Analysis of Household Food Insecurity and the Implication of Measurement Error, Mandera County, Kenya

ANALYSIS OF HOUSEHOLD FOOD INSECURITY AND THE IMPLICATION OF MEASUREMENT ERROR, MANDERA COUNTY, KENYA

MWENJERI G WAITHAKA. (BSC, MSC) A99/22787/2011

A Research Thesis Report Submitted In Fulfillment of the Requirement for the Degree of Doctor of Philosophy in Agricultural Economics, Kenyatta University

April 2015

DECLARATION

This thesis report is my original work and has not been presented for a degree in any other university.

Signature…………………………………………Date……………………………….. Mwenjeri Gabriel Waithaka (A99/22878/2011) Department of agribusiness management and trade Kenyatta University

SUPERVISORS

We confirm that the work reported in this thesis was carried out by the candidate under our supervision and has been submitted with our approval as the university supervisors.

Signature…………………………………………Date……………………………….. Prof. Bernard Njehia Department of agribusiness management and trade Kenyatta University, Nairobi

Signature…………………………………………Date……………………………….. Prof. Samuel Mwakubo Department of agricultural Economics and Resource Management Moi University, Eldoret

iii

DEDICATION

To my parents Mr. John Mwenjeri and Mrs. Margaret Wanjiru who planted me firmly on this path.

ACKNOWLEDGEMENTS

Even though the PhD. corridor is a solo march; I have benefited enormously from the guidance and support of several people along the way.

I am especially indebted to the 325 respondents who willingly and freely gave me their time and needed information. Special thanks to my field assistants, Halkano Jarso,

Charles Mwangi and the rest of the team for their selfless effort during the process of data collection.

I wish to acknowledge with gratitude the guidance of my University supervisors, Prof

Bernard Njehia, Department of Agribusiness management and Trade, Kenyatta

University and Prof. Samuel Mwakubo of Moi University, Department of Economics and Agricultural Resource Management.

I also wish to acknowledge my wife Safia, children Philip, James and Frances whom without their love, patience and understanding, I would never have come this far. Thank you for your continued support.

Special thanks to my parents Mr John Mwenjeri and Mrs Margaret Wanjiru for their pecuniary support, God Bless you.

TABLE OF CONTENTS Title page i Declaration ii Dedication iii Acknowledgements iv Table of contents v List of Tables viii List of Text Boxes ix List of Figures x List of Appendices xi Abbreviations xii Abstract xiii

Chapter 1: Introduction 1 1.0 Preamble 1 1.1. Background Information 1 1.2 Problem Statement 6 1.3 General Objectives 7 1.4 Specific Objectives 7 1.5 Hypothesis 7 1.6 Significance of the study 8 1.7 Study Area 9

Chapter 2: Literature review 11 2.0 Introduction 11 2.1 Concept of food insecurity measurement 11 2.2 Estimation of consumer demand and household expenditure 13 2.3 Empirical measurement of food insecurity 16 2.4 Research Gaps 20

Chapter 3: Methodology 22 3.0 Introduction 22 3.1 Conceptual Framework 22 vi

3.2 Theoretical Framework 25 3.3 Instrumental variable in General Method of Moments (IV-GMM) 28 3.4 Cost of Basic Needs Method (CBN) 32 3.4.1 Surplus/Shortfall Index 33 3.4.2 Adult equivalent factors 34 3.5 Food insecurity depth analysis 35 3.6 Data collection 36 3.6.1 Sampling Design and Sample size determination 36 3.6.2 Data collection methods 37 3.6.3 Data types and sources 38 3.7 Data Analysis 39

Chapter 4: Results and Discussion 41 4.0 Introduction 41 4.1 Sex of household head 42 4.2 Major sources of Food 44 4.3 Major Household Income sources 45 4.4 Food budget for households in Mandera County 46 4.5 Relationship between food share and Household expenditure – Non Parametric analysis 48 4.6 Food insecurity Levels for Mandera households 48 4.7 Food insecurity estimates with measurement error- Standard OLS regression results 52 4.8 Food insecurity measurement with corrected Measurement Error- IV-GMM Estimates 53 4.9 Measurement error in household insecurity analysis 57 4.9.1 Food insecurity gap for Households in Mandera 59 4.10 Responsiveness of the household food expenditure estimates 60 4.11 Implication to Food security Policy 62

Chapter 5: Conclusions and Policy Recommendations 69 vii

5.0 Introduction 69 5.1 Summary of findings 69 5.2 Conclusions 71 5.3 Policy Recommendations 72 References 75

viii

LIST OF TABLES

Table 3.1 Adult-equivalent conversion factors 34

Table 3.2 Data Design 38

Table 4.1 Gender of the household head 41

Table 4.2 Household food sources 42

Table 4.3 Household Income sources 43

Table 4.4 Food expenditure for household in Mandera County 44

Table 4.5: Cost of Basic needs for Mandera households 48

Table 4.6 Household food insecurity indices for the Mandera County 49

Table 4.7 Food insecurity estimates with measurement error –

OLS robust regression estimates 52

Table 4.8 Food insecurity estimates with corrected measurement error

-IV -GMM regression estimates 54

Table 4.9 Food Insecurity Gap for Households in Mandera County 59

Table 4.10 Responsiveness of the household food expenditure estimates 61

LIST OF TEXT BOXES Box 1. The role of district steering committee in food insecurity assessments 4

LIST OF FIGURES Figure 3.1 Conceptual frameworks of food access and consumption 24

Figure 4.1 Relationship between food budget and household expenditure- Non- Parametric Analysis 47

LIST OF APPENDICES

Appendix 1: Map of Mandera County 83 Appendix 2: Questionnaire 84 xii

ABBREVIATIONS

AIDS Almost Ideal Demand System

CBN Cost of Basic Needs

FAO Food and Agriculture Organization

GMM Generalised Method of Moments

GoK Government of Kenya

HES Household Expenditure Survey

IV Instrumental variable

LES Linear Expenditure System

MDG Millennium Development Goals

OLS Ordinary Least Squares

PRSP Poverty reduction strategy paper

UK United Kingdom

UNICEF United Nations International Children Educational Fund

US United States

USAID United States of America International Development

USDA United states department of Agriculture

WFP World Food Programme

xiii

ABSTRACT The objective of this study was to analyze food insecurity, underlining the significance of accurate measurement to formulate the required policies for addressing food deprivation. The need for accurate measurement of food requirement is essential to generate adequate information to support decision making especially in areas vulnerable to food shortages and famine. Using random sampling techniques and employing Fisher’s formula, a total of 323 households were selected for the study. Informed by demand theory as articulated by Engel’s law of inverse relationship between total household income and the expenditure on food, plus adding a quadratic term in the equation, the study sought to estimate the magnitude of food insecurity in Mandera County. The cost of basic needs (CBN) method was employed to provide preliminary estimate for the households’ food expenditure level. In order to deal with the problem of measurement error econometric models including ordinary least squares and using instrumental variable in generalized method of moment (IV-GMM) techniques were applied to quantitatively analyze data on quadratic Engel curve. The study established that Mandera County experiences food deprivation of significant magnitude. The study has revealed that, observed household expenditure is not a perfect measure of the actual food insecurity situation. This is because microeconomic data are contaminated by measurement error which reduces reliability of parameters and if not addressed will result to erroneous conclusion in economic analysis. The results show negative and significant quadratic coefficients for both OLS and IV-GMM. Accordingly the results shows that for the estimator that corrects for measurement error 81% of the households are food insecure as opposed to 64%. In this study it is observed that measurement error reduces parameter reliability by 32% which leads to underestimation of food insecurity by about 17%. Among the recommendations resulting from the study include; first it is easy to underestimate the proportion of food insecure households if they are incorrectly estimated and therefore superior statistical and sampling techniques should form the basis of quantifying food insecurity to facilitate decision making process. Secondly, the study supports for policy formulation that is guided by economic limitations not only as a gauge to measure food insecurity but also to guide intervention and evaluating policies aimed at alleviating it. Lastly, to increase food availability and reduce food insecurity, sound data-based analysis anchored on statistical theory that provides inferential basis for guiding policy and program interventions in of paramount importance. 1

CHAPTER ONE: INTRODUCTION

1.0. Preamble

This chapter provides background information on food insecurity analysis, highlights the analytical approaches and briefly expound on challenges on policy formulations emanating from these analysis. Section 1.2 presents the problem statement on information gaps on which this study is anchored, and the objectives guiding the study are presented in sections 1.3 and 1.4. Section 1.5 states the hypothesis of the study.

Lastly, the chapter presents the significance of the study and describe the study area in sections 1.6 and 1.7 respectively.

1.1 Background Information

Food insecurity continues to be a major challenge facing the world today. New determinants of food insecurity, such as capricious food price changes and climate change, are combining with previously identified namely; poverty, inequality and weak governance to expose ever-growing numbers of hungry people to an all time high (WFP,

2010). While millennium Development Goals (MDGs) have injected new momentum towards eradicating food insecurity, (UNICEF, 2005; United Nations, 2006) their achievement is progressively proving difficult especially in developing countries. The situation is further worsened by lack of comprehensive data to quantify and monitor progress. It is evident that tracking the realization of MDG No.1 continues to rely on highly uncertain assumptions (Keyzer et al, 2006) that have led to statistical misrepresentation on the scale of food insecurity. For example, reports show that in 2009, more than a billion people were undernourished (FAO,2010), and more than 1.3 billion

2 people were living on less than US$1.25 a day per capita, with almost half the world’s population on less than US$2 a day (World Bank, 2010a). Similar reports in Kenya indicate that about a third (10 million) of the population suffer from chronic food insecurity based on dietary energy supply (GoK, 2008). As pointed out by Keyzer et al

(2006), many governments and organizations such as FAO, derive their estimates from aggregate consumption data, based on food production and international trade balances, a practice that has been criticized as being sensitive to assumptions. This kind of data contains a number of gaps that compromise the results thereby resulting to interventions that are inapt. Surprisingly the situation has not provoked the required expert attention.

Certainly, the problem of food insecurity requires better customized interventions that are well grounded on accurate information to address it. Unfortunately, such strategies cannot be designed when they are based on wrong targets, insufficient information or both. Reliable information is critical in answering the following essential questions in addressing food insecurity, as pointed out by Smith (2006);

• Who are the hungry?

• How many people are hungry?

• What are the causes of hunger?

Addressing these questions will provide the necessary guiding principle important for targeting assistance and formulating suitable policies and planning interventions for helping people. It is very important to know how many people there are that face food shortages, because policy and aid decisions depend on that number and very often their lives depend on that number.

In Kenya, food security interventions have been directed towards improving the supply side through food production in pursuit by the government to ensure food availability primarily at the national level (GoK, 2007). The government has employed enormous resources in enhancing productivity through technological advances such as improved farm inputs including seeds, fertilizers etc, and improving access to credit and market information to enhance production. Paradoxically, these efforts have not been successful in ensuring food security for all citizens. The country continues to bear large proportion of the hungry afflicted population especially in arid and semi-arid areas which are characterized by low and inefficient production systems. Many households are continuously supported by the government and development partners through relief aid.

Amatya Sen in his seminal work on food entitlements and deprivation, (Sen, 1981), sought to characterize food security as access to food (demand side) rather than food availability. He argued that food availability at the national level does not directly translate to food security both at household and individual levels. This brought to the forefront the fundamental indicator of food insecurity as the inability of the households to access food. This necessitates for paradigm shift in the approaches that the policy makers needs to adopt to contain hunger among the affected in the population. There is need to think outside the supply-based box and include the demand based policies in order to solve the problem of food insecurity in the population.

Household expenditure and consumption have widely been used in economic and poverty analysis as indicators of welfare of households as well as inform on food policies. Analogous to the Sen’s observations, food consumption data from expenditure provides a credible measure of the household ability to access food. They project how

4 food demand is influenced by policies due to changes in prices and income levels (Dunne and Edkins, 2005). Food consumption data is particularly important in developing countries where the budget allocation on food is comparatively larger to other household expenditures. Ironically, household expenditure surveys in Kenya for example, have been undertaken (though not so frequent) specifically to provide a basis for poverty and welfare analysis among the population (GoK, 2001; 2003; 2007). However, food insecurity assessments continue to rely on aggregate estimates on prevailing situations in different areas through food security committees at district level (Kamau et al, 2011).

This includes information on food available at the national stores, at farm levels and expected food supply from crop harvests.

Box 1: The role of district steering committee in food insecurity assessments

District Food security steering committee is a Multi-disciplinary team from the Technical District Steering Group (DSG) under the leadership of the office of the president. It is composed of government ministries and other stakeholders related to

food security at the district level. The committee verifies existing information on the current food security situation and immediate prospects. They arrive at consensus when food security information/interventions are disputed. Food steering committees collate information on food security at the district level including:

 Food available at farm levels  Expected supply from crop harvests  Food available in national stores

 Food available in the market and prices  Crop condition.  livestock condition  Pests and diseases. They assist with broad identification of the types and magnitude of interventions to

reduce food insecurity and prepare current food security status report and forward to the government for decision making.

Source: GoK, 2005. Capacity building manual for district steering groups

This information is passed to the government decision making component to form the basis for food security interventions. Nevertheless, these aggregate estimates do not shed light on households’ access to food, and therefore presents a major drawback in formulating food security policies.

While microeconomic data is important in assessing food access, they are often contaminated by mis-measured variables that lead to biased and inconsistent parameter estimates resulting to erroneous conclusions in economic analysis (Cameron et al, 2005).

Groves (1989) highlights that, cross-sectional expenditure data collected from surveys usually suffers from limitations such as interviewer's errors, errors due to respondents, and so on. This underlines the significance of correcting for such errors while answering the above questions for the development of appropriate policy instruments. However, it should be noted apriori that this study is not focused on eliminating measurement error, but highlights the importance of its correction in the analysis of demand studies to minimize biases in variable estimations.

The specification of Engel curves has gained prominence in modeling of demand estimations in household budget studies. According to Deaton (1997), Engel curves provide an opportunity to classify goods and services, and it provides a basis for understanding how the allocation of the budget over goods changes with levels of living.

Though traditionally many studies employed linear approximation of Engel curves (e.g.

Deaton and Muellbauer, 1980), the presence of non-linear relationships in budget equations has been recognized to exist especially in non-food demand items. This research work focused on household food demand using quadratic food Engel curves to exemplify the effect of neglecting measurement error in measuring food insecurity.

1.2 Problem Statement

Eradicating food insecurity continues to be a socio-economic public policy mirage in many countries. This challenge is complicated by insufficient diagnostic approaches to provide accurate information on severity, magnitude and underlying causes of food insecurity. Currently, the approaches applied in determining food insecurity raise a number of concerns (Smith et al, 2006). The method most widely employed for measuring food consumption is aggregate data on food availability based on the supplies at the national level as opposed to the information indicating people’s access to food.

This method’s reliability has been the subject of considerable debate (Smith et al, 2006).

In other cases consumption data where food budget shares are presumed to be linear functions of the total expenditure have been used to assess food insecurity (Deaton and

Muellbauer, 1980; Mwenjeri, 2009; Ananda et al, 2003; Agbola, 2003). But, recent research findings in non-food items have confirmed that, linear specifications fails to explain non-linear interactions in various budget share equations (Lewbel, 1991) consistent with the observed consumer behavior. Furthermore, despite its wide application in food security measurement household consumption data are normally prone to measurement error (Smith et al, 2006). This reduces parameter reliability thereby undermining policy design and implementation. However, to accurately target assistance and to develop the necessary strategies to address the food insecurity problem, correct information is critical. Literature presents inadequate empirical work that describes the nature and consequence of measurement error in determining food insecurity, thus true extent and magnitude of household food insecurity remains unknown.

1.3 General Objective

The objective of this study was to analyze food insecurity situation in Mandera

County using quadratic Engel curve and households expenditure data, to highlight the importance of accurate measurement and derive policy implications required to counter food poverty.

1.4 Specific Objectives a) To estimate the magnitude, level and extent of food insecurity problem among the households in Mandera County. b) To determine the variance caused by the measurement error in the total household expenditure. c) To assess the implications of measurement error on household food insecurity estimation for policy formulation.

1.5 Hypotheses a) The degree of food insecurity experienced by households in Mandera County is

not significant. b) The variance caused by measurement error in the estimation of household food

demand is not significant. c) Measurement error in food insecurity estimation has no significant implication on

food policy.

1.6 Significance of the study

In Kenya the methods of assessing food insecurity are based on the availability of food supplies in the country (national level) without due regard on accessibility at the households and individual levels. However, food availability at national level does not guarantee food security and households or individual levels. This has led to a large population being ignored in food security decisions as a result of inapt assessment.

Accurate understanding of the extent of food insecurity is critical for situational analysis and to guide policy response.

Motivated by Engel’s theory and classical test theory, this study presents an in-depth analysis of the food insecurity situation in Mandera County. It provides timely and useful insight for measuring food insecurity in this county for appropriate action. The study is unique in a number of ways. First, it employs household consumption data which is hardly ever used in measuring food insecurely in Kenya. Household expenditure data enables the assessment of food accessibility at the household level. Secondly, the study employs models that account for measurement error which is inherent with micro data.

Previous research has shown that measurement error reduces parameter reliability thereby leading to erroneous conclusions. Correcting for measurement error is important when dealing with household data to achieve accurate estimates. Thirdly, continued application of linear equations in food insecurity analysis, has failed to account for non-linear relationships of budget share equations which are identified with consumer behavior.

Non-linear equations allow food to be classified as to be classified as luxuries at low income levels and necessities at higher income levels. This is important to guide useful

9 policy discussions that are led by economic considerations that are prerequisite to food security.

The study provides a solid reference point in food security analysis by highlighting the effect of measurement error when using microeconomic data. It presents another milestone in applying quadratic Engel curves in measuring food insecurity and makes valuable contribution in understanding non-linear relationships in household expenditure.

1.7 The study area

Mandera County is one of the two counties in the north eastern province and it makes the northern part of the province. It lies between latitude 20 11’ north and 40 17’ and longitude 390 47’ east and 410 48’ east. The county shares international borders to the

Ethiopia to the north, Somalia to the east and south east and border Wajir County to south and south west (GoK, 2004). It covers an area of 25,871 km2. The population in the district stood at 250, 372 according to 1999 population census (NB: The 2009 population census figure was disputed due to malpractices).

Water and pasture determine the population distribution and density. The county is characterized by low lying rocky hills between 400 and 700 above sea level which are located on plains from the south to the north. The river Daua passes over half of the county border with Ethiopia as it flows to the east to Somalia. The county is characterized by ‘Lagas’ (dry river beds) which get filled up by run off during the rainy season.

Rainfall is scanty and erratic with an average of 25mm. Mandera is usually very hot with mean annual temperature of 28.30c with hottest average at 38 0 c.

Mandera County falls under ecological zone VI which is characterized by very low

10 unreliable rainfall and very high temperatures. Most of the district is rangeland except for a strip potential for irrigation along the river Daua. The rangeland is used widely for supporting of livestock (cattle, camel, goats and sheep) which is the main economic activity in the county (GoK, Mandera district PRSP report 2001-2004).

The choice of study area has been motivated by the fact that Mandera County is one of the leading food poor regions and has persistently suffered from food insecurity characterized by incessant famines. This coupled with dwindling income sources and inefficient production systems, has led to inadequate purchasing power for the households to access food. The County food poverty is estimated at 83.5% adult equivalent and about 70.8% households classified as food poor with household food expenditure averaging at 80% of total household expenditure (GoK, 2007). The region has continued to rely on food support from the government and other organizations almost every year (GoK, 2004). Nevertheless, accurate measurement and analysis of household economic indicators is paramount for the policy makers to succeed in eliminating the development challenges posed by food insecurity. The geographical and income profiles in this county can be a basis for targeting to ensure they benefit from quality policy interventions. The study endeavors to provide accurate food poverty estimations that will guide the development of robust policy instruments in Mandera

County and other food poor areas.

CHAPTER TWO: LITERATURE REVIEW

2.0 Introduction

The chapter on literature review is centered on three sections related to food insecurity analysis with household as the central unit for measurement. The three sections are presented as follows; detailed background information on food insecurity analysis; theoretical background involving food insecurity analysis based on household expenditure and analysis of empirical work on methods employed in food insecurity analysis. The chapter concludes with a summary highlighting the gaps that the study endeavors to address.

2.1 Concepts of food insecurity measurement

Food security is an essential, universal measurement of household and personal welfare. Its absence demonstrated by food insecurity and hunger is detrimental to the wellbeing and also is potential antidotes to nutritional, health, and developmental problems.

Food Security exists when all people, at all times, have physical and economic access to sufficient, safe and nutritious food to meet their dietary needs and food preferences for an active and healthy life (World food Summit, 1996). The concept of food security is complex and requires elaborate measures to accomplish. At first the definition focuses on the daily consumption of food where distribution systems ensure a continuous availability of food. Secondly, the concept of access to sufficient and safe food includes the continuous physical availability of food, and thirdly, it implies the sustained economic ability to acquire food through the supply system. This system includes: food produced

12 for own-consumption; food commercialization schemes; public distribution systems with subsidized prices; or institutionalized food aid. Social access refers to acceptable food products supplied for consumption by population groups based on their cultural preferences.

Correspondingly, food insecurity refers to limited or uncertain availability and/or uncertain ability to acquire acceptable foods in socially acceptable ways (Bickel et al,

2000). Manifested through breakdown or inefficiency of food supply systems leading to food poverty, and marked by the inability of the households to access their food requirements.

Hunger is a persistent predicament especially in developing countries, worsening people’s wellbeing, productivity and often their continued existence (Smith et al, 2006).

For instance, the 2007/2008 food price crisis and subsequent global economic recession, have pushed the number of hungry to historical levels; exceeding one billion people worldwide (FAO & WFP, 2009). Surprisingly, some schools of thought indicate that world food production per capita has constantly increased since the beginning of sixties and will continue doing so till at least 2030 (FAO, 2002). However, world poverty and income inequality have unambiguously fallen over the last three decades (Sala-i-Martin,

2006), and thereby posing serious challenges in addressing food insecurity.

In view of the above concepts there is need to employ indicators that are suitable and reliable for identifying food insecure in the population. Accurate information on the extent of food insecurity is essential to anchor sound policy and decision making platforms for eradicating hunger and malnutrition. Decisive steps towards assessing relative trends in food insecurity are paramount in order to understand the magnitude of

13 food deficiencies and to identify the sections of population who are food insecure and their locations to enhance interventions.

Paradoxically, accurate data and information on the magnitude of food insecurity is hard to achieve especially in developing countries where it is needed most. In Kenya for instance, food insecurity information is generated through estimates based on unstructured observations of extension workers rather than formal statistical process and survey techniques (GoK, 2011). However, these aggregate estimates do not shed light on access to food by households, and therefore presents a major drawback in formulating food security policies. This puts the country on precarious position as far as addressing food insecurity problem is concerned. The present study highlights the significance of sound diagnostic actions that are essential to policy analysts as they seek to solidify their resolve on addressing food insecurity.

2.2 Estimation of consumer demand and household expenditure

Demand theory has widely been applied to model individuals’ or households’ consumption behavior. Household expenditure is one of the important aspects of aggregate demand used to model household consumption behavior. It can be broken down to specific categories of major spending such as food, clothing or fuel consumption, to model the expenditure trend of the households.

In 1857 Ernest Engel (1821 - 1896) achieved to model household food consumption behavior and came up with Engle’s law, which states that “the proportion of consumer’s budget spent on food declines with income” (Mundlak, 2000).

He observed that the proportion of income spent on food declines as income increases, implying that "food is a necessity whose consumption rises less rapid than does income" (Nicholson, 1992). Engel curve describes the relationship between a household's expenditure on food and the total household expenditure. Particularly, this study is guided by the principle of consumer demand theory, whereby the consumer will arrange his pattern of expenditure in a rational manner to maximize his satisfaction (Nicholson,

1992).

In the past, many studies have employed models in which budget allocation for individual good are assumed to be linear functions of the total expenditure (Household

Budget). For instance, in 1943 Working advanced the log-linear budget share specification which is known as Working Lesser model. Deaton and Muellbauer (1980) adopted this working-lesser specification, to derive a flexible concept of demand system useful for estimation of consumption behaviors and came up with Almost Ideal Demand

System (AIDS). Other demand models that are widely used in demand analysis include

Linear expenditure system (LES) of stone and Rotterdam model as proposed by Barten

(1964) and Theil (1965). These classes of demand systems are linear or quadratic in total expenditure has been favored because of their exact aggregation. But according to

Lewbel (1996) these specifications fails to correct for measurement error while using two stage least squares. This has led to recent studies focusing on Engel curve that allow more flexible curvature. For example Hausman et al (1991); Lewbel (1991); Blundell and

Duncan (1998) and Girma et al (2003) found that quadratic terms are significant in the modeling consumption and income relationships. The inclusion of quadratic term in the

Engel Curve analysis, allows the income-expenditure relationship to categorize goods as

15 luxuries and necessities at different levels of income. This is important for policy makers in food insecurity measurement since food being a necessity, it is imperative to understand the level at which household income is not enough to guarantee economic food access.

Analogous to consumption expenditures, measurement errors have been recognized in the estimation of Engel-curves (Hasegawa and Kozumi, 2001). The significance of measurement error on the implications of economic theories can be traced back to the theory of consumption advanced by Milton Friedman (Friedman, 1957). In his theory of permanent income hypothesis, indicated that consumption and income are composed of two components, namely; permanent and transitory, this can result from measurement error or a valid variation. He argued that both permanent consumption and income components are associated with marginal propensity to consume (MPC). He showed that attenuation bias causes the slope coefficient of a regression of observed income to lead to underestimation of the marginal propensity to consume (Chen, 2012).

Likewise, Lewbel (1996), observed that in a regression equation if total expenditure

(independent variable) is measured with error, there is high probability that the share of individual good (dependent variable) is also likely to be measured with error. i.e. measurement errors in the dependent and independent variables need not to be independent.

Lewbel (1996) using fuel expenditure in UK proposed the use of generalized method of moments (GMM) to correct for measurement error. Household food expenditure surveys are used to collect comprehensive data about the quantity expenditures of food consumed, economic and socio-demographic characteristics of households. Therefore,

16 estimating demand relationships from cross sectional survey data, accounting for measurement error is of paramount interest.

2.3 Empirical measurement of food insecurity

Reducing food insecurity in the developing world remains an important economic challenge, and one that is complicated by scarcity of information on the location, severity, and causes of food insecurity (Smith et al, 2006). Numerous approaches have been put forward in attempt to provide the necessary information on which to anchor food security policies.

The indicator for measuring the degree of food insecurity that is most widely employed by policymakers is the measure of “undernourishment,” or the percentage of a country’s population that does not consume sufficient dietary energy, by the Food and

Agriculture Organization of the United Nations (FAO). This approach focus on food supplies at the national level as opposed to the information indicating food accessibility at either individual or household levels. This mostly have been necessitated by lack of reliable information at the household or individual level in national surveys, however concerns have been raised that has led to action gaps in policymaking and program planning (smith et al, 2006).

One source of concern is that not all food acquired by the household is consumed.

This discrepancy will lead to overestimation of energy consumption which means downward bias to food energy deficiency estimates. Bouis et al (1992), using data generated from Kenya in 1985-87, shows a -4.6 percentage difference between energy acquisition and energy consumption. This means that any food insecurity measurement using the undernourishment method, and uses food acquisition as a bases of calculating

17 household energy requirement, will be presenting inaccurate estimation. The FAO method of undernourishment relies on mean energy requirement in estimating the depth of hunger in a country or region. However, this population group approach disregards the individual own requirement which makes it inappropriate to conclude if an individual is consuming adequate dietary energy. In so doing, this method may lead to overestimation of food insecurity in some regions or underestimate others.

The report by the Central bureau of statistics of Sudan (2010) shows that though the national average is 2180 Kilo calories, rural and urban areas had different levels of 2140 and 2270 daily Kilo calories respectively. Any food insecurity measurement based on the national average would present inaccurate depth of hunger in some regions. Similar observation by Smith et al (2006) in their study on food insecurity in sub-Saharan Africa, found that methods based on aggregate food availabilities have limited use for understanding the magnitude of food insecurity and exclude information on the degree of food accessibility.

The suitability of individual-level information on dietary intake has been central in food insecurity estimation. The method uses consumption data on an individual basis, using interviews on a 24-hour recall, dietary history or food frequency questionnaires.

The focus is on the information of food intake at individual level. The method has been credited to provide quality information on food insecurity at individual level. However, its application may be limited to specific groups within the population such as pregnant women and elderly persons. Therefore it is challenging to apply it at household level where food is not equally distributed according to specific individual needs (Scaccin et al,

1992).

Turning to body conditions, anthropometric indicators have been widely used to give an indication of food insecurity situation. This includes measuring body tissues with an objective to assess the impact of hunger and malnutrition. This can only be measured at an individual level as opposed to household level (Thompson et al, 1998). This method is commonly applied in children and amongst the most vulnerable adults in the society.

On account of food policy, anthropometric indicators poses a big challenge, in that it is difficult to determine whether the problem indicated by the body tissues is actually a food insecurity problem or from another source such as public health or diseases not related to nutrition.

On the contrary, smith et al (2006) advocates for household expenditure surveys which they argue provide reliable information and the indicators of quantity and access can be measured. The study further observed that there are discrepancies between HESs and FAO estimates, whereby the estimates of the former are found to be more strongly associated with other MDG indicators of food insecurity such as poverty than the later estimates. One major advantage of the Household Expenditure Survey (HES) method is due to the fact that food security decisions are taken at household level and the data collected represents both food choices and the extent to which households enjoy physical and economic access.

In addition, the proportion of expenditure dedicated to food, provides a relevant and valid measure of food deprivation. This is more meaningful when all sources of food are evaluated and estimated to account household access to food. Nevertheless, the shortcomings associated with this method include the fact that HES surveys are not undertaken frequently due to the cost involved in terms of time, financial resources and

19 technical skills. Similarly, though household expenditure surveys are reliable in estimating food insecurity; they are prone to bias due to errors of measurement.

As pointed out by Smith et al (2006), non-sampling errors in the measurement of household food acquisition arise especially during the data collection. They identified two important types of systematic bias that may arise during data collection; first the inability of respondents recalling their food acquisition over the survey period and the second bias is as a result of telescoping i.e. whereby a respondent include events that occurred before the recall period, thus inflating estimates of the household food acquisition. Also Lewbel (1996) reported that discrepancy due to storage and wastage resulting from the time of acquiring food to consumption is a form of measurement error.

Studies recognize various sources of measurement error, but this study will focus on the respondents’ inadequate ability to accurately recall their consumption levels, telescoping bias and discrepancy due to storage and wastage as the major sources of measurement error. This is according to Sudman and Bradburn (1973) who observed that recall errors in survey expenditure are based on omission and telescoping errors generated by respondents forgetting what they consume and incorrect remembrance when they actually consume.

Previous works have well documented the existence of measurement error in the use of household expenditure to study consumption behaviors. Using the data on Israeli family budget data Litivian (1961) reported that neglected measurement error induces bias in

OLS estimates of linear Engel curve parameters. While studying US consumer expenditure data, Hausman (1995) examined parameters of Engel curves and reported that about 41% of the variance of measured expenditure is due to measurement error.

Lewbel (1996), in an empirical application for fuel consumption in the United Kingdom, observed that correcting measurement error changed parameter estimates by more than

15%. The instrumental variable estimation for polynomial errors-in-variable due to

Hausman et al (1991, 1995) is one approach used to deal with measurement error.

However, this approach corrects for measurement error on the right hand side

(Independent variable) of the demand equation thereby disregarding the left side

(dependent). However, as alluded by Lewbel (1996), if the independent variable (sum of goods consumed) is measured with error, subsequently the dependent variable (one or more individual goods) is also affected, since their measurement is dependent of each other.

2.4 Research gaps

Food insecurity measurement is effectively realized when anchored on the three pillars of food availability, access and stability. The above review highlights a number of drawbacks in the quest for researchers and policy makers to gauge the magnitude of food insecurity experienced by the population. First the literature presents a grim picture of inadequate information that is not based on sound statistical and analytical approaches as antidotes for making sound policy and programming decisions. On the contrary, food insecurity analysis requires solid information on which to built interventions, as exemplified by the present study.

Secondly, as deduced in the above literature there are many approaches for measuring food insecurity that have been adopted with varying degrees of accuracy. In comparison, household expenditure surveys have been appraised as one of the methods that provide

21 strong lens through which prognosis of food insecurity problem can be achieved (Smith et al, 2006). However, due to the microeconomic nature of the data involved, the accuracy of the method is compromised by the problem of measurement error.

Furthermore, the literature above reveals that consumer demand is typically measured with error, which means that when estimating demand models, measurement error is either neglected or it can be dealt with by using models that correct for measurement error (Lewbel 1996). Consequently, to fill this analytical gap, this study employed a model that correct for measurement error both at dependent and independent variables to eliminate biased estimation in the context of food insecurity measurement.

Finally, the review further revealed that a number of studies usually adopt the use of linear associations between food expenditure and household income and so they tend to ignore the non-linearity relationship that exists which is associated with consumer behavior. In contrast, this study endeavors to typify demand for food by households by taking advantage of quadratic Engel curves that employ non-linear variable estimators to understand and characterize household’s food demand economics in their quest to reduce food insecurity.

CHAPTER THREE: METHODOLOGY

3.0 Introduction

The chapter presents both conceptual and theoretical frameworks underpinning the present study. Section 3.1 explains the conceptual framework on the consumption and access to food at household level. Section 3.2 examines the theory of consumer demand with reference to

Engle’s law. The chapter further expounds on quadratic Engle curve and its application to food demand analysis along with the problem of measurement error. The section also discusses the cost of basic needs approach in analyzing food insecurity. Lastly, the chapter details the process of data collection including data types and sources, methods of data collection, sampling and sample size determination and concludes with data analysis.

3.1. Conceptual framework

In this section the study focuses on description and definition of food security concepts and functional relationship that exist in food systems. The definitions of food security focus on processes as well as outcomes. They define not only the outcomes of food security that are useful for formulating policies and guiding actions, but the processes that lead to expected outcomes as well (FAO, 2000). The framework described below (figure 3.1) offers a holistic approach to food security processes and outcomes. The performance of these processes is essential in determining the probability of achieving food security.

In this model the study endeavors to capture the main activities and elements that encompass the course of food security. Figure 3.1 illustrates how food correct data analysis is useful in formulating policies and programming. It is grounded on the activities related to

23 production and distribution which leads to outcomes related to food security including food availability, food access and food stability.

Food availability focuses on the supply side of food security, which mainly depends on food production and exchange (trade) that boosts the stock levels that household can depend on. In food security analysis information of household productive capacity is essential in determining the degree of food availability. The amounts of food items that a household produces are expected to give an indication of the capability to meet food availability.

Accessibility to food is characterized by the economic capacity of the household and convenience of food distribution points such as markets. Insufficient food accessibility has led to shift policy focus to incomes, expenditure, markets and prices in achieving food security objectives (IFPRI, 2012). Adequate information on production based entitlement will provide the factors for policy development.

The concept of stability refers to both the availability and access dimensions of food security (FAO, 2006). Aspects that influence economic factors including expenditure, income and food prices are of major concern. Insufficient understanding of these factors in food security system, will lead to inadequate access to food at all times. Food security estimates are measured on the three pillars of availability, accessibility and stability.

To achieve this pillars of food security sound data collection and analysis is paramount The figure below (Fig. 3.1) shows the sources of food security related data from environment of production, exchange (trade), Livelihood systems household characteristics and socio institutions.

Food security environment

National level Reduced food Data Analysis Natural resource accessibility With Reduced food Food endowment Ineffective

Policy formula

Data Analysis measurement stability Insecure Agricultural sector error Policy design Reduced food Households Market conditions availability

Food related Data collection Improved level of Sub-national level tion Without Effective food accessibility Food Socio institutions measurement Policy design Improved level of Secure error Livelihood systems food availability Households Cultural attitudes Improved level of food stability Household characteristics

Source: Modified from FAO 2000

Figure 3.1: Conceptual framework of food access and consumption

Figure 3.1 presents a process that provides statistical strategies that enables generation of information to guide policy development. Discreet data analysis will result to sound policies that lead food secure households through ensuring food availability, accessibility and stability. Measurement error is inherently associated with microeconomic data and its correction will lead to solid background to anchor the necessary food related policies. On the contrary, disregarding the problem of measurement error will lead to deficient food related policies resulting to reduced food availability, accessibility and stability thereby causing households to be food insecure. The framework provides a better understanding of how food security processes can be used for policy-making and intervention planning.

It highlights the relationship of how food data collection and analysis can be improved to obtain more reliable, consistent and appropriate food security information. Data collection and analysis will always guide the policy formulation that determine the households being food secure or food insecure.

3.2 Theoretical Framework

The Neoclassical consumer demand theory holds that a negative relationship exists between the quantity demanded for a particular product and that product's price. The basic axiom of the utility maximization process is that a rational consumer will always choose a most preferred bundle of goods from the feasible set of consumption bundles allowed by his budget. Neoclassical demand theory holds that individual consumer’s rank-order their preferences from a set of options for what to purchase and select the combination of goods and services that will bring them the greatest amount of utility

(Nicholson, 2008). Consumer demand, then, is driven in part by what individual consumers seek to maximize their utility. This study employs the Engel curve as a

demand function for food, keeping the prices constant, to depict the consumption behavior of households with different levels of total expenditure. In line with demand theory, this study was anchored on two principles. The first is Engle’s law which describes the non-linear relationship between household food expenditure and total household expenditure. The second is the classical test theory of measurement that is concerned about the accuracy of observed variables which important in decision making.

Engel’s law describes how household expenditure of a good varies with household income. According to Engel (1857) “the poorer the family, the greater the proportion of its total expenditure that must be devoted to the provision of food” i.e. as income increases, the share of expenditure for food declines, demonstrating the shares of income spent on food are inversely related to income levels (Chen and Wallace, 2009). The theory does not imply that food budget decreases with increase in income, but rather that the proportion of income devoted to food increases at a slower rate than increase in income. Informed by this theory, this study employed household expenditure data to put forward a case for Mandera County in Kenya in estimating the extent of food insecurity.

Until recently many studies have used linear functions of budget shares based on working-Lesser specification (Deaton 1980, Mwenjeri, 2009), however studies using non-linear relationships in budget share specifications are being done especially in non- food items (Hausman et al, 1995, Lewbel, 1991).

Employing similar arguments, this study estimated non-linear Engel curves in form of quadratic equations to study household food items. As suggested by Blundell et al (1997), including quadratic terms in the Engel curve is important to account for goods being luxuries at some income levels and necessities at others. In addition, both Roy (2001) and

Skoufias (2003) have suggested that non-linearity characteristic of the income elasticity on calories intake is important in understanding the relationship in a household.

The study was also guided by the classical test theory of measurement which states that an observed variable is composed of a true value and an error value (Kane, 2010).

That is, an individual observed variable is modeled as composed of the true value of the unit of measurement and a measurement error. This is an important characteristic to the study of Engel curves which is subject to problems of measurement errors, given that measuring household income/expenditure is difficult and is subject to errors. Errors in variables induce non-zero correlation between the contaminated regressors and the equation disturbances, so that OLS estimates are biased and inconsistent (Fuller, 1987).

In the present study measurement errors are due to non-sampling errors in the measurement of household food acquisition that arise especially during the data collection. These include inability of respondents recalling their food acquisition over the survey period and telescoping bias i.e respondent include events that occurred outside the recall period, thus inflating estimates of the household food acquisition. The discrepancy due to storage and wastage resulting from the time of acquiring food to consumption is also as source of measurement error.

The use of accurate measure model can provide efficient and relatively accurate measures of variables that are relevant to the decision making. This study employs the concepts of measurement error both in dependent and independent variables and to account for this, the study uses instrumental variable approach according to Lewbel et al

(1996) and Battistin et al (2012).

3.3. Instrumental Variable in General Method of Moments (IV-GMM)

Application of the demand theory of the household can be specified as follows; qi = ƒn (x,p)……………………………………………………………….3.1

I = 1, 2 ….n

Where qi is the quantity demanded, p is the price; x is total household expenditure

(income). Suppose the true model underlying the data analysis is the working-lesser equation of the form:

wih = ai + bi ln xh + uih ……………………………………………………3.2 where the share of the food item (wih) is linear function of the log of the total household expenditure per adult equivalent (xh)

According to classical test theory, the observed variable X in the above model is composed of a true value and a measurement error. That is: x= x* + v (measurement error)………….…………………………………..3.3

Where X is observed variable; X*- true measurement; v - measurement error

The framework used in the in the study is the Engel curve estimation, which is: xh = ƒn (yh)………………………………………………………………...3.4

th where Xh is monthly expenditure of food by the household h , yh denotes monthly total household expenditure.

* * From equation 3.4 let yh and xh represent the correct total consumption expenditure and expenditure on food items respectively for household h=1……., H.

* * Likewise, let yh and xh be the measured values of yh and xh respectively. Therefore the measurement error in xh, denoted as vh can be given as;

* Vh = xh / xh …………………………..…………………………………….3.5

In log terms yields to

* ln vh = ln xh – lnxh …………………………...... 3.6

Likewise according to Lewbel, (1996) if the total expenditure (Independent variable) is measured with error food expenditure share (dependent variable) is also measured with error; Therefore, the correct and measured food expenditure is

* * * Wh = xh /yh ………………………………….……….……………………3.7

The observed food expenditure,

Wh = xh/yh……………………………………….……………………...... 3.8

Consequently, the measurement error in the observed food expenditure is given by,

* ωh = wh – wh ……………………………..………………………………...3.9

Applying budget shares that are higher than first degree polynomial, equation 3.2 is replaced by the following quadratic food share model

* * * 2 Wh = β0+ β1ln xh +β2 (ln xh ) + µh………………..………………………3.10

µh error term of mean zero,

This according to Hausman et al (1995), who suggest that for some goods equation 3.10

(quadratic), is adequate specification than equation 3.2(linear).

Combining equation 3.10 with equations 3.6 and 3.9 leads to the estimating model as follows

2 Wh =β0 +β1 lnxh+β2 (ln xh) + εh…………………………….……………...3.11

The compound error term εh is therefore expressed as

2 εh =ωh + µh –β1 ln vh +β2 (ln vh) - 2β2 (ln vh ln xh)………………………...3.12

Assumption: E(µh) =E(ωh) = E(vh) = 0

Although the above assumptions hold, the analysis will correct for measurement errors

both at dependent and independent variables of equation 3.11, since as measured food expenditure yh is part of total expenditure xh (Lewbel, 1996). Using Lewbel (1996) the study adopted a two step approach to arrive at consistent estimators for the demand model. Step one involved multiplying equation 3.11 by mis-measured regressor (xh) and then employed GMM estimation technique to obtain consistent estimates of the transformed model. The second step involved developing the relationship of the structural and the reduced model to recover consistent estimates of the former. This strategy was accomplished as follows:

Equation 3.11was multiplied by xh and using the household income, income squared and the interaction terms between income and log of income as instrumental variables, denoted as zh, results in the following moment conditions equation; to be estimated using

GMM technique.

* 2 zh xhwh = a0 E( zh xh)+ b1E(zh xh ln xh) +c2 E(zhxh ln xh) + vh...... 3.13

The following set of assumptions as identified by Lewbel (1996) provided the basis for the identification;

z) ≠ 0׀i) E(x)

z) = 0׀ii) E(εi)

(iii) E(vi) = 0

Assumptions (i) and (ii) ensure the validity of the instruments and (iii) implies that measurement errors are independent of total expenditure.

This means taking the conditional expectation with respect to z there is;

* 2 zh xhwh = a0 E( zh xh)+ b1E(zh xh log xh) +c2 E(zhxh log xh) + ρ3E(xhηz)...... 3.14

η was estimated through a non parametric regression of the observed log x on the

instruments z in the following specification;

Log x = g(z) + η……………………………………………………………….3.15

According to Battistin et al (2012), a GMM regression of equation 3.14, would consistently estimate the quadratic coefficient β2 (equation 3.11) being the coefficient of x log x2. The linear curve of the above equation can be estimated by the following equation

Battistin et al (2012),

* zh xhwh = a0 E( zh xh)+ b1E(zh xh log xh) + ρ3E(xhηz) ……………………..3.16

According to Battistin and Nadai (2012), instrumenting for endogeneity without adjusting for the non-linearities, introduced by measurement error will result to inconsistent parameters of interest as follows;

z]………………………………………….………3.17׀z] = α0 + bi0E[log x׀E[wi

The GMM regression of wi on log x using z instruments will result to biased estimate of b1 (equation 3.16). Therefore the parameters of the transformed model (3.13); a, b and c were estimated through GMM and the parameters of the Engel curve, equation 3.11 recovered by making the following relationships (Battistin et al, 2012).

2 2 β0 = a +b E (vh log vh) + c {[vh (logvh) ]- 2E[vlogc] } + ρicov(v, logv)…....3.18

β1 = b + 2cE [vhlog vh]…………………………………………………….…3.19

β2 = c………………………………………………………..……………….…3.20

2 To estimate E (vh log vh) and E [vh (log vh) moments the study adopted the following equations as suggested by Battistin et al (2012). Under log normality assumption and the fact that E [V] = 1 the following expression was used;

2 E(VlogV) = σ v/2…………………………………………………….3.21

2 Cov(V,logV) = σ v...... 3.22

Assuming that v is log normally distributed, the ratio of 3.16 and 3.17 a method of

2 moments estimate for the variance of log v(σ v) is obtained through

2 σ v = log (βi0/βi1)…………………………………..………………………….3.23

The values were then substituted back into the equations 3.18 and 3.19 for estimating a and b respectively. To estimate the magnitude of the bias the study exploited the following equation;

1 + var[V] = eσ2v ………………………………………………………….…3.24

Employing the argument of Battistin et al (2012), the magnitude of bias is approximately proportional to the variance of the measurement error. Therefore assuming the log normality of v, the variance of the measurement error was estimated as follows;

1 - eσ2v = var[v]………………………………………………………….….3.25

To evaluate the effect of measurement error the study examined the rate at which the rise in total household expenditure leads to a decrease in the budget share of food using the elasticity estimates for both OLS and GMM. The elasticity estimates were calculated as follows;

Β1 + 2β2X ……………………………………………………………...... 3.26

3.4. The Cost of Basic Needs Method

The cost of basic needs (CBN) method was employed in this study as a standard technique of assessing the extent and magnitude of food insecurity used by the government. The method has been widely used in measuring food poverty related studies

(Gok, 2007). The CBN method specify a consumption bundle supposed to be adequate for ‘basic consumption needs’, and then estimates the cost of this bundle in reference to

requisite prices The method is based on minimum nutritional requirement or threshold.

An individual is considered food poor if the nutritional consumption is less than determined threshold. The cost of basic method was calculated based on the following steps.

To begin with calorie requirement was defined based on the minimum intake of 2250 kilocalories, recommended by the FAO, which is also adopted by the Kenyan government (GoK, 2006). Secondly, a food bundle was determined using a diet that reflects the eating habits of Mandera households. Lastly, the cost of meeting the calorie requirement using that food bundle was estimated. The cost of buying the food bundle is the food poverty line, which is used to determine the proportion of the population that is food insecure.

3.4.1. Surplus/Shortfall Index

The Index is given as:

Where

P = Surplus/Shortfall Index; L = Recommended daily per capita requirements (2250Kcal.);

Gj = Calorie deficiency faced by householdj;

Xj = Per capita food consumption available to householdj; N = Number of households that are food secure (for Surplus index) or food insecure (for Shortfall index).

The tool was used to measure the extent to which a household is food secure or insecure. The households’ food security was defined based on the minimum intake of

2250 kilocalories, recommended by the FAO, which is also adopted by the Kenyan government (GoK, 2006). The nutrient composition of commonly eaten foods in Kenya was used to estimate the calorie intake of households using the National Public Health

Laboratory Services (1993) report.

3.4.2. Adult-Equivalent Conversion Factors

To determine the adult-equivalent reference scale, the study estimated the mean calorie requirements for men and women from 0 to above 60 years of age, resulting from a reference value of 2,250 kilo calories. The fractions are presented in the table below.

Table 3.1: Adult-Equivalent Conversion Factors

Adult equivalent Age in years factors Children 0 – 5 0.55 Boys 5 – 12 0.93 12 – 18 1.20 Girls 5 – 12 0.83 12 – 18 0.97 Men 1.14 18 -30 1.18 30-60 0.95 >60 Women 0.91 18 – 30 0.93 30-60 0.84 >60 Source: Estimates by the author data 2012

The adult-equivalent fraction assigned to each individual was determined by the ratio

between the calorie requirements (according to age and gender) and the estimated adult reference value (2,250kcal). The accessible daily energy for the households was estimated and adequacy of energy to meet the dietary needs of household members determined. The food quantities consumed by the household were identified (per individual food item) and converted to kilocalories by applying conversion tables

(National Public Health Laboratory Services (1993) report). The kilocalories for all food items available to the household were summed-up and analyzed to obtain energy intake per household. This was achieved by dividing total calories for the household per day by the adult equivalent persons eating from the household. The extent of household food insecurity in the County was measured by estimating the surplus/shortfall index from the cost-of-calorie formula. The surplus index and shortfall index indicating the degree to which a household is food secure and insecure respectively.

3.5. Food Insecurity Depth Analysis

One common dimension of understanding food insecurity incidence is the measurement of the depth and severity of food poverty. The depth of food insecurity shows the extent to which households are from the food security line. The severity relate to the distance separating the poorest households from the food security line. To achieve this, the study adopted Foster, Greer, Thorbecke (1984) index widely used in studies for measuring poverty. This was employed with minor modification of using household expenditure per adult equivalent for households (FAO, 2006). The food insecurity index of the FGT was defined as follows:

where Z = is the cut-off level of household expenditure per adult equivalent used to categorize a household as food secure or not, q is the number of households below the food security line, N is the total number of households in the sample, Yi is the food expenditure of household i, and P is the degree to which a household is food insecure

(food insecurity gap short fall index).

3.6 Data Collection

3.6.1 Sampling Design and Sample Size Determination

Sampling was constructed based on the 2008 integrated households’ budget survey report. The target population being all households in the Mandera County, random sampling methodology was employed proportionately to select a study sample based on fisher’s formula (Fisher et al, 1998) as follows:

where nf is the desired sample when the population is less than 10,000; n is the sample when the total population is more than 10,000; and N is the estimated population of the households in Mandera county (40,599). In Fisher’s formula, n was determined as follows:

Where n is the sample size when the population is more than 10,000, t is standard normal deviation at the required confidence level (1.96), p is the proportion of interest i.e. households who are food insecure, which stand at approximately 70% of the households

(GoK, 2007). The variable d is the significance level and is set at 0.05 because 95% confidence level was used as a cut-off point for significance in this study. The calculation of n was achieved as follows:

n = [1.962 x 0.70 x 0.30] = 323 0.052 The ratio of the households in each district to the total households in the county was used to define the sample sizes in all administrative districts as follows; Mandera west 73,

Mandera Central 102 and Mandera east 148

3.6.2 Data Collection Methods

The study employed questionnaires, interviews and review of secondary data as the main tools of data collection. The selection of the tools was guided by the nature of data and the objectives of the study. The study employed personal interviews in which the household members were required to recall quantities acquired and/or expenditures made over a seven day recall period. Data on different foods consumed by the households was obtained to guide the analysis. Data on household’s income including sources and amounts based on various household activities was collected.

To ensure quality data collection, the questionnaires were pre-tested on about 15 households selected from neighboring Wajir district. Information collected included; household total expenditure, expenditure on food, household income, as well as other household characteristics such as household sizes. Interviews were held to elicit more information from the households, NGOs representatives and government officers.

Secondary information was obtained through perusal of the government annual reports, statistical abstracts and other reports from NGOs.

3.6.3 Data Types and Sources

The study used both primary and secondary data. The primary data collected include

household total expenditure both food and non-food expenditure, household income and

household demographic characteristics as shown in table 3.2. Secondary data including

background information on population and households’ budget profiles.

Table 3.2: Data Design

Variable Definition Measurement Dependent variables

Wh Annual household food expenditure  Household food daily purchases  households' in-kind foods expenditure provided by the government (e.g. relief food) Amount  household's own account consumption of outputs (Kshs.) produced (e.g. own-consumption of milk produced on a farm)

Independent Variables

Xh Household annual expenditure Including  households' purchases of products for their everyday needs (e.g. food, clothing, rents, personal services)  households' in-kind expenditure for products provided by Amount the government (e.g. relief food) (Kshs)  households' payments (e.g. School fees, Health expenses)  household's own account consumption of outputs produced (e.g. own-consumption of milk produced on a farm) Zh Income earned by household members  Formal employment Amount  Small businesses (Kshs)  Sale of farm produce  Income received in-kind Ih Household demographic characteristics  Size of the household (members the household cook for) Number  Age structure of the household

3.7 Data Analysis

The present study, undertakes an in-depth analysis to determine the magnitude of food insecurity in Mandera County highlighting the consequences of neglecting measurement error in household expenditure data. The study employed both quantitative and qualitative data analysis techniques.

Qualitative tools were used in the study to provide a necessary and detailed descriptive characterization of variables that affect food insecurity in Mandera County based on the selected sample. This analysis presents potential relationships that exist between food insecurity situation and household characteristics. Measures of central tendency were explicitly used in this study.

The cost of basic needs (CBN) was used to provide preliminary analysis of food insecurity situation including the magnitude and the extent of food insecurity of Mandera households. In addition, the method is normally used by the government in estimating the wellbeing of the Kenyan citizens and therefore provides a comparative basis for assessing measurement error.

To study the effect of measurement error the quadratic Engel curve equation 3.11 was analyzed using Ordinary Least squares (OLS) and instrumental variable approach of general method of moments (GMM).

Ordinary least squares was used to offer a basis for which to estimate the variance caused by measurement error. This was done to show that estimators that do not correct for measurement error are biased and inconsistent for estimating microeconomic data coefficients as pointed out by Cameron et al (2005). The inconsistency of OLS in this study is due to measurement error in household expenditure data. To overcome this challenge the study employed instrumental approach using general method of moments

(GMM) as proposed by Lewbel (1996) and Battistin and Nadai (2012). This involved two- step approach; first by transforming the structural model into a reduced form and secondly by developing a relationship between the structural and reduced model to obtain consistent estimators as explained by the Denadai (2005). Data analysis was done using

STATA 12 software.

To exemplify the statistical difference of quadratic coefficients resulting from both

GMM and OLS estimates, the Hausman test was conducted. The test was to assess the implication of measurement error in the estimation of household food demand. The test of over-identifying restriction was also carried out to assess the validity of the instruments.

To account for consumption levels due to different household compositions, estimation was done using expenditure per adult equivalents, since as pointed out by

Deaton (1997) per capita expenditure has the effect of overestimating food budget share by households of larger sizes. This was achieved by dividing household expenditure by units of adult equivalents.

CHAPTER FOUR: RESULTS AND DISCUSSION

4.0. Introduction

This chapter presents descriptive and inferential statistics in the context of household food insecurity. First descriptive statistics results are presented. The results demonstrate the potential factors that directly or indirectly contribute to households experiencing food insecurity. These include sex of the household head, food sources, income sources, household age structure and household total expenditure. The findings of cost of basic needs method on the extent of food insecurity are presented in section 4.6. Lastly the chapter provides a discussion on the consequences of not accounting for measurement error and concludes with an in depth analysis of policy implication to inaccurate food insecurity measurement.

4.1. Sex of the Household Head

The sex of the household head goes hand in hand with decision making process in a household. For instance, it was observed that female headed households are biased towards purchasing certain food items for household consumption (Mwenjeri, 2009). In

Mandera County, the distribution of the households with respect to sex of the household head is skewed towards males. About 78% and 22% of the sampled households are headed by male and female respectively (Table 4.1).

Table 4.1: Gender of the Household Head

Sex of the Frequency Percentage Household Female 71 22% Male 252 78% Total 323 100% Source: Field survey data August 2012

These results concur with other studies undertaken in sub-Saharan Africa which supports the fact that family constructions are mainly patrilinear (Mwakubo et al., 2004;

Akinleye. 2009; Mwenjeri, 2009). These strongly suggest that household decisions in

Mandera County are mainly made by men. However, when it comes to food security decisions taken by men, households have been found to experience nutritional inadequacy (Mwenjeri, 2009).

4.2: Major Sources of Food

Despite the importance of livestock to the national economy, pastoralists in Kenya are becoming poorer. A recent livelihood study by NGOs (SCUK, 2007)1 in North-eastern

Kenya found a notable increase in poverty levels from 45-50% of the population to 50-

60% in the last five years. Pastoralist communities continue to suffer chronic food insecurity in the country experiencing high malnutrition rates that are consistently above international emergency thresholds (AFDB, 2014). The table below (Table 4.2) shows household food sources and percentages of the sampled households in Mandera County.

1 SCUK – Save the children UK

Table 4.2: Household Food Sources

Food source Percentage contribution by food source Crop farming 0.4% Livestock Products 8.6% Food purchase and exchange 54.8% Food aid/Relief 36.2% Source: field survey data August 2012

Table 4.2 shows that majority of households in Mandera County rely mainly on markets for their food requirements. This poses a clear challenge to the achievement of food security considering that Mandera County is classified as poor with a poverty index of 72% (GoK, 2007). This is not surprising considering that Mandera is characterized by very low and unreliable rainfall and comprising of rangeland type of climate that mainly supports pastoralism as the main economic activity. The region has been affected by recurrent droughts, which has in return led to a drastic and continuous decrease of economic assets, and an ever growing number of households that are dependent on external assistance (GOK, 2001). Following candid focus group discussions in the county it emerged that, normal market operations are frequently disrupted by insecurity mainly ignited across the border with Somalia and Ethiopia, ethnic conflicts, out migration of livestock and high transport costs which affects the inflow and outflow of food commodities. This explains the rationale of food aid contributing a substantial percentage of food sources for the households.

Crop farming in the county is minimal and mainly concentrated on the banks of river

Daua along the Kenya-Ethiopia Border thereby accounting for a very small source of household food requirements. This being a pastoralist economy, livestock and livestock products comprise a notable contribution to food basket for the people of Mandera.

4.3: Major Household Income Sources

Table 4.3 presents the sources of cash income for the households in Mandera County.

From the table it is observed that self employment contributes half of the income for the households followed by sale of livestock and livestock products.

Table 4.3: Household Income Sources

Income Source % of contribution by income source Crop sales 1% Sale of Livestock and products 25% Labour and employment 13% Self employment, small businesses and Trade 50% Other Income (Remittances, Gifts, Loans) 11% Source: field survey data August 2012

Mandera County normally experiences recurrent droughts which have negatively affected the livestock sector, which is the main economic mainstay of the households

(GoK, 2001). This has led to households diversifying their income sources to low return small businesses and trade and casual employment. This means that many households in

Mandera are faced with untenable income sources to support their expenditures. About

72% of the households in the county spend less than Kshs. 1440 per adult equivalent yet most of them have seven members in a household or more (GoK, 2013). The income for food purchases is earned from small businesses with meager earnings and related activities which are not sufficient to meet household expenditures. This impacts food security bearing in mind that studies show there is a strong and positive correlation between household income and food security (GoK, 2007; Deaton, 1997).

4.4 Food Budget for Households in Mandera County

Table 4.4 provides statistical summary of the household variables that were used in the study. On average the study shows that food share accounts for approximately 88% of the total household expenditure.

Table 4.4: Food Expenditure for households in Mandera County

Variable mean Std. dev. Min. Max.

Food Share 0.881 0.074 0.72 0.99 Total expenditure (Kshs.)* 1348.5 649.7 276.2 3719.1 Total Income (Kshs)* 944.7 586.8 112.5 3505.9 Household size (No)* 7.6 2.7 3.3 15.45 Sample size 323

Source: Field Survey data 2012; * Adult equivalent values

The household expenditure on food is observed at 88.1% this is comparable to GoK

(2007) report that puts it at 80%. However, this level of food expenditure is high compared to the national average of 51.1%. In comparison to national statistics2,

Mandera County differs in a number of ways. This can be attributed to a number of factors; high on the list is the high poverty levels. According to government reports,

Mandera County is categorized as poor with a poverty index of 72% (GoK, 2007). This is consistent with the Engel’s law that low income households’ expenditure on food is proportionally higher in the total household expenditure (Thompson and Metz, 1998).

Secondly, Mandera County relies heavily on markets and food aid as the major food source which stands at 54.8% and 36.2% respectively. At national level, 53.9% of food

2 Kenya National Bureau of Statistics: Kenya Integrated Household Budget Survey 2005/06 Basic report and Basic Report Well-Being in Kenya 2005/06.

consumed in rural areas comes from purchases (GoK, 2007). Moreover, the situation has been compounded by inefficient markets mainly due to poor road infrastructure, insecurity disruptions and food aid dependency (SCUK, 2007). This has made the prices of various food commodities to increase. Thirdly, the sample also shows a higher household size averaging about 8.9 as compared to the national of 5.1 as per GoK report of 2007 (GoK, 2007). Besides, the county experiences a high dependency ratio, about

53% of the population aged between 0-14, compared to the national average of about

43% (GoK, 2007).

4.5: Relationship between Food Budget and Household Expenditure - Non

Parametric Analysis

By way of varying the analysis, the study employed non-parametric regression analysis to enhance data description. Non parametric regression analysis is important since it permits the data to take a ‘local’ shape of the conditional mean relationship.

The smoothed food share values (bandwidth = 0.8) at 95% confidence interval is presented in figure 4.1 below. The result of the non-parametric regression analysis explains quadratic relationship between food share and log of household expenditure.

Engel curve provides valuable information of community’s consumption behavior at various levels of total expenditure and for different family compositions. According to

Bhanoji (1981) Engel curve shows how proportion of food expenditure increases, reaches a maximum and then declines, and suggested that the turning point could be taken as the threshold for food poverty. Hassan (2012) suggested that it’s important to account for the curvature of the Engel curve to capture the expenditure variability of low income people

suffering from a negative income shock.

. 9

Food Share

3.6 3.8 4 4.2 4.4 Log of household expenditure

Figure 4.1: Relationship between food budget and household expenditure- Non- Parametric Analysis

This preliminary analysis supports the rationale of this study in employing quadratic analysis to estimate the relationship between household food expenditure and the total expenditure for Mandera households. Similar results were observed by Girma and Kedir (2002), in their study in Ethiopia thus signifying the importance of employing quadratic relationship in this study. In addition, the quadratic term3 in the regression results is significant which indicates an improvement of the model. Against the backdrop of low-income households incapable of meeting their nutritional requirements and

3 From table 4.7 the quadratic term β2 =-- 0.0713 (-2.18) is significant.

spending almost all of their additional income on food, Hassan (2012) suggests that the food Engel curve is quadratic in developing countries.

Quadratic functional form provides analysis that is consistent with the observed consumer behavior, in which case the expenditure of food does not rise proportionately with household income. The food budget share increases at low income levels and declines at higher income levels (Nicholson et al, 2008). This characteristic allows food to be classified as luxuries at low income levels and necessities at higher income levels.

This is significant for policy formulation since it makes it possible to understand the economic levels at which poor households experience food insecurity thus necessitating the requisite interventions.

4.6: Food Insecurity Levels for Mandera Households

Food insecurity in a household can be analyzed from two dimensions namely: a problem of acquisition and a problem of utilization. The aspect of acquisition is based on the economic status of the household. Mandera County is categorized as one of the poorest in Kenya. With a poverty index of about 72%, the households in the county face an acute economic challenge of food acquisition.

Tables 4.5 and 4.6 presents the cost of minimum energy requirement and the extent of food insecurity estimates for Mandera County respectively. The cost of minimum energy requirement was calculated by analyzing the household food expenditure using Cost-of- basic needs (CBN) method proposed by Ravallion (1994, 1998).

The objective of the analysis presented in table 4.5 is to generate evidence of food insecurity status of households from expenditure data using the standard technique frequently used by the government of Kenya (GoK, 2007). However, this approach

although it uses expenditure pattern does not account for measurement error (Haughton et

al, 2009).

Table 4.5: Cost Of Basic Needs Estimates for Mandera Households4

Food Item Share in food Price Cost per Kshs for basket Kcal (1 kg) (Kshs) kcal 2,250 Maize 0.04 3530 35 0.01 22.31 Posho 0.07 2640 65 0.02 55.40 Wheat flour 0.10 3200 90 0.03 63.28 Beans 0.09 3140 85 0.03 60.91 Oil 0.16 9000 150 0.02 37.50 Milk 0.07 720 70 0.10 218.75 Pasta 0.09 1110 85 0.08 172.30 Potatoes 0.09 810 80 0.10 222.22 Sugar 0.07 3750 70 0.02 42.00 Rice 0.08 3460 75 0.02 48.77 Meat 0.21 1450 200 0.14 310.34 Cost of minimum energy requirement per adult equivalent 1253.78 Source: field survey data 2012;

Nevertheless it provides a basis for this study to determine the extent to which

measurement error can affect the accuracy of analytical parameters thereby resulting to

policy deformation.

The study investigates the extent of food insecurity based on diet quality indicators

that are central in food insecurity analysis. First the study examined the proportion of

households consuming inadequate dietary energy. Secondly, the depth of hunger, as

measured by the degree to which food intake falls below the minimum level of dietary

energy requirement. The results present background information for which to evaluate the

4 The column under share in food basket indicates the expenditure share in the basic needs bundle and the Kilo calorie column provides the number of edible kilocalories (Kcal) per 1 kg. The price column shows the average community prices of the various food items. While the column under the cost per calorie computes amount in Kshs a household would have to spend on one Kilo calorie of each item. The last column estimates the cost in Kshs to meet the minimum daily per adult equivalent calorie requirement of 2,250 Kcal. At the existing prices the total cost of purchasing the minimum daily adult equivalent calorie requirement amounts to Kshs 1253.78 per month.

effect of measurement error in estimation of food insecurity levels.

Table 4.6: Household Food Insecurity Indices for the Mandera County

Recommended Daily energy Level 2250Kcal.

cost of minimum energy requirements per adult equivalent (Per Kshs. 1,253.78 month) food secure households 0.326 5Head count ratio Food insecure households 0.673

6Surplus Index 0.12

Shortfall Index 0.34

Percentage food secure households 33%

Percentage food insecure Households 67%

Source: field survey data August 2012.

The above indices were calculated based on the recommended daily energy levels of

2250 Kilo Calories (Kcal), (FAO, 1993). The food security level (Z) was estimated at about Kshs. 1253.78 per month. These results indicate some similarity when compared to the one estimated by the government. The national average is about Kshs. 988 per month

(GoK. 2007) though a higher food poverty line is observed in this study. The discrepancy can mainly be described on the basis of both time and area consideration

5 Percentage of food insecure/secure households was computed based on food expenditure below and above recommended daily per capita energy requirement. 6 Shortfall/surplus index was calculated using P = 1/NΣGj where N is Number of households that are food secure (for Surplus index) or food insecure (for Shortfall index); Gj-is calorie deficiency/surplus faced by the household [(G =X-L)/L] L is recommended daily per capita requirement (2250 Kcal) and X is per capita food consumption available to the household.

used for analysis. The government estimates were calculated about 5 years earlier and presented a whole country analysis to arrive at the average food poverty line.

However, comparing the two estimates it is clear that the cost of food has increased by more than 25% and more people are expected to slip into food insecurity taking into account that expenditure does not rise proportionately with income (Thompson et al,

1998). Based on the above analysis, kshs.1253 corresponds to 67th percentile of the raw data. Thus 67% of Mandera households are food insecure subsisting on less than 2250

Kcal (recommended daily per capita calorie requirement) while only 33% are food secure. This observation is in line with government estimates which reported that

Mandera households are 74.6% food insecure (GoK, 2007).

A shortfall index of 0.34 and a surplus index of 0.12 indicate that food insecure households fell short of the recommended calorie intake by 34% while food secure households exceeded the calorie requirement by 12%. The shortfall index quantifies the extent to which food insecure households are below the average food requirement with reference to dietary energy. It is a useful measure that can be used to assess the amount of food needed to eliminate deficits, and can also be used to estimate the expenditure to facilitate the necessary food interventions and strategies.

The food insecure households are considerably below the estimated food poverty line and the food secure households are marginally above the estimated food poverty line.

But as established in the subsequent sections the technique fails to account for measurement error. This makes it inefficient in estimating the magnitude of food insecurity accurately to necessitate the formulation of policy responses to address the problem.

4.7: Food Insecurity Estimates with Measurement Error – Quadratic OLS Regression Results The study employed OLS regression procedure to emphasize the significance of correcting for measurement error in the analysis of survey data. The OLS procedure is based on certain assumptions some which refer to the relationship between the variables.

In the relationship between variables OLS assumes that the explanatory (independent) as well as dependent variables are measured without error (Wooldridge, 2006).

Table 4.7: Food Insecurity Estimates with Measurement Error - OLS Regression

Results

(Dependent Variable = Household expenditure on food)7 Log of total household expenditure 0.4140 (3.41)* Square of log of the total household Expenditure - 0.0713 (-2.18)* Log of household size 0.8804 (2.61)* Constant -0.0544 (3.03)* R-Squared 0.8701 8Turning point (95% confidence interval) 1,216.60

Source: Computed from Field Survey, August 2012; t-test in parenthesis; * significant at 95% level.

7 2 Wh =β0 +β1 lnxh+β2 (ln xh) + εh; wh is the household food expenditure; xh is the total household expenditure.

8 The value of x that defines the extremum of the relationship between wh and xh was derived using θ = -β1/2β2. Where, θ is the maximum value of x with measurement error.

However, as reviewed in the literature, microeconomic data is contaminated by measurement errors and therefore OLS estimators are both biased and inconsistent.

Table 4.7 presents the estimated ordinary least squares (OLS) regression parameters for the Food share and household expenditure adult equivalent relationship. The regression parameters are significant at 95 % level. The results display the quadratic relationship which justifies its application in this study. The turning point is estimated at

Kshs. 1,216.60 adult equivalent, which means with OLS regression, food is considered a luxury beyond this level of expenditure. This means that any policy to address food insecurity in this County will focus at this being the minimum level of income per month.

This is comparable to the Kshs. 1400 reported by the government but below the national average of Kshs. 2270 per adult equivalent in 2013 inequality report (GoK, 2013).

4.8: Food Insecurity Estimates with Corrected Measurement Error –IV- GMM Results

The measurement error corrected estimates are presented in table 4.8. From the table, the results are demonstrating presence of quadratic Engel curve relationship of the estimated model. The negative and significant coefficient of the square log of the household expenditure supports the use of quadratic relationship in this study. To test the validity of the instruments, the test of over-identifying restriction (p–values) was conducted and confirmed to be appropriately uncorrelated with the disturbance process.

The calculated p-value of 0.1867 is larger than the preferred significant level of 0.05, thereby accepting the null hypothesis of the validity of the instruments.

Table 4.8: Food Insecurity Estimates with Corrected Measurement Error - GMM Estimates

(Dependent Variable = Household expenditure on food) Log of total household expenditure 0.5876 (5.32)* Square of the log of total household Expenditure -0.0902 (-5.01)* Log of household size 0.3211 (1.37)* Constant -0.2364 (-2.80)* 9Turning point (95% confidence interval) 1762.5 P-value of the over-identifying restriction 0.1867 P-value of test for endogeneity 0.0005

Source: Field Survey, August 2012 (t-test in parenthesis; *significant at 95% level)

The study also employed Durbin-Wu-Hausman test for endogeneity to expose any differences between the estimation methods i.e. Ordinary Least Squares (OLS) and the

General method of moments (GMM). The comparison in this test was restricted to the point estimate and the standard error of the endogenous variable (Total household expenditure). The result of the Hausman test statistic shows the small p-value of (0.0005) which is less than the preferred at significant level of 0.05, indicates that OLS results are not consistent, and thereby rejects the exogeneity of the total household expenditure variable. In other words, the test shows that OLS method of estimation will result to biased and inconsistent parameters, and supports the use of instrumental variables in the

9 The value of x that defines the extremum of the relationship between wh and xh was derived using θ = -β1/2β2. Where, θ is the extremum value of x without measurement error.

estimation model.

The results in tables 4.7 and 4.8 show negative and significant estimates for both OLS and IV-GMM error corrected quadratic coefficients. The results suggest that food is a luxury to many households in Mandera and any economic shock would lead to serious food insecurity. Similar results were observed by Girma and Kedir (2002) in their study about when food stops being a luxury in urban households in Ethiopia. In quadratic functions, the usual interpretation of coefficients that the effect of one a unit change in its associated variable holding all other variables constant is not applicable. This because it’s

2 not possible Xh to change without changing x h, one has to differentiate the equation with respect to Xh, one obtains the change in Wh per unit change in Xh (Wooldridge, 2006).

But looking at the coefficients of the two equations (OLS and GMM) there is noticeable difference in magnitude with the OLS coefficients being smaller. The marked difference can be traced to the effect of measurement error, which causes the coefficient of the Xh (household expenditure) to be biased downwards, that is, smaller in magnitude

(Baum, 2006).

Turning the attention to the discussion of the measurement error corrected quadratic

Engel curve model, real income was used to construct the instrumental variables. This is because income as an instrument is correlated with endogenous variable (i.e. Total household expenditure); it is uncorrelated with error term and is assumed not to enter the main equation (i.e., does not explain the dependant variable, the food share).

One may argue that income may as well be measured with error, but as pointed out by

Lewbel (1996), there is no valid justification that reported income is correlated with measurement error in total household expenditure. He further argues that, under the

standard assumptions, income should be independent of the demand model errors.

Quadratic expression is one way of capturing the diminishing effect of a good. The

Engel curve of the corrected measurement error indicates that the food share increases with the total household expenditure up to the turning point at Kshs. 1762.50 per adult equivalent. This corresponds to 81st percentile of the raw data, indicating that for this proportion of households, their food requirements are classified as luxury. In contrast, the

Ordinary Least Squares estimates the food share start displaying signs of necessity in terms of budget proportion of Kshs. 1216.60 per adult equivalent. This corresponds to

64th percentile of the raw data the level beyond which food stops being a luxury in these households. According to inequality report (GoK, 2013), Mandera is subsisting at Kshs.

1,400 per adult equivalent. This means that in comparison with the present study using measurement error corrected estimates this measure underestimates household food expenditure level by about 20% per adult equivalent which can be attributed to the measurement error.

Accordingly, the Ordinary Least Squares and Cost of Basic Needs method estimates, understates the food insecurity situation of the households in Mandera County by about

17% and 14% respectively. This means that any policy formulation that relies on approaches that do not account for measurement error will underestimate the problem of food insecurity and exclude many households in the pursuit of addressing this challenge.

Turning to the coefficients of the household size, the results show marked difference between the OLS and GMM estimates. Comparatively OLS estimates are larger than the ones presented through GMM analysis. This difference can be traced to the presence of measurement error in OLS as opposed to GMM estimates. According to Gibson (2006),

measurement error in household expenditure is correlated with household size which inflates the OLS coefficients. He further argues that food expenditure data obtained through recall process shows reporting errors that are correlated with household size

(Gibson, 2003).

4.9: Measurement Error in Household Food Insecurity Analysis

According to reliability theory a measure is reliable if it is consistent in its measurement, that is, the degree to which analysis provides dependable estimates

(Trochim, 2006). Normally, a perfect measure has reliability value of 1 while imperfect one is zero (Trochim, 2006). The difference between the true value and the observed value is the measurement error that decreases the level of reliability of the estimates. In the present study the value of measurement error depicts the proportion of variability in the measure attributable to error. The results show that about 32% of the total variance in the total household expenditure per adult equivalent is due to measurement error10. This means that the reliability of OLS estimates (with measurement error) is reduced by 32%

(i.e. they are 68% reliable) and therefore inconsistent. This observation is reminiscent of

Webb et al, (2006) when they suggested that coefficients are considered sufficient reliable to make decisions if they are above 80%, but higher values perhaps more the

90% are preferred for decisions with significant consequences such as food which is normally described as human right.

To test for measurement error in the independent variable the study conducted the

Hausman test in section 4.8. The results of the test shows a smaller p-value (0.0005)

10 σ2v 2 The variance of the measurement error was estimated using 1 - e = var[v]. Where σ v is the variance of 2 log v, [ σ v = log (βi0/βi1) the ratio of βi0/βi1 equations 3.16 and 3.17 pp.31]; e is a constant (2.718282)

thereby rejecting the null hypothesis of no measurement error in the household expenditure variable at 5% significance level. The test confirms that there is significant difference between GMM and OLS coefficient estimates. This supports the suggestion that measurement error in household expenditure is significant to affect data analysis.

Consequently, the following inferences can be derived. First, the variance artificially narrow the distribution of the true value total household expenditure, making the measurement appear smaller than it is. Secondly, measurement errors flatten the quadratic relationship between food expenditure and total household expenditure. This observation corresponds to Kuha and Temple (1999) who pointed out that the flattening of the quadratic relationship between total household expenditure and the food expenditure is induced by measurement error. Thirdly, the error in the measurement of total household expenditure produces downward-bias and inconsistent parameter estimates of its effect, while inadequately controlling for the confounding effects of this variable on the well measured variables. These conclusions make reference to the assumption of the study that measurement errors are additive and statistically independent of the true values. It should be clear that measurement errors have a significant effect on data analysis since their variance is non-negligible fraction of the variance of the true values of the total household expenditure. The data used for this study was also examined for the presence of outliers to ascertain that there were no extreme points to ensure that the measurement error was not derived by the outliers.

4.9.1. Food Insecurity Gap for Household in Mandera County

To measure the food insecurity gap, the study employed a modified version of the

Foster, Greer, Thorbecke (FGT) index that is usually used for measuring poverty.

Household expenditure was used to assess the extent to which are far off from achieving food security. The food poverty gap was considered to represent the depth of food insecurity deficit of the households. The idea of the food insecurity deficit is to estimate the resources that would be required to strengthen food insecure households. The study localized the FGT index to suit the analysis by use of household expenditure levels developed in section 4.8 and 4.9. To this end FGT index was used to give an indication of the effect of measurement error in understating the enormity of food insecurity. The results in table 4.9 shows a marked difference between food insecurity gap when analyzed with and without the measurement error.

Table 4.9: Food Insecurity Gap for Households in Mandera County

CBN OLS IV-GMM with measurement with measurement measurement error error error corrected Food security 1253 1216 1762.50 Level (Kshs). Food Insecure 67% 63% 81% Households Depth of food 19% 14% 33% insecurity Calculations from field survey, August 2012

In economic terms the results show that on average income needed to uplift food poor household based on the OLS, CBN and IV-GMM analysis is 14%, 19% and 33% respectively for their particular levels of household expenditure. The result shows that measurement error not only underestimates food insecurity levels but also the depth and resources required to support the food poor households. The above analysis shows that based on OLS and CBN food insecure households in Mandera County would require

about Kshs. 178 and 203 per month per adult equivalent respectively to enhance their food purchases as compared to Kshs. 621 when using measurement error corrected estimates. In other words, any policy aimed at reducing food insecurity in Mandera

County that is based on techniques that do not account for measurement error would underestimate the benefits up to about 60% of the required economic assistance.

4.10: Responsiveness of Household Food Share

Expenditure elasticity provides valuable information on responsiveness of food budget share as a result of increasing total household expenditure.

Table 4.9 reports elasticity estimates11 of the total expenditure distribution for the

OLS and GMM measurement error corrected estimators. From the results it’s clear that the distribution takes a quadratic formation, whereby the rate of household expenditure rises to a maximum and then declines.

However, results show that the rate at which the expenditure rises is different between the two estimators. The OLS estimator starts to decline at about 19th percentile and the IV estimator starts to decline between 40th and 41st percentile. Similar results were observed by Girma and Kedir (2002) in their study in Ethiopia. The negative elasticity implies that food ceases to be a normal commodity. Consequently, with OLS estimates food remains a normal commodity to 19th percentile whereas measurement error corrected estimates rises to 41st percentile. Therefore, food security decisions that are based on the measurement error affected estimates would ignore the wellbeing of about a fifth of food insecure households.

11 The elasticity estimates were calculated using е = β1+2β2X, where е is the elasticity at a Xi.

Table 4.10: Responsiveness of Household Food Share

Expenditure percentile OLS Measurement error corrected

5th 0.061 0.129

10th 0.034 0.096

15th 0.024 0.084

19th 0.000 0.054

20th -0.003 0.051

25th -0.06 0.047

30th -0.017 0.033

35th -0.032 0.016

40th -0.043 0.002

41th -0.051 -0.008

75th -0.142 -0.119

95th -0.225 -0.212

Calculations from field survey, August 2012

In line with Kamau et al (2011), interventions that are anchored on sloppy analysis and not guided by current understanding on food expenditures are likely to be unsuccessful in the drive to eliminate hunger. Lack of crucial information such as the fraction of population that is food insecure and the severity of hunger often translates to ineffective planning, targeting and packaging of assistance and evaluation of interventions. Inadequate information implies inability to gather the necessary evidence and therefore necessitate for more effective options for measuring food insecurity in households.

4.11. Implication to Food Security Policy

“Evidence-based policy-making exists when policy decisions are based on careful, rigorous and robust analytical procedures using sound and dependable data. Sound and reliable statistics are essential for effective policy-making – a necessary part of the enabling environment for improving development outcomes” (Scott, 2005).

One of the overriding aims of the Kenyan government public policy is to ensure citizens enjoy sufficient food both in quantity and quality to meet their nutritional needs.

Furthermore, Kenya is signatory to millennium development goal of reducing hunger and malnutrition by half by the year 2015. Accurate measurement and analysis of food insecurity conditions is the antidote to providing information for successful interventions to contain food poverty in Kenyan households. The Kenyan Constitution provides one of the principles of national security is to be free from hunger and to have adequate food of acceptable quality for all (GoK, 2010). Accurate information is central in guiding the government to devise the strategies to ensure the right to food will be achieved to the population as enshrined in the constitution.

Past studies have suggested that to obtain the full range of food insecurity it is best to categorize it as a demand concern (Sen, 1981). The inability of household to access enough food is an obvious indicator towards food insecurity. On the contrary, the methods adopted by the government have failed to use this concept and therefore have not yielded much in terms of defining food insecurity for proper planning and interventions. Every year several households in the country are faced with food insecurity levels of varying proportions especially in less productive regions of the country.

In the effort to estimate food insecurity the government has continued to focus on the supply side of the food equation as opposed to demand side (GoK, 2011). This is done through disaggregated information generated through estimates based on unstructured observations of extension workers and food security committees at the district level

(GoK, 2011). The information is transmitted to the decision making organs of the government through the ministry of Agriculture to guide planning and interventions. This information normally describes estimates of food stored at the farm level and projected supplies from the crops in the field. In addition the information also includes the food available at the national storage facilities i.e. NCPB. These aggregated estimates of food insecurity have serious limitations in informing decision making (Kamau et al, 2011).

First, the information does not facilitate identification of food insecure households which is critical in guiding policy formulation. Secondly, the information is mainly based on the cereals production and other few major food commodities available. Thirdly, this information focuses so much on food availability but does not give indication of food accessibility by the households. This leads to inadequate information that is important to guide the necessary policy initiatives.

To improve food insecurity assessment, the government has recently employed multi- sectoral approaches. This involves joint teams comprising of government officers and other players in food security sector (GoK, 2013). The focus is mainly on the current overview of the food situation, the status of ongoing food programmes and key challenges affecting food production. Likewise the information gathered through this approach is scanty in elucidating the magnitude of food insecurity at household level and lacks the authority to define the degree of food access which is significant in alleviating

food poverty, decision making and planning interventions. In addition these assessments provide information for the very short-term decisions. This complicates both the situational and response analysis, with response planning often guided by vague estimates leading to inappropriate strategies that effectively address food insecurity quagmire.

In the last few years, the government has been guided by an approximated figure of about a third of the Kenyan population suffering chronic and nutrition insecurity (GoK,

2012). Following discussion with food security staff in the ministry concerned (Ministry of devolution and national planning) it emerged that government normally responds to food insecurity situations reactively rather than proactive approach. Based on this, the government distributes relief food guided by the respective population levels in affected

Counties.

In the year 2012, the government distributed about 625,000 kgs of rice against a population of about 257,735 that was recognized as food insecure, this accounts for about

25% of the total population in Mandera County. Comparatively, this study estimated food insecure population using the cost of basic needs12 approach (method used by the government) in the same year and found that about 623, 035 people were food insecure.

However, when computed using the method that corrects for measurement error the problem of food insecurity is compounded and the affected population rises to approximately 743, 143. This means that the resources allocated by the government could only support about 32 % of food insecure population by providing food relief, thereby leaving out about 68% of the population without adequate food assistance. From the above analysis it is observed that estimating food insecurity without correcting for measurement error, approximately 542, 355 (68%) of the food poor population were

12 Using 73.8% food insecurity level from table 4.6

statistically ignored and their wellbeing jeopardized. This depicts the depth of food deprivation in Mandera County and how far-off the government is in addressing the problem of food poverty. This to a certain extent explains why the food poverty in this

County appears to be inherently associated with the population.

The above analysis shows that amorphous assessment leads to skewed food security decisions that disregard food poor population which is statistically invisible. As observed in section 4.8, the Hausman test shows the presence of measurement error in the total household expenditure variable and if neglected leads to the estimates that are significantly biased and potentially capable of distorting food policies. This supports the hypothesis that measurement error in household expenditure is significant enough to distort food policy decisions.

Kenya has adopted the current food security concept that was approved at the World

Food Summit in 1996. This concept characterizes food security to exist when all people, at all times, have physical and economic access to sufficient, safe and nutritious food that meets their dietary needs and food preferences for an active and healthy life (GoK, 2011).

This food security concept is articulated in the food National Food Security and Nutrition

Policy (FSNP) of 2009. Strategy number one of the policy emphasizes on the key food dimensions of availability, accessibility and nutritional adequacy. The objective is to increase the quantity and quality of food available and accessible in order to ensure that all Kenyans have an adequate, diverse and healthy diet. The policy statement is very comprehensive in the quest to address food insecurity, however the implementation success reflected in declining hunger and food poverty is proving elusive to achieve

especially in arid and semi arid areas. Poor targeting occasioned by inadequate information presents a clear challenge.

Typically the government argues that it is faced with limitation in terms of resources allocation to support food security activities (MAFAP, 2013). In reality, the government and other organizations may be constrained by the resources even with the correct estimates. Budgetary limitation is a common experience especially in the developing countries. But it is imperative to note that, correct information can be the catalyst through which more resources can be harnessed to mitigate against food insecurity. Correct data can be used for advocacy to attract more resources from the international community through the government or the NGOs. Furthermore accurate information is a strong prerequisite for any government to declare hunger as a national emergency.

Likewise, it is important to note that inability to employ accurate information on policy design and implementation reduces transparency and accountability of strategies and programmes aimed at alleviating food poverty. Transparency enhances equity and efficiency while accountability ensures effectiveness (Scott, 2005). These are essential in management of food security matters which are embroiled with political interests, emotionally explosive and normally motivated by human rights. Therefore, accurate statistics are imperative for the government, NGOs and other organizations involved in food security matters to efficiently and effectively utilize the resources allocated and to support food related decisions. Lack of rigorous data analysis as a guide to plan for the necessary strategies statistically leaves sizeable food poor population without appropriate interventions.

The present study corroborates previous work on food security that has established the significance of household expenditures surveys for estimating food insecurity (Smith et al, 2006). However, in Kenya similar studies on expenditure are few; these include welfare monitoring survey of 1998 and wellbeing report of 2007. For instance, the latter provided an overview of the food poverty statistics based on household expenditures.

According to this report food poverty for Mandera stood at 74.6 % households (GoK,

2007). But, as it is the case with many similar studies the problem of measurement error was not addressed which means the parameter estimates were contaminated thereby leading to erroneous conclusions. This presents a clear challenge to the food policy

(FNSP) in which one of the aims is to guide the provision of quality information to support food security and nutritional programmes and strategies.

Absence of accurate information is an obstacle in tracking progress towards achievement of key food security objectives such as MDGs goal No.1 of halving hunger, constitutional mandate of right to food for all, government strategy of food access and the welfare of the citizens. In other cases lack of precise information can be disastrous especially in times of emergencies when the available data underestimates the magnitude of problem facing the population. It is worth noting that the benchmark of robust food policy is the correct statistical revelation of food poverty and the axis of its achievement is accurate information. It is imperative that policy direction should be informed by systematic and rigorous use of statistics to provide evidence that is reliable to guide implementation, evaluation and impact assessment. “Evidence-based policy-making is the only way of taking public policy decisions which are fully consistent with a

democratic political process characterized by transparency and accountability” (Scot,

2005).

CHAPTER FIVE: SUMMARY, CONCLUSIONS AND POLICY

RECOMMENDATIONS

5.0: Preamble

This chapter presents summary of the findings, conclusion and policy implications that result from the study underpinning the consequences of inappropriate analytical techniques in measuring food insecurity.

5.1: Summary of the Findings

To estimate the extent of food insecurity in Mandera County the study exploited the quadratic relationship between food expenditure and total household expenditure. The food share was the dependent variable while total household expenditure (independent variable) was squared to complete a quadratic relationship. The study also undertook to assess the effect of measurement error normally associated with microeconomic data and to establish their implication in food policy formulation.

Preliminary findings show that households in Mandera depend mainly on markets and relief food for their food needs. The study also established that indeed households in

Mandera allocate a substantial percentage of their household budget, 88.1% on food synonymous with low income households as pointed out in Engel’s law.

The study used cost of basic needs (CBN) approach, a standard techniques employed by the government to provide a preliminary analysis of the food insecurity problem in

Mandera County. The result further shows that majority of the households are far below the food security line (recommended calorie intake) and the ones which are considered food secure are just marginally above the line. This means that in the event of economic shocks these households will certainly slide to food insecurity status.

By means of robust specifications including non-parametric, ordinary least squares and measurement error corrected regressions the study has established that Engel food curves are non-linear and it’s imperative to study the curvature to precisely appreciate the relationship in budget share equations, in estimating the magnitude of household food insecurity problem. This assertion contrasts previous findings (Deaton and Muellbauer,

1980; Mwenjeri, 2009; Ananda et al, 2003; Agbola, 2003) which are based on linear specifications such as working-Lesser models.

The study considered the implication of measurement error in expenditure data and found that microeconomic data are contaminated by measurement errors and if not addressed will lead to unreliability of the estimates. In this study it is observed that measurement error reduce the reliability of the OLS estimates by about 32%. Thus the reliability of the methods that do not correct for measurement error is not guaranteed.

To emphasize the prospective policy ineffectiveness by failure of not accounting for measurement error, the study estimated income elasticity at different levels of household expenditure. The two estimators OLS and GMM produced contrasting findings looking at the variation of the expenditure elasticity in relation to household expenditure and the decrease of food budget share. According to OLS results, food share starts to decline at

19th percentile and IV results show decline at 41st percentile. The study reveals the analytical limitation that needs to be addressed to improve the information reliability. For instance, it’s observed that on the basis of OLS, results would overestimate the food security situation by up to 20%, thereby leaving out about a fifth of the population out of policy consideration.

The results further show that measurement error affects both the depth of food poverty

and the resources to support the poor households. The study found that the depth of food poverty based on OLS and IV-GMM analysis is 14% and 33% respectively.

Lastly, the study has pointed out a number of policy implications resulting from inaccurate food insecurity assessment. First, disregarding food insecure households that are statistically invisible and poor targeting of food assistance due to inadequate information presents a clear impediment to policy formulation and implementation.

Secondly, transparency and accountability and the essence of tracking progress in food security achievement will be compromised by lack of quality information. Finally, inaccurate food insecurity analysis can be disastrous in times of emergencies since they can lead underestimation of the magnitude of the problem facing the population.

5.2: Conclusions

Based on the results emanating from this study the following conclusions can be drawn on the findings:

First it has established that indeed Mandera County experiences food deprivation of significant magnitude. Food insecure households are considerably below the estimated food poverty line and the few food secure households are marginally above the estimated food poverty line which means most of the households in the county are vulnerable to food shortages.

Secondly, the study has revealed that, observed household expenditure is not a perfect measure of the actual food insecurity situation. The fact that significant variance in total household expenditure is due to measurement error demonstrates the contamination of microeconomic data. Therefore, robust analytical techniques are important in

understanding food poverty as the basis of sound intervention approaches. It has demonstrated the significance of superior food insecurity data analysis to guarantee quality information as backdrop for policy formulation.

Thirdly, the study has found out that using the techniques that do not account for measurement error, would underestimate the scale of food insecurity and any intervention to reduce food poverty in Mandera County would ignore the welfare of about a fifth of the population.

Fourth, the study also revealed that the depth of food poverty is affected by the measurement error, and therefore limit the allocation of the resources that would be needed to lift the food poor out of deprivation through perfectly targeted assistance.

Finally, the study has brought out that the ineffectiveness of interventions to address food insecurity is significantly contributed by statistical bias, caused by erroneous techniques used in its determination resulting to underestimation. Food insecurity measurement should yield reliable estimates that guide policies effectively to enhance the dynamism to eradicate food deprivation.

5.3: Policy Implications

The findings of this study leads to the following policy recommendations in attempt to improve food insecurity measurement:

First, the study asserts that is it is easy to ignore food insecure households if they are statistically invisible and therefore suggests use of sound data-based analysis, to estimate the extent of food insecurity for definitive quantitative evidence that contributes to helpful policy discussions for the purpose of planning, developing and targeting food

security interventions. Based on household expenditure, the study has established that measurement error leads to underestimation of the magnitude of food insecurity problem.

For instance any economic policy aimed at addressing food poverty in Mandera County, will target over 81% of the households. This is distinctly different from policy recommendations from approaches that ignore measurement error i.e. ordinary least squares and cost of basic needs. They both imply that about 63% and 67% of households respectively experience food insecurity, thereby underestimating the food insecurity situation in Mandera County by about 20%.

Secondly, the study proposes for the formulation of policy instruments that are guided by income limitations as a credible measure of food insecurity for shaping the attention on hunger in Mandera County. Augmenting household incomes has the potential of enhancing the existing mechanisms of food access; this can be achieved by adjusting the price and income policies. From the findings, households in Mandera allocate over 88% of their budget on food. This means that these households are more severely affected by food price increases because it decreases access to food and lead to a reduction in the diversity and quantity of diets. It is important to note that the higher the budget allocation to food, the stronger the effect of food price changes on household income. Furthermore, the study shows that many households in Madera are far below economic food security line and therefore more vulnerable and food insecure. This is important for targeting of interventions and the design of policies intended to safeguard food security for such households.

Third, the study further advocates for meticulous analysis that quantify the effects and appraise policies and programs designed to address food insecurity. By collecting

information on households expenditure helps to understand how far populations exceed or fall short of being food secure. This will guide targeting of food assistance to the needy and to formulate policies that will lead to informed decisions with positive food security impact. For instance, a targeted food subsidy is an option that can be explored.

Through food-mediated income the food insecure households will benefit from increased income which leads to improved food security. This will also have a positive reduction on costs and budgetary burdens associated with general food subsidies.

Lastly, the study envisages policy formulation and implementation that is anchored on reliable statistical facts for improved transparency and accountability. This is critical for addressing food insecurity problem which is human right issue and normally entangled with political interests and prejudice. Experience has shown that food issues are explosive and emotional in nature, therefore accurate information is a requirement to provide the needed empirical evidence to support the decisions taken by the government and other stakeholders. A food-security strategy that relies on accurate information, leads to greater policy predictability and general openness to interventions and hence more effective in addressing food poverty.

REFERENCES

Africa Development Bank Group (2014). Kenya country Strategy Paper 2014-2018

Agbola, F. W. (2003). Estimation of Food Demand patterns in South Africa Based on a Survey of Households. Journal of Agricultural and applied Economics, 35_3 (December 2003):663-670. Southern African Economics Association

Ananda W, Nyange D, Tsujii H (2003). Food Demand Patterns in Tanzania: A Censored Regression Analysis of Microdata. Sri Lankan J. Agric. Econ. 5(1):9-27

Banks, J., Blundell, R., Lewbel, A. (1997). Quadratic Engel curves and consumer demand. The review of economics and statistics, Vol. 79, No. 4. (Nov., 1997), pp. 527-539

Barten, A.P., 1964, Consumer demand functions under conditions of almost additive preferences. Econometrica 32, 1-38.

Battistin, E., Michele N., (2012). Identification and estimation of Engel curves with endogenous and unobserved expenditures: identification and estimation of Engel curves, Journal of applied econometrics

Baum, C F,. (2006). An introduction to modern econometrics using stata

Bickel, G., M. Nord, C. Price, W. Hamilton, and J. Cook. (2000). “Guide to Measuring Household Food Security, Revised, 2000.” Technical report, U.S. Department of Agriculture, Food and Nutrition.

Bhanoji Rao, V.V. (1981) measurement of deprivation and poverty based in the proportion spent on food: an exploratory exercise. World Development 9(4), 337- 353.

Blundell, R., Duncan, A. (1998). 'Kernel Methods in Empirical Microeconomics', The Journal of Human Resources, 33(1):pp. 62{87, 1998. ISSN 0022166X. URL http://www.jstor. org/stable/146315

Bouis, H., L. Haddad, and E. Kennedy. 1992. Does it matter how we survey demand for food? Evidence from Kenya and the Philippines. Food Policy17 (5): 349– 360.Bound, J. Brown, C., Mathiowetz, N. (2000) Measurement Error in Survey Data; Report No. 00-450

Cameron A, C., Trivedi P, K. (2005). Micro econometrics: Methods and applications. Cambridge University Press

Carroll, R., J. Ruppert, and L. Stefanski (1995), Measurement Error in Nonlinear Models (Chapman and Hall, London)

Chen, S., Wallace S. (2009). Food consumption in Jamaica: A household and social behavior International Studies Program, Andrew Young School of Policy Studies Georgia State University

Chen, X. (2012). Nonlinear Models of Measurement Errors, Journal of Economic Literature. Clarendon Press, Oxford CSAE, University of Oxford

Deaton A. (1997). The Analysis of household surveys: A microeconometric approach to Development policy. John Hopkins University Press

Deaton, A., Muellbauer, J. (1980). Almost ideal demand system; The American Economic Review, Vol. 70, No. 3. (Jun., 1980), pp. 312-326

Dunne, P. and B. Edkins, (2005) “The Demand for Food in South Africa.” Presented at the Economics Society South Africa Conference, Durban, September, 2005

FAO and WFP. (2009). The state of food insecurity in the world: economic crises- Impacts and lessons learned; Rome.

FAO. (2010). State of the food insecurity in the world 2010: Addressing protracted crises: issues and challenges, Rome

FAO. (2002). World agriculture: Towards 2015/2030, Summary report, FAO Rome.

FAO. (2006). Food security and agricultural development in sub Saharan Africa.

Food and Agriculture Organisation (FAO), 2006. Food security. In FAO Policy Brief: FAO of the United Nations.

Foster, J., J. Greer and E. Thorbecke. 1984. A class of decomposable poverty measures. Econometrica, 52(5): 761-766.

Fuller, Wayne A. (1987). Measurement Error Models.John Wiley and Sons, New York.

Gibson, J. (2003). Does Measurement Error Explain a Paradox about Household Size and Food Demand? Evidence from Variation in Household Survey Methods. Paper prepared for presentation at the American Agricultural Economics Association Annual Meeting, Montreal, Canada, July 27-30, 2003

Gibson, John , “Why Does the Engel Method Work? Food Demand, Economies of Size and Household Survey Methods,” Working Paper, Department of Economics, University of Waikato, New Zealand, 2002.

Girma., S and Kedir., A,M,. (2002). When Does Food Stop Being a Luxury? Evidence from Quadratic Engel Curves with Measurement Error. Centre for Research in Economic Development and International Trade, University of Nottingham

GoK .(2001). Central bureau of statistics, second report on poverty in Kenya, Vol. III Welfare indicators Atlas

GoK, (2013). Exploring Kenya’s Inequality: Pulling apart or pooling together

GoK. (2001). Mandera district Poverty Reduction Strategy Paper report 2001-2004 Ministry of Finance and planning

GoK. (2005). Capacity building manual for district steering groups

GoK. (2007). Basic Report on wellbeing in Kenya; Kenya Bureau of Statistics

GoK. (2007). Vision 2030, The National Economic and Social Council of Kenya (NESC)Office of the President

Gok. (2010). The constitution of Kenya

GoK. (2011). National Food and Nutrition Security Policy

GoK. (2011). Population Dynamics and Food Security in Kenya; National Coordinating Agency for population and Development; Policy briefs

GoK. (2012). Food security outlook; Kenya Food Security Steering Group

GoK. (2013). Food Security Assessment Report; Agriculture Sector working Group

Grooves, (1989). Survey errors and Survey costs, New yolk: John Wiley

Hassan, S A., (2012) Engel Curves and Equivalence Scales for Bangladesh. ASARC Working Paper Series

Hasegawa H., Kozumi H. (2001). Bayesian Analysis on Engel Curves Estimation With Measurement Errors and an Instrumental Variable, Journal of Business & Economic Statistic.

Haughton J., Khandker S R. (2009). Handbook on Poverty and Inequality. The International Bank for Reconstruction and Development/The World Bank

Hausman, J., Newey, W., Ichimura, H. and Powell, J. (1991). ‘Identification and estimation of polynomial errors-in-variables models’, Journal of Econometrics, Vol. 50, pp. 273–295.

Hausman, J., Newey,W. and Powell, J. (1995). ‘Nonlinear errors in variables estimation of some Engel curves’, Journal of Econometrics, Vol. 65, pp. 205–233.

IFPRI. (2012). Improving the Measurement of Food Security Journal of Human Resources, Vol. 33, pp.62-87

Kamau M., Githuku J., and Olwande J. (2011). Food Security in Urban Households: An Analysis of the Prevalence and Depth of Hunger in Nairobi and its Relationship to Food Expenditure, Tegemeo Institute of Agricultural Policy and Development, Egerton University

Kane, M T, (2010). Errors of measurement, Theory and Public policy, Policy Evaluation

and Research Division Policy Information Center Princeton, NJ 08541-0001

Kang’ethe W G. (2004). Agricultural Development and Food Security in Kenya, A paper prepared for Food and Agriculture Organization (FAO)

Keyzer, M A., Nubé, M., Wesenbeeck, L (2006). Estimation of under nutrition and mean calorie intake in Africa: methodology, findings and implications for Africa’s record, Centre for World Food Studies

Kuha, J. and Temple, J. (2003). ‘Covariate measurement error in quadratic regression’, International Statistical Review, Vol. 71, pp. 131–150.

Lewbel,A. (1991). ‘The rank of demand systems: theory and non-parametric estimation’, Econometrica,Vol. 94, pp. 979–1000.

Lewbel, A. (1996). ‘Demand estimation with expenditure measurement error on the left and right handside’, Review of Economics and Statistics, Vol. 78, pp. 718–725.

Liviatan, N. (1961). ‘Errors in variables and Engel curve analysis’, Econometrica, Vol. 29, pp. 336–362.

MAFAP (2013). Review of food and agricultural policies in Kenya. MAFAP Country Report Series, FAO, Rome, Italy.

Mundlak, Y. (2000) Agriculture and Economic Growth: Theory and Measurement. Cambridge, Massachusetts: Harvard University Press.

Mwakubo, S.M., Obare,G,A., Omiti,J., Mohamend,L. (2004). The influence of social capital of sustainable agriculture in marginal areas of Kenya: A case study of Machakos and Taita Taveta Districts.IFPRI Eastern Africa Food Policy Network, Network Report 13. Kampl, Uganda: IFPRI November 2004.

Mwenjeri G, W. (2009). Analysis of household food demand patterns in Laikipia district, Kenya; Unpublished Master’s thesis, Moi University

National Public Health Laboratory Services (1993) National Food Composition Tables

and the Planning of Satisfactory Diets in Kenya. Nairobi

Nicholoson, W., Snyder, C.(2008). Microeconomic theory: Basic principles and extensions, 10Th edition.

Nicholson W. (1992). Microeconomic theory: basic principles and extensions, 5th edition.

Ravallion, M. (1994) Poverty Lines in Theory and Practice. LSMS Working Paper No. 133, Washington, D.C. The World Bank

Ravallion, M. (1998) Poverty Comparisons. Fundamentals of Pure and Applied Economics, Volume 56, Chur, Switzerland: Hardwood Academic Press.

Roy, N. (2001). ‘A Semiparametric Analysis of Calorie Response to Income Change Across Income Groups and Gender?’. Journal of International Trade & Economic Development , 10 (1): 93-109

Salai-i-Martin, X. (2006). The World Distribution of Income: Falling Poverty and … Convergence, Period. In Quarterly Journal of Economics, CXXI(2): pp. 351-397.

Salois M, Taffin R, Balcombe G. (2009). Calorie and nutrient consumption: A cross country analysis. Department of Food Economics and Marketing, University of Reading

Save the children UK, (2007). Vulnerability and dependency in 4 livelihood zones of North Eastern province, Kenya

Scaccini C, Sette S, Mariotti S, Verdecchia A, Ferro-Luzzi A. Nutrient adequacy of dietary intakes of elderly. Age and Nutrition, 1992, 3 :41-47

Scott, C. (2005). Measuring up to the measurement problem: The role of statistics in evidence-based policy-making. Paris: PARIS21, Organisation for Economic Co- operation and Development. Available from: http://www.paris21.org

Sen, A.K. (1981). Poverty and Famines: An Essay on Entitlement and Deprivation, Clarendon Press, Oxford.

Skoufias E., (2003) “Is the Calorie-Income Elasticity Sensitive to Price Changes? Evidence from Indonesia,” World Development, Vol. 31, No. 7

Smith L. C., Alderman H., Aduayom D.,(2006). Food insecurity in sub-Saharan Africa: New estimates from household expenditure surveys. IFPRI Research Report Number 146.

Sudman, S. and Bradburn, N. (1973). ‘Effects of time and memory factors on response insurveys.’ Journal of the American Statistical Association, 68(344): 805- 815.Thiel, H., (1965). ‘The information approach to demand analysis In: Econometrica Vol. 33.

Thompson A., Metz,M. (1998) Implications of economic policy for food security: A training Manual.

Trochim, W. M., (2006). The Research Methods Knowledge Base, 2nd Edition. Internet WWW page, at URL: http://www.socialresearchmethods.net/kb/

UNICEF (2005). The State of the World’s Children 2006: excluded and invisible. New York: UNICEF, pp 51.

United Nations. 2006. "The Millennium Development Goals Report: 2006." United Nations Development Programme, www.undp.org/publications/MDGReport2006.pdf

Webb, N. M., Shavelson, R. J., & Haertel, E. H. (2006). Reliability Coefficients and Generalizability Theory. In C. R. Rao & S. Sinharay (Eds.), Handbook of Statistics, Vol. 26 (pp. 81-124). (PDF)

WFP (2010) Revolution: From Food Aid to Food Assistance. Rome: WFP.

Wooldridge J M., (2006). Introductory econometrics: A modern Approach

World Bank. (2010a). World development indicators 2010 Washington DC

World Food Summit, (1996). Rome Declaration on World Food Security

Working, H.(1943).Statistical laws and family expenditure. Journal of the American Statistical Association 38:43-5

APPENDIX ONE

MAP OF MANDERA COUNTY

APPENDIX TWO

QUESTIONNAIRE

1District Village Serial Number Gender of respondent Ref year: July Interviewee Name 09 – June 2010

Interviewer(s) Date Interviewee position in HH

a). Household/Family size and composition

Gender of the household head Number of people No. of wives Number of children below 5 Boys in HH Girls living/eating at No. of children Number of children age 6-12 Boys home daily Girls (include number Number of children age 12-18 Boys of wives/children Girls if polygamous + Household members age 18-30 Men extra dependents) Women Household members age 30-60 Men Women Additional Household members age >60 Men dependants Women

(3). LIVESTOCK PRODUCTION (milk, meat,)

Consu No. of lactat Averag Total Quant Pric Cas Other Balanc mptio animals ion e milk producti ity e h use (e.g. e n and milked (in product on per sold per inco gifts, consu sale of (A) days) ion per season/ or unit me payment med period = milk, (B) animal (A) x excha sol for (note per day (B) nged d labour) sour or (C) x(C) (note fresh) sour or fresh) ** High Produc tion Period Camel s milk Low Produc tion Period

High Produc tion Period Cow’s milk Low Produc tion Period

High Produc tion Period Goats milk Low Produc tion Period

Consumption Total Meat Total Sold or Price Cash Other Balance and sale of number of per meat exchan per inco use consum meat (from animals carcass (kg) ged unit me (e.g. ed own livestock) slaughtered (kg) sold gifts) Camel

Cattle

Goat

Sheep

SALE OF LIVESTOCK Total Sold Which Price per Cash income (e.g. camels, cows, goats, months? unit sold (or sheep, chickens) profit for livestock Remember to separate local fattening and export quality livestock below) sales

Camels – local quality

Cows – local quality

Goats – local quality

Sheep – local quality

Chickens

Donkey

OTHER INCOME FROM LIVESTOCK: e.g. livestock rental, livestock fattening (cattle, goats, sheep bought and sold within the same year), hides, eggs

Total Income

3). PURCHASE AND EXCHANGE of staple and non-staple FOOD for consumption (not for trade)

Unit of Quantity Frequency Total kilos Price per Total cost measure purchase purchased unit (cash Commodit and d Per week Per (= quantity or y weight (*52) month purchased*fre livestock (*12) quency) (if (e.g. exchange cereals, d)) pulses, oil, sugar, meat, wild food)

Sugar Beans

Pulses

Meat

Oil

Rice

Pasta

Wheat flour Maize posho Maize grain Milk

Salt

Tea leaves

Others ______

Purchased food & cost total 

(4). FOOD RELIEF (including free distributions and food-for-work)

Description Quantit Frequen Total Quantit Price Cash Other Balanc y cy (per receive y sold per income use e (and week or d unit (e.g. consum unit of month) sold gifts, ed measure exchang Relief grain ) e)

Pulses

Oil

CSB

School Feeding

Gifts / Zakat

Total 

3). CASUAL LABOUR / EMPLOYMENT

Unit of Number Frequency Duration Payment Total Activity / work (e.g. of people (no. of per unit cash i day, acre) doing this Per week Per month weeks or of work income n activity (*52) (*12) months) per year c o m e

s o u r c e **** Total 

**** Checklist: agricultural labour (clearing fields, preparing land, planting seeds, weeding, harvesting, threshing), digging pit latrines/wells, construction, brick making, skilled casual labour (e.g. carpentry), salaried employment, domestic work, livestock herding, pension).

1. SELF-EMPLOYMENT / SMALL BUSINESS / TRADE

Unit of Number of Frequenc Duration Price or Total cash Activity / income measure people y (no. of Profit per income source (e.g. doing this (per week weeks or unit sold †††† bundle, activity or month) months) sack, )

Total 

2. OTHER INCOME (Both Food and Cash) SOURCES – GIFTS / ZAKAT / LOANS / REMITTANCES IN CASH AND IN KIND

Unit of Frequency Duration Quantity Price per Total Activity / measure (per week or (no. of weeks unit sold cash i (e.g. bundle, month) or months) income n sack, ) per year c o m e

o u r c e Total 

†††† Checklist for self-employment: collection of firewood, charcoal, grass, handicrafts, sand collection, gum/resins, thatch/poles. Checklist for small business/trade: petty trade, trade, rental/hire, kiosks and shops.

3. EXPENDITURE PATTERNS: Obtain quantified information on the main expenditure items. Some categories are suggested below. Remember to ask about seasonal variations in expenditure.

Typical Annual Amount Spent Main Expenditure Categories Quantity (unit) purchased Frequency Price per unit Total = [a] purchased [b] [c] [a] x [b] x [c] Main food items  copy total from section 5   Other food item: Other food item: Household items Soap Toiletries Kerosene/paraffin Grinding of grain Water for humans Firewood/charcoal Jerrycans Utensils/pots/mats Inputs Livestock drugs: Salt for animals Water for animals Livestock feed Livestock investment Labour for herding Seeds, tools Fertilizers, pesticides Land rental Irrigation, pump fuel Other inputs (ag Sociallabour) services School - fees School - uniforms School - stationery Medicine – Medicineconsultation – drugs Medicine – traditional Other expenditure Loan repayments Clothing/shoes Clan Giftscontribut / Zakations/taxes Transport Festivals

Other EXPENDITURE TOTAL (REMINDER: cross check with total income) 

4. QUALITY OF INTERVIEW (confidence of informants, knowledge of area, consistency of information, etc):