)

(0.000)

(0.072)

(0.001)

Own a Stove (Gas or Electric)

0.019**

0.023*

0.037**

0.023

0.000

(0.000)

(0.017)

(0.000)

(0.118)

(0.972)

Own a Bicycle

Own a Hotplate

0.003

0.005

0.011

0.007

0.010

(0.519)

(0.496)

(0.235)

(0.677)

(0.465)

Own a Mobile Phone

0.012*

0.004

-0.001

0.022

0.034**

(0.022)

(0.672)

(0.914)

(0.077)

(0.008)

Highest Completed Grade of
Education in Household

0.001

-0.001

0.002

0.002

0.001

(0.482)

(0.511)

(0.275)

(0.331)

(0.796)

0.000

0.001

0.001**

-0.001*

0.000

(0.136)

(0.064)

(0.000)

(0.050)

(0.533)

Age of Household Head
Female Head of Household
Lean Season dummy variable

0.001

0.012

-0.012

-0.028*

0.021

(0.884)

(0.131)

(0.162)

(0.021)

(0.070)

0.042**

0.037**

0.040**

0.037**

0.066**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

195

Table 3-A-7: Estimates of the marginal effects of the determinants of high processed food in
total food consumption (OLS)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
0.015**
(0.000)

(0.000)

-0.143**

-0.181**

(0.000)

(0.000)

(0.000)

-0.143**

-0.131**

(0.007)

(0.002)

(0.000)

0.034

-0.018

(0.597)

(0.723)

Mansa
0.026**
(0.000)

-0.188**

-0.239**

(0.000)

(0.000)

Lusaka dummy variable

-0.030**

Kitwe dummy variable

-0.045**

Kasama dummy variable

-0.054**

Percentage Working Age Females
Employed in Nonfarm Labor

Share of High Processed Food
Lusaka
Kitwe
Kasama
0.015**
0.015**
0.023**

(0.000)
(0.000)
(0.000)

0.003

-0.006

0.010

-0.004

0.001

(0.621)

(0.578)

(0.329)

(0.755)

(0.936)

Dependency Ratio

-0.003

-0.011

0.009

0.000

0.012

(0.780)

(0.554)

(0.651)

(0.996)

(0.692)

Adult Equivalents

0.007**

0.011**

0.006**

0.002

0.006

(0.000)

(0.000)

(0.002)

(0.435)

(0.081)

Own a Car

-0.033**

-0.031*

-0.027

-0.029

-0.070**

(0.000)

(0.023)

(0.058)

(0.268)

(0.006)

-0.015

0.019

-0.063

-0.008

0.011

(0.568)

(0.719)

(0.146)

(0.886)

(0.855)

-0.029*

Own a Motorcycle

0.000

0.010

-0.002

0.020

(0.946)

(0.286)

(0.860)

(0.056)

(0.028)

Own a Bicycle

-0.020**

-0.024*

-0.003

-0.003

-0.055**

(0.000)

(0.046)

(0.765)

(0.740)

(0.000)

Own Farmland

-0.011

-0.028

0.013

-0.032

-0.013

(0.062)

(0.063)

(0.197)

(0.001)

(0.316)

Own a Stove (Gas or Electric)

-0.016*

-0.033**

-0.015

-0.015

0.019

(0.015)

(0.006)

(0.168)

(0.317)

(0.292)

Own a Hotplate

-0.006

-0.014

-0.019

-0.014

0.028

(0.351)

(0.144)

(0.098)

(0.387)

(0.103)

-0.009

Use Public Transit

Own a Mobile Phone
Highest Completed Grade of
Education in Household

0.001

-0.002

0.012

0.007

(0.922)

(0.844)

(0.257)

(0.614)

(0.611)

0.000

0.003

-0.003

0.001

-0.001

(0.816)

(0.053)

(0.120)

(0.727)

(0.697)

Age of Household Head

-0.002**

-0.002**

-0.002**

-0.002**

-0.002**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

Female Head of Household

-0.019**

-0.024

-0.006

-0.006

-0.042**

(0.001)

(0.018)

(0.582)

(0.611)

(0.006)

Lean Season dummy variable

-0.049**

-0.051**

-0.052**

-0.039**

-0.043**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

196

Table 3-A-8: Second stage of two stage least squares; estimating the share of processed food in
total food consumption (2SLS)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
0.010**

Lusaka
0.009**

(0.000)

(0.000)

-0.032

0.022

(0.791)

(0.853)

(0.000)

-0.179

0.343

(0.312)

(0.139)

(0.981)

0.275

2.833**

(0.188)

(0.004)

Mansa
0.020
(0.762)

0.575**

0.825*

(0.001)

(0.036)

Lusaka dummy variable

0.042**

Kitwe dummy variable

0.047**

Kasama dummy variable

-0.047**

Percentage Working Age Females
Employed in Nonfarm Labor

Share of Processed Food
Kitwe
Kasama
0.012**
0.003

(0.010)
(0.003)
(0.001)

0.007

-0.001

0.022*

0.017

(0.301)

(0.898)

(0.046)

(0.289)

(0.016)

Dependency Ratio

-0.008

-0.005

0.014

-0.048

-0.041

(0.633)

(0.813)

(0.486)

(0.221)

(0.279)

Adult Equivalents

0.006**

0.006*

0.005*

0.015*

0.007

(0.003)

(0.025)

(0.025)

(0.014)

(0.072)

Own a Car

-0.015

-0.013

-0.005

-0.222**

-0.018

(0.119)

(0.296)

(0.708)

(0.008)

(0.352)

Own a Motorcycle

0.007

0.031

-0.064*

-0.101

-0.032

(0.766)

(0.393)

(0.021)

(0.062)

(0.300)

0.008

0.014

0.003

-0.013

-0.006

(0.274)

(0.164)

(0.765)

(0.396)

(0.698)

Own a Bicycle

-0.005

-0.010

0.013

0.014

-0.046**

(0.490)

(0.366)

(0.232)

(0.380)

(0.002)

Own Farmland

-0.037**

-0.040**

-0.035**

-0.023

-0.034

(0.000)

(0.006)

(0.003)

(0.139)

(0.052)

-0.004

-0.014

0.018

-0.118*

-0.027
(0.444)

Use Public Transit

Own a Stove (Gas or Electric)
Own a Hotplate
Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head

0.047*

(0.631)

(0.284)

(0.150)

(0.023)

-0.013**

-0.018

-0.002

-0.051

0.012

(0.090)

(0.060)

(0.859)

(0.064)

(0.627)

0.015

0.012

0.018

0.039

0.013

(0.096)

(0.310)

(0.182)

(0.052)

(0.590)

-0.005

0.001

0.002

-0.002

-0.020*

(0.709)

(0.455)

(0.450)

(0.018)

(0.140)

-0.002**

-0.002**

-0.001

-0.002**

-0.002*

(0.000)

(0.000)

(0.075)

(0.000)

(0.017)

Female Head of Household

-0.012

-0.007

-0.022

-0.010

-0.031

(0.161)

(0.543)

(0.083)

(0.690)

(0.111)

Lean Season dummy variable

-0.006

-0.005

-0.011

0.000

0.036**

(0.304)

(0.496)

(0.145)

(0.992)

(0.009)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

197

Table 3-A-9: Second stage of two stage least squares; estimating the share of low processed food
in total food consumption (2SLS)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
-0.004**
(0.000)

(0.000)

-0.171

-0.060

(0.163)

(0.616)

0.124

(0.045)

(0.647)

0.049

2.996**
(0.001)

Mansa
-0.006
(0.902)

0.729**

0.885**

(0.000)

(0.006)

Kitwe dummy variable

0.074**

Dependency Ratio

(0.847)

(0.796)

Lusaka dummy variable

Percentage Working Age Females
Employed in Nonfarm Labor

(0.186)

-0.351*

0.059**

Kasama dummy variable

Share of Low Processed Food
Lusaka
Kitwe
Kasama
-0.005**
-0.002
-0.022

(0.000)
(0.000)

0.005
(0.662)

0.001

-0.002

-0.002

0.020

(0.929)

(0.831)

(0.814)

(0.128)

0.043*
(0.009)

0.014

0.024

0.009

-0.049

-0.048
(0.122)

(0.337)

(0.230)

(0.652)

(0.151)

-0.003

-0.004*

-0.001

0.014**

0.005

(0.055)

(0.047)

(0.506)

(0.005)

(0.234)

0.035**

0.043**

0.022

-0.217**

0.039*

(0.001)

(0.001)

(0.165)

(0.004)

(0.037)

Own a Motorcycle

-0.014

-0.017

-0.021

-0.082

-0.054

(0.538)

(0.543)

(0.659)

(0.250)

(0.300)

Use Public Transit

0.011

0.020*

0.004

-0.032**

0.018

(0.096)

(0.030)

(0.721)

(0.004)

(0.130)

Adult Equivalents
Own a Car

0.013

0.015

0.013

0.031*

0.001

(0.077)

(0.182)

(0.176)

(0.044)

(0.933)

-0.025**

-0.012

-0.046**

0.003

-0.026

(0.000)

(0.302)

(0.000)

(0.830)

(0.060)

Own a Stove (Gas or Electric)

0.022*

0.017

0.040**

-0.097*

-0.045

(0.013)

(0.185)

(0.001)

(0.025)

(0.110)

Own a Hotplate

0.001

0.000

0.011

-0.042*

-0.023

(0.847)

(0.976)

(0.422)

(0.039)

(0.225)

0.009

0.009

-0.001

0.030

0.026

(0.278)

(0.460)

(0.966)

(0.081)

(0.158)

Own a Bicycle
Own Farmland

Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household
Lean Season dummy variable

0.001

0.001

0.004

-0.021**

-0.001

(0.343)

(0.517)

(0.064)

(0.008)

(0.852)

0.000

0.000

0.001**

-0.001

0.000

(0.151)

(0.462)

(0.000)

(0.070)

(0.505)

-0.003

-0.001

-0.009

0.010

0.006

(0.700)

(0.950)

(0.493)

(0.635)

(0.661)

0.037**

0.037**

0.033**

0.037**

0.072**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

198

Table 3-A-10: Second stage of two stage least squares; estimating the share of high processed
food in total food consumption (2SLS)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
0.014**
(0.000)

(0.000)

0.139

0.082

(0.376)

(0.605)

0.220

(0.422)

(0.456)

(0.490)

0.227

-0.163

(0.308)

(0.842)

Mansa
0.026
(0.264)

-0.154

-0.061

(0.421)

(0.903)

Lusaka dummy variable
Kitwe dummy variable

-0.027

Percentage Working Age Females
Employed in Nonfarm Labor

(0.000)

0.172

-0.017

Kasama dummy variable

Share of High Processed Food
Lusaka
Kitwe
Kasama
0.013**
0.014**
0.025

(0.295)
(0.098)

-0.052**
(0.000)

0.007

0.001

0.024*

-0.003

0.004

(0.403)

(0.940)

(0.030)

(0.814)

(0.867)

Dependency Ratio

-0.022

-0.029

0.005

0.001

0.007

(0.191)

(0.194)

(0.817)

(0.965)

(0.860)

Adult Equivalents

0.008**

0.010**

0.006**

0.001

0.002

(0.000)

(0.000)

(0.005)

(0.775)

(0.586)

Own a Car

-0.050**

-0.056**

-0.026

-0.006

-0.058**

(0.000)

(0.001)

(0.162)

(0.933)

(0.009)

0.021

0.048

-0.043

-0.019

0.022

(0.573)

(0.397)

(0.303)

(0.778)

(0.655)

Use Public Transit

-0.003

-0.006

0.000

0.019

-0.024

(0.659)

(0.540)

(0.972)

(0.087)

(0.136)

Own a Bicycle

-0.018*

-0.025*

0.000

-0.017

-0.048**

(0.021)

(0.029)

(0.989)

(0.220)

(0.005)

Own Farmland

-0.012

-0.028

0.012

-0.026*

-0.008
(0.668)

Own a Motorcycle

Own a Stove (Gas or Electric)
Own a Hotplate
Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household
Lean Season dummy variable

(0.169)

(0.095)

(0.374)

(0.037)

-0.026**

-0.030*

-0.022

-0.021

0.019

(0.007)

(0.028)

(0.132)

(0.603)

(0.669)

-0.014

-0.019

-0.013

-0.009

0.035

(0.088)

(0.082)

(0.349)

(0.680)

(0.241)

-0.014

0.006

0.004

0.019

0.009

(0.536)

(0.769)

(0.151)

(0.499)

(0.544)

-0.001

0.000

-0.005*

0.001

-0.004

(0.615)

(0.862)

(0.034)

(0.843)

(0.204)

-0.002**

-0.002**

-0.002**

-0.001**

-0.002**

(0.000)

(0.000)

(0.000)

(0.005)

(0.004)

-0.009

-0.006

-0.013

-0.020

-0.038*

(0.335)

(0.622)

(0.351)

(0.172)

(0.044)

-0.043**

-0.042**

-0.044**

-0.037**

-0.036**

(0.000)

(0.000)

(0.000)

(0.000)

(0.007)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

199

Table 3-A-11: Fractional probit estimates of the determinants of household share of supermarket
purchases in total food consumption (CRE)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Lusaka dummy variable

Pooled
0.005**
(0.000)

Share Supermarket in Total Food
Lusaka
Kitwe
Kasama
0.006**
0.003**
0.010**
(0.000)

(0.000)

Mansa
0.015**

(0.000)

(0.000)

0.004
(0.230)

Kitwe dummy variable

-0.079**

Kasama dummy variable

-0.051**

(0.000)
(0.000)

Percentage Working Age Females
Employed in Nonfarm Labor

-0.003

0.011

0.004

0.013

-0.006

(0.367)

(0.449)

(0.658)

(0.433)

(0.675)

Dependency Ratio

0.000

0.036*

-0.003

0.014

0.011

(0.210)

(0.032)

(0.748)

(0.413)

(0.684)

Adult Equivalents
Relative Distance to Supermarket
in Lusaka (km) (log)
Relative Distance to Supermarket
in Kitwe (km) (log)
Relative Distance to Supermarket
in Kasama (km) (log)
Relative Distance to Supermarket
in Mansa (km) (log)
Own a Car
Own a Motorcycle
Use Public Transit
Own a Bicycle
Own Farmland
Own a Stove (Gas or Electric)
Own a Hotplate
Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household
Lean Season dummy variable

0.015

0.004

-0.001

-0.002

0.005

(0.056)

(0.151)

(0.546)

(0.373)

(0.081)

-0.008**

-0.010**

(0.000)

(0.000)

-0.004**

-0.005**

(0.000)

(0.000)

-0.005**

-0.002

(0.009)

(0.338)

-0.007

-0.006*

(0.059)

(0.027)

0.002

0.006

-0.005

0.048

-0.042*

(0.264)

0.003

(0.693)

(0.485)

(0.087)

(0.023)

0.035

-0.003

0.001

-0.048

(0.821)

(0.218)

(0.901)

(0.945)

(0.054)

0.017

0.007

-0.007

-0.002

-0.008

(0.295)

(0.456)

(0.324)

(0.814)

(0.456)

0.001

0.002

0.009

0.001

-0.019

(0.860)

(0.892)

(0.317)

(0.960)

(0.307)

0.001

-0.014

0.004

0.008

-0.014

(0.922)

(0.104)

(0.578)

(0.426)

(0.384)

0.002

0.013

0.002

0.018

0.012

(0.810)

(0.496)

(0.842)

(0.243)

(0.537)

0.011

0.010

0.004

-0.011

0.028

(0.386)

(0.399)

(0.624)

(0.364)

(0.131)

0.009

0.000

0.012

-0.015

0.006

(0.289)

(0.995)

(0.176)

(0.393)

(0.722)

-0.009*

0.010

0.006**

0.002

0.000

(0.333)

(0.003)

(0.468)

(0.898)

(0.033)

0.004**

0.000

0.000

0.000

-0.001

(0.007)

(0.206)

(0.816)

(0.350)

(0.053)

0.024

0.016

0.016*

-0.006

0.002

(0.052)

(0.117)

(0.046)

(0.554)

(0.919)

-0.017**

-0.001

-0.006

-0.001

-0.001

(0.000)

(0.837)

(0.053)

(0.819)

(0.890)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

200

Table 3-A-12: First stage of control function regression; OLS estimates of the determinants of
the share of supermarket purchases in total food consumption (CRE)
Variable
Total Daily Household Expenditure
per Adult Equivalent

Pooled
0.005**
(0.000)

Lusaka dummy variable

-0.050**

Kitwe dummy variable

-0.050**

Kasama dummy variable

-0.020**

Percentage Working Age Females
Employed in Nonfarm Labor
Dependency Ratio
Adult Equivalents
Relative Distance to Supermarket
in Lusaka (km) (log)
Relative Distance to Supermarket
in Kitwe (km) (log)
Relative Distance to Supermarket
in Kasama (km) (log)
Relative Distance to Supermarket
in Mansa (km) (log)

Share Supermarket in Total Food
Lusaka
Kitwe
Kasama
0.006**
0.003**
0.010**
(0.000)

(0.000)

Mansa
0.015**

(0.000)

(0.000)

(0.000)
(0.000)
(0.003)

0.008

0.009

0.003

0.008

-0.006

(0.359)

(0.492)

(0.696)

(0.494)

(0.628)

0.027*

0.040*

-0.002

0.016

0.022

(0.033)

(0.025)

(0.820)

(0.390)

(0.412)

0.001

0.002

-0.001

-0.004

0.002

(0.651)

(0.549)

(0.492)

(0.091)

(0.421)

-0.019**

-0.021**

(0.000)

(0.000)

-0.011*

-0.008**

(0.017)

(0.000)

-0.014

-0.004

(0.054)

(0.441)

-0.014**

-0.007*

(0.003)

(0.050)

0.005

0.012

-0.016

(0.826)

(0.675)

(0.349)

(0.276)

(0.078)

0.033

0.055

0.011

-0.010

-0.074

(0.131)

(0.097)

(0.778)

(0.865)

(0.276)

0.000

0.007

-0.007

0.002

-0.007

(0.966)

(0.547)

(0.185)

(0.841)

(0.513)

Own a Bicycle

-0.001

-0.004

0.010

0.006

-0.021

(0.964)

(0.859)

(0.194)

(0.736)

(0.265)

Own Farmland

-0.016

-0.030

0.001

0.002

-0.006

(0.078)

(0.061)

(0.882)

(0.797)

(0.682)

Own a Stove (Gas or Electric)

0.016

0.018

0.005

0.025

0.019

(0.295)

(0.430)

(0.568)

(0.317)

(0.516)

Own a Car
Own a Motorcycle
Use Public Transit

Own a Hotplate

0.049

-0.077

0.012

0.013

0.007

-0.021

0.039

(0.326)

(0.413)

(0.468)

(0.318)

(0.148)

0.000

-0.001

0.006

-0.012

0.007

(0.926)

(0.819)

(0.350)

(0.274)

(0.520)

Highest Completed Grade of
Education in Household

0.004*

0.005*

0.001

-0.001

-0.008

(0.017)

(0.013)

(0.648)

(0.785)

(0.058)

Age of Household Head

0.000

0.000

0.000

0.000

-0.001*

(0.296)

(0.248)

(0.697)

(0.964)

(0.039)

0.014

0.014

0.014

-0.009

0.000

(0.073)

(0.193)

(0.110)

(0.465)

(1.000)

-0.004

-0.002

-0.005

0.000

-0.003

(0.291)

(0.683)

(0.124)

(0.932)

(0.702)

Own a Mobile Phone

Female Head of Household
Lean Season dummy variable

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

201

Table 3-A-13: Second stage of control function regression; estimating the share of processed
food in total food consumption (CRE)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa
Lusaka dummy variable
Kitwe dummy variable
Kasama dummy variable

Pooled
0.010**

Lusaka
0.009**

(0.000)

(0.000)

0.015

0.028

(0.890)

(0.817)

Share of Processed Food
Kitwe
Kasama
0.011**
0.006
(0.000)

-0.085

0.388

(0.470)

(0.115)

(0.996)

0.164

2.892**

(0.216)

(0.008)

Mansa
0.021
(0.646)

0.409**

0.796

(0.001)

(0.068)

0.025
(0.080)

0.030*
(0.021)

-0.048**
(0.000)

Percentage Working Age Females
Employed in Nonfarm Labor

0.013*

0.002

0.043*

-0.028

0.052

(0.013)

(0.890)

(0.024)

(0.238)

(0.187)

Dependency Ratio

-0.009

-0.010

0.010

-0.048

-0.036

(0.574)

(0.660)

(0.634)

(0.233)

(0.331)

Adult Equivalents

0.006**

0.006*

0.005*

0.018**

0.007

(0.001)

(0.014)

(0.041)

(0.004)

(0.060)

Own a Car

-0.026

-0.037

0.014

-0.019

-0.006

(0.122)

(0.074)

(0.497)

(0.861)

(0.931)

Own a Motorcycle

-0.019

0.002

-0.071

-0.003

-0.075

(0.600)

(0.971)

(0.277)

(0.939)

(0.444)

Use Public Transit

-0.006

-0.003

-0.001

-0.031*

-0.013

(0.559)

(0.858)

(0.956)

(0.044)

(0.531)

Own a Bicycle

-0.011

-0.028

0.004

0.032

0.021

(0.496)

(0.238)

(0.827)

(0.173)

(0.504)

0.013

0.019

0.006

-0.003

-0.034

(0.398)

(0.475)

(0.722)

(0.884)

(0.224)

0.006

-0.002

0.024

-0.041

0.026

(0.730)

(0.916)

(0.217)

(0.379)

(0.571)

Own a Hotplate

-0.014

-0.019

0.006

0.020

-0.029

(0.336)

(0.316)

(0.783)

(0.652)

(0.233)

Own a Mobile Phone

0.018

0.028

0.004

0.012

-0.013

(0.231)

(0.187)

(0.818)

(0.689)

(0.694)

0.005

0.006

0.005

0.001

0.003

(0.103)

(0.149)

(0.234)

(0.825)

(0.698)

-0.001**

-0.002**

-0.001

-0.002**

-0.002*

(0.000)

(0.000)

(0.115)

(0.000)

(0.023)

Female Head of Household

-0.011

-0.006

-0.022

-0.007

-0.038

(0.195)

(0.587)

(0.079)

(0.773)

(0.047)

Lean Season dummy variable

-0.005

-0.004

-0.012

0.003

0.041**

(0.380)

(0.574)

(0.093)

(0.742)

(0.002)

Endogenous Error from 1st Stage
Regression

-0.186

-0.134

-0.450

-2.695*

-0.679

(0.094)

(0.280)

(0.072)

(0.014)

(0.114)

Own Farmland
Own a Stove (Gas or Electric)

Highest Completed Grade of
Education in Household
Age of Household Head

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

202

Table 3-A-14: Second stage of control function regression; estimating the share of low processed
food in total food consumption (CRE)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
-0.004**

Share of Low Processed Food
Lusaka
Kitwe
Kasama
-0.004**
-0.003
-0.015

(0.000)

(0.002)

-0.099

-0.074

(0.398)

(0.555)

(0.214)

-0.205

0.096

(0.118)

(0.743)

(0.983)

-0.069

2.589**

(0.593)

(0.007)

Mansa
-0.004
(0.919)

0.271

0.749*

(0.018)

(0.026)

Lusaka dummy variable

0.037**

Kitwe dummy variable

0.053**

(0.005)
(0.000)

Kasama dummy variable

-0.008

Percentage Working Age Females
Employed in Nonfarm Labor

0.018*

0.013

0.033

-0.014

(0.018)

(0.483)

(0.078)

(0.496)

(0.029)

Dependency Ratio

0.017

0.025

0.008

-0.046

-0.045
(0.128)

(0.427)

0.052*

(0.248)

(0.213)

(0.684)

(0.202)

-0.002

-0.004

-0.002

0.014**

0.004

(0.109)

(0.068)

(0.309)

(0.009)

(0.281)

0.050**

0.051*

0.042

0.041

0.028

(0.003)

(0.017)

(0.147)

(0.565)

(0.675)

Own a Motorcycle

-0.020

-0.023

0.012

0.079

-0.050

(0.580)

(0.582)

(0.882)

(0.089)

(0.588)

Use Public Transit

0.010

0.027

-0.012

-0.031*

-0.004

(0.350)

(0.075)

(0.486)

(0.024)

(0.812)

0.014

0.015

0.006

0.006

-0.017

Adult Equivalents
Own a Car

Own a Bicycle

(0.281)

(0.421)

(0.749)

(0.778)

(0.442)

Own Farmland

-0.014

-0.005

-0.044*

0.003

-0.027

(0.325)

(0.840)

(0.029)

(0.843)

(0.157)

Own a Stove (Gas or Electric)

-0.003

-0.002

-0.005

-0.083*

-0.002

(0.836)

(0.936)

(0.828)

(0.026)

(0.958)

Own a Hotplate

-0.004

0.007

-0.042

0.044

-0.042

(0.792)

(0.702)

(0.070)

(0.311)

(0.066)

Own a Mobile Phone

-0.001

0.008

-0.019

0.054*

-0.044*

(0.968)

(0.689)

(0.372)

(0.032)

(0.024)

Highest Completed Grade of
Education in Household

0.005

0.006

0.007

-0.003

0.005

(0.097)

(0.129)

(0.089)

(0.605)

(0.296)

0.000

0.000

0.001**

-0.001*

0.000

(0.124)

(0.410)

(0.000)

(0.035)

(0.452)

-0.001

0.000

-0.007

0.011

0.002

(0.945)

(0.986)

(0.575)

(0.619)

(0.900)

0.041**

0.042**

0.034**

0.040**

0.075**

(0.000)

(0.000)

(0.000)

(0.000)

(0.000)

0.111

0.173

-0.024

-2.416*

-0.487

(0.349)

(0.186)

(0.935)

(0.012)

(0.143)

Age of Household Head
Female Head of Household
Lean Season dummy variable
Endogenous Error from 1st Stage
Regression

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

203

Table 3-A-15: Second stage of control function regression; estimating the share of high
processed food in total food consumption (CRE)
Variable
Total Daily Household Expenditure
per Adult Equivalent
Household shops at a supermarket
in Lusaka
Household shops at a supermarket
in Kitwe
Household shops at a supermarket
in Kasama
Household shops at a supermarket
in Mansa

Pooled
0.014**

Share of High Processed Food
Lusaka
Kitwe
Kasama
0.013**
0.014**
0.022

(0.000)

(0.000)

0.115

0.097

(0.383)

(0.515)

(0.000)

0.114

0.291

(0.428)

(0.345)

(0.944)

0.271*

0.055

(0.047)

(0.935)

Mansa
0.028
(0.177)

0.105

-0.096

(0.473)

(0.863)

Lusaka dummy variable

-0.011

Kitwe dummy variable

-0.017

(0.493)
(0.271)

Kasama dummy variable

-0.049**

Percentage Working Age Females
Employed in Nonfarm Labor

-0.003**

-0.010

0.011

-0.010

0.003

-(0.003)

(0.619)

(0.600)

(0.615)

(0.926)

Dependency Ratio

-0.026

-0.034

-0.002

0.000

0.003

(0.111)

(0.126)

(0.940)

(0.997)

(0.947)

Adult Equivalents

0.009**

0.009**

0.007**

0.004

0.005

(0.000)

(0.000)

(0.005)

(0.386)

(0.312)

Own a Car

-0.064**

-0.076**

-0.022

-0.049

-0.039

(0.000)

(0.000)

(0.469)

(0.364)

(0.464)

0.000

0.024

-0.069

-0.057

-0.033

Own a Motorcycle

(0.000)

(0.993)

(0.740)

(0.208)

(0.289)

(0.620)

Use Public Transit

-0.016

-0.030

0.011

-0.001

-0.010

(0.169)

(0.095)

(0.404)

(0.963)

(0.648)

Own a Bicycle

-0.024

-0.041*

-0.004

0.025

0.031

(0.101)

(0.049)

(0.808)

(0.273)

(0.324)

Own Farmland
Own a Stove (Gas or Electric)
Own a Hotplate
Own a Mobile Phone
Highest Completed Grade of
Education in Household
Age of Household Head
Female Head of Household

0.028

0.025

0.052**

-0.003

-0.007

(0.080)

(0.364)

(0.004)

(0.864)

(0.811)

0.009

-0.001

0.029

0.047

0.033

(0.652)

(0.982)

(0.253)

(0.166)

(0.404)

-0.009

-0.023

0.049*

-0.025

0.017

(0.529)

(0.213)

(0.049)

(0.363)

(0.586)

0.021

0.022

0.024

-0.044

0.033

(0.189)

(0.308)

(0.223)

(0.052)

(0.328)

-0.003

0.001

0.001

-0.001

0.004

(0.841)

(0.861)

(0.794)

(0.489)

(0.750)

-0.002**

-0.002**

-0.002**

-0.001*

-0.002**

(0.000)

(0.000)

(0.000)

(0.015)

(0.004)

-0.010

-0.005

-0.014

-0.021

-0.039*

(0.298)

(0.667)

(0.344)

(0.131)

(0.037)

Lean Season dummy variable

-0.044**

-0.045**

-0.046**

-0.037

-0.036**

(0.000)

(0.000)

(0.000)

(0.000)

(0.004)

Endogenous Error from 1st Stage
Regression

-0.274*

-0.281

-0.397

-0.053

-0.064

(0.046)

(0.076)

(0.203)

(0.938)

(0.907)

Source: Zambia Urban Consumption Survey 2007/2008
Note: Primary font indicates the estimated coefficient. Parentheses indicate the associated p-value. ** p < 0.01, * p < 0.05

204

REFERENCES

205

REFERENCES

Asfaw, A. (2008). Does supermarket purchase affect the dietary practices of households? Some empirical
evidence from Guatemala. Development Policy Review, 26(2), 227-243.
Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel curves and consumer demand. Review of
Economics and statistics, 79(4), 527-539.
Battersby, J., & Peyton, S. (2014, June). The geography of supermarkets in Cape Town: Supermarket
expansion and food access. In Urban Forum (Vol. 25, No. 2, pp. 153-164). Springer Netherlands.
BerdeguĂŠ, J. A., Carriazo, F., Jara, B., Modrego, F., & Soloaga, I. (2015). Cities, Territories, and Inclusive
Growth: Unraveling Urban–Rural Linkages in Chile, Colombia, and Mexico. World Development.
Bettencourt, L. M., Lobo, J., Helbing, D., KĂźhnert, C., & West, G. B. (2007). Growth, innovation, scaling,
and the pace of life in cities. Proceedings of the national academy of sciences, 104(17), 7301-7306.
Christiaensen, L., Weerdt, J., & Todo, Y. (2013). Urbanization and poverty reduction: the role of rural
diversification and secondary towns1. Agricultural Economics, 44(4-5), 435-447.
D’Haese, M., & Van Huylenbroeck, G. (2005). The rise of supermarkets and changing expenditure patterns
of poor rural households case study in the Transkei area, South Africa. Food Policy, 30(1), 97-113.
Ecker, O., & Qaim, M. (2011). Analyzing nutritional impacts of policies: an empirical study for Malawi.
World Development, 39(3), 412-428.
FerrĂŠ, C., Ferreira, F. H., & Lanjouw, P. (2012). Is There a Metropolitan Bias? The relationship between
poverty and city size in a selection of developing countries. The World Bank Economic Review,
lhs007.
Goldman, A., Ramaswami, S., & Krider, R. E. (2002). Barriers to the advancement of modern food retail
formats: theory and measurement. Journal of Retailing, 78(4), 281-295.
Goldman, A., & Hino, H. (2005). Supermarkets vs. traditional retail stores: diagnosing the barriers to
supermarkets’ market share growth in an ethnic minority community. Journal of Retailing and
Consumer Services, 12(4), 273-284.
GĂłmez, M. I., Barrett, C. B., Raney, T., Pinstrup-Andersen, P., Meerman, J., Croppenstedt, A., Carishma,
B., & Thompson, B. (2013). Post-green revolution food systems and the triple burden of
malnutrition. Food Policy, 42, 129-138.
GĂłmez, M. I., & Ricketts, K. D. (2013). Food value chain transformations in developing countries: Selected
hypotheses on nutritional implications. Food Policy, 42, 139-150.
Harrison, K., Henley, D., Riley, H., & Shaffer, J. (1974). Improving food marketing systems in developing
countries: experiences from Latin America (No. INVES-ET E14p H321). Michigan State Univ.,
East Lansing, Mich. (EUA). Latin American Studies Center.

206

Hayes, A. F., & Preacher, K. J. (2010). Quantifying and testing indirect effects in simple mediation models
when the constituent paths are nonlinear. Multivariate behavioral research, 45(4), 627-660.
Hichaambwa, M., M. Beaver, M. Weber, and A. Chapoto. 2009. Patterns of Urban Consumption and
Expenditure in Zambia: A General Report Based on the CSO/MACO/FSRP Urban Consumption
Survey in Urban Areas of Lusaka, Kitwe, Mansa, and Kasama, 2007-2008. FSRP Draft Working
Paper. Lusaka, Zambia: Food Security Research Project.
Hino, H. (2014). Shopping at different food retail formats: Understanding cross-shopping behavior through
retail format selective use patterns. European Journal of Marketing, 48(3/4), 674-698.
Kennedy, E., & Reardon, T. (1994). Shift to non-traditional grains in the diets of East and West Africa: role
of women's opportunity cost of time. Food Policy, 19(1), 45-56.
Minten, B., Reardon, T., & Sutradhar, R. (2010). Food prices and modern retail: The case of Delhi. World
Development, 38(12), 1775-1787.
Moyer, R., & Hollander, S. C. (Eds.). (1968). Markets and marketing in developing economies. Homewood,
IL: Richard D. Irwin, Inc.
Neven, D., & Reardon, T. (2004). The rise of Kenyan supermarkets and the evolution of their horticulture
product procurement systems. Development policy review, 22(6), 669-699.
Neven, D., Reardon, T., Chege, J., & Wang, H. (2006). Supermarkets and consumers in Africa: the case of
Nairobi, Kenya. Journal of International Food & Agribusiness Marketing, 18(1-2), 103-123.
Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response variables with an
application to 401(K) plan participation rates. Journal of Applied Econometrics, volume 11, 619632.
Papke, L. E., & Wooldridge, J. M. (2008). Panel data methods for fractional response variables with an
application to test pass rates. Journal of Econometrics, 145(1), 121-133.
Pingali, P. (2007). Westernization of Asian diets and the transformation of food systems: implications for
research and policy. Food policy, 32(3), 281-298.
Popkin, B. M., Adair, L. S., & Ng, S. W. (2012). Global nutrition transition and the pandemic of obesity in
developing countries. Nutrition reviews, 70(1), 3-21.
Reardon, T., Timmer, C. P., Barrett, C. B., & BerdeguĂŠ, J. (2003). The rise of supermarkets in Africa, Asia,
and Latin America. American journal of agricultural economics, 85(5), 1140-1146.
Reardon, T., Timmer, P., & Berdegue, J. (2004). The rapid rise of supermarkets in developing countries:
induced organizational, institutional, and technological change in agrifood systems. Electronic
Journal of Agricultural and Development Economics, 1(2), 168-183.
Reardon, T., & Timmer, C. P. (2007). Transformation of markets for agricultural output in developing
countries since 1950: How has thinking changed?. Handbook of agricultural economics, 3, 28072855.

207

Reardon, T., Barrett, C. B., BerdeguĂŠ, J. A., & Swinnen, J. F. (2009). Agrifood industry transformation and
small farmers in developing countries. World development, 37(11), 1717-1727.
Rischke, R., Kimenju, S. C., Klasen, S., & Qaim, M. (2015). Supermarkets and food consumption patterns:
The case of small towns in Kenya. Food Policy, 52, 9-21.
Senauer, B., Sahn, D., & Alderman, H. (1986). The effect of the value of time on food consumption patterns
in developing countries: evidence from Sri Lanka. American Journal of Agricultural Economics,
68(4), 920-927.
Smith, L. C. (Ed.). (2003). The importance of women's status for child nutrition in developing countries
(Vol. 131). Intl Food Policy Res Inst.
Smith, R., and R. Blundell (1986). An Exogeneity Test for a Simultaneous Equation Tobit Model with an
Application to Labor Supply. Econometrica 54, 679–685.
Staiger, D. O., & Stock, J. H. (1994). Instrumental variables regression with weak instruments.
Strydom, J. W. (2011). Retailing in disadvantaged communities: the outshopping phenomenon revisited.
Journal of Contemporary Management, 8, 150-172.
Thiele, S., & Weiss, C. (2003). Consumer demand for food diversity: evidence for Germany. Food
Policy, 28(2), 99-115.
Tschirley, D., Reardon, T., Dolislager, M., & Snyder, J. (2015). The rise of a middle class in East and
Southern Africa: Implications for food system transformation. Journal of International
Development, 27(5), 628-646.
Turrell, G., & Kavanagh, A. M. (2006). Socio-economic pathways to diet: modelling the association
between socio-economic position and food purchasing behaviour. Public health nutrition, 9(03),
375-383.
Veeck, A., & Veeck, G. (2000). Consumer segmentation and changing food purchase patterns in Nanjing,
PRC. World Development, 28(3), 457-471.
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press.
World Bank, (2009). World Development Report 2009: Reshaping Economic Geography, World Bank,
Washington, DC.

208

CONCLUSIONS

These essays highlight multiple consumption patterns to show that the continuation of
rising incomes, rapid urbanization and food retail modernization will induce growth and change
within the agrifood system.
Essay one supports the hypotheses that with rising incomes households purchase greater
shares of the food that they consume and follow the product transition as stated by Bennett’s
Law. I find that the surprisingly high levels of purchased shares of food in total household
consumed food value across ESA (greater than 40% at any expenditure level in all four
countries) are a result of households transitioning food consumption towards purchased food at
subnumerary levels of income. This finding highlights the fact that transitions in food
consumption patterns are not solely a middle-class story, as conventionally assumed, but poor
households are accessing markets to purchase greater shares of their food and are consuming
greater shares BSF foods as their incomes rise. Spatial considerations of increasing city size and
reduced distance to cities have significant positive effects on purchased share, affirming the
expectation that households will purchase more food (compared to consuming own production)
when they have increased access to markets.
Essay two further highlights the heterogeneity in the effects of urbanization on food
consumption patterns across income levels, city size and distances to cities, but with a focus on
the consumption of processed foods. Processed foods have penetrated the diets of rural (36% of
total food expenditure including consumed own production) and urban (63%) households at all
levels across the income distribution of ESA. I find that income-induced diet change that begins
among the poor contributes to these high levels of processed food consumption. The income-

209

induced transition towards processed food is driven by the consumption of highly processed food
that begins to rise rapidly at subnumerary levels of income. Given the large portion of the
population in ESA currently below the international poverty line, coupled with the pattern of
increased processed share with income growth, these findings signal a strong future demand for
increased food market infrastructure.
As anticipated, larger cities and lesser distance to cities result in greater consumption
shares of processed food, but of note is the high consumption of processed food even by
households who reside at considerable distances to urban areas.
Essay three incorporates food retail modernization into the food consumption pattern
analysis, leading to the following three findings. First, supermarket use positively affects the
consumption of processed food, which is expected given that on average eighty-five percent of
the food value purchased from supermarkets is processed food. Second, and more surprising,
conditional on the presence of a supermarket, households in smaller cities consume greater
shares of food, by value, from supermarkets than households in larger cities. My conceptual
model provides for this possibility by recognizing that congestion is greater in larger cities and
that differences in investment cost of supermarkets and non-supermarkets could increase the
relative density of non-supermarkets in larger cities, both of which would incentivize households
to consume greater shares of their food from outlets other than supermarkets. Third, although
supermarkets positively affect the consumption share of processed food, households that
consume some food from supermarkets continue to purchase eighty percent of their processed
food from non-supermarket retailers. The findings of this essay indicate that while food retail
modernization within the agrifood system affects food consumption patterns, such as rising

210

consumption of processed food, these patterns are not solely dependent on food retail
modernization.
Together, these three essays provide detail to how the macro trends of income growth,
rapid urbanization and the beginning of food retail modernization significantly affect the
consumption patterns of ESA. The fact that the diet transformation is happening among poor
households, not just the minority above the poverty line, lends urgency to the need for improved
public marketing infrastructure and regulatory approaches to accommodate and shape it.

211