2010-2011

INFRASTRUCTURE ACCESS AND HOUSEHOLD WELFARE IN RURAL GHANA: THE EMPIRICS OF THE NEXUS

Student: Mensah, Emmanuel Joseph

Promotor: Prof. Marilyne Huchet-Bourdon

Thesis submitted in partial fulfilment of the requirements for the joint academic degree of International Master of Science in Rural Development from Ghent University (Belgium), Agrocampus Ouest (France), Humboldt University of Berlin (Germany) and University of Cordoba (Spain) in collaboration with Wageningen University (The Netherlands), Slovak University of Agriculture in Nitra (Slovakia) and the University of Pisa (Italy).

This thesis was elaborated and defended at Agro-Campus Ouest, Rennes, France within the framework of the European Erasmus Mundus Programme Erasmus Mundus International Master of Science in Rural Development " (Course N 2004-0018/001- FRAME MUNB123)

ABSTRACT Access to public infrastructure services is generally identified in the development literature as critical for economic transformation. At the household level, infrastructure access can be essential for improved socio-economic well-being. However, empirical evidence on the extent to which access to different infrastructure services contribute to enhancing household welfare, especially in the developing world, remains limited. In a comprehensive review of the literature on this subject, Ayogu (2007) aptly surmise that overall the question is not about whether infrastructure matters but precisely how much it matters in different contexts? The present thesis responds to this knowledge gap by providing some quantitative understanding of how much infrastructure matters to economic development and in the context of Ghanas development experience since the 1990s. This is achieved by investigating, empirically, the direction and size of the relationship between access to three different public infrastructure services and household welfare. In so doing, the study presents a new construction of the Sustainable Livelihood Framework. This is based on the concept of relative cumulative effect of institutions in the determination of livelihood outcomes and the elaboration of livelihood assets by private and public types. This framework is tested under the premise that infrastructure services are publicly provided. Access to these public resources is therefore exogenous to household decisions though they complement household capital endowments in ways that eventually influence welfare outcomes. Welfare is thus modeled dynamically as a function of: 1) private capital endowment; 2) a vector of public assets that transpire through the manifestation of the prevailing institutions; 3) households characteristics; and, 4) households livelihood vulnerabilities. Based on pseudopanel modeling and using three waves of nation-wide livelihood surveys, the empirical findings suggests that across Ghanas rural economy, access to public transport, electricity and water infrastructure have important but differential impact on household welfare, ceteris paribus. Whereas the impact of access to public transport system and electricity is found to be significant and consistently positive, that of water depends on the households endowment in private assets and the opportunity costs it faces. Focusing on agricultural households, access to public transport shows significant and positive impact on welfare, irrespective of asset endowment. Development policy in Ghana should therefore prioritize public investment in functional rural transport systems. To maximize the welfare impact, provision of electricity and water is recommended to be targeted at rural communities with sufficient transport networks and private capital endowments.
i

CONTENT
ABSTRACT............................................................................................................................................. i LIST OF TABLE ................................................................................................................................... iii LIST OF FIGURES ................................................................................................................................ iii LIST OF ACRONYMS .......................................................................................................................... iv DECLARATION .................................................................................................................................... v ACKNOWELDGEMENT ...................................................................................................................... vi 1.0 BACKGROUND ......................................................................................................................... 1 1.1 Problem Statement ................................................................................................................... 4 1.2 Objectives of the Thesis ........................................................................................................... 4 1.3 Justification and Rationale of the Study .................................................................................... 5 1.4 Organization of the Thesis ........................................................................................................ 7 2.0 CONCEPTUAL FRAMEWORK ................................................................................................. 8 2.1.1 On the Determination of Welfare.............................................................................................. 8 2.1.2 The Sustainable Livelihood Framework.................................................................................. 11 2.1.3 Livelihood Assets in the SLF: An Elaboration ........................................................................ 13 3.0 LITERATURE REVIEW ........................................................................................................... 18 3.1 Ghana: A Country Overview .................................................................................................. 18 3.1.1 Ghana: Background to the Policy & Development Experience since the 1980s................ 19 3.2 Infrastructure Services and the Development Nexus ............................................................... 22 3.3 The State of Ghanas Rural Economy ................................................................................. 26 4.0 METHODOLOGY .................................................................................................................... 31 4.1 The Generic Model ................................................................................................................ 31 4.2 The Pseudo-Panel Modeling Framework: An Overview.......................................................... 32 4.2.1 The Pseudo-Panel Modeling Framework: The Basic Model .................................................... 34 4.2.2 The Pseudo-Panel Modeling Framework: The Dynamic Model .............................................. 39 4.3 The Empirical Model, Definition of Variables and the Postulated Hypotheses ........................ 41 4.2 Construction of an Index of Household Private Asset Endowment .......................................... 47 4.3 Data, Data Sources and the Construction of the Pseudo-Panel Database.................................. 48 4.4 The Estimation Method .......................................................................................................... 52 5.0 DISCUSSION OF RESULTS .................................................................................................... 54 5.1 Qualitative Analyses........................................................................................................... 54 5.2 Quantitative Analysis: Estimated Models............................................................................ 56 5.2.1 The Principal Model for Rural Ghana ............................................................................. 56 6.0 POLICY RECOMMENDATIONS AND CONCLUSIONS........................................................ 64 6.1 Policy Recommendations ....................................................................................................... 64 6.2 Limitations of the Thesis and Suggestions for Future Research ............................................... 64 6.3 Conclusions ........................................................................................................................... 66 7.0 REFERENCES .......................................................................................................................... 67 8.0 APPENDIX ............................................................................................................................... 71

ii

LIST OF TABLE
TABLE 4-1: AN EMPIRICAL TRANSLATION OF THE CONCEPTUAL FRAMEWORK & MODEL HYPOTHESIS ............................47 TABLE 5-1: THE E VOLUTION OF NATIONAL AND RURAL POVERTY IN GHANA, 1991 - 2005 .............................................54 TABLE 5-2: INFRASTRUCTURE ACCESS & HOUSEHOLD WELFARE: WITHIN ESTIMATION OF MODEL (11) FOR RURAL GHANA .........................................................................................................................................................57 TABLE 5-3: INFRASTRUCTURE ACCESS & HOUSEHOLD WELFARE: GMM ESTIMATION OF MODEL (11) FOR RURAL GHANA .........................................................................................................................................................58 TABLE 8-1: THE E VOLUTION OF GHANAS ECONOMIC STRUCTURE SINCE THE 1980S ...................................................71 TABLE 8-2: TRENDS IN AGRICULTURAL LAND ALLOCATION AND PRODUCTIVITY IN GHANA, 1994 - 2006 .........................72 TABLE 8-3: PROPORTION OF HOUSEHOLDS MAIN SOURCES OF WATER IN RURAL GHANA, GLSS 3 5 (%) ....................73 TABLE 8-4: PROPORTION OF HOUSEHOLDS MAIN TOILET FACILITY, GLSS 3 5 BY AGRO-ECOLOGICAL ZONE (%) ........73 TABLE 8-5: SUMMARY OF SOME KEY LITERATURE ON THE ESTIMATION OF PSEUDO-PANEL MODELS .............................74 TABLE 8-6: PRINCIPAL COMPONENT ANALYSIS OF HOUSEHOLD ASSET OWNERSHIP, SUMMARY RESULTS ........................75 TABLE 8-7: SAMPLE DISTRIBUTION OF RURAL HOUSEHOLDS BY ECOLOGICAL ZONE/GENERATION COHORTS ..................76 TABLE 8-8: DESCRIPTIVE STATISTICS OF KEY VARIABLES RURAL SAMPLE POOLED DATA ............................................76 TABLE 8-9: DESCRIPTIVE STATISTICS OF KEY VARIABLES RURAL SAMPLE PSEUDO-PANEL ..........................................77 TABLE 8-10: SAMPLE DISTRIBUTION OF AGRO-HOUSEHOLDS BY ECOLOGICAL ZONE/GENERATION COHORTS .................77 TABLE 8-11: SAMPLE DISTRIBUTION OF NON-AGRO-HOUSEHOLDS BY ECOLOGICAL ZONE/GENERATION COHORTS .........78 TABLE 8-12: MODEL ESTIMATES OF INFRASTRUCTURE ACCESS & HOUSEHOLD WELFARE, RURAL GHANA ......................79 TABLE 8-13: MODEL ESTIMATES OF INFRASTRUCTURE ACCESS & HOUSEHOLD WELFARE, AGRO-HOUSEHOLDS .............80 TABLE 8-14: MODEL ESTIMATES OF INFRASTRUCTURE ACCESS & HOUSEHOLD WELFARE, NON-AGRO-HOUSEHOLDS......81

LIST OF FIGURES
FIGURE 2-1: THE SUSTAINABLE LIVELIHOOD FRAMEWORK: A RECONSTRUCTION .........................................................16 FIGURE 3-1: GHANA: MAP OF LOCAL & INTERNATIONAL ADMINISTRATIVE BOUNDARIES ..............................................19 FIGURE 5-1: TRENDS IN THE HEADCOUNT OF INDIVIDUALS WITH INFRASTRUCTURE ACCESS, 1991 TO 2005, PERCENT ....56 FIGURE 8-1: THE SUSTAINABLE LIVELIHOODS FRAMEWORK (BASED ON THE DFID SCHEMA) ...................................71 FIGURE 8-2: TRENDS IN INTERNATIONAL TRADE AND FINANCE (% OF GDP) SINCE THE 1980S ......................................72 FIGURE 8-3: PROPORTION OF HOUSEHOLD CONNECTED TO THE NATIONAL POWER GRID, GLSS 3 5 (%)....................73

iii

LIST OF ACRONYMS AFD AfDB DfID FAO GLSS GSS IFAD IFPRI LSMS MT PCA SLA SLF SSA UN UNDP UNECA UN-MDG USD EC ISIC IV UNIDO Agence Franaise de Dveloppement African Development Bank Department for International Development Food and Agriculture Organization Ghana Living Standards Survey Ghana Statistical Service International Fund for Agricultural Development International Food Policy and Research Institute Living Standard Measurement Survey Ministry of Transport Principal Component Analysis Sustainable Livelihood Approach Sustainable Livelihood Framework Sub-Sahara Africa United Nations United Nations Development Fund United Nations Economic Commission for Africa United Nations Millennium Development Goals United States Dollars European Commission International Standard Industrial Classification Instrumental Variables United Nations Industrial Development Organization

iv

ACKNOWELDGEMENT

To Him who is able to do much more than we can think or dream of be all Glory, Honour and Praise, Amen!

This thesis has benefitted immensely from the wisdom and encouragement of a considerable number of people. I am particularly grateful to my Supervisor, Prof. Marilyn Huchet-Bourdon for the warm reception of the thesis and mobilization of support for the purpose. I am also grateful to Professors Laure Latruffe, Alain Carpentier and Angela Cheptea for the time and comments that have helped shape this study.

My appreciation also goes to the European Unions Erasmus Mundus program, under which I have received financial and technical assistance in pursuing my studies. I am greatly indebted to the Professors and staff of the IMRD/Atlantis network across Europe, especially Professors Van Huylenbroeck, Guy Durand, Konrad Hagedon , Anna Bandlerova, Dr. Marijke DHaese and Gerrit Stassyns. Also, to Professors Jonathan Haughton, Tillman Brueck, Boris Bravo-Ureta, Edward Clay, Carlos Bozzoli and all Professors/Researchers at DERG/INRA, Rennes; I can only say, Thank You.

I must also express my sincere appreciation to my friends from IMRD program and in Agrocampus, Ghent and Humboldt, and of course the massive network of paddies across Africa, Europe, Asia and the US for your continued company and kind support. Thanks Joan and CDlele lol!

Last but definitely not the least, much love and appreciation goes to my Mom, brothers and sister. I know you know and you know I know that you are much appreciated in my life. Thanks a billion, always!

vi

1.0

BACKGROUND

Ghana, in recent years, has been credited to be one of the strong performing economies in SubSahara Africa (SSA). The country lies along the Gulf of Guinea to the south, sandwiched on land by Togo to the east, Cote dIvoire to the west and Burkina Faso to the north. It is situated on a total land area of 238,533 sq km with a population of about 24 million and annual intercensal growth rate of 2.7 (GSS, 2011). According to the World Bank (2011) and Mensah et al. (2009a), the West African country has maintained considerable stability in economic growth and political governance since the late 1980s. Its economy has grown in real terms by over 5 percent since 2000 and continues to exhibit good resilience to especially adverse developments from the external sector, including the recent economic and financial crisis.

In 2005, the countrys index of extreme poverty was estimated at 18 percent (GSS, 2006). Relative to the estimated 37 percent of the population that were identified to be extremely poor in 1991, Ghana effectively achieved the United Nations Millennium Development Goal (UNMDG) of halving extreme poverty, well ahead of the 2015 target year. However, sustaining these trends and sharing the benefits of the development experience is considerably constrained by a number of factors. Prominent among these is the pervasiveness of weak physical infrastructure provision and the associated inequality in spatial distribution and access by households.

Generally, the development literature argues for improved access to physical infrastructure services for sustained and rapid development. The Commission for Africa (2005) has observed, for instance, that infrastructure access is critical for improving competitiveness in trade and commerce. This is especially linked to improved access to markets and mitigation of transaction costs arising from market fragmentation, information asymmetry and geographic disadvantages. Better access to infrastructure is also argued by the paper to help enhance opportunities for human and capital formation, the creation of productive employment opportunities and deepening of scale economies for business and industrial growth. As further argued by HuchetBourdon et al. (2010), economic infrastructure projects, while stimulating inflow of intermediate goods could also engender productivity growth of local industries through linkages with the domestic sector. Stiglitz (2011) stresses that widespread weaknesses in infrastructure access form
1

more critical barrier to trade in the third world. Thus, for developing economies such as those in Africa generally and Ghana in particular, improved access to infrastructure is just fundamental for the agenda of sustained growth and development.

However, the extent to which these economies can deliver access to infrastructure services in the right amounts and quality, and towards achieving an optimal mix for catalyzing their development processes remain a major challenge. Underlying this challenge is the limited and highly over-stretched fiscal resource endowments for such investments. In particular, the delivery of physical infrastructure services1 is exceptionally capital intensive. In the absence of cost-effective, home-grown technologies for the construction and manufacturing sectors of many African economies, infrastructure service provision is made even more expensive. This is in terms of both the direct and the indirect costs arising especially from the use of foreign exchange reserves in the acquisition of the requisite technology and expertise.

As a result of this and accounting for the existing deficit, the Commission for Africa (op. cit), for example, recommends an annual expenditure of about 5 percent of GDP in the development of new infrastructure projects and 4 percent of the same for operations and maintenance, for periods between 2005 and 2015. For Africa as a whole, this sums up to about USD 20 billion. Even more recent estimates from an AFD/ World Bank country diagnostic study suggests a total annual requirement of USD 93 billion for up to the year 2020 (AfDB, 2011; Foster, 2010).

In Ghana, the annual expenditure requirement is estimated by the government to be about USD 1.6 billion, for a period of 10 years (World Bank, 2011). In relation to this requirement, the country already faces a huge financial gap. The medium-term shortfall in this expenditure requirement is currently estimated to be 5-6 percent of GDP, whereas the short-term deficit is estimated at 3-4 percent of GDP. Needless to say, this is a major fiscal challenge.

Throughout this study, access to infrastructure service is used to connote existence, in close proximity, of functional public physical infrastructure such as roads and public transport, electricity, portable water and communication. The magnitude of investment required to build such investment allows one to hold as true the assumption that private households generally depend on the provision of the public sector in accessing such infrastructure services, either free or under user-pay arrangements.

In the view of the present thesis, addressing such challenge requires efficient and strategic allocation of the available resources while deepening the resource base for increased financing. In the context of contemporary development policy and practice, efficiency in the targeting of resource allocation, especially to assure optimal returns without compromising on equity considerations, is itself an outstanding challenge in many developing economies.

As already observed by Masika and Baden (1997), considerable biases arise in the provision of infrastructure services in many economies of the world. In tropical Africa, Bates (1981) labels such trend as reflecting the urban-bias in the sharing of the development experience. In Ghana, Mensah et al. (2009) document evidences of spatial inequity in the distribution of basic public infrastructure services. IFPRI (2005) also makes similar observations. More recently, the MT and GSS (2008) have provided some current estimates that shows disproportionate allocation of public investment in favour of infrastructure projects for the urban sector, relative to the rural economy. This trend is generally consistent with the dominant arguments postulated in the public policy and political economy literature. These suggest that in the presence of relatively weaker collective actions from the rural sector, development policy will consistently be biased in favour of the urban sector. Hence, without proper targeting, the provision of public infrastructure services and related investments, rather than reducing poverty and bridging any existing rural/urban economic inequality, may instead exacerbate it.

Safeguarding such scenarios to ensure equity in access and benefits of infrastructure services on household socio-economic well-being therefore requires an understanding of the dynamics of the linkage between infrastructure access and household welfare. Here, the present study submits that different households may use different infrastructure services differently, and the consequent impact in terms of the direct and indirect benefits on welfare could differ significantly. Achieving equity in both access and actual benefits therefore requires better understanding of the dynamics of household welfare determination. In the provision of different mix of infrastructure services for the different household types in an economic system, such understanding must enable policy-makers to determine how much different infrastructure services materialize in benefits to different households. This, in turn, will inform decisions on how to better target the

provision of such services, within the scope of the already scarce and the ever fungible development resources.

For Ghana, empirically robust evidence on the direction and size of the relationship between infrastructure services and household welfare is sine qua non for sustaining its development agenda. This is particularly so as such knowledge is critical for enhancing policy efficiency and optimal application of the available development budget

1.1 Problem Statement

Empirical arguments on the extent to which access to different infrastructure services contribute to enhancing household welfare, especially in the developing world, remains limited. For instance, empirically, what is the quantitative impact of access to infrastructure services on household welfare? Do different infrastructure services impact on the well-being of households equally? And, is there any important differential impact on welfare of access to different infrastructure services by different categories of households in Ghana? These are some questions that the contemporary literature provides limited insight. They therefore constitute the core research questions that the present thesis attempts to address, keeping particular focus on rural Ghana.

1.2 Objectives of the Thesis

In the light of the above, the primary objective of this thesis is to quantify the differential impact of infrastructure access on household welfare, focusing on rural Ghana. The specific objectives are to: 1. establish the direction and magnitude of the relationship between access to different infrastructure services and household welfare in Ghana; and, 2. evaluate whether there is any important differences in impact of infrastructure services on the welfare of different household types in the country.

1.3 Justification and Rationale of the Study

The present study draws its significance from three main dimensions; that is, the specific contribution the study makes to the consolidation of the existing literature; the development of some empirical knowledge for informing economic policy; and, the value addition to existing investments in socio-economic research infrastructure in Africa. These are elaborated in more details below.

In terms of contemporary development research, an important gap is already identified in the existing literature, to which the results of this study are expected to make significant contribution in bridging. This gap is identified particularly by the work of Ayogu et al. (2007). Based on a comprehensive review of the existing literature on the link between infrastructure and economic development, their paper draws a terse conclusion which establishes the view that overall the question is not about whether infrastructure matters but precisely how much it matters in different contexts? (p. 1). While acknowledging the extensive investments in both theoretical and empirical research on the question of the link between infrastructure and economic development generally, Ayogu et al. observe that focus has largely been on responding to the question of whether infrastructure matter at all in economic growth and development. This, the paper argue, has been done at the expense of providing rigorous, policy-relevant knowledge on exactly how much infrastructure access matters and in what context of the development process.

The consequence is the existing paucity of knowledge on the extent of impact of infrastructure in different contexts of the socio-economic transformation processes. In seeking to understand the dynamics of impact of different infrastructure services in the determination of household welfare in Ghana, this study readily responds to the challenge of helping bridge this knowledge gap.

In achieving this, the study further makes extensive application of varying socio-economic concepts and theories. In the particular instance of the core analytical framework employed in the thesis (that is, the Sustainable Livelihood Approach), the thesis applies existing knowledge to provide a new construction of the framework by emphasizing the role of institutions and asset types in the determination of household welfare. The study then proceeds further to provide
5

theoretically robust and empirically meaningful application of the approach for household welfare analysis. This is no mean contribution to knowledge formation.

Further, the scale of analysis of the study allows the results to be evaluated at the country-level, which could also be replicated for different countries using a widely credible and available database and for a study period spanning over two decades. In relation to the analytical approach, pseudo-panel modeling is innovative and rigorous methodology. The modeling approach optimizes contemporary scientific understanding about the use of both cross-sectional and panel data observations to draw inferences about a population and the dynamics of household behavior. It therefore enables efficient use of data resources from independent cross-sectional studies to evaluate dynamic relationships that hitherto is usually possible with genuine panel observations. The literature on the application of this methodology is still evolving and imbued with considerable potential. The present study is therefore situated at the forefront of this process of knowledge formation and makes important contribution in this course.

In terms of development policy and practice, the existing literature still fail to provide adequate empirical understanding of the channel by which different households utilize different infrastructure services in income formation. Particularly, within the context of the heterogeneity in household endowment and overall capability, the use, and therefore withdrawal, of the benefits of infrastructure could materialize to different households at different levels and scales.

As an example, a resource-poor, food-crop subsistence farmer could be postulated to use a road or communication infrastructure at a scale much lower than a resource-rich, cash-crop producer. Such trend could also be the case for the withdrawal of the indirect benefits (including the existence value) of the infrastructure to these different households. To what extent does the difference in resource endowments and economic activity condition the use and impact of public infrastructure assets to these different households? What specific public infrastructure services can policy deliver for optimal economic returns to different household groups? To the extent that these questions remain central to any important use of scientific knowledge in policy decisionmaking, the inability of the current literature to adequately address these questions represent an important knowledge-gap that must be bridged. For developing economies, this is made even
6

more compelling by the varying, country-specific socio-economic and political conditions that tend to create different dynamics of policy outcomes.

For rural Africa, the results of this study are expected to contribute to forging better appreciation of policy effectiveness and efficient allocation of public resources. In particular, by investigating the different contexts under which different infrastructure services impact on the welfare of different groups of the rural population in Ghana, the study will be furnishing policy with critical knowledge required for better targeting of scarce public investments.

The thesis makes extensive use of existing data resources in Ghana. Especially, massive investments are required to generate multi-subject, nationwide, representative household surveys of the kind that the Living Standard surveys follow. And while such surveys are financed from both local and international public resources, their justification and consequent sustenance depends so much on the extent to which previous data resources are utilized to inform policy, enhance knowledge and demonstrate in practical terms the relevance of such large scale exercises. The results of the thesis are expected to contribute to providing such visibility to these investments. Admittedly, the use of these data at much disaggregated levels is very tasking and requires extensive knowledge of the instruments and data management skills. From the perspective of this thesis, this is an outstanding challenge for the analyst and counts as an important dimension of the value addition that this thesis makes to such public investments.

Hence, the core rationale and justification underlying this thesis are very significant and truly non-trivial.

1.4 Organization of the Thesis

The thesis is structured into six main chapters. In Chapter two, the conceptual framework of the study is presented, and then a review of the relevant literature is presented in Chapter three. The methodology of the study is presented in Chapter four, while Chapter five discusses the results. The key recommendations for policy, limitations of the study and suggestions for future research are then presented in Chapter six, alongside the thesis conclusions.
7

2.0

CONCEPTUAL FRAMEWORK

The analysis of this thesis is set within the framework of the Sustainable Livelihood Approach (SLA). In adopting this framework, two key issues are first clarified below.

First, the SLA has been discussed and treated in the development discourse variously as an analytical framework, a development objective and even an approach to policy decision-making (Clarke and Darney, 2008; Maunder, et al, 2001; Ashley and Carney, 1999). It must be clarified that for the present study, the approach is treated purely as a framework for analysis. Its relevance and use in the present study is therefore grounded on the utility of the framework as a basis for disaggregating the obviously complex socio-economic interactions under investigation. Here, one will agree with Ashley and Carney (op. cit.) on the view that the SLA is an analytical structure that provides a way of thinking about livelihoods that is more representative of a complex, holistic reality but is also manageable (Ashley and Carney, op. cit., p. 47).

Second, while the background and popularity of the SLA is linked to the work of DfID (1999), earlier studies, in particular Sen (1981) and Chambers and Conway (1991) present relatively good starting points for a good analytical preview of the approach.

The discussion of the analytical framework will therefore begin from these two papers, followed by the more familiar framework contained in DfID (1999) and the subsequent studies and applications that have evolved around it. The critical essence of this preview is to communicate the basic fact that the SLA has evolved from other conceptual discourse. It is therefore not sacrosanct. It is indeed amenable to any analytical review which will be deemed necessary for the purpose of the present study while preserving its core utility as an analytical framework.

2.1.1 On the Determination of Welfare

Conceptual analyses of the factors that condition the determination of welfare have been achieved in the economic literature with varying approaches. At the height of global policy and research debate on poverty and famine in the early 1980s, Sen (1981) contributed a seminal
8

study that seems to have extensively influenced the discourse on the subject, even up to the present time.

Particularly, narrowing the argument down to the pattern of incidence of famine in some selected developing economies, the paper argues that famine (and therefore poverty) in the developing world is not explained by the long-held conception of food supply inadequacies or what had been described broadly as the argument of food availability decline (FAD). Rather, the paper proposes the exchange entitlement argument. This argument contends that access to basic needs is determined in society by legally acquired ability to obtain life needs through exchange with own produce and capabilities, subject to the mediating role of any prevailing institutions and processes that define the socio-economic order. At the individual level, this perspective then implies that an individuals exchange entitlement is dependent on the level of asset endowment.

From this view, Sen (op. cit.) thus postulates the assets and capabilities as the single fundamental factor defining the life-choices and strategies that determine individuals welfare outcomes, not discounting the mediating role of the policy environment that conditions the exercise of this entitlement. To this perspective, Chambers and Conway (1991) reemphasized the capabilities argument and introduced it alongside the equity and sustainability dimensions of livelihood to provoke further thinking on what has been commonly referred to as the Sustainable Livelihood approach or framework (SLA/SLF)2.

In building their thesis, Chambers and Conway (op. cit.) first rejected the then prevailing conception that poverty and famine are problems of production inadequacies, unemployment and the monetization of socio-economic wellbeing (or what is described generally as the poverty line thinking). They argue that these approaches are reductionist in nature and fail to account for the varied dimensions of good livelihood, both in terms of the diversity of life and life needs and intergenerational dynamics. On the contrary, they propose a conception of livelihood that integrates the fundamental attributes of capability, equity and sustainability.

SLA and SLF are synonymous in references and therefore used interchangeably in this thesis

Following from Sen (1981), Chambers and Conways use of capability is expressed as the ability by households and individuals to access and exploit livelihood resources, construct and pursue livelihood strategies towards achieving a desired welfare outcome. Capability is also argued to embody the capacity to adjust to new conditions and situations, respond favorably to shocks and pro-actively adjust to the dynamics of livelihood trends and conditions. The absence of these increases households vulnerabilities to famine and low living standards (poverty).

Equity, on the other hand, addresses the trend and distribution of the quality of life arising from the welfare outcomes of livelihood strategies. The concept is therefore adopted to debate the pattern of access to livelihood resources and opportunities. It therefore discusses the disproportionate distribution in well-being across the broad spectrum of a population. The most important issue is how such distributional inequity could pre-condition the incidence and degree of exposure to poverty and famine, as experienced by some segments of the population.

On the concept of sustainability, the paper expresses its use to connote responsible exploitation of livelihood resources in a manner that assure intergenerational equity in access and use. It therefore brings into perspective, the need to maintain adequacy in asset levels and quality over time, while addressing present needs and development challenges. As noted by the paper, the use of the sustainability concept could therefore be regarded as the social dimension of the same concept used in the environmental resource disciplines to discuss global phenomena such as deforestation, exploitation of natural resources, pollution and global warming, among others.

All together, these three core concepts (of capability, equity and sustainability) constitute the sustainable livelihood paradigm of development thinking. This, the paper defines as comprising: the capabilities, assets (stores, resources, claims and access) and activities required for a means of living: a livelihood is sustainable which can cope with and recover from stress and shocks, maintain or enhance its capabilities and assets, and provide sustainable livelihood opportunities for the next generation; and which contributes net benefits to other livelihoods at the local and global and in the short and long term (Chambers and Conway, 1991, p. 6). Generally, the SLF can be judged to have achieved considerable level of success. At least, on the basis of the framework, development policy and research since the late 1990s have shifted
10

significantly towards the interpretation of poverty and economic deprivation as results of weak capabilities and assets of the affected economic groups and households. Extensive application of the framework is reported at both local and global levels. Some prominent international development actors that have actively employed the framework include among others, the UN system (especially UNDP, FAO, UNECA), Oxfam, the European Commission, CARE International, World Bank, International Fund for Agricultural Development (IFAD), (Hussein, 2002).

This notwithstanding, some criticisms persist on the SLF. In the specific context of the present thesis, a submission will be made on the conceptual treatment of some aspects of the framework, especially assets and institutions. This is to enable proper attribution of functions and effect of assets, especially in informing households livelihood choices and the determination of welfare outcomes. This discussion is however preceded by some detailed outline of the standard arguments in the SLF, based on DfID (1999).

2.1.2 The Sustainable Livelihood Framework

The SLF posits that households possess different levels of resource endowment and capabilities, endure different scales of exposure to the institutions and policies that condition the environment in which they operate, and that the interaction of these factors determine their livelihood choices and the consequent differences in welfare outcomes. In the different applications of the SLF therefore, considerable emphasis is placed on the core issue of individual and household endowments. Figure 8-1 presents a schema of the SLF as reported in DfID (1999).

According to Ashley and Carney (1999), DfID (1999) and Scoone (1997), in the SLF individuals and households are the focus of the analysis. In the different uses and adaptations of the framework, these papers identify the vulnerabilities of the poor in societies as the core challenge in the design and implementation of development interventions. In so doing, SLF identifies five broad categories of resources from which individuals may determine their production possibilities, especially within the context of the shocks, trends and seasonality of their

11

livelihood and in the light of the institutional structures and processes that they confront. These resource groups are: 1. Natural resources including soil, water, genetic materials and environmental services; 2. Social capital, which embodies the social networks, affiliations and associations; 3. Human capital such as labour resources, skills and knowledge base; 4. Physical capital including infrastructure and production equipments; and, 5. Financial capital, encompassing cash, credit and debts, savings and other assets.

Depending on the level of endowment in these resource groups, individuals construct and identify possible livelihood strategies that would yield optimal returns in welfare outcomes such as increased income, reduced vulnerability to economic shocks and improved food security. However, decisions on such choices of livelihood strategies are not independent of the institutional processes and structures that dictate the order of economic interactions. Some of these include formal laws and social expectations, culture and societal sensitivities, legislative regimes and rules of economic exchange. In fact, beyond own endowments, the framework observes that these institutional arrangements, political organizations and power relations may generate on their own different levels of access to these aforementioned livelihood resources. This in turn will determine different combinations of livelihood activities to be pursued and their possible outcomes. The key role of institutional and policy factors in the framework is therefore the extent of their influence on access to livelihood resources, construction of livelihood portfolios and the eventual determination of livelihood outcomes (Scoone, 1998).

On the merit of the SLA therefore, the welfare of household groups is postulated to be a function of the household resource endowments, the trends, conditions and context of the livelihood formation processes as well as the institutional and policy environment that condition the economic and social exchanges. The definition of physical resources in the framework as including physical infrastructure then implies that conceptually, welfare could specifically be postulated to be a function of the different levels of access that these households have to physical infrastructure services such as electricity and roads. An empirical understanding of the nexus between welfare and infrastructure access could therefore be achieved in this framework.

12

2.1.3 Livelihood Assets in the SLF: An Elaboration

It is easy enough to make models on stated assumptions. The difficulty is to find the assumptions that are relevant to reality. The art is to set up a scheme that simplifies the problem so as to make it manageable without eliminating the essential character of the actual situation on which it is intended to throw light (Robinson, 1971, p. 141).

Before proceeding on the adoption of the SLF in the specification of the analytical model for the current study, an alternative elaboration of the conceptual aggregation of livelihood resources in the SLF and how these transpire in the determination of the welfare outcomes is presented below. This is motivated by Robinson (1971) in the statement quoted above and based on the need to provide an explicit distinction between two sets of livelihood assets necessary for the intended analysis of the present study; viz, those assets that households are able to hold and control, whose levels and user rights are directly determined by the decisions and behavior of households themselves, and, those assets that occur mainly as outcomes of policy decisions, whose levels and access are exogenous to the decisions and behavior of households and to which no (explicit) private user rights or control can be exercised by the household. The former set of assets is described here as household private endowments. The latter set embodies those that will be regarded as public goods and services (or public capital). In particular, it is contended that a better approach to the analytical treatment household endowments and their relations with the institutions is the further elaboration of resources as originating either from the public sector or in private ownership. Here, the original SLF does not make any explicit distinction between resources to which individuals have private, legitimate tenure right and those of public nature. Resources in the framework are broadly treated as though they remain in the domain of households and that the prevailing institutional arrangement only mediates their access/ use. As mentioned earlier, this thesis proposes the view that some of these resources are public in nature; they are themselves outcomes of the institutional structure, social arrangements and political processes. In many respects, they are in reality, the manifestation of these institutional structure and processes (both formal and informal). They may be as tangible as physical infrastructure and their distribution across the urban and rural segments of an economy and as intangible as the meanings and interpretations given to gender, ethnic differences or macroeconomic policy regime of a society.

13

It should also be noted that such public resources are also facilitative (or complementary) in function and do not occur as the core resource-base that households maintain, control and exploit for production and exchange. As complementary assets, they inform households appreciation of the real value and utility of their asset endowment, the opportunities and livelihood choices available and the actual outturn in overall welfare outcomes. Therefore, the access and utility of this resource category to households assume a form and character that is systematically distinct from those resources to which these agents can exercise private tenure rights. Even more, the level and access possibilities of these resources are just exogenous to the decisions and behavioral patterns of households. Hence, a framework for a critical analysis of livelihood must explicitly invoke and apply such distinction in assets3. For instance, in the treatment of resource endowment, an implicit assumption is made that these resources are at the disposal of households and could be commandeered for production and other uses at will (as can be done with private resources). In reality, this is not the case. While households and microcredit schemes can do so much as to build financial capital for livelihood improvements, the competitiveness of the macroeconomic environment (especially in terms of price stability and market incentives) depends on the decisions of the public authority. Similarly, the quality and productivity of private natural resources such as agricultural land and aquifers are ones that depend on the actions and decisions of the household with the tenure right. But the supporting landscape such as the air and water pollution sink exist as public resources whose use (and misuse) depends on the state of public policy and the enforcement regime. Expressed in a more radical sense, this thesis submits that institutions per se do not matter at all in the determination of household livelihood4. Rather, it is how they manifest themselves by way of the quality of governance, facilitation of economic exchange, interpretation and enforcement of rules and the creation of public capital that is of practical relevance to the determination of
3

While the difference between public and private assets may appear natural and even trivial from the face value, the analytical essence could be likened to the differences in the roles of governments in capitalist and communist economic regimes. For livelihood analysis, this is very significant and cannot be taken lightly. 4 Here, one will agree with Udry et al. (2005, p.2) on the view that irrespective of the form in which institutions emerge in socio-economic interactions, their actual operations may be quite different than intended. Hence, the mere existence and motives of a Central Bank, for example, is in-material to a households livelihood decisions; rather, it is how the operations of the Central Bank manifests in price levels and asset values that households will deem material in constructing asset portfolio and forming livelihood strategies.

14

household welfare. Further, for any institution that manifest in the determination of livelihood outcomes, such practical relevance is conceived by households over time as the relative cumulative effect of the given institution. That is, the totality of the material impact of an institution on households real (net) livelihood outturn. On the same basis, households feed into the evolution of a given institution/policy process depending on the magnitude and direction of this relative cumulative effect. Here, one will hold as true the assumption that households are rational economic agents. In pursuit of their livelihood expectations, the welfare-maximizing tendencies of these households should therefore be expected to follow Garreth Hardins type of economic agents (see Harding [1968] in the Tragedy of the Commons). Institutions that exhibit negative relative cumulative effect on livelihood outcomes would thus be resisted (or undermined), generally. To adapt, such institutions would respond along one of three possible evolutionary pathways; namely, 1) transform themselves to follow the larger expectations of households; 2) neutralize their existence; and, 3) work to bring the expectation of households in tune with their effects on livelihood outcomes. This is the second thesis underlying the elaboration of the original SLF, as suggested earlier.

Specifically, in the original framework, sustainability is assumed to arise from household welfare expectations and therefore exogenous (see Figure 8-1). As Harding (1968) demonstrates, this assumption is inconsistent with the time-proven knowledge that households are inherently welfare-maximizing. This thesis corrects for this analytical error by postulating the sustainability of any development path as arising endogenously in the framework via a balance between households livelihood expectations and the direction of institutional evolution. This mechanism is explicitly introduced in the current framework as the cumulative feedback from households to institutions, which is itself driven by the relative cumulative effect of institutions on livelihood outcomes5. Figure 2-1 provides a schematic outline of these elaborations (following the schema reported in Figure 8-1).

This cumulative feedback is also argued to assume 5 different distinctions as livelihood assets. That is, physical, financial, social, natural and human. In the current context, increased tax revenue to local and national (public) authorities, electoral support or solidarity actions, are typical examples.

15

Figure 2-1: The Sustainable Livelihood Framework: A Reconstruction

Source: Author

Conceptually, it can therefore be postulated within the framework of the SLA and in the light of the elaborations proposed above that household welfare is a function of 1) private capital held by households; 2) a vector of public capital that transpire to households through the manifestation of the prevailing institutional structure/ policy processes; 3) households characteristics (attributes); and, 4) factors that define the vulnerability context of the livelihoods under consideration. It can further be argued that an institutional structure and development policy can be pro-poor and effective if it is able to perceive heterogeneity in household capital endowment and properly target provision of public capital in ways that optimally complement households endowments.

In the broad context of livelihood analysis, the elaboration argued above reinvigorates the SLF as a micro-macro analytical tool. In particular, to the extent that a households access, use and utility of public resources are defined by factors exogenous to these households (and contrary to what will be expected of a households own endowment), the distinction brings into better perspective the role of institutions and the macro-sector in the determination of household
16

welfare. Here, the criticism that the SLF is too micro, too household focused is clearly avoided (Clarke and Darney, 2008). In fact, since both resource types are identified at the same scale and relevance in Figure 2-1, one cannot be emphasized over the other. This treatment also helps in addressing some of the other outstanding criticisms of the SLF. For instance, as noted by Clarke and Darney (2008), notwithstanding the integrative, crosssectoral approach that the SLF may be argued to advocate, the idea of maintaining people and the priorities of the poor at the centre of policy thinking consistently creates a loss of balance. At the end, a sustained, generalized disposition towards building household assets as a response to poverty and economic vulnerabilities dominate. Here, this thesis argues that such tendency arises mainly because assets in the framework are narrowly presented. That is, where assets are dominantly viewed purely from the perspective of what households are able to construct and administer, then the frequent reality of the corrosive effects on private assets of institutional failures and macro-policies inadequacies will always be discounted. In terms of development research and practice, a reflection of this tendency is documented in Clark and Darney (2008). Based on a review of the experiences of DfID in the implementation of the SLA since its adoption in the late 1990s, the paper suggests that this loss in balance led to the redundancy of the Health/Education and Private Sector Development groups in the implementation of DfIDs interventions using the framework. In more recent times, the inflexibility in adapting the approach to changing trends in international development assistance, especially as related to the increased shift towards institutional building, the MDGs and sectorwide support programmes, could explain the relegation of the framework to the periphery.

Ironically, this set of activities is the very kind that consolidates knowledge, builds public capital and creates better complements to household private assets. They also provide the critical basis for institutional development to balance households welfare expectations, upon which the sustainability of a societys development path are established. With the elaboration above, the proposed framework should help development actors escape this analytical trap.

17

3.0

LITERATURE REVIEW

The current chapter provides an overview of the existing literature on Ghana and the evolution of economic policy since the 1980s. The 1980s is chosen as the base for the review as it reflects the era of the onset of the structural adjustment program (of the World Bank and the International Monetary Fund (IMF)), which have generally defined economic policy and development path of the country even up till the present time. A review of the literature on infrastructure and the development nexus then follows.

3.1 Ghana: A Country Overview

Ghana is a dominantly agrarian, low income developing economy. The country is located in the West of SSA. Its total land area is 238,539 sq km, of which the proportion under agricultural use has grown consistently from 53 percent in the early 1980s to over 65 percent since 2003 (World Bank, 2011).

Administratively, Ghana is organized into ten regions (as revealed by the Map Figure 3-1), under a multi-tier local governance system headed by a central government. Following its independence from British colonial rule in 1957, the country evolved into a fully-fledged parliamentary Republic in 1960. From 1960 until the inception of the current democratic Republic in 1992, the country endured intermittent disruption of its constitution through military coups dtats. These disruptions are particularly noted to have accounted for the considerable instability in the socio-economic and political environment experienced during those periods. The current dispensation of stability and peaceful transfer of political authority through multiparty democratic processes, beginning from 1992, could therefore be considered as the longest and most stable period of the countrys post independence policy environment.

According to GSS (2011), the countrys population is provisionally estimated at 24.23 million people, which is equivalent to a 28 percent rise since 2000. This implies an increase in population density from 79.23 people per sq. km in 2000 to 101.58 people per sq km in 2010.

18

Figure 3-1: Ghana: Map of Local & International Administrative Boundaries

Source: Ghana Statistical Service (2011)

Over the same period, the countrys intercensal growth rate is estimated at 2.4 percent. This trend also reflects a consistent decline in annual population growth rate, from the average of 2.3 percent between 2000 and 2004 to 2.1 percent and 1.96 percent in 2005 and 2007, respectively (World Bank, 2011). In the ensuing section, some understanding of the evolution of the countrys socio-economic policy and development strategy since the 1980s is provided, alongside some reflections on the state of rural development in that country.

3.1.1 Ghana: Background to the Policy & Development Experience since the 1980s

Until the onset of the Economic Reform program (ERP) in 1983, Ghanas development process and policy environment had endured considerable instability from frequent changes in
19

government, both civilian and military. These changes, beginning from the overthrow of the countrys first Republic in 1966, culminated in the near collapse of the countrys economy in the early 1980s. For instance, the World Bank estimates that Ghanas macroeconomic policy environment during the 1970s and early 1980s deteriorated to the extent that the countrys per capita income and import volume declined by over 30 percent. Real export earnings also plummeted by 52 percent, representing a decline from 21 percent to just 4 percent of GDP. Over the same period, domestic savings also declined from 12 percent to 3 percent while investments rate declined from 14 percent of GDP to 2 percent of GDP. Quite reflective of the distortions in the macro-economy was also the estimated decline in real wages of up to 80 percent.

In halting this trend, the ERP was adopted to provide short term stability for the economy while providing a basis for the implementation of the Structural Adjustment Program (SAP). The latter was aimed at engendering long term structural transformation of the economy and involved a significant shift from state-controlled, industrialization policy towards an export-led, market economy. These policies also involved the liberalization of the national economy with the aim of enhancing the efficiency of the public sector and a simultaneous shift towards private sector-led economic transformation. Among the key sectors targeted for reduced state intervention and enhanced presence of the private sector included the banking and finance industry, communication and postal services, manufacturing and natural resource exploitation and international trade (that is, export and import services). The international trade and macroeconomic policy regime, especially the exchange rate, import and export controls and monetary policy were also re-oriented significantly to eliminate state support and enhance international competitiveness. In the process, divestiture of public enterprises in a range of private sector activities became an essential implementation strategy. Most of these exercises also involved retrenchment in the public sector. Cleary (1989) estimate that a total of 45,000 public sector employees were affected by the exercise.

The implementation of the SAP also involved the elimination of subsidies and price controls, especially on agricultural inputs and products. A sector that was less affected by the price liberalization policy was the cocoa sub-sector, as it remained an important source of export tax revenue. Even then, considerable changes were implemented that led to significant reduction in
20

the size of the responsible state institution. These policy changes, which were further enhanced by the transition from military administration to a democratic, parliamentary Republic in 1992, provided the basis for a sustained focus on building a stable economic and political environment.

Under various policy and support programs including the Enhanced SAP, Vision 2020 and Ghana Poverty Reduction Strategies (GPRS I and II), the core basis of the ERP policy direction has evolved over time into various phases of sector-based programs. Some of these include the Financial Sector Adjustment Program (FINSAP) and the Agricultural Services Sub-Sector Investment program (AgSSIP). An important outcome of these policies has been the considerable improvement in the macroeconomy and general environment for sustained growth and development. Publicly available economic and social indicators reflect these trends.

Particularly, at the macroeconomic level, the country is noted to have achieved considerable diversification of the structure of the economy. This is revealed especially in the shift from extensive dependence on agriculture to services and industry. Macroeconomic trends also suggest considerable stability and growth prospects. This view is presented by the World Bank (2011) in a review of the economic performance of the country since the past two decades. A similar observation is made by Mensah et al. (2009) while UNECA (2010) also documents the countrys stride in sustaining growth, poverty reduction and increasing the pace of good governance over the same period.

Indeed, notwithstanding the recent global financial and economic crisis and its destabilizing effect on the world economy, Ghanas macroeconomy is found to be increasingly stable and has maintained an average annual GDP growth rate of about 6 percent since 2005 (BoG, 2011).

In terms of the structure of the economy, agriculture still dominates. The sector is noted to employ over 60 percent of the countrys labour force. Until 2000, the share of the sectors value added in the countrys GDP was well over 40 percent. This has declined consistently to a level of about 30 percent in 2009 whereas that of the service sector is noted to have experienced a reverse trend over the same period. In fact, in 2000, the contribution of the service sector value added to GDP was 32 percent, which has grown to over 49 percent in 2010. The industrial sector has
21

maintained an average contribution of 20 percent since 2006. Further details on the evolution of the structure of the economy since the 1980s are presented in Table 8-1 in the Appendix.

Similarly, trends in international trade and finance suggest significant improvements since the 1980s. In many ways, these reflect the increasing competitiveness and integration of the countrys economy in the global market As revealed in Figure 8-2, the countrys total imports (which is a proximate reflection of the realized capacity of a small, open economy to meet its domestic demand for foreign products and services) has grown considerably, from a low of 9 percent of GDP in 1980 to over 40 percent of GDP over the last two decades. Similarly, earnings from the export of goods and services have risen from about 8 percent to over 30 percent of GDP, for the same period. As a major source of external financing to meet domestic capital needs (alongside the potential technological accompaniments), foreign direct investment (FDI) has also increased considerably, albeit at a much slower pace than the growth in imports expenditure and export revenue. Relative to GDP, net inflow of FDI to Ghana has witnessed relatively low but stable growth from the 1980s to mid-2000. Beyond this period, the recorded growth has been considerable, peaking at over 6 percent of GDP at the end of 2009. For FDI in particular, the observed trend seem to depict the well-noted sensitivity of FDI to especially instability in policy and political environment6.

3.2 Infrastructure Services and the Development Nexus

The importance of access to physical infrastructure in engendering growth and sustaining economic transformation is not discounted in the development literature. In particular, access to infrastructure services such as electricity, water and sewerage, markets, transport network, telecommunication, education, health care, etc are noted to be critical for stimulating the growth and development processes of any economy. In this regard, economies with better access to infrastructure services are argued to be better positioned to drive up and sustain a growth and development trajectory that reinforce future growth prospects.
6

The transition from military administration to parliamentary democracy occurred in 1992, albeit, under the control of the members of the military regime. Thus, the peaceful transfer of power to the opposition party in 2000 and the subsequent constitutional changes in civilian administrations since 1992 could be cited as an important precursor of the renewed confidence in the countrys investment climate, thus the dramatic increase in net FDI inflows.

22

The Commission for Africa (2005) suggests that infrastructure access is critical for improving competitiveness in trade and commerce as it helps reduce market fragmentation, information asymmetry and geographic disadvantages. Altogether, these provide critical basis for mitigating transaction cost. Huchet-Bourdon, et al. (2010) have also argued that investments in economic infrastructure tend to stimulate better linkages in local economies while contributing to productivity growth.

In other studies reported in Brennerman and Kerf (2002), improved investment in physical infrastructure services contributes to the expansion of opportunities for inclusive growth processes, generalized growth in productivity and the transformation of rural economies. Indeed, other analysts such as the AfDB (1999), AfDB (2010) and the AFD/ World Bank (2010) reiterate this view and further submit that significant improvement in infrastructure access is paramount for scaling up access to regional markets and competitiveness of developing economies in the global economy.

More recently, Stiglitz (2011) has observed that poor infrastructural services in third world economies remain the most pervasive internal barrier to international trade. Compared to more traditional, artificial barriers such as tariffs and custom regulations from trading partners, the author suggests that poor roads, ports and other production structures presents more critical constraints to the capacity of developing economies to exploit opportunities from international trade and commerce. This evidence, Stiglitz (op. cit.) argues, provides the justification for a shift in paradigm in international development policy towards Aid for Trade in the south.

Thus, for developing economies generally, improved access to infrastructure is just fundamental for the agenda of engendering rapid growth and accelerating poverty reduction.

In Ghana, post-independence development policies have always maintained the sustained improvement in public infrastructure as a core component of the economic development strategy. More recently, the Ghana Poverty Reduction Papers (GPRS I and II), the countrys medium to long term development programme from 2001 to 2009, identified improvement in public
23

infrastructure as a strategic support investment for accelerated poverty reduction (NDPC, 2003, 2005).

However, the extent to which the country is able to deliver access to these services in the right amounts and quality, and towards achieving an optimal mix for catalyzing its development processes remains a major challenge. Underlying this challenge is the limited and highly overstretched fiscal resource endowments.

In particular, the delivery of physical infrastructure services is exceptionally capital intensive. Unlike many other kinds of public sector commitments, the provision of physical infrastructure requires extensive and disproportionate allocation of public resources vis--vis other competing needs. On this, the private sector is yet to prove itself as a dependable source of financing. As amply argued by the Commission for Africa (op. cit.), past conceptions that the private sector could provide such scale of financing have only proved to be a failed (development) strategy.

As a result of this and accounting for the existing deficit, the Africa Commission (op. cit.) estimates an annual expenditure of up to 5 percent of GDP in the development of new infrastructure projects and about 4 percent of the same for operations and maintenance, from 2005 till 2015. For Africa as a whole, this sums up to about USD 20 billion. Even more recent estimates project much higher fiscal requirement. According to an AFD/World Bank country diagnostic study of infrastructure services in Africa, African economies require up to USD 93 billion annually to build, operate and maintain infrastructure services adequate for their developmental needs until 2020 (AfDB, 2011; Foster, 2010).

In Ghana, the annual expenditure requirement is estimated by the government at USD 1.6 billion, for a period of 10 years (World Bank, 2011). In relation to this requirement, the country already faces a huge financial gap. The medium-term shortfall in this expenditure requirement is currently estimated to be 5-6 percent of GDP, whereas the short-term deficit is estimated at 3-4 percent of GDP (or USD 350 to 430 million). Obviously, this is a major fiscal challenge.

24

In relation to the rural sector, Masika and Baden (1997) already document considerable biases in the distribution of the limited investment in these infrastructure services. In Ghana, IFPRI (2005) and Mensah et al. (2009) also document pertinent evidence, which is reflected particularly in the spatial inequity in the distribution of basic public infrastructure services in favour of the urban sector. Quite consistent with the dominant argument in public policy literature, Masika and Baden (op. cit.) further observe that more recent shift in African economic policies towards market liberalization have implied that African rural communities, as an example of economic groups, will continue to benefit less from infrastructure service provision than their urban counterparts. This is more so as the sheer size of demand for these services in the urban areas continue to overwhelm the overall demand in rural communities. Bates (op. cit.) has documented these trends as the urban-bias in development policy in tropical Africa.

Among rural communities, the allocation of infrastructure may inadventedly favour specific groups or interests. Most common is the resource-rich constituents. Kessides (1993) provide some empirical evidence to support this assertion. In its review of the contribution of infrastructure services to economic development, the paper observes that in rural communities with both poor (and sometimes landless) farmers and rich farmers (with land and capital resources), the construction of an irrigation facility benefits the latter group of households more directly than the former. Thus, without proper targeting, the provision of quality infrastructure facilities, rather than reducing poverty and bridging any pre-existing income gap, may instead exacerbate the situation.

In the broader context of achieving sustained growth and development, a better understanding of the direction and size of the relationship between the type of and access to infrastructure services and household welfare is critical in ensuring effective and efficient policy targeting. The need for this knowledge emerges strongly from the review of the existing literature on the subject. Most comprehensive of such review is the work by Ayogu (2007).

In its study, Ayogu (op. cit.) notes that previous investigations on the role of infrastructure in economic growth in Africa have largely responded to the question of whether infrastructure matter at all. This is achieved both in terms of theoretical and empirical studies. This question,
25

the paper observes is largely over-flogged and has limited value addition. Rather, more significant and important is the extent to which infrastructure matters in the development process.

More recently, Dilion et al. (2011) provided some insight on this question by deploying a cocktail of methodologies on different measures of household welfare and infrastructure types in Nepal. The study is able to provide some interesting empirical understanding on the issue, especially the fact that rural roads, irrigation and extension services show important positive impact on welfare. Indeed, while recommending increased public investments in these infrastructure types and service, the authors still suggest that increased investment in research is needed to understand precisely this nexus. In the view of the present paper, a rigorous response to such knowledge gap requires as a first, critical step an analytical tool/framework that is comprehensive and sufficiently detailed but still simple enough in establishing the complexities of the mechanisms at play. As Robinson (op. cit.) will argue, it is only then that models could be formulated in a more meaningful and realistic way to distill in a more precise manner, the infrastructure and household welfare nexus. The present thesis deals with this first, critical step with the reconstruction of the SLF as provided earlier.

3.3 The State of Ghanas Rural Economy

The state of Ghanas rural economy could simply be described as under-developed and unstructured. This is largely explained by the fact that this sector has not evolved from any mainstream development plan, at least, during the period under review. Hence, the experience of the sector in the countrys development process has been defined by the programs and interventions targeted at other sectors such as agriculture, health, education, water and sewerage.

In relation to public institutions and capacity building, the decentralization program implemented since 1992 represents the core policy action in the development of rural institutions in Ghana. This program is aimed at the devolution of administrative authority from central government to the district and other local institutions. Nonetheless, the continued dominance of central planning, budgeting and financing of policy programs and the limited devolution of fiscal
26

authority to these decentralized bodies have implied that the pace of institutional capacity building and overall direction of development in rural Ghana remains confined to sector-wide, macro policies.

In fact, even among the non-governmental entities involved in rural development activities in the country (including international development agencies), the design of development interventions in the rural sector have been guided more by international development concerns. The influence of state-driven development plan is marginal. This observation is buttressed by the evidence reported in the NDPC (2003). According to that report, notwithstanding the considerable inflow of resources for development interventions, the absence of credible influence from state policy has induced considerable distributional inequity. Most affected is the rural sector.

Given this scenario, the discussion of the state of Ghanas rural sector is achieved on the basis of broad-based inferences to sectoral trends, as against a structured analysis on the basis of the evolution of the countrys rural economy and policy environment.

Like in most developing economies, Ghana continues to experience increasing decline in the share of the rural population in the national population. From a level of 60 percent in 1995, Ghanas rural population declined to 56 percent in 2000. In 2009, this was estimated to have declined further to 49 percent, partly reflecting the increasing urbanization of the countrys population. These trends notwithstanding, the countrys rural sector remains central to its medium to long term socio-economic transformation.

In particular, rural Ghana is argued to be the hub of the countrys domestic production, especially in relation to the primary production sectors. This arises from the strong linkage between the rural economy and agriculture and natural resource exploitation. For instance, NDPC (op. cit.) estimates that nearly 90 percent of the countrys agricultural value added is generated by smallholder farmers. These are generally rural-based. They also form the backbone of the countrys food security and natural resource conservation efforts. Beyond the economy, rural Ghana is pivotal to the countrys social and political transformation. As the principal base for most traditional authorities, rural Ghana remains the locus of the countrys cultural heritage;
27

the embodiment of the culture and tradition of the majority of indigenous Ghanaian societies. It therefore represents the environment within which the history, music, food, social norms and lifestyle of elemental Ghanaian societies are preserved.

These features have also implied that Ghanas rural economy remains trapped in low value production, with the associated implications on productivity growth and technology infusion. As an instance, the World Bank (2011) estimates that agricultural machinery to total arable land in the country was just 5 pieces per 100 sq. km in 2000, which declined even further to 4.5 per 100 sq. km in 2005. Reflecting this trend is also the extent of use of fertilizer in crop production. This is estimated to have increased from about 4 kg per hectare of arable land in the early 2000s to 20 kg in 2006, before plummeting back to the level of 6.4 kg reported in 2008. The extensive dependence of rural production on the natural weather and low technology, especially rainfall and low skilled labour, respectively, have also implied that investments in the countrys rural sector confront high risk and low productivity.

Consequently, rural Ghana is generally characterized by low income and vulnerability to extremities of weather and natural disasters. GSS (2000) provides another important example. Following its review of poverty trends in the country since the early 1990s, this paper identifies poverty in that country as predominantly a rural phenomenon. Associated with this are also the low infrastructural base and non-farm employment opportunities. The pervasiveness in limited infrastructure access is particularly argued to account for the significant differences in livelihood standards between the rural and urban sectors, thus fuelling a surge in migration to urban Ghana.

In the predominant production sector (agriculture), overall productivity is acknowledged to be very low. Diao (2010) observes that much of the growth in this sector is driven by expansion in land areas rather than productivity. As an instance cited by the paper, productivity in land-use across all crops declined steadily from 155 cedis 7 per hectare in 1994 to 112 cedis per hectare in 2000. In 2006, this reversed marginally to 116 cedis per hectare. However, for the food crop, the trend was worse, as land productivity in 2006 fell below the level decades ago.

Cedis is the unit of the local currency of Ghana.

28

On the other hand, the total area of land under cultivation is observed to have increased across all categories of crops, both cash and non-cash. Table 8-2 provides some details of the differentials in agricultural land allocation and productivity, over the period 1994 to 2006. An important implication from this trend is the considerable decline in the countrys forest cover. In 1990, Ghanas forest cover was estimated at 32.73 percent of the countrys total land area. This declined to 26.25 percent in 2000 and to 24.25 percent in 2005. Current estimate suggests a further decline to about 21 percent.

In relation to infrastructure, the countrys rural sector continues to lag in access to public infrastructure services. Across all three agro-ecological zones in the country and over the period 1991/92 and 2005/06, water from natural sources accounted for the largest share of water sources for rural households in Ghana. Indeed, the proportion of households that receive water through public sector investments such as pipe-networks for in-house plumbing, public stand pipes and even protected wells (boreholes) remain very low. Compared with the national average, the trend even suggests deterioration in the fraction of rural households that obtained water from public services, especially between 1998 and 2005 (as reported in Table 8-3).

Similarly, the national average of the proportion of households that have access to improved toilet facilities differ significantly from the levels reported in rural Ghana. Focusing specifically on public toilet provisions (which is mainly via the Kumasi Ventilated Improved Pit (KVIP) system), rural households continue to show extensive dependence on unimproved systems such as the pit latrine and pan (see Table 8-4). In Figure 8-3, access to the national grid is also shown to consistently differ between the rural sector and the overall national estimate. Estimated strictly in terms of a households connection to the national grid, rural households are reported to have limited access to electricity, across all three agro-ecological zones and over the three crosssections. This bias in distribution is worst manifested in the rural savannah, where households connection to electricity is noted to have grown from less than 5 percent during the 1991/92 and 1998/99 cross-sections to just 15 percent in 2005/06.

Among others, these trends highlight the precarious state of rural under-development in Ghana. It reflects a state of considerable inadequacy and a development gap that require extensive public
29

intervention. Considered within the broad context of the overall development needs of the country, efficiency in resource use and good targeting are critical. In the particular instance of public infrastructure investments, this need can only be overwhelming and urgent but should be pursued in a manner that assures sustainability. This then is the basis for the urgent demand for a rigorous, empirical understanding of the differential impact of infrastructure access and household welfare. The present thesis provides some direct response to this demand.

30

4.0

METHODOLOGY

Advances in micro-econometric modeling provide varied methodologies for analyzing the research problem at hand. For the present thesis, pseudo-panel modeling is employed. 4.1 The Generic Model On the premise of the discussions preceding this chapter, the welfare (say = , , , =1, , N), at a given time ( = 1, , ) is modeled generically and in a dynamic sense as: ), of a household (

(1) refers to

where,

refers to a vector of variables defining household-level characteristics;

households private asset endowment;

refers to a vector of indicators of household access to refers to a vector of

public capital (focusing here on public physical infrastructure); and, variables defining the livelihood vulnerability context of the household.

In the present context, access to public infrastructure (capital) is used to imply the existence, within reasonable proximity, of the opportunity to make regular use of functional, public physical infrastructure. As has been argued by Gasner (1998), this may not imply an actual use of the service but a prevailing condition that, when desired, a household can actually use the resource as would be expected. A consequent argument arising from this conception is that at any given point in time, all households in an economic system could be ascribed to one of two possible categories; that is, those with access and those without access to physical infrastructure. Further, for the former category of households, benefits from the resource may arise from direct use (at zero cost of access) or indirectly through the positive economic externalities created by the access to the infrastructure (especially through the market and price dynamics) or even the knowledge of (and the related possible utility that could be derived from) the existence of the resource. The latter is discussed in other disciplines as the existence value of the resource. In the same vein, households without access cannot be argued to generate any important benefits from such resource, and even if they do, this can occur only at some economic cost. In fact, for such disadvantaged households, aside the user-cost, an actual need to access the resource would first entail the expenditure of other productive resources (constituting the access cost),
31

which would be disproportionately higher than the zero access cost enjoyed by those households with access. Indirect benefits of the infrastructure to such households are assumed to be zero, especially as proximity to a given resource is fundamental to the ability of a household to capture any potential positive externality8. The potential net return to household livelihood outcomes from access to such public infrastructure is therefore hypothesized to be positive and systematically higher for households with access than those without it. This is the core hypothesis of the thesis. In more specific terms, different levels of access to infrastructure set in motion transmission mechanisms that feed dynamically into household welfare; either through price and wage changes (as may be expected of households that gain better access to new markets and transport infrastructure), employment generation (such as in the case of increased access to financial services or the emergence of cottage industries arising from access to a power grid) or even better conditions of living arising from improved access to portable water, sewerage or sanitation services and health care (Winters, et al., 2009). As this type of assets only tend to complement the primary resource endowment from which households construct their livelihood choices, the ability of households to optimize benefits from infrastructure access is therefore conditioned by other factors that may vary from household to household. As postulated by the discussion on the SLA in the previous chapter, such other factors may include the households capital endowments, shocks and related livelihood vulnerabilities (such as climate changes, socioeconomic differentiations, the macro-economic and political regime and history (Scoone, 1998)).

Arising from this conception, access to public infrastructure is generally expected to have systematic and positive effect on household livelihood outcomes, ceteris paribus. Among different infrastructure, such effects are also expected to differ in magnitude, ceteris paribus. 4.2 The Pseudo-Panel Modeling Framework: An Overview

In providing a quantitative evaluation of such intricate relationship, pure panel data models have typically been adopted. Unlike cross-sectional data models that are static in structure, pure panels

This is the same as arguing that a household residing in close proximity to the resource would more likely endure any negative externality, which will be systematic and significantly higher than their counterparts residing far off.

32

allow changes in household welfare to be tracked over time, and such as to properly account for the dynamics of the relationship and in attributing cause. As noted in the literature, the use of panel data models however has its limitations. Among these are problems related to the attrition of the sample over time, changes in the demographic dynamics and representation, and the cost and accessibility of such data sets. A brief discussion of these problems is outlined below.

Attrition problems: Panel data approaches tend to suffer from non-random attrition of observations. This problem is particularly serious for long term panels where respondents may exit from the survey for reasons ranging from migration to fatigue and loss of interest. A major consequence is the loss of true representativeness of the survey observations and the consequent erosion of the inherent statistical power. And even when attrition problems are catered for by replacement with new and similar households, such replacements generate errors in the measurement of the variables of interest. This is more so since all households are unique in many respect and those similar households are typically not the same as the households they replace. By drawing a random sample of individuals from the population, pseudo-panels clearly avoid this challenge and propose analytical techniques appropriate for handling such pooling of the independent cross-sections (Deaton, 1995; Koutsoyiannis, 2001).

Demographic dynamics and representation: A problem related to the representativeness of the sample observations in long term panels relates to demographic changes during the span of the survey. As sample observations are drawn at the onset of the panel survey and followed over the duration of the panel, accounting for non-random changes in the structure and composition of the population is impossible in pure panels. This heightens the problem of representativeness of the sample at latter stages of the survey and could seriously undermine the internal validity of the statistical design and related analysis based on such surveys. Indeed, implicitly, genuine panels assume constant, non-changing structure and demographic distribution of the population; an assumption that is quite strong to uphold especially in long term panel surveys. On the contrary, pseudo-panels allow new and random samples to be drawn from the population such as to capture changes in the underlying distribution and size of the given population, and in so doing, reflect better the demographic structure and related socio-economic dynamics. Though this may

33

be argued to complicate the analytical assumptions of pseudo-panel data modeling, it still provides a better representation of the population at any given point in time.

Cost and accessibility: Panel data surveys, especially socioeconomic surveys such as the one with which this study is concerned, are expensive to conduct and manage. Tracking the same household over a period of time, which may involve investing time and efforts to follow households to new geographic locations and residences, is practically challenging and costly. In many instances, the benefits of panel observations have not been enough to justify the investments required. Pure panels are therefore difficult to access. In many developing economies including Ghana, such data hardly exist.

Consequently, pseudo-panel data models provide plausible alternative in the analysis of the present thesis. In particular, the existence of waves of independent countrywide household survey datasets in Ghana allows for the use of pseudo-panel modeling in estimating the relationships of interest without missing out on the dynamics and key theoretical efficiencies obtainable from panel data models. The present study therefore adopts dynamic pseudo-panel modeling in the quantitative evaluation of the household welfare and infrastructure access nexus.

4.2.1 The Pseudo-Panel Modeling Framework: The Basic Model This section presents a general discussion of the estimation approaches of pseudo-panel. Firstly, the basic model is presented, followed by the more complicated issues with the estimation of dynamic pseudo-panel models. Following an introduction to the data, the specific model of the present study is then finally provided. Indeed, as the literature on this subject is relatively young and still evolving, this structure is expected to better highlight the issues at hand and the plausible options to be adopted for the estimation of the proposed model. In Table 8-5 below, a summary of the leading analytical literature is provided.

34

Generally, the analyses of pseudo-panel data9 follow two broad methodological strands. The first is the estimation of models based on pooled Ordinary Least Squares (OLS), which treats the panel observations as simple pooling of independent cross-sections. Woodridge (2003) and Cameron and Trividi (2009) provide some discussion on this. The second strand of modeling generally follows error component (or random effect) modeling. The seminal paper by Deaton (1985) follows this approach. In fact, subsequent studies have followed Deatons paper, in both theoretical and applied contexts such as those by Verbeek (1993), Moffit (1993), Collado (1997), Verbeek et al. (2000), McKenzie (2000, 2001, 2004), Verbeek (2007), Inoue (2008), among others. The works by Moffit (1993), Collado (1997) and McKenzie (2000, 2004) especially deal with dynamic pseudo-panel models, with some recent contribution coming from Inoue (2008).

In respect of the first strand of estimation, it is observed that cross-sectional data are typically drawn from a random, representative sample of the same defined population, over different time periods. These data are also typically large in sample size. As such, it is feasible to pool these together to achieve even larger sample observations of the same underlying population. Such independently pooled cross-section of information could then be used in OLS regression analysis. On the assumption that the sample observations are truly independent across the pool, serial correlation is assumed away and OLS estimation is argued to yield even more precise, consistent estimates and robust test statistics. An issue that arises is the need to account for the potential differences in the underlying distribution of the population for the different sampling years. This, Woodridge (2003) notes, is easily achieved with dummy explanatory variables for the survey years.

However, a major drawback of this simple treatment is that where time-invariant individual effects are present and significantly correlated with other explanatory terms, the models error term cannot be argued to be independent and identically distributed (Woodridge, op. cit.; Cameron and Trivedi, op. cit.). This is the basis for the error component modeling.

The term pseudo panel data is used generally to refer to the pooling of multiple cross-section dataset In the literature, it is also referred to as repeated cross-section, time series of cross-section, etc but all implying the multiple time dimension of cross-sectional observations. It therefore does not imply any specific analytical method or estimation approach.

35

In particular, it is argued that since the error factor of the model accounts for the effect of all unobserved determinants of household welfare, it is not feasible to assume that it is truly idiosyncratic and therefore uncorrelated with the other explanatory variables in the model. Wooldridge (op. cit.) notes especially that among the unobserved factors captured by there

are also unobserved time-invariant factors such as history, intrinsic qualities of the household and its members and even some geographic conditions. These factors are not measureable, do not change over time but could still be relevant in the determination of welfare. An OLS estimation that fails to account for such effect will therefore be systematically biased. In this sense, an error component, static form of (1) could be specified generally as: = + , +

where, (

is the unobserved time-invariant explanatory variable (also called the fixed effect), while

)=

; (

) = 0;

(2)

) 0; and,

is

the true idiosyncratic error component of the model.

In Deaton (1985), the argument brought advanced is that since households in the pseudo-panel are known to differ from one survey to another, standard fixed effect estimation (which is necessary for eliminating ) is not appropriate. Rather, error-in-variable estimation approach

could be used such that a taxonomic grouping of the population sample could be constructed and tracked in the various survey years. Once such groups (or cohorts) are well-defined on the basis of some time-invariant attribute of the population, they would be fixed in membership and identifiable anytime they show up in a population sample.

Thus, an aggregation of the individual observations to cohort averages makes possible the tracking of the latter in different cross-sections. These cohort observations, including any presumed (unobserved) cohort fixed effect, could thus be handled just like pure panel observations albeit under varying cohort design and related asymptotic.

In respect of the asymptotic dimensions, a preliminary point to be noted here is that depending on the cross-section and time dimensions of the data set, the choice of cohort design give rise to three broad classes of asymptotes (Verbeek, 2007), which are:
36

1. 2. 3. where, and

with fixed , thus as and T , thus and

is fixed; and, is fixed; is the time period, is the number of cohorts

are fixed such that

is the total sample of the cross-section,

is the observations per cohort.The estimation approaches proposed by the existing

literature therefore generally follow one of these asymptotes. For instance, the error in variables estimator proposed by Deaton (op. cit.) is identified to follow conditions of large cross-section and time dimensions, akin to the second asymptotic argument reported above.

Specifically, Deaton (op. cit.) postulates a model based on panel observations of cohort averages such that given a cohort (for = 1, .., C) to which a household belongs, could be re-specified = + +

on the basis of the sample cohort means for any given time , as:

(3)

Here, assuming that the unobserved population cohort model is specified such that:

(4)

where the superscript * indicates the unobserved population cohort mean of the variable, and assuming further that errors in the observations at the household level have zero mean and are normally distributed such that: = =

+ +

(5) then, by invoking the law of large numbers, the errors in the sample cohort means could also be assumed to be normally distributed with a mean that converges to the population cohort mean and variance that tend to zero as cohort observations ~

increase as specified below: . (6)

37

However, where

is small (as assumed by Deaton), the sample cohort means approximate the

population cohort means with a margin of error. In correcting for this, Deaton (op. cit.) proposes the error-in-variables estimator such that the errors in the sample cohort means are directly accounted for in the estimation of the model. The model parameters are therefore identified such that: = where,

(7)

is the measurement error estimated from the covariance matrix.

In Verbeek et al. (1993), the contention is made that even when the number of cohorts tend towards infinity, the error-in-variables estimator may still not consistently identify the parameters of the model when the number of survey periods , is small. Rather, the paper

identifies a consistent estimator of the parameters where a fraction of the estimated measurement error is corrected by a factor , where the size of this factor is estimated as a function of the time dimension of the panel such that: =( )

(8) )

By this, Verbeek et al. (op. cit.) propose a consistent estimator of the model parameters, where: =

(9)

Here, it will be seen that if an index of the fraction of the measurement error that is corrected ( ) is defined to be such that = [0, 1], then the estimator in (9) will be of the following form: =

(10) = 1. On the

Implicitly, this estimator converges to Deatons error-in-variable estimator when other hand, when

application of the proposed estimator therefore requires a decision on the optimal level of the

= 0, the estimator converges to the standard fixed effect estimator. Empirical

measurement error to be corrected for in the model. According to Verbeek et al. (op. cit.), the size of that is chosen is inversely related to the variance of the estimated coefficients but

38

directly related to the mean square error. A decision on and variance of the estimated model.

is therefore a trade-off between bias

4.2.2 The Pseudo-Panel Modeling Framework: The Dynamic Model Following the works of Deaton (1985) and Verbeek et al. (1993), the literature on the estimation of pseudo-panel models has been enriched with some studies that have tended to focus on the estimation of dynamic pseudo-panel models. Leading this set of studies is the work by Moffit (1993).

Fundamentally, aside the problem of measurement error and individual heterogeneity, dynamic pseudo-panel models also confront the problem of the measurement of the lagged dependent variable (more so as different individuals are observed in the different cross-sections) and the presence of this variable as an explanatory term in the model. The literature in this category of pseudo-panel modeling provides different responses to this problem, following the same asymptotic as the static model.

Briefly stated, Moffit (1993) provides an instrumental variable (IV) interpretation of the error-invariables estimation approach proposed in Deaton (1985). In so doing, the paper introduces the estimation of dynamic pseudo-panel modeling based on two-Stage Least Squares (2SLS) estimation. Here, the paper proposes the estimation of the unobserved lagged dependent variable with the first stage being an OLS estimation of welfare using information on individuals observed at time 1. Under the condition of fixed and , Moffit argues that a second-

stage OLS estimation will yield consistent identification of the parameter estimates.

However, in responding to this argument, Verbeek et al. (2000) have argued that OLS estimation of the dynamic linear model will generally be inconsistent when time-varying (or autocorrelated) exogenous regressors are introduced in the prediction of the lagged dependent variable. Another criticism is based on the view that the paper fails to provide an explicit theory on the distribution of the proposed estimator (McKenzie, 2004).

39

Thus, under the same asymptotic condition, McKenzie (2004) proposes OLS estimation of the dynamic model where the lag of the sample cohort mean is used to replace the unobserved lagged dependent variable. This treatment is consistent with the approach of Collado (1997) and also follows the arguments of Verbeek et al. (1993) that under the condition of large number of observations per cohort, the measurement error in the sample cohort mean diminishes.

Here, McKenzie (2004) observes that as long as the cohort-level component of the measurement error does not vary over time and assumed to have converged to zero, OLS estimation of the dynamic pseudo-panel model would be consistent. Verbeek and Vella (2005), McKenzie and Antman (2005) and Verbeek (2007) provide further substantiation of this argument. In this sense, indicator variables of the cohorts are explicitly introduced in the model as proxies for the cohort fixed effects. By definition, this estimation converges to the standard fixed effect (within) estimation, where the cohort effects are assumed fixed and thus eliminated by differencing.

In Collado (1997), an alternative estimation of the dynamic pseudo-panel model is provided. Unlike the previous approaches, Collado (op. cit) establishes an error-corrected Generalized Method of Moments (GMM) estimator that provides consistent estimation of the dynamic model even under more relaxed conditions of fixed and , and as . As argued by McKenzie

(2004), differences in cross-sectional observations of the pseudo-panel model imply that the lagged dependent variable, irrespective of how it is measured, is measured with error in small sample. This measurement error is established to correlate with the error term of the model, which violates the condition for consistent estimation of the model by OLS.

The estimation of the model in Collado (op. cit.) therefore proceeds by exploiting the instrumental variable procedure proposed in Arrellano and Bond (1991) to achieve errorcorrected first-difference GMM estimator. Relative to the estimator proposed by Moffit, Collado (op. cit.) establishes that under conditions of fixed and and as , the GMM estimator and with

performs better. McKenzie (2004) further explores the consistency of this estimator under varying asymptotic and distributional assumptions. The paper observes that as

fixed , OLS converges to the within estimator, which yields more robust results and therefore preferred. On the other hand, as , the GMM estimator performs better. In Roodman
40

(2006), the latter estimator is generalized in a manner that allows for the estimation of similar models in the presence of multiple, sequentially exogenous regressors. The routine is also enhanced with options for augmenting the proffered internal instruments with variables outside the model. Additionally, the estimation proceeds under assumptions that allow the cohort effects to be distributed arbitrarily.

4.3 The Empirical Model, Definition of Variables and the Postulated Hypotheses In specifying the empirical model of the study, an outstanding argument to be made concerning the generic model in (1) above is the choice of conception and measure of household welfare. Here, welfare economics presents varied conceptions, which are all based on the quantitative evaluation of the standard of living of the economic agent (typically, individuals or households). As noted by Deaton (1997) and Haughton (2001), a plausible conception of welfare is the command over goods and services towards a desired standard of living.

In implementing this conception, a desired standard of living is constructed taking into account the socio-economic setting within which the household makes its livelihood, the structure and composition of the household and the basket of goods and services that could be deemed sufficient to achieve a minimum living standard, in absolute terms. Once this desired standard of living is defined and constructed, an objective determination of welfare could then assume the form of the amount of resources that an individual could possibly employ to command the needed goods and services. This perspective thus identifies household total income (for any defined period of time) as an important measure of welfare.

However, the use of income is limited by its high volatility even when measured over a relatively short period. The argument is also presented that as households minimize vulnerability to economic shocks by spreading the expenditure of their income over a relatively longer time period, welfare could best be measured by actual consumption (or expenditure) of the household10. This measure therefore captures, generally, the more stable fraction of a households income base. Beyond expenditure from regular income, this also permits
This argument is best captured by the life cycle theory of consumption identified in the works of especially Friedman (1957) and Modigliana and Brumberg (1954).
10

41

measurement of welfare as households adopt different livelihood strategies such as the consumption of own production, withdrawal from farm stocks and financial savings. In more severe instances, households may also borrow or liquidate capital assets (starting from the more liquid to the less liquid)

A related issue is also the view that information on households actual consumption or livelihood expenditure tends to be more accurate and reliable than information on income (Filmer and Pritchett, 2001). Hence, a households total consumption per equivalent adult per year is widely used in both the theoretical and empirical literature as a measure of welfare11.

The present analysis follows this trend and proceeds to model welfare functionally as: = + + + + + + + + + + ++

11

where, all variables are estimated as the cohort average for the different cross-sections such that , for example, becomes , which is the cohort average of the at time . Here, welfare is

logarithmic transformation of the welfare index of a given cohort

measured as total household consumption expenditure per equivalent adult, in real terms (that is, adjusted for variation in prices across space and time, using the 1999 constant prices of Accra) (GSS, 2000). The s are the partial effects of the explanatory terms, s are the parameters of the explanatory variables in the model. In the log-

level functional form specified above, the

which are interpreted as the proportionate change in household welfare following a unit change in the relevant explanatory variable.

The definition of the independent variables and their construction are explained below:
11

Notably, households are conceived as private entities and their consumption expenditure is therefore identified to include actual and imputed expenses on food and housing, non-food items and inter-household transfers. In other words, welfare is measured as the total amount of goods and services per capita that is actually consumed by a household, over the survey period.

42

is age of the head of household, in years. To identify possible non-linearities in the effect of age on household livelihood, the square of the age variable is also introduced and estimated in the model as . The argument here is that below a certain threshold, age is expected to have

negative effect on household welfare as younger households are expected to maintain relatively low asset levels and income generation potential. Beyond this threshold, age is expected to impact positively on household welfare based on increased potential for higher income generation, better risk management skills through life experiences, improved asset endowment and enhanced portfolio of livelihood strategies; is the gender of the head of household, which is constructed as a dummy variable, assigned a value 1 if the head is male, 0 otherwise. At the cohort level, it describes the proportion of households within a cohort that is headed by a male, relative to the probability of a female head. A priori, male headed households are expected to maintain higher levels of welfare than female headed households, especially based on the generally labour-intensive and manual nature of income generation activities in rural Ghana. Social and cultural norms that may limit females from engaging in some high-risk but relatively high-return economic activities could also be argued to hamper the capacity of female-headed households to maintain welfare levels comparable with male-headed households (Ellis, 2000). Aside these, female-headed households could be argued to be systematically more vulnerable than their male counterparts in the traditionally patriarchal societies found in most Ghanaian rural communities; is the adult size of the household to proxy the actual labour resources available to the household. This variable is measured as the proportion of household members aged over 18 years in the total household size12. An alternative measure of this variable is the equivalent adult size of the household, especially to better capture the effect of the households potential labour resources on welfare outcomes. In most empirical studies, the actual size of the household (that is, the total number of household members) is used for this purpose. This thesis departs from this argument mainly on the basis that households of the same size do not necessarily have the same composition and therefore potential labour resources. Thus, a five-member household might be made up of two adults and three infants, the same way as it could be made up of four adults and one infant, with potentially different welfare implications. The use of the proportion of adults (or
One could view household heads aged 18 years and below as economically self-dependent. Single-member households with heads whose age falls below 18 years are therefore treated in the analysis as single-adult households.
12

43

even equivalent adult) should therefore provide more objective measure and enhance the visibility of this factor. This variable is expected to have positive effect on household welfare, all things being equal; is the cohort average of the index of households endowment in private capital, which is constructed as the factor score of the first principal component of the correlation matrix of indicators of household ownership of selected physical assets. The methodology of the Principal Component Analysis used here is described briefly in the ensuing sub-section and based largely on insights drawn from Filmer and Pritchett (2001), McKenzie (2004) Vyas et al. (2006) and Winters et al. (2009). While the total value of household asset endowment provides an option for measuring this variable, self-valuation of assets by respective households during the survey is considered here quite subjective and arbitrary. As the number of the assets owned is not consistently measured across the three cross-sections, indicators of ownership are used. Assets are hypothesized in the model to have positive impact on welfare, more so as households capital investments are argued to form the principal basis upon which households solicit complementary resources such as public infrastructure to formulate livelihood choices; , and p are indicator variables for access to infrastructure. At the cohort

level, they reflect the degree of access to physical infrastructure by households within a given cohort. Three different physical infrastructure are investigated. These are public transport ( ), water ( ) and electricity ( ). At the level of individual households,

these variables are measured as dummies, assuming a value 1 if a household has access to these infrastructure from own home, in the community of residence or within a distance of 1 km or less from the residence), and 0 otherwise13. Thus, the variables assume a value 0 if households cover more than 1 km to gain access, 0 otherwise. The choice of a distance of 1 km is informed by current discussions on the UN-MDG, which generally deems such distance to access a resource as reasonable. This is postulated to have direct relationship with welfare, more so as the closer the infrastructure to households, the more savings in access-cost such as time, energy and the related opportunity costs. Also, proximity to infrastructure implies increased opportunity to

For water in particular, the actual distance travelled could have been used in the model. This may however introduce non-linearities in the relationship between distance and its welfare impact. Thus, a way of keeping the model simple is the use of dummy variables to measure access using the 1 km threshold.

13

44

capture any positive externality that may arise (including the existence value)14. These variables are interacted with household asset endowment ( , and

) and introduced in the model to evaluate the degree of complementarity

between infrastructure access and households capital endowment in influencing welfare outcomes. All three interaction terms are postulated to have positive effect on welfare; is the number of years of education and skill training attained by the head of the household. It is a proxy variable to capture the potential effect of education and skill level of the household on welfare. As the number of years of education and skill training is likely to show an increasing effect on the quality of household decision and livelihood formation, a nonlinear relationship is postulated such that at lower levels of educational attainment, the returns on welfare of education is hypothesized to be negative. Beyond a given threshold, education is then expected to show positive and significant impact on welfare, all things being equal. A quadratic term of this variable is therefore introduced as and estimated in the model. In some

studies, this variable is measured as the highest level of schooling completed. However, since the country has implemented different educational systems since the 1990s, with some notable overlaps in the number of years required for completing each cycle, this thesis considers such construction naturally inconsistent. The choice of the number of years spent in formal education and skill training is therefore considered to be a more plausible option and adopted in this study. is a proxy dummy variable representing the dominant sector of employment of the household head. Generally, seven sectors are identified15. These are Food Crop production (including animals and fishery), dubbed agfoodcrp, Cash Crop production (especially agricultural production for exports, agcashcrp), the public sector (public), formal private (pfmlemploy) and informal private (pinfmlemploy) sectors, self-employment (selfemploy) and non-working (that is, unemployed or retired). The latter group of households is generally considered to be economically inactive and therefore dropped from the analysis. The analysis therefore proceeds with the remaining six sectors, which are appropriately identified with dummy variables following the standard procedure of allowing the variable to assume a value 1 for the relevant sector, and 0 otherwise;
While differences may exist in the quality of infrastructure, it would be assumed in the present context that public infrastructure services are homogeneous. 15 The sectors of employment used for the analysis are generally drawn from the UNIDOs International Standard Industrial Classification (ISIC) of economic activities.
14

45

represents the status of the household head as a migrant or indigene of the community of residence, assuming a value 1 if the head is a migrant, 0 otherwise. Like the choice of economic sector ( ), migration is identified in the literature as an important livelihood

strategy (Ellis, op. cit.). Households may move for reasons ranging from economic, social or even political. In the present context, migration as an economic decision is interpreted as a

rational decision exercised by the household in optimizing welfare outcomes. It is therefore argued to impact on welfare directly; is the lag of the logarithmic transformation of the welfare index. As argued elsewhere in this thesis, welfare is modeled dynamically to reflect the argument that shocks to past levels of welfare (such as bumper harvest or declines in wage income) do not terminate instantaneously but rather ripples through time as households smooth their consumption. Higher levels of welfare in the past are therefore hypothesized to impact on current welfare outcomes positively. A lag of the sample cohort means is therefore used as the estimate of the lag of the cohort population means, following the argument of Collado (op. cit.) and applied extensively in the pseudo-panel estimation literature; is defined as the cohort-fixed effect, which arises naturally as the orthogonal projection of the individual fixed effects. In Verbeek et al. (2007), this is estimated explicitly in OLS estimation by the cohort indicators, interacted with time while other literature simply proxies this variable as cohort dummies; is a dummy variable to proxy time-effects. It assumes a value 1 for a given crosssection, and 0 otherwise; is the stochastic error term.

In relation to the conceptual framework of the study, a summary of the empirical model and the hypothesized direction of effect of each explanatory variable are reported in Table 4-1. Here, it is shown that four broad categories of explanatory variables are hypothesized to explain welfare in this study. That is, household attributes, levels of private asset endowment, levels of access to public infrastructure, and the vulnerability context of the households livelihood.

46

Table 4-1: An Empirical Translation of the Conceptual Framework & Model Hypothesis
Livelihood Outcome Conceptual Arguments Household Attributes Livelihood Explanatory Factors Household socioeconomic features Household Private Asset Endowment Public Capital Assets Household Livelihood Vulnerability Context Empirical Measures Age Gender Migrant Sector Size Education Assets Power Water Public transport Lag Welfare Time Effect A Priori Expectation - (+)* + + ** + - (+)* + + + + + **

Measures of Household Welfare: - Consumption Expenditure - Income - Poverty Status

Household Livelihood Assets

Shocks, Trends, Seasonality

Note: (+)* refers to the expected sign of the quadratic term; ** implies indeterminate expectation, as the expected direction of effect is dependent on the reference category employed.

4.2 Construction of an Index of Household Private Asset Endowment The Principal Component Analysis (PCA) procedure is adopted in the construction of an index of household private asset endowment. The PCA is a statistical technique applied in the reduction of a set of correlated variables to a limited set of uncorrelated components, representing a linear weighted combination of the underlying variables. McKenzie (2004) has argued that since the first principal component accounts for the largest share of the variations in the original data, it suffices for capturing the most important dimensions of the information contained therein. The thesis adopts this argument and implements the procedure in the STATA software, which is based on the eigen decomposition of the correlation matrix of the indicator variables. The indicator variables are constructed from information provided by households on whether they own a given asset with some market value or not, assuming a value 1 if yes, 0 otherwise. This information is captured at the household level.

Given the choice of the assets identified for the procedure, the index is used here to provide a single, homogenous measure of the long term capital endowment of the household. This is particularly informed by Filmer and Pritchett (op. cit.). The authors note that generally, the
47

eigenvectors from PCA of household asset endowment tend to reflect better the long term rather than short term view of the households asset endowment. Hence, assets deemed most appropriate for the present analysis are those that reflect households long term capital holdings (or less liquid asset endowment). The analysis therefore leaves out household transitory farm stocks like animals and crop stocks and focuses on durable goods including furniture, sewing machine, stove (kerosene, gas and electric), refrigerator (including deep freezer), air conditioner and fan. The others are radio and radio cassette player, television, car, tractor, plough, carts, spraying machine and land.

Also, McKenzie (2004) and Vyas et al. (2006) have argued that a key principle for achieving a good index in PCA is to limit the selection of assets to those that are relatively more correlated and whose distribution varies significantly across the sample population. On this basis, regular agricultural inputs such as cutlass and other basic tools are left out of the analysis. For those assets selected, consideration is given to the consistency with which the measurement is made across all the three cross-sections of the survey data, as revealed by the summary statistics. The factor score of the first principal component from the procedure is predicted and used as the measure of the capital endowment, .

In Filmer and Pritchett (op. cit.), the weight of each asset from the PCA is estimated as the ratio of the score of the principal component and the associated standard deviation. This provides a measure of the unit change in the households asset index following the acquisition of a given asset It therefore reflects the degree of importance of a given asset within the overall set of assets identified for the analysis. This measure is also estimated and reported.

4.3 Data, Data Sources and the Construction of the Pseudo-Panel Database Data for the present study is obtained from the latest three rounds of the Ghana Living Standard Survey (GLSS). The GLSS is a nationally representative household and community survey conducted by the Ghana Statistical Service (GSS). The structure of this survey follows from the World Banks Living Standard Measurement Survey (LSMS), which was first conducted in 1987, with a subsequent round in 1988. Following the experiences of these two waves,
48

corrections to the principal survey instruments and methodologies have been undertaken and adopted for the conduct of the subsequent surveys. These corrective measures are argued to have improved significantly the comparability of the dataset from these latter waves. An earlier effort to obtain similar survey data conducted in 2009 by GSS, for and under a collaborative arrangement with the Ministry of Transportation, was unsuccessful. The thesis therefore focused on those latest three waves of the GLSS, viz, GLSS 3 (1991/92), GLSS 4 (1998/99) and GLSS 5 (2005/06).

In terms of the structure and content, the GLSS is organized into four main components (Coulombe and McKay, 2008). These are: 1. The household survey instruments; 2. Non-farm household survey instruments; 3. Rural community survey instruments; and, 4. Prices of food and non-food items survey instruments. The first two sets of instruments are collected at the level of individuals, with households constituting the sampling unit. Thus, in the 1991/92 wave of the survey, a net total of 4,523 households were sampled and reported in the survey database, whereas the 1998/99 survey reports on 5,998 households. Data points for the 2005/06 wave covers a net total of 8,688 households. The fractions of the rural sub-sample in the total sample for the three cross-sections are 65.1 percent, 63.3 percent and 58.4 percent, respectively. Among the key information obtainable from these instruments are the demographic composition and structure of households, income and expenditure patterns, health, education and skill training as well as household assets, savings and patterns of access to key infrastructure services. Information on migration, remittance, employment and household production activities (including agriculture) are also reported in this component of the overall survey instrument, while all non-farm income generating activities are reported in the non-farm household instrument.

The latter two instruments are collected as community-level information. The rural community survey instrument presents information on community facilities and infrastructure, socioeconomic and political organization of the sampled communities. As sampled households are drawn from these communities, the data is structured such as to allow especially information on
49

all rural households in the survey to be matched with information from communities from which these households were drawn.

Finally, the survey instrument on prices of food and non-food items allows for the assessment of the price levels for key expenditure items of households, during the period of the surveys. This enables price information to be collated at the level of communities, enumeration areas and regions to help construct the relevant price indices.

The construction of the pseudo-panel data followed a pooling of the three independent waves of the GLSS data. This achieved a total sample size of 19,209 households. As the analysis focused on rural Ghana, households residing in urban Ghana were dropped, leading to the loss of 6,735 observations (that is, 1,464 households in 1991/92, 2,020 in 1998/99 and 3,251 in 2005/06). Also dropped were households with economically inactive heads (as will be explained later), which totaled 954 individual households (made of 181, 271 and 502 households in GLSS3, 4 and 5, respectively). The pooled panel therefore contains a net total of 11,519 rural households.

Cohorts of generations of households were then constructed from this data in two stages. First, the year of birth of household heads was constructed and used as the first level of aggregation of individual observations into groups or cohorts. This followed the approach of Asiedu et al. (2009), where 8 generations of household groups were identified. These cohorts comprised the generation of household heads born before 1932, those born between 1932 and 1938, and in similar order of 7-year gap until the last and much younger generation of households heads born after 1969. This approach helped retain all the 11,519 rural households available in the pool. Second, these cohorts were further disaggregated by location of residence in respect of the three agro-ecological zones in Ghana. This disaggregation is quite useful in enhancing the homogeneity of household observations within each cohort, more so as natural resources remain an important element in rural household livelihood formation. For the proposed model, this approach to cohort formation therefore already controlled for the potential effect of generation and natural resources (and environment) in household welfare formation. Table 8-7 details the distribution of the sample observations for these rural household cohorts, for each period and for the overall pool. Generally, the cohorts contain over 100 households. The distribution reveal that
50

the pseudo-panel achieved is balanced with a maximum of 567 observations and average number if observations of 159 households. Error in the measurement of the cohort means could therefore be generally ignored (Deaton, 1985; Verbeek, 2007). The estimation of the principal model therefore proceeded along the first asymptotic assumptions discussed earlier. As these cohorts were disaggregated further by household types (that is, agricultural and non-agriculture households), the observations per cohort declined considerably, though the agricultural cohorts still retain some reasonable number of observations. The analysis and interpretation of the model for these household groups therefore proceeded with some caution.

Lastly, some comments about missing data points are in order. Towards optimizing the available sample information, missing data points were imputed using auxiliary information available from the pool; unless such information are unavailable, in which case the missing points were declared missing, generally16. This was the case for those observations of access to electricity and public transport where community information were missing and household-level information provided no insight on the level of access. For access to electricity, 36 observations were affected in the 1998/99 cross-section, while another 231 observations were affected in the 2005/06 crosssection. Affected observations on access to public transport were 136 in 1998/99 and 401 in the 2005/06 cross-section. These were declared missing using the STATA procedure for handling missing values. Thus, the cohort means were estimated based on 11,252 and 10,982 observations on access to electricity and public transport variables, respectively.

However, in the presence of relevant auxiliary data points, it was deemed more appropriate to impute the missing values than to generally declare such values missing. Schlomer et al. (2010) provides some discussion of the best practices. For variables such as the number of years of educational attainment of the household head, missing cases were imputed in multiple steps as the median value for the enumeration area if the proportion of the households literate adults forms 50 to 80 percent of the overall household size. The value of the heads education for households with adult literacy below 50 percent is predicted as the 10th centile of the enumeration areas (EA) and the 90th centile of the same for those households with over 80
16

Missing cases in survey data are not uncommon and could arise for reasons ranging from inapplicable questions, non-response, etc. For the present purpose, these differences are in-material and thus treated generally as missing observations in the analysis.

51

percent adult literacy. Similarly, the asset score for those households with missing values were predicted as the 10th centile if the households welfare index lies between the extreme and higher poverty lines for the respective years, 90th centile if it lies above the higher poverty line and the 10th centile if it lies below the extreme poverty line. This way, one avoids the arbitrary use of mean values in handling such missing data points or the extreme option of excluding affected observations in the analysis (Schlomer et al., 2010; Vyas et al., op. cit.; Gwatkin, 2000).

4.4 The Estimation Method The pseudo-panel data at hand is identified to assume the structure of weakly balanced panel data. For the principal model, both the OLS and within estimators are established to yield consistent, efficient estimates of the model parameters under these conditions (McKenzie (2004), Verbeek and Vella (2005), McKenzie and Antman (2005), Verbeek (2007) and Cameron et al. (2010)). The study therefore proceeds to estimate the empirical model following this argument. In relation to public infrastructure assets, one can assert potential endogeneity17 of these variables from two possible sources. The first relates to the potential effect of wealth in households access to infrastructure services. This argument is addressed by asserting that within the framework of this analysis and by definition, private households necessarily depend on the public sector for access to infrastructure. Thus, unlike private assets, access to public infrastructure is held exogenous to households decisions and behaviour. Wealth effect may therefore matter only in the instance of the actual use of the resource. This is not the measure used here (otherwise total expenses from the use of these assets would have been employed).

The second issue arises from the argument of Wouterse (2010) according to which households, in pursuing migration decisions, may self-select into communities along such criteria as ease of access to livelihood resources, better employment opportunities and other features of the destination community. It therefore controls for such potential endogeneity by using measures of community characteristics such as ethnicity, dominant economic activity, etc. In the present
The argument of endogeneity is generally used here to include instances of sequential exogeneity of variables in the data generation process. This is because when households self-select into destination communities, infrastructure access occurs as factors determined prior to the current period (that is, predetermined rather than strictly endogenous in the model). Similar argument applies to past levels of welfare.
17

52

context, one will observe that as welfare constitute the principal motive for a households migration decision, any factor that influences the choice of households destination community would correlate with welfare, and strongly.

Thus, community-level characteristics cannot meet the stringent requirements of a good instrument18. The present analysis therefore provides an alternative estimation of the postulated model using the lag of the variables as instruments in first difference estimation. Given the preceding argument, public assets are best conceived to be sequentially exogenous in the model. The estimation therefore follows the standard GMM-style instrumentation (see Roodman, 2006). For completeness, the lagged dependent variable is also treated similarly. Private endowments and its interactions and the other explanatory terms are held exogenous, as usual.

For the agricultural and non-agricultural household models, the limited sizes of cohort observations imply the preponderance of measurement-error in the estimation of the population cohort means (Deaton, op. cit.; see Table 8-10 and Table 8-11 in the Appendix). This provides further justification for accounting for the potential endogeneity of the lagged dependent variable, following the classical pseudo-panel argument of measurement-error rather than serial correlation (Collado; op. cit; McKenzie, 2004). The estimation of the two models thus follows the same approach as the principal model though the discussions are limited to the difference GMM estimates.

The STATA software is used to implement these estimators. The GMM estimates for all the models are also corrected for small-sample error in the covariance matrix, more so as it is used as the principal estimator of the agriculture and non-agriculture household models. The relevant test statistics are also specified to facilitate the evaluation of the robustness of the estimated models.

A good instrument exhibits, simultaneously, strong correlation with households access to public infrastructure and no relationship with welfare. The initial approach was to consider augmenting the proffered instruments of Arellano and Bond (1991) with community-level variables such as population, degree of homogeneity (in terms of ethnicity and religion) and main economic activity. However, the argument on how these variables could influence a households access to public infrastructure without any relevant influence on welfare is difficult to advance. The analysis therefore proceeded with the recommended treatment of Roodman (2006).

18

53

5.0

DISCUSSION OF RESULTS

The results of the analysis are presented here based on two broad outputs. These are the qualitative or descriptive analysis and the quantitative (econometric) outputs.

5.1

Qualitative Analyses

Poverty in Ghana since the 1990s is described as predominantly a rural phenomenon (GSS, 2000). A review of the pooled panel confirms this observation as largely valid. As noted in Table 5-1, trends in the countrys poverty indicators show high prevalence of welfare deprivation in the rural sector, relative to the national average since the 1990s. In particular, while the countrys headcount poverty is estimated to be about 52 percent of the national population in the early 1990s, that for the rural sector is found to be 64 percent (using the higher poverty line) of the rural population. It is observed in particular that throughout the focal periods of the study, poverty reduction in rural Ghana lagged consistently behind national poverty rates by at least 10 percentage points, while the poverty gap index for the rural economy lagged at levels 4-5 percentage points higher than the national estimate.

Table 5-1: The Evolution of National and Rural Poverty in Ghana, 1991 - 2005 Poverty Index (%) Higher Headcount Poverty Poverty (%) Line Poverty Gap (%) Lower Poverty Line Headcount Poverty (%) Poverty Gap (%) GLSS 3 Rural National 63.58 23.98 47.13 14.84 51.69 18.48 36.5 11.05 GLSS 4 Rural National 49.45 18.15 34.44 11.24 39.45 13.88 26.85 8.33 GLSS 5 Rural National 39.33 13.52 25.62 8.14 28.56 9.59 18.11 5.67

Note: Estimates based on the national poverty line for each respective population (i.e. rural and national) Source: Authors computation from the Pooled Cross-section of GLSS 3, 4 & 5

The rate of extreme poverty is observed to follow a similar distributional pattern. Specifically, while 36.5 percent of the national population was identified to be poor from the GLSS 3 cross54

section, the fraction in the rural sector was over 47 percent. In the subsequent cross-sections (that is, GLSS 4 and GLSS 5), these reduced to about 34 percent and 26 percent in the rural sector, respectively, while the national average stood at 27 percent and 18 percent, in the same order.

Thus, while overall poverty levels show a generalized decline across the country, the continued non-convergence of the rural and national poverty levels provide some reflection of the inequity in the distribution of welfare outcomes within the country. Indeed, given the significance of the rural sector in the country economy, this trend reinforces the argument of persistent urban-bias in national economy. Underlying this trend is also the low access levels to public physical infrastructure in rural Ghana.

Focusing on public transport, water and electricity, there are considerable levels of stagnation in access in rural Ghana. Particularly, between the early and late 1990s, access to public transport services is identified to have increased from 50 percent to about 71 percent but declined to 67 percent in the mid-2000. Access to the national power grid, on the other hand, increased from about 19 percent in 1991/92, 33 percent in 1998/99 to 43 percent in 2005/06, as reported in Figure 5-1. It ought to be noted that these trends are reported based on the current definition of infrastructure access as the proximity to the resource point and not necessarily access within the residence of rural households.

Similarly, access to water within a distance of 1 km is noted to have increased across the survey periods. The analysis reveal that the fraction of rural population that travel distances of more than 1 km to obtain water for general household use declined from 23 percent in 1991/1992 to about 14 percent in 1998/99. The period between GLSS 4 and GLSS 5 did not see any important transformation, as similar fraction of the population in 1998/99 reported having such access to water. These access levels, however, also conceal any understanding of the quality of water accessed by households, more so as public sector investments in rural water supply systems have also included the construction of boreholes and covered wells. Thus, for many households, the principal dimension of improvement in these access levels is the reduction in the distances by households to obtain water.

55

Figure 5-1: Trends in the Headcount of Individuals with Infrastructure Access, 1991 to 2005, Percent
100 90 80 70 60 50 40 30 20 10 0 85.98 76.79 70.7 49.75 42.89 33.34 18.63 66.59 86.43

(%)

GLSS 3 Public Transport

GLSS 4 Water Poor (=< 1000m)

GLSS 5 Power

Note: Estimates based on the respective sample observations (i.e. rural and national) Source: Authors computation from the Pooled Cross-section of GLSS 3, 4 & 5

In proceeding to discuss the quantitative output, a summary result of the PCA analysis employed in the determination of household asset endowment is presented in Table 8-6 in the Appendix. Most importantly, the Kaiser-Meier-Oklin (KMO) measure suggest sufficiency in sampling adequacy, across each asset type and overall. The weights of furniture, aircondition, tractor, refrigerator and car identify the ownership of these assets as very important in enhancing a households private capital endowment. The ownership of television, stove, fan and video player are also noted to be very significant in households capital endowment.

5.2 Quantitative Analysis: Estimated Models The results of the principal model of the thesis are presented and discussed here. Those for the agricultural and non-agricultural disaggregated models are then subsequently discussed, jointly.

5.2.1 The Principal Model for Rural Ghana The results of the principal econometric model for rural Ghana are reported in Table 5-2. Table 5-3 also report estimates for the same model but accounting for the potential endogeneity of infrastructure access to household welfare outcomes. These results are summarized alongside the
56

OLS estimates and reported in the Appendix (as Table 8-12), for the purposes of comparison. Consistent with the diagnostic guidelines provided in McKenzie (2000), all three estimates show very strong similarities. As expected, the R^2 statistic for both the OLS and within estimates are noted to trend towards unity. Given the sufficiency of the sample observations per cohort, the ensuing discussion will follow the results from the within effect and relate to the OLS and IV outputs, wherever necessary. The models are estimated with robust standard errors, thus correcting for potential autocorrelation within the household groups and heteroscedasticity. Other test statistics including the F-test show that the estimated model is good and has sound explanatory power.
Table 5-2: Infrastructure Access & Household Welfare: Within Estimation of Model (11) for Rural Ghana
Number of obs = 48 Number of groups = 24 Obs per group: min = 2 avg = 2.0 max = 2 F(21,23) = 3956.08 corr(u_i, Xb) = -0.9633 Prob > F = 0.0000 (Std. Err. adjusted for 24 clusters in cohortzone) -----------------------------------------------------------------------------| Robust lnwelfare | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+----------------------------------------------------------------Lnwelfare.L1 | .4819056 .0435159 11.07 0.000 .391886 .5719252 power | .5146498 .1322518 3.89 0.001 .2410661 .7882336 pubtrans | 2.116579 .1043332 20.29 0.000 1.90075 2.332409 water | -.395847 .1163194 -3.40 0.002 -.6364719 -.1552220 assets | -4.251321 .224347 -18.95 0.000 -4.715418 -3.787224 power*assets | .5573921 .0211949 26.30 0.000 .513547 .6012372 pubtrans*as~s| .7879666 .046059 17.11 0.000 .6926863 .8832468 water*assets | 4.052313 .2198443 18.43 0.000 3.59753 4.507095 size | .4512599 .0761102 5.93 0.000 .293814 .6087057 age | -.0501559 .0105759 -4.74 0.000 -.0720339 -.028278 agesq | .0009486 .0000561 16.90 0.000 .0008325 .0010647 educ | -.2095756 .0162304 -12.91 0.000 -.2431507 -.1760005 educsq | .0377376 .001857 20.32 0.000 .0338961 .0415792 male | .3814539 .0792904 4.81 0.000 .2174293 .5454785 migrant | 1.397164 .1054522 13.25 0.000 1.179019 1.615308 public | 1.082491 .1495981 7.24 0.000 .773024 1.391958 pfmlemploy | -9.998839 .3946495 -25.34 0.000 -10.81523 -9.182444 pinfemploy | -7.175434 .4150221 -17.29 0.000 -8.033972 -6.316895 agcashcrp | 5.324293 .1734609 30.69 0.000 4.965461 5.683124 agfoodcrp | 1.590848 .0708467 22.45 0.000 1.44429 1.737405 period2 | .4792846 .0506385 9.46 0.000 .3745309 .5840383 _cons | 2.500703 .8924136 2.80 0.010 .6546051 4.346801 -------------+---------------------------------------------------------------sigma_u | 1.2123498 sigma_e | .01567354 rho | .99983289 (fraction of variance due to u_i) -----------------------------------------------------------------------------Fixed-effects (within) regression Group variable: cohortzone R-sq: within = 0.9987 between = 0.0779 overall = 0.0848

57

Table 5-3: Infrastructure Access & Household Welfare: GMM Estimation of Model (11) for Rural Ghana
Dynamic panel-data estimation, one-step difference GMM -----------------------------------------------------------------------------Group variable: cohortzone Number of obs = 24 Time variable : year Number of groups = 24 Number of instruments = 24 Obs per group: min = 1 F(21, 24) = 1191.90 avg = 1.00 Prob > F = 0.000 max = 1 -----------------------------------------------------------------------------| Robust lnwelfare | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------Lnwelfare.L1 | .4819056 .0792797 6.08 0.000 .3182804 .6455309 power | .51465 .2409434 2.14 0.043 .0173672 1.011933 pubtrans | 2.116579 .1900798 11.14 0.000 1.724274 2.508885 water | -.3958469 .2119168 -1.87 0.074 -.8332216 .0415278 assets | -4.251321 .4087273 -10.40 0.000 -5.094893 -3.40775 power*assets | .5573921 .0386141 14.43 0.000 .4776965 .6370876 pubtrans*as~s| .7879667 .0839127 9.39 0.000 .6147793 .9611541 water*assets | 4.052313 .4005242 10.12 0.000 3.225671 4.878954 size | .4512598 .1386616 3.25 0.003 .1650763 .7374432 age | -.0501559 .0192678 -2.60 0.016 -.0899227 -.0103892 agesq | .0009486 .0001023 9.28 0.000 .0007376 .0011597 educ | -.2095756 .0295693 -7.09 0.000 -.2706038 -.1485475 educsq | .0377376 .0033832 11.15 0.000 .030755 .0447203 male | .3814539 .1444554 2.64 0.014 .0833127 .6795951 migrant | 1.397164 .1921184 7.27 0.000 1.000651 1.793677 public | 1.082491 .2725458 3.97 0.001 .5199844 1.644998 pfmlemploy | -9.998839 .7189935 -13.91 0.000 -11.48277 -8.514909 pinfemploy | -7.175434 .7561095 -9.49 0.000 -8.735967 -5.614901 agcashcrp | 5.324293 .3160204 16.85 0.000 4.672059 5.976527 agfoodcrp | 1.590848 .1290722 12.33 0.000 1.324456 1.85724 period2 | .4792847 .0922559 5.20 0.000 .2888779 .6696915 -----------------------------------------------------------------------------Instruments for first differences equation Standard D.(assets poweriassets pubtransiassets wateriassets size age agesq educ educsq male migrant public pfmlemploy pinfemploy agcashcrp agfoodcrp period1 period2 period3) GMM-type (missing=0, separate instruments for each period unless collapsed) L(1/.).(L.lnwelfare power pubtrans water) -----------------------------------------------------------------------------Sargan test of overid. restrictions: chi2(3) = 24.00 Prob > chi2 = 0.000 (Not robust, but not weakened by many instruments.) Hansen test of overid. restrictions: chi2(3) = 7.79 Prob > chi2 = 0.051 (Robust, but can be weakened by many instruments.) ------------------------------------------------------------------------------

Focusing first on the key variables of interest, the results suggest that in the absence of access to public infrastructure, private capital investments in rural Ghana yields negative effect on welfare. This effect is found to be significant even at the 1 percent level. In respect of public physical capital, some important differences are found, in terms of both the direction and magnitude. In the absence of private capital, households with access to water show negative and significant effect of such access on welfare. While this is contrary to expectation, the OLS and IV estimates

58

even suggest this effect is not significant, statistically. On the other hand, the partial effects of access to electricity and public transport services are positive and statistically significant in determining household welfare. This suggests that irrespective of households endowment in private capital, access to public transport and electricity in rural Ghana have a generally positive and significant effect on livelihood outcomes.

Further, interacted with household private capital endowment, access to all three public infrastructure is observed to have positive and statistically significant impact on welfare (even at the 1 percent level). These results are consistent with the a priori expectation of the thesis and provide some important empirical evidence of the degree of complementarity between key public capital investments and household private assets in rural Ghana. This result is novel in the contemporary literature. It suggests that the impact of access to water on households economic welfare is largely dependent on the level of the households private capital endowment, whereas access to public transport and electricity has positive and significant effect for all households, irrespective of endowment in private assets.

A useful interpretation of this evidence is that the opportunity cost of the distance traveled to access water in rural Ghana increases significantly with the level of the households private capital investments. On the other hand, the effect of access to public transport and electricity infrastructure on household welfare is positive and independent of private capital, though it still increases with increasing levels of households private capital endowment. Here, one can observe that whereas electricity and public transport may feed both directly and indirectly into households income generation and overall economic livelihood outcomes, the effect of improved access to water is largely indirect. The economic impact of the former two public assets would therefore be immediate and revealed in both market and non-market channels, whereas that of the latter flows through the opportunity cost channels. Thus, where the opportunity cost of time is low (as we will find in the case of households with low endowment in private capital investments, for example), the savings in travel time to access water from public water facilities may not sufficiently offset the user cost.

59

Nonetheless, in the presence of private capital investments, all three public assets investigated are observed to enhance the marginal returns of private capital on household welfare. And these effects are highly significant, buttressing the earlier arguments.

At the mean, it is noted that the marginal returns of public infrastructure on household welfare are generally positive and larger for access to public transport systems relative to access to the national grid and water (that is, 1.47, 0.21 and -0.13, respectively). This effect is not surprising as improved access to public transport also connotes enhanced connectivity to local and regional markets. This would also imply reduced transaction cost and competitive prices (in both the supply and demand markets of households). Improved access to public transport therefore provides overwhelming opportunities for generating higher returns on capital investments and asset productivity. The effect on the livelihood outcomes of households across Ghanas rural sector would therefore be positive and high, as established by the results.

Access to electricity also implies greater opportunity for value addition and new jobs, especially in cottage industries. The extent of impact however depends on the scale of households direct involvement in off-farm income generation activities. For water, the most important source of impact is the opportunity cost of time and energy. Relative to the direct cost of use, the negative marginal effect on welfare of access suggests the former outweigh the latter.

Thus, across Ghanas rural economy, these results establish that access to public infrastructure services have important but differential impact on household livelihood outcomes, ceteris paribus. Indeed, whereas the impact of access to public transport system and electricity is significant and generally positive, that of water is conditional on households endowment in private assets and the opportunity cost that it faces.

In respect of the other explanatory terms, the estimated model reveals that past levels of welfare and the size of the household both have direct and significant effect on household welfare outcomes. This is consistent with the thesis hypotheses. The effects of age and education on household livelihood are also observed to show increasing returns such that at low levels of age and years of formal training, these variables have adverse effect on welfare. These effects reverse
60

and turn positive in their impact on household welfare from the age threshold of about 26 years and a critical point in formal education that is equivalent to at least lower primary schooling19.

Again, consistent with the hypothesis of the thesis, the difference in welfare between male and female headed households is noted to be positive and significant in rural Ghana, all things being equal. A similar observation is identified for the migrant status of the head of the household. Thus, migration in rural Ghana is found here to provide an important avenue for enhancing household welfare outcomes. Indeed, as the model already instruments differences in natural resource endowments across households and accounts for differences in access to public infrastructure on welfare as well, this result provides a quantitative estimate of the contribution of realized migration to livelihood outcomes for reasons relating especially to employment and job search, education, marriage and family.

Finally, household livelihood outcomes are also found to show significant differences among the different sectors of employment of the household head. The results reveal that relative to households with self-employed heads, those households with heads employed in formal or in informal private sectors experience lower welfare outcomes. The estimated coefficients reveal welfare differences that are nearly twice those of the self-employed heads. On the other hand, households with heads in agricultural employment (both cash and non-cash) as well as in public employment experience livelihood outcomes that are positive and significantly higher than the reference category. For households in cash (or export) agricultural sub-sector, this difference is found to be very large, reflecting the high returns identified with export-oriented agroproduction.

5.2.2 Agriculture and Non-Agricultural Household Models

Aside the overall model for Ghanas rural population, the analysis proceeded to investigate potential differences in the effect of public infrastructure assets across two broad categories of households in rural Ghana. These are agricultural households, comprising both cash and non19

This specific threshold is potentially influenced by the large number of household heads with little or no formal education in rural Ghana. Yet still, given the low level of technical investment in agriculture and the other production activities, one can consider such threshold as reasonable.

61

cash agricultural sectors, and the non-agricultural households, constituting public sector employees, formal and informal private sector employees and the self-employed. The estimated model for these two household types are summarized and reported in the Appendix (as Table 8-13 and Table 8-14, respectively). Though the model for is specified to correct for small-sample errors as well as heteroscedasticity and autocorrelation within the cohort observations, the results for the non-agricultural households should be considered with caution due to the combined inadequacy of cohorts size and sample (see Table 8-10 and Table 8-11). Focus is maintained on the IV estimates as explained earlier.

Unlike the non-agricultural household model where none of the variables in the model show any important effect on household welfare, all three alternative methods of estimation, some variables of the agricultural household model reveal some effects of analytical importance.

In particular, access to public transport reveals a positive and significant effect on welfare, irrespective of the households asset endowment. Interacted with the households capital endowment, this effect turns negative and marginally significant. Access to electricity shows negative but non-significant effect in the absence of capital investments but a positive effect on welfare when interacted with the households capital investment. Access to water, on the other hand, reveals a negative and significant effect on welfare, at the 10 percent significance level. Upon interaction with households asset endowment, the effect is still negative and significant, though marginally.

Altogether, the trend in these results suggest that among Ghanas rural agricultural households, access to public transport services have a positive and significant impact on livelihood outcomes, irrespective of the households endowment in private assets. The overall marginal effect is estimated to be 1.51, which is large and suggests improvement in public transport systems in rural Ghana as a potent poverty reduction strategy. The marginal effects of access to water and electricity on the welfare of these households are however noted to be negative (that is, -0.49 and -1.91, respectively). These two assets also complement households asset endowment poorly.

62

In explaining the observed negative marginal effect of infrastructure access on agricultural household welfare, one can simply point out that access to and potential use of electricity and water infrastructure implies increased responsibility in user-cost. This may include contributions at the community level. Thus, where the economic returns on such access do not outweigh the cost, households without access may prove to be better off in welfare than those with access. The fact that electricity and public water sources are not fundamental for agricultural production may reinforce this inverse effect. Given the dominance of low skill, low technology and low valueadded small scale production in the countrys agricultural sector, such negative economic returns is plausible20.

In fact, this could also explain the relatively high marginal returns of capital endowment on welfare outcomes among these households, where a percent increase in households private capital is estimated to yield over 80 percent returns on household welfare, all things being equal. In other words, the scope and scale of the countrys rural-based agricultural production allows the injection of private capital investments and the provision of public infrastructure to generate disproportionately high positive returns on household livelihood outcomes, whereas improved access to water and electricity do not appear to generate sufficient returns to offset any user cost associated with these assets.

Among the other explanatory factors, the size of the households human resources and migration are noted to show important positive effect on welfare. This is consistent with the a priori expectation and reveals the extent of dependence of the countrys agricultural households on human labour (especially drawn from within the household). The age, years of formal education and the gender of the household head do not show any important effect on livelihood outcomes, statistically.

20

As an example, a recent estimate of post-harvest losses in rural Ghana ranged from 20-50 percent (USDA, 2009). Due to weaknesses infrastructure access, transport costs are also suggested to account for about 60 percent of consumer prices. This implies that in the absence of strong connection to markets via efficient transport network and marketing opportunities, the impact on welfare of any productivity increases from access to water and electricity would only be whittled away by such losses.

63

6.0

POLICY RECOMMENDATIONS AND CONCLUSIONS

In this chapter, the recommendations from the above findings and the limitations and suggestions for future research are provided. These are concluded with a summary of the key findings.

6.1 Policy Recommendations The empirical findings suggest that public physical infrastructure show differential impact on the welfare of rural households. This is to the extent that except public transport services whose impact is consistently positive, that of electricity and water depends on private asset endowment.

Thus, towards the agenda of sustained poverty reduction and efficient targeting of development budgets, public policy in Ghana must prioritize the development of functional transport network in rural Ghana. To optimize the welfare impact, policy must also target the provision of electricity and water services to rural communities under conditions where the levels of private capital holdings are sufficient. This will be reasonable where such community already maintains sufficient access to functional transport networks.

Among rural agricultural households, policies aimed at sustaining households capacity to deepen private capital endowment while increasing connectivity to public transport systems are recommended, as such assets show strong positive effect on welfare outcomes. The level of complementarity between private capital on one hand and electricity and water infrastructure on the other is noted to be low (and actually negative). Electricity and water infrastructure therefore have relatively low utility in enhancing livelihood outcomes among these households while public transport and private capital show high utility. Policy must prioritize the latter set.

6.2 Limitations of the Thesis and Suggestions for Future Research A weakness of this study is the choice of real household consumption per capita as the index of household welfare. This is because consumption welfare does not necessarily measure other important aspects of household well-being such as happiness, health and social security. Future studies could employ more comprehensive index of welfare in evaluating the nexus.
64

Public physical assets exist at different levels of quality. The present study focused on access levels while assuming homogeneity in quality. However, it may be reasonable to think that differences in the quality of services such as the quality of water or reliability of electricity and public transport services could impact differentially on household welfare. This is an issue of significant implication for development policy, which is worth investigating in future studies.

On different platforms, some questions have been raised about the quality of data from the LSMS in most developing countries. While such questions may find basis from different dimensions of cross-sectional studies, this thesis maintains the view that no known crosssectional databases are able to achieve the standards applied in LSMS surveys in many developing economies. Particularly, the design and sampling processes for LSMS are based on well-established sampling frames usually drawn from census data and nationwide household listing exercises. The survey design and sampling is therefore well structured and internally valid. Further, concepts, measures, questions and the overall survey instruments are developed to follow standard definitions that allow comparison of responses across time and over space. Thus, many key but basic technical concepts in microeconomic analysis such as the definition and identification of household units and head, age, occupational classifications, farm, units of measurement of weights, as well as even income and expenditure measures are well defined and never left to the interpretation of the enumeration personnel.

As admitted elsewhere, the use of this data (especially in an extensive form achieved in this thesis) is very tasking and requires extensive knowledge of the instruments and data management skills. This, the thesis maintains remain the challenge of the analyst. Nonetheless, the World Bank is currently known to be involved in a process of harmonizing the LSMS across countries. It is suggested that some focus must be given to enhancing the quality of the data for much disaggregated use. Here, the community surveys deserve special mention.

Lastly, given the small sample sizes of cohorts in the non-agricultural household category, the estimated model for those households were not found robust enough to provide any meaningful insight. It is suggested that future studies should focus on this household type.
65

6.3 Conclusions In the context of providing some rigorous, empirical understanding of precisely how much infrastructure matters to economic development and poverty reduction, this study investigated the direction and size of relationship between access to three different public infrastructure services and household welfare in rural Ghana. This has been based on pseudo-panel modeling, using three waves of household livelihood surveys.

In so doing, the study provided a new construction of the SLF and proceeded to test this framework on the premise that infrastructure services generally occur as among those public assets whose availability and access levels are exogenous to household decisions but nonetheless complement household capital endowments in ways that eventually influence welfare outcomes. Indeed, the empirical findings corroborate this argument and reveal that across Ghanas rural economy, access to different public infrastructure, specifically public transport, electricity and water infrastructure have important but differential impact on household livelihood outcomes. The extent of complementarity between these assets and household private endowment are also found to be important and shows differential impacts on welfare as well.

Specifically, across the rural sector, access to public transport and electricity is positive and significant, while that of water depends on households private endowments. Among agricultural households, access to public transport system shows significant and positive impact on welfare, irrespective of the households asset endowment. The marginal effect of water and electricity is nonetheless negative though private capital has substantial positive impact on welfare. Public investment in functional rural transport systems is therefore identified to be a potent instrument for poverty reduction in rural Ghana. Development policy must prioritize this service. To maximize the welfare impact of investments in electricity and water infrastructure, it is recommended that these services be targeted at rural communities with sufficient access to transport network and private capital endowments.

66

7.0

REFERENCES

AFD/World Bank. 2010. Africas Infrastructure: A Time for Transformation. A Report of the Agence Franaise de Dveloppment and the World Bank. World Bank, Washington, DC. AfDB. 1999. African Development Bank Report 1999 Summary. Oxford University Press, New York. AfDB. 2010. Infrastructure Deficit and Opportunities in Africa. Economic Brief, Vol. 1, Issue September 2010. The African Development Bank Group, Chief Economist Complex, Tunisia AfDB. 2011. The Africa Infrastructure Knowledge Program. A web-based resource of the African Development Bank Group, accessed February 21, 2011. http://www.infrastructureafrica.org/about Arrelano, M and S. Bond. 1991. Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. The Review of Economic Studies, 58, pp. 277 297. Ashley, Caroline and Diana Carney. 1999. Sustainable Livelihoods: Lessons from Early Experience. Department for International Development, London. Ayogu, Melvin. 2007. Infrastructure and Economic Development in Africa: A Review. Journal of African Economies, 16 (supplement 1): 75 126, Oxford University Press. Balgati, Badi H. 2005. Econometric Analysis of Panel Data, 3e. John Wiley and Sons Ltd, West Sussex, UK. Bates, Robert H. 2001. Markets and States in Tropical Africa. University of California Press BoG. 2011. Statistical Bulletin. February 2011 Edition. Bank of Ghana, Accra. Brennerman, Adam and Michel Kerf. 2002. Infrastructure and Poverty Linkage: A Literature Review. The World Bank, Washington, DC. Cameron, Colin A. and Pravin K. Trivedi. 2010. Microeconometrics Using STATA. STATA Press Publication, College Station, Texas. Chambers, Robert and Gordon R. Conway. 1991. Sustainable Rural Livelihoods: Practical Concepts for the 21st Century. IDS Discussion Paper 296, IDS, Brighton. Clarke, Jane and Diana Darney. 2008. Sustainable Livelihoods Approaches: What Have We Learnt? A Seminar Report, ESRC, London. Cleary, Seamus. 1989. Structural Adjustment in Africa, Trcaire Development Review, p. 4159, Dublin. Collado, M. Dolores. 1997. Estimating Dynamic Models with Time-Series of Independent Cross-sections, Journal of Econometrics 82, p. 37 62. North London. Commission for Africa. 2005. Our Common Interest. Report of the Commission for Africa. Coulombe and McKay. 2008. The Estimation of Components of Household Incomes and Expenditures: A Methodological Guide based on the Llast Three Rounds of the Ghana Living Standards Survey, 1991/1992, 1998/1999 and 2005/2006. A Report of the Ghana Statistical Service, GSS, Accra. Deaton, Angus. 1985. Panel Data from Time Series of Cross-sections. Journal of Econometrics 30, 109 126, Elsevier Science Publishers, North Holland. DfID. 1999. Sustainable Livelihood Guidance Sheet: Introduction. Department of International Development, UK

67

DfID. 1999. Sustainable Livelihoods Guidance Sheet. Department of International Development (DfID), London. Diao, Xinshen. 2010. Economic Importance of Agriculture for Sustainable Development and Poverty Reduction: Findings from a Case Study of Ghana. A Research Paper for the Working Party on Agricultural Policy and Markets, Global Forum on Agriculture. OECD, Paris. Dillon, Andrew, Manohar Sharma and Xiaobo Zhang. 2011. Estimating the Impact of Infrastructure and Extension Services in Rural Nepal. Research Monograph, International Food Policy Research Institute, Washington, DC. Ellis, Frank. 2000. Rural Livelihood and Diversity in Developing Countries. Oxford University Press. Foster, Vivien and Cecilia Briceno-Garmendia. 2010. Africas Infrastructure: A Time for Transformation. Agence Francaise de Development and the World Bank. Friedman, Milton. 1957. The Theory of the Consumption Function. Princeton University Press. Gasner, Katharina. 1998. An Estimation of UK Telephone Access Demand Using Pseudo-Panel Data. Utility Policy 7, p. 143 154. The European Institute, London School of Economics, London. GSS. 2000. Poverty Trends in Ghana in the 1990s. Ghana Statistical Service, Accra. GSS. 2008. Ghana Living Standard Survey: Report of the Fifth Round. Ghana Statistical Service, Accra. GSS. 2011. 2010 Population and Housing Census: Provisional Results Summary of Findings. Ghana Statistical Service, Accra. Gwatkin, D.R., Rutstein S., Johnson K., Suliman E., Wagstaff, A. and Agbesi, A. 2007. Socioeconomic Differences in Health, Nutrition and Population within Developing Countries. The World Bank. Gyimah-Brempong, Kwabena and Elizabeth Asiedu. 2009. Remittances and Poverty in Ghana. A Research Paper presented at the 4th African Economic Conference, Addis Abbaba, Ethiopia. Hardin, Garreth. 1968. The Tragedy of the Commons. Science, Vol. 162, p. 1243 1248. Huchet-Bourdon, Marilyn, Mariana Vijil and Chantel Le Mouel. 2010. Aid for Trade: A Survey. 4th Journal of Research in Social Sciences. The French Society of the Rural Economy, France. Hussein, Karim. 2002. Livelihoods Approaches Compared: A Multi-Agency Review of Current Practice. Overseas Development Institute (ODI) and the Department for International Development (DFID), London. IFPRI. 2005. Increasing Access to Infrastructure for Africas Rural Poor. International Food Policy Research Institute, Washington, DC. Inoue, Atsushi. 2008. Efficient Estimation and Inference in Linear Pseudo-Panel Data Models. Department of Agricultural and Resource Economics, North Carolina State University. Kessides, Christine. 1993. The Contributions of Infrastructure to Economic Development : A Review of Experience and Policy Implications. World Bank Discussion Paper 213, The World Bank, Washington DC. Koutsoyiannis, A. 2001. Theory of Econometrics, 2e. Palgrave Publishers Ltd, New York. Masika, Rachel and Sally Baden. 1997. Infrastructure and Poverty: A Gender Analysis. A Report prepared for the Gender Equality Unit, Swedish International Development Agency (SIDA), BRIDGE, Institute of Development Studies, University of Sussex, UK.
68

Maunder, D, A. Davies, Bryceson D., J. Howe, T. Mbara and T. Onweng. 2001. Sustainable Livelihoods, Mobility and Access Needs in Urban and Peri-urban Areas. A paper presented to the 20th Annual South African Transport Conference 16 July 2001 Meeting the transport Challenges in Southern Africa, CSIR International Convention Centre, Pretoria. McKenzie, David and Francisca Antman. 2005. Earnings Mobility and Measurement Error: A Pseudo Panel Approach. World Bank Policy Research Working Paper 3745. Washington, DC. McKenzie, David J. 2000. Estimation of AR(1) Models of Unequally Spaced Pseudo Panels. Econometric Journal, Volume 4, p. 89-108. Royal Economic Society, USA. McKenzie, David J. 2004. Asymptotic Theory for Heterogeneous Dynamic Pseudo-Panels. Journal of Econometrics, Volume 120, p. 235 - 262. Royal Economic Society, USA. McKenzie, David. 2004. Measuring Inequality with Assets. Department of Economics, standford University, USA. Mensah, Emmanuel Joseph, Ernest Aryeetey and George Owusu. 2009. An Analysis of Poverty and Regional Inequality in Ghana. GDN Working Paper Series, Number 27, Global Development Network (GDN), New Delhi, India. Mensah, Emmanuel Joseph, Matthew Geddes, Luisa Natali and Peter Quartey. 2009a. OEC DAC-PDE Thematic Study on Untied Aid: Ghana Country Study. Overseas Development Institute, London. Modigliani, Franco and Richard Brumberg. 1954. Utility Analysis and the Consumption Function. An Interpretation of Cross-section Data. Rutgers University Press. Moffit, Robert. 1993. Identification and Estimation of Dynamic Models with a Time-Series of Cross-sections. Journal of Econometrics 59, p. 99 123. North Holland. MT and GSS. 2008. Transport Indicator Database: Statistical and Analytical Report (2000 2006). Ministry of Transport and the Ghana Statistical Service, Accra. Murray, Janet and Mary Fergusson. 2001. Women in Transition out of Poverty. Women and Economic Development Consortium, Toronto. NDPC. 2003. Ghana Poverty Reduction Strategy Paper, 2003 2005, National Development Planning Commission (NDPC), Accra, Ghana NDPC. 2005. Growth and Poverty Reduction Strategy Paper, 2006 2009, National Development Planning Commission (NDPC), Accra, Ghana Robinson, Joan. 1971. Economic Heresies: Some Old Fashioned Questions in Economic Theory. London Roodman, David. 2006. How to to xtabond2: An Introduction to Difference and System GMM in STATA. Working Paper No. 103. Centre for Global Development. Schlomer G.L., Bauman S. And Card N.A. Best Practices for Missing Data Management in Counselling Psychology, Journal of Counseling Psychology, Vol. 57, No. 1, p. 1-10. American Psychology Association. Scoone, Ian. 1997. Sustainable Rural Livelihoods: A Framework for Analysis. IDS Working Paper 72. IDS, UK. Sen, Amartya. 1981. Ingredients of Famine Analysis. Availability and Entitlements, The Quarterly Journal of Economics, Vol. 96, No. 3 (p. 433 464), MIT Press. Stiglitz, Joseph. 2011. Fair Trade for All: How Trade Can Promote Development. Carnegie Council for Ethics in International Affairs, accessed online on August 10, 2011 via: http://www.cceia.org/resources/transcripts/5339.html/:pf_printable? Udry, Chris and Rohini Pande. 2005. Institutions and Development: A View from Below. Centre
69

Discussion Paper No. 928, The Economic Growth Centre, Yale University, New Haven. UNECA. 2010. Economic Report on Africa 2010: Promoting High-Level Sustainable Growth to Reduce Unemployment in Africa. United Nations Economic Commission for Africa, Addis Ababa, Ethiopia. USDA Foreign Agricultural Service. 2009. Assessment of Local Production for School Feeding in Ghana. United States Development Agency, Ghana, Accra. Verbeek, Marno and Francis Vella. 2000. Estimating Dynamic Models from Repeated Crosssections. Centre for Economic Studies, K.U. Leuven, Belgium. Verbeek, Marno and Theo Nijman. 1993. Minimum MSE Estimation of a Regression Model with Fixed Effects from a Series of Cross-sections. Journal of Econometrics 59, p. 125 136. North Holland. Verbeek, Marno. 2007. Pseudo Panels and Repeated Cross-sections. RSM Erasmus University and Nestpar, Department of Financial Management, The Netherlands. Vyas, Seema and Lilani Kumaranayake. 2006. Constructing Socio-economic Status Indices: How to use Principal Components Analysis. Oxford University Press. Winters, Paul, Benjamin Davies and Gero Carletto. 2009. Assets, Activities and Rural Income Generation: Evidence from a Multicountry Analysis. World Development 2009, pp. 1435 1459. Wooldridge, M. Jeffrey. 2003. Introductory Econometrics: A Modern Approach, 2e. SouthWestern, Thompson Publishers, USA. World Bank. 2004. World Development Report 2004: Making Services Work for the Poor. World Bank, Washington, DC. World Bank. 2011. African Development Indicators. A Publication of the World Bank Group, Washington, DC. World Bank. 2011. As Ghana Grows, Demand for Water Flows. A World Bank Country Newsletter, Issue January 3, 2011. Accra, Ghana. World Bank. 2011. Ghana: Overview. A web-based resource on Ghana by the World Bank Group, Washington, DC, accessed February 22, 2011. http://go.worldbank.org/WEBSITE/EXTERNAL/COUNTRIES/AFRICAEXT/GHANAEXT N Wouterse, Fleur. 2010. Internal Migration and Rural Service Provision in Northern Ghana. IFPRI Discussion No. 00952. Development Strategy and Governance Division, IFPRI.

70

8.0

APPENDIX

Figure 8-1: The Sustainable Livelihoods Framework (based on the DFID Schema)

Source: Sustainable Livelihoods Guidance Sheet (DFID,1999) Table 8-1: The Evolution of Ghanas Economic Structure since the 1980s

Year 1980 1985 1990 1995 1999 2000 2001 2004 2005 2006 2007 2008 2009

Agriculture

Service

Industry

Manufact uring

GDP [constant 2000] (million US$) 2,639.87 2,583.66 3,266.89 4,028.92 4,799.89 4,977.49 5,176.59 6,009.52 6,364.08 6,771.38 7,208.79 7,816.53 8,180.60

(% GDP)
60.06 48.43 45.07 42.70 39.93 39.41 39.33 41.55 40.94 30.40 29.05 30.96 31.67

(% GDP)
27.64 33.58 38.08 30.55 31.71 32.20 32.54 31.39 31.60 48.80 50.20 48.61 49.47

(%GDP)
12.31 17.99 16.86 26.74 28.36 28.39 28.14 27.06 27.46 20.80 20.75 20.42 18.86

(% GDP)
8.10 12.44 9.81 10.27 10.07 10.08 10.05 9.57 9.46 10.24 9.15 7.94 6.90

GDP Growth Rate (%) 0.47 5.09 3.33 4.11 4.40 3.70 4.00 5.60 5.90 6.40 6.46 8.43 4.66

Inflation (%) 50.07 10.31 37.26 59.46 12.41 25.19 32.91 12.62 15.12 10.92 10.73 16.52 19.25

Source: Authors Compilation from World Banks World Development Country Data (2011)

71

Figure 8-2: Trends in International Trade and Finance (% of GDP) since the 1980s
80 70 60 50 40 30 20 10 0

Imports (% GDP) International Receipts (Exports & Remittance, % GDP) FDI, Net Inflow (% GDP) Note: International receipts are estimated to include revenue from exports and remittances Source: World Bank Country Data (2011)

Table 8-2: Trends in Agricultural Land Allocation and Productivity in Ghana, 1994 - 2006
1994 2000 112 87 121 6100 1500 4600 2006 159 188 149 7195 1835 5360 1994-06 0.91 1.56 0.69 4.1 7.01 3.31 1994-99 -4.77 -4.97 5.39 13.62 3.59 2000-06 5.97 13.67 3.62 2.79 3.42 2.58

Land Productivity (/hectare)

Crop and Cocoa Cocoa Crops other than cocoa 155 162 154 4500 687 3813

Land Allocation (in 1000 hectares)

Cultivated Land Cocoa land Crops other than cocoa Source: Diao, 2010

72

Table 8-3: Proportion of Households Main Sources of Water in Rural Ghana, GLSS 3 5 (%) 1991 1998 2005

National Rural Coast

National

Source Inside Plumbing Public Standpipe Borehole Neighbor Water Vendor Well Natural Sources Total

2.7 18.6 2.7 7.4 1 18.5 49.2 100

3.3 4.9 28.8 0.6 0.1 14.1 48.2 100

1.4 5.9 24.6 1.1 0.4 17.1 49.6 100

15.1 10.2 14.6 9.2 1.5 14.1 35.2

100

4.2 26.7 15 5.6 3.3 19.7 25.5 100

4.6 9.8 31 8 0.5 10.1 36 100

0 0 36 0.8 2.3 17.7 43.2 100

14.7 12.3 18.5 14.1 3.6 12 24.9

100

4.2 14.6 27.7 11.4 2.4 10.2 29.5 100

2 7.2 55.5 2.9 0.1 11.9 20.3 100

2.2 1.2 53.4 2.5 8.7 32 100

16.3 10.7 30.4 14.3 2.2 10.3 15.8

100

Source: GSS (2007)

Table 8-4: Proportion of Households Main Toilet Facility, GLSS 3 5 by Agro-ecological Zone (%) 1991 1998 2005

Rural Savannah

Rural Savannah

Rural Savannah
0.7 20.8 9.2 0.3 69

National

National

Rural Forest

Type Flush Toilet Pit Latrine Pan/Bucket KVIP Other

1.4 56.2 3.5 6.2 32.7

1.9 81.8 5.4 3.2 7.6

0.6 32.8 2.1 2.3 62.2

7.1 50.2 11.3 6.8 24.6

Rural Coast

2.1 47.5 4.5 23 22.9

1.5 60.7 4.7 23.3 9.8

0.3 16.7 0.3 11.2 71.5

6.6 35.4 6.8 29.1 22.1

1.4 43.5 27.2 0.1 27.8

1.1 57.6 33.5 0.3 7.5

10.2 31.5 37.4 1.3 19.6

Total
Source: GSS (2007)

100

100

100

100

100

100

100

100

100

100

100

100

Figure 8-3: Proportion of Household Connected to the National Power Grid, GLSS 3 5 (%)

Source: GSS (2007)

73

National

Rural Coast Rural Forest

Rural Coast

Rural Forest

National

Rural Coast Rural Forest Rural Savanna h

Rural Forest Rural Savanna h

Rural Forest Rural Savanna h

Rural Coast

Table 8-5: Summary of Some Key Literature on the Estimation of Pseudo-Panel Models Author(s) Year Title Key Issue
Deaton, Angus Verbeek, Marno & Theo Nijman 1985 1993 Panel Data from Time-Series of Cross-sections Minimum MSE Estimation of a Regression Model with Fixed Effects from a Series of Cross-sections Proposes error-in-variables estimation based on sample cohort means of the variables in the model. Follows Deaton but proposes the adjusted error-invariables estimator. Concludes with recommendation that adjusted error must be less than the total variance (to achieve balance between precision and min. MSE) Introduces IV interpretation of Deatons work. Proposes 2SLS estimation for estimation of dynamic models. Ignores potential fixed effect in model estimation Builds on Deaton to propose consistent, errorcorrected GMM estimation (especially for finite T), then measurement error-corrected within group estimator. Finds GMM estimation more robust. Corrects for cohort fixed effect Builds on Collado (1997), evaluates conditions for consistency in dynamic model estimation.

Remarks
Static, error component model. Consistent as , fixed. Consistent even under small and . Converges to fixed effect and error-in-variable for error adjustment factor, = (0, 1), respectively Dynamic model based on predicted values of lagged dependent variable. No distribution theory. Error-corrected GMM, consistent as , for fixed & . Error-corrected within group estimator, which is consistent as T . Dynamic models, follows from Collado. Assumes large N and fixed T asymptotic. Presents OLS as consistent estimator, under condition of and fixed T. Dynamic models from unequallyspaced surveys. Presents nonlinear LS as more consistent Proposes the augmented IV estimation Proposes consistent and rigorous estimates and test statistics for existing estimators. Maintains stringent identification condition.

Moffit, Robert

1993

Identification and Estimation of Dynamic Models with a Time-Series of Cross-sections Estimating Dynamic Models with Time-Series of Independent Crosssections

Collado, M. Dolores

1997

Verbeek, Marno & Francis Vella McKenzie, David J. McKenzie, David J. Verbeek, Marno Inoue, Atsushi

2000

Estimating Dynamic Models from Repeated Cross-sections

2004

2001

2007 2008

Asymptotic Theory for Heterogeneous Proposes OLS, GLS and IV estimation of dynamic Dynamic Pseudo-Panels models. Discusses empirical application to data on inter-generational Taiwanese consumption growth Estimation of AR(1) Models with Proposes (NL LS, OMD and 1-step) estimation Unequally Spaced Pseudo-Panels approaches for dynamic pseudo-panel models with non-linear parameter restrictions Pseudo-Panels and Repeated CrossProvides overview of the literature on the sections estimation approaches and the asymptotics Efficient Estimation and Inference in 1. Provides further evidence of inconsistency of Linear Pseudo-Panel Data Models OLS estimation in pseudo-panels; 2. Shows t-test is not valid and proposes alternative and more robust t-statistic; 3. alternative GMM estimator with better precision and valid t-test

74

Table 8-6: Principal Component Analysis of Household Asset Ownership, Summary Results Variable Furniture Sewing Machine Stove Fridge Aircondition Fan Radio Cassetter Record Player Video Player 3-in-1 Radio System Television Tractor Plough Cart Sprayer Car Overall Stats: Mean 0.549 0.253 0.069 0.060 0.000 0.095 0.363 0.337 0.024 0.026 0.019 0.110 0.000 0.004 0.002 0.016 0.011 N:1049 Std. Dev. 0.498 0.435 0.254 0.238 0.022 0.293 0.481 0.473 0.154 0.158 0.135 0.313 0.020 0.061 0.039 0.124 0.106 Scoring Factor 0.213 0.157 0.300 0.427 0.078 0.444 0.037 0.197 0.238 0.360 0.081 0.443 0.028 0.005 0.019 0.047 0.148 Rho: 0.179 Weight 0.428 0.360 1.183 1.791 3.550 1.517 0.076 0.416 1.548 2.283 0.602 1.414 1.444 0.084 0.495 0.377 1.391 KMO* 0.866 0.821 0.898 0.835 0.773 0.820 0.550 0.761 0.814 0.854 0.740 0.834 0.668 0.600 0.576 0.621 0.864 0.789

* Kaiser-Meier-Oklin (KMO) index of sampling adequacy

75

Table 8-7: Sample Distribution of Rural Households by Ecological Zone/Generation Cohorts

| Year Group(zone birth) | 1991 1998 2005 | Total --------------------+---------------------------------+---------Coastal, <= 1931 | 184 106 54 | 344 Coastal, 1932-1938 | 61 114 46 | 221 Coastal, 1939-1945 | 95 86 68 | 249 Coastal, 1946-1951 | 79 121 70 | 270 Coastal, 1952-1957 | 99 101 119 | 319 Coastal, 1958-1963 | 99 147 103 | 349 Coastal, 1964-1969 | 62 104 114 | 280 Coastal, >=1970 | 19 81 241 | 341 Forest, <= 1931 | 268 209 103 | 580 Forest, 1932-1938 | 105 176 134 | 415 Forest, 1939-1945 | 148 190 205 | 543 Forest, 1946-1951 | 187 221 199 | 607 Forest, 1952-1957 | 192 246 257 | 695 Forest, 1958-1963 | 217 345 271 | 833 Forest, 1964-1969 | 187 264 299 | 750 Forest, >=1970 | 31 253 629 | 913 Savannah, <= 1931 | 182 102 105 | 389 Savannah, 1932-1938 | 83 92 143 | 318 Savannah, 1939-1945 | 110 100 197 | 407 Savannah, 1946-1951 | 146 130 181 | 457 Savannah, 1952-1957 | 120 144 244 | 508 Savannah, 1958-1963 | 108 170 280 | 558 Savannah, 1964-1969 | 82 121 305 | 508 Savannah, >=1970 | 14 84 567 | 665 --------------------+---------------------------------+---------Total | 2,878 3,707 4,934 | 11,519

Table 8-8: Descriptive Statistics of key Variables Rural Sample Pooled Data
Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------Lnwelfare | 11519 13.85909 .7227655 10.74189 17.09211 power | 11519 .2973694 .4571208 0 1 pubtrans | 11519 .6234748 .4845362 0 1 water | 11519 .8765518 .3289653 0 1 assets | 11519 .0162566 1.724994 -1.115927 15.44519 power*assets | 11519 .1869879 1.236608 -1.115927 12.9606 pubtrans*a~s | 11519 .0989351 1.437253 -1.115927 12.9606 water*assets | 11519 .0278773 1.650608 -1.115927 15.44519 migrant | 11519 .4144457 .4926475 0 1 size | 11519 .5654734 .2637366 .1 1 age | 11519 45.82195 15.36032 16 99 agesq | 11519 2335.57 1554.844 256 9801 educ | 11519 3.550048 5.325805 0 24 educsq | 11519 40.96458 75.54846 0 576 male | 11519 .7307926 .4435672 0 1 public | 11519 .054345 .226707 0 1 pfmlemploy | 11519 .0282143 .1655916 0 1 pinfemploy | 11519 .0331626 .1790688 0 1 agcashcrp | 11519 .095668 .2941482 0 1 agfoodcrp | 11519 .5950169 .4909101 0 1 selfemploy | 11519 .1935932 .3951309 0 1

76

Table 8-9: Descriptive Statistics of key Variables Rural Sample Pseudo-Panel

Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------lnwelfare | 72 13.85313 .314662 13.30721 14.41828 power | 72 .2846443 .1871267 0 .755102 pubtrans | 72 .6109453 .1885165 .1483517 .9090909 water | 72 .8540309 .106593 .4736842 .9708738 assets | 72 -.0491143 .4571438 -.7529871 1.358263 power*assets | 72 .1514874 .2065785 -.1963543 .8514639 pubtrans*a~s | 72 .0595832 .2632722 -.3874367 .8895311 water*assets | 72 -.0383753 .4049382 -.7026674 1.054433 migrant | 72 .3920744 .1167157 .1428571 .6108949 size | 72 .5787157 .0916099 .4210857 .7956989 age | 72 47.36629 16.37999 20.07143 81.6 agesq | 72 2517.098 1640.356 404.2143 6705.848 educ | 72 3.410585 2.000857 .3356643 7.645161 educsq | 72 37.27102 20.75479 3.843137 77.56061 male | 72 .7119079 .1199017 .3703704 .9285714 public | 72 .0530587 .0442085 0 .1875 pfmlemploy | 72 .0265492 .0266951 0 .1120332 pinfemploy | 72 .0313374 .0313794 0 .1410788 agcashcrp | 72 .0784115 .1030948 0 .407767 agfoodcrp | 72 .6030362 .154884 .2962963 .8666667 selfemploy | 72 .2076069 .0931472 .0686275 .4567901

Table 8-10: Sample Distribution of Agro-Households by Ecological Zone/Generation Cohorts

| Year Group(zone birth) | 1991 1998 2005 | Total --------------------+---------------------------------+---------Coastal, <= 1931 | 125 64 35 | 224 Coastal, 1932-1938 | 33 73 36 | 142 Coastal, 1939-1945 | 60 45 47 | 152 Coastal, 1946-1951 | 36 50 49 | 135 Coastal, 1952-1957 | 50 51 80 | 181 Coastal, 1958-1963 | 42 63 58 | 163 Coastal, 1964-1969 | 33 33 62 | 128 Coastal, >=1970 | 9 24 112 | 145 Forest, <= 1931 | 214 173 91 | 478 Forest, 1932-1938 | 77 145 114 | 336 Forest, 1939-1945 | 98 140 161 | 399 Forest, 1946-1951 | 118 146 142 | 406 Forest, 1952-1957 | 113 140 189 | 442 Forest, 1958-1963 | 132 214 196 | 542 Forest, 1964-1969 | 112 150 219 | 481 Forest, >=1970 | 21 151 362 | 534 Savannah, <= 1931 | 154 94 93 | 341 Savannah, 1932-1938 | 64 76 122 | 262 Savannah, 1939-1945 | 86 80 164 | 330 Savannah, 1946-1951 | 116 109 149 | 374 Savannah, 1952-1957 | 84 114 195 | 393 Savannah, 1958-1963 | 81 125 234 | 440 Savannah, 1964-1969 | 65 95 239 | 399 Savannah, >=1970 | 11 67 451 | 529 --------------------+---------------------------------+---------Total | 1,934 2,422 3,600 | 7,956

77

Table 8-11: Sample Distribution of Non-Agro-Households by Ecological Zone/Generation Cohorts

Group(zone birth) | 1991 1998 2005 | Total --------------------+---------------------------------+---------Coastal, <= 1931 | 59 42 19 | 120 Coastal, 1932-1938 | 28 41 10 | 79 Coastal, 1939-1945 | 35 41 21 | 97 Coastal, 1946-1951 | 43 71 21 | 135 Coastal, 1952-1957 | 49 50 39 | 138 Coastal, 1958-1963 | 57 84 45 | 186 Coastal, 1964-1969 | 29 71 52 | 152 Coastal, >=1970 | 10 57 129 | 196 Forest, <= 1931 | 54 36 12 | 102 Forest, 1932-1938 | 28 31 20 | 79 Forest, 1939-1945 | 50 50 44 | 144 Forest, 1946-1951 | 69 75 57 | 201 Forest, 1952-1957 | 79 106 68 | 253 Forest, 1958-1963 | 85 131 75 | 291 Forest, 1964-1969 | 75 114 80 | 269 Forest, >=1970 | 10 102 267 | 379 Savannah, <= 1931 | 28 8 12 | 48 Savannah, 1932-1938 | 19 16 21 | 56 Savannah, 1939-1945 | 24 20 33 | 77 Savannah, 1946-1951 | 30 21 32 | 83 Savannah, 1952-1957 | 36 30 49 | 115 Savannah, 1958-1963 | 27 45 46 | 118 Savannah, 1964-1969 | 17 26 66 | 109 Savannah, >=1970 | 3 17 116 | 136 --------------------+---------------------------------+---------Total | 944 1,285 1,334 | 3,563

78

Table 8-12: Model Estimates of Infrastructure Access & Household Welfare, Rural Ghana
+--------------------------------------------------------------+ | Variable | OLS | Fixed Effect | FD GMM | |--------------+---------------+---------------+---------------| | L.lnwelfare | .48190559* | .48190559*** | .48190565*** | | power | .51464981 | .51464981*** | .51464999* | | pubtrans | 2.1165793** | 2.1165793*** | 2.1165792*** | | water | -.39584698 | -.39584698** | -.39584689 | | assets | -4.2513212** | -4.2513212*** | -4.2513213*** | | poweriassets | .55739208** | .55739208*** | .55739207*** | | pubtransia~s | .78796658** | .78796658*** | .78796668*** | | wateriassets | 4.0523126** | 4.0523126*** | 4.0523127*** | | size | .45125989 | .45125989*** | .45125977** | | age | -.05015593 | -.05015593*** | -.05015592* | | agesq | .00094861** | .00094861*** | .00094861*** | | educ | -.2095756** | -.2095756*** | -.20957563*** | | educsq | .03773762** | .03773762*** | .03773763*** | | male | .38145388 | .38145388*** | .38145388* | | migrant | 1.3971635** | 1.3971635*** | 1.3971636*** | | public | 1.0824912* | 1.0824912*** | 1.0824912*** | | pfmlemploy | -9.9988386** | -9.9988386*** | -9.9988389*** | | pinfemploy | -7.1754338** | -7.1754338*** | -7.1754339*** | | agcashcrp | 5.3242926*** | 5.3242926*** | 5.3242926*** | | agfoodcrp | 1.5908477** | 1.5908477*** | 1.5908477*** | | period2 | .47928461* | .47928461*** | .47928474*** | | _Icohort* | YES | | | | _cons | 1.1735 | 2.5007032* | | |--------------+---------------+---------------+---------------| | N | 48 | 48 | 24 | | r2 | .99986167 | .99868076 | | | r2_a | .99783289 | .99761521 | | | rmse | .01567354 | .00532404 | | +--------------------------------------------------------------+ legend: * p<0.05; ** p<0.01; *** p<0.001 Note: FD GMM refers to First-Difference GMM estimates

79

Table 8-13: Model Estimates of Infrastructure Access & Household Welfare, Agro-Households
+--------------------------------------------------------------+ | Variable | OLS | Fixed Effect | FD GMM | |--------------+---------------+---------------+---------------| | L.lnwelfare | .38607572 | .38607572 | .77931137 | | power | -1.2890096 | -1.2890096 | -1.4950374 | | pubtrans | 2.1660881** | 2.1660881* | 2.0975897** | | water | -2.6028071* | -2.6028071 | -4.5627233* | | assets | 3.1695147** | 3.1695147* | 4.3667713* | | poweriassets | .12032624 | .12032624 | -.14102758 | | pubtransia~s | -.95994653* | -.95994653* | -1.0033016* | | wateriassets | -1.9677923* | -1.9677923 | -3.0531641* | | size | 2.2657302*** | 2.2657302** | 2.4709559*** | | age | .03111944 | .03111944 | .07383093 | | agesq | -.00015552 | -.00015552 | -.000268 | | educ | .18333543** | .18333543 | .13068363 | | educsq | -.0043373 | -.0043373 | .00070262 | | male | .68219749 | .68219749 | .18548455 | | migrant | 1.1937419** | 1.1937419* | 1.3812056* | | agcashcrp | .58042303 | .58042303 | -.5014834 | | _Icohort* | YES | | | | period2 | -.63011621 | -.63011621 | -.3972555 | | _cons | 6.6714735 | 6.5120878 | | |--------------+---------------+---------------+---------------| | N | 48 | 48 | 24 | | r2 | .99136902 | .94355735 | | | r2_a | .94204911 | .62102793 | | | rmse | .08174579 | .08174579 | | +--------------------------------------------------------------+ legend: * p<0.05; ** p<0.01; *** p<0.001 Note: FD GMM refers to First-Difference GMM estimates

80

Table 8-14: Model Estimates of Infrastructure Access & Household Welfare, Non-Agro-households
+--------------------------------------------------------------+ | Variable | OLS | Fixed Effect | FD GMM | |--------------+---------------+---------------+---------------| | L.lnwelfare | -.5505256 | -.5505256 | -.539235 | | power | -.75029366 | -.75029366 | -.7401504 | | pubtrans | .7510475 | .7510475 | .7394957 | | water | -.05542715 | -.05542715 | .0553982 | | assets | .24963744 | .24963744 | .2508854 | | poweriassets | -.07898575 | -.07898575 | -.0769178 | | pubtransia~s | -.03425416 | -.03425416 | -.038222 | | wateriassets | .1634618 | .1634618 | .162874 | | size | .84164289 | .84164289 | .8317842 | | age | -.01565597 | -.01565597 | -.0152643 | | agesq | .00105585 | .00105585 | .0010401 | | educ | -.47335586 | -.47335586 | -.4668492 | | educsq | .02794171 | .02794171 | .0275605 | | male | .3905294 | .3905294 | .3782502 | | migrant | .46487102 | .46487102 | .472309 | | public | -.02310199 | -.02310199 | -.027407 | | pfmlemploy | .97044523 | .97044523 | .9659123 | | pinfemploy | 3.1987463 | 3.1987463 | 3.163481 | | _Icohort* | YES | | | | period2 | 1.9739357 | 1.9739357 | 1.948178 | | _cons | 14.709965** | 17.556315* | | |--------------+---------------+---------------+---------------| | N | 48 | 48 | 24 | | r2 | .97614214 | .88205121 | | | r2_a | .77573607 | -.1087186 | | | rmse | .13320244 | .13320244 | | +--------------------------------------------------------------+ legend: * p<0.05; ** p<0.01; *** p<0.001 Note: FD GMM refers to First-Difference GMM estimates

81