Determinants of Public Support for Social Protection in Ghana. A Micro-level Analysis

No. 76 Callistus Akachabwon Agbaam λογος UA Ruhr Studies on Development and Global Governance Determinants of Public Support for Social Protection in Ghana: A Micro-level Analysis UA Ruhr Studies on Development and Global Governance Band 76 UA Ruhr Studies on Development and Global Governance vormals UAMR Studies on Development and Global Governance, Bochum Studies in International Development und Bochumer Schriften zur Entwicklungsforschung und Entwicklungspolitik Band 76 The UA Ruhr Graduate Centre for Development Studies is a collaboration project between the Institute of Development Research and Development Policy (IEE), Ruhr-University Bochum, the Institute of Political Science (IfP) and the Institute for Development and Peace (INEF), both located at the University Duisburg-Essen. The Centre is part of the University Alliance Ruhr (UA Ruhr) which aims at establishing the region as a cluster of excellence in research and training. Herausgegeben f¨ ur das UA Ruhr Graduate Centre for Development Studies von: Prof. Dr. Thushyanthan Baskaran, Prof. Dr. Matthias Busse, Prof. Dr. Tobias Debiel, Prof. Dr. Christof Hartmann, Prof. Dr. Markus Kaltenborn, Prof. Dr. Wilhelm L¨ owenstein Callistus Akachabwon Agbaam Determinants of Public Support for Social Protection in Ghana A Micro-level Analysis Logos Verlag Berlin λογος UA Ruhr Studies on Development and Global Governance Herausgegeben von: Prof. Dr. Thushyanthan Baskaran, Prof. Dr. Matthias Busse, Prof. Dr. Tobias Debiel, Prof. Dr. Christof Hartmann, Prof. Dr. Markus Kaltenborn, Prof. Dr. Wilhelm L¨ owenstein Institut f¨ ur Entwicklungsforschung und Entwicklungspolitik Ruhr-Universit¨ at Bochum Universit¨ atsstr. 150 D-44801 Bochum Telefon: +49(0)234/32-22418 Telefax: +49(0)234/32-14294 E-mail: IEEOf[email protected] http://www.uar-graduate-centre.org Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. Bochum, Univ., Dissertation 2022 This work is licensed under the Creative Commons license CC BY-NC-ND (https://creativecommons.org/licenses/by-nc-nd/4.0/). Logos Verlag Berlin GmbH 2025 ISBN 978-3-8325-5746-1 ISSN 2363-8869 Logos Verlag Berlin GmbH Georg-Knorr-Str. 4, Geb. 10 12681 Berlin Tel.: +49 (0)30 / 42 85 10 90 Fax: +49 (0)30 / 42 85 10 92 http://www.logos-verlag.com v Abstract In recent times, social protection reforms have gained significant momentum, particularly in lowand middle-income countries. However, a large chunk of the existing scholarship on these programmes tend to focus predominantly on their impact on various dimensions of poverty and human welfare in general. To date, not much has yet been done to understand the factors influencing or shaping citizens support or otherwise for these programmes especially in a development context. Focusing on Ghana, this study seeks to analyze the factors that determine public support or otherwise for different social protection mechanisms at the individual level. Specifically, using data from an attitudinal field survey, the study examines how factors such as economic self-interest, beliefs concerning the causes of poverty, institutional trust, and knowledge influence individual preferences or support for the Livelihood Empowerment Against Poverty (LEAP) social cash transfer programme and the National Health Insurance Scheme (NHIS) respectively. The results of the study show that with respect to the LEAP social cash transfer programme, economic self-interest, beliefs concerning the causes of poverty, institutional trust, and knowledge are relevant factors shaping public support. In contrast, for the NHIS, only beliefs concerning the causes of poverty, institutional trust, and knowledge tend to be relevant determinants. Therefore, taken together, the results of this study suggest that preferences for social protection are influenced by different factors, and that these factors tend to differ either based on the type of programme or the particular kind of risk being addressed. Based on these findings, the study concludes by recommending, among others, the institution of policy measures aimed at enhancing trust in public institutions especially those responsible Abstract vi for the implementation of social protection programmes, and the provision of adequate information on social protection programmes to citizens since these tend to be very strong correlates of support or otherwise for the social protection programmes in Ghana. vii Acknowledgement First and foremost, I wish to express my deepest gratitude to Prof. Dr. Katja Bender, my first supervisor, who has been a central figure in my academic journey for over a decade. This work would not have been possible without her unwavering guidance, thoughtful advice, and consistent support throughout every stage of my PhD. I am profoundly thankful for her mentorship. My sincere appreciation also goes to Prof. Dr. Wilhelm Löwenstein, my second supervisor, whose outstanding contributions to the field of development cooperation continue to inspire me. His insightful comments and feedback have significantly enriched this dissertation. I am also immensely grateful to all members of the Institute of Development Research and Development Policy (IEE), both academic and administrative staff, with special thanks to Dr. Gabriele Bäcker for her invaluable support throughout my studies. I also extend my heartfelt thanks to the DAAD for the financial support provided through the Graduate School Scholarship Programme, and to the RUB Research School Plus for funding several aspects of my PhD work. To my beloved family especially my parents and siblings, your enduring love, sacrifices, and unwavering belief in me have been the foundation upon which I’ve built this journey. I am forever indebted to you. To my family away from home, particularly the Alhassan’s, Markus, Arndt, and Sussane von Itter, thank you for embracing me as one of your own. Your kindness, support, and generosity have meant more than words can express. To all my friends and colleagues who stood by me, encouraged me, and supported me in one way or another over these past years, I say a very heartfelt thank you. May the good Lord richly bless you all. Contents xiv 2.3.4 Civil Society Organisations/NGOs � � � � � � � � � � � � � � � � � � � � � � �18 2.3.5 International Organisations � � � � � � � � � � � � � � � � � � � � � � � � � � � � �19 2.4 Major Global Policy/Legal Frameworks for the Development of Social Protection . . . . . . . . . . . . . . . . . . . . . . . .19 2.5 Chapter Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �25 3.0 Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25 3.1 Defining Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25 3.2 Factors Determining Preferences for Redistribution . . . . . . . . . .27 3.2.1 Self-interest Related Motives � � � � � � � � � � � � � � � � � � � � � � � � � � � �27 3.2.1.1 The Standard Median Voter Model: The Effect of Income Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 3.2.1.2 The Effect of Labour Market Characteristics . . . . . . . . . 31 3.2.2 Other-regarding Preferences: The Effect of Beliefs � � � � � � � � � � � � �37 3.2.3 The Impact of External Institutions on Preferences for Redistribution � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �39 3.2.3.1 The Nature of Institutions . . . . . . . . . . . . . . . . . . . . . . . . 39 3.2.3.2 Institutional Quality and Support for Redistribution . . . 40 3.2.4 Knowledge of Policies and Support for Redistribution � � � � � � � � � � �42 3.3 Chapter Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .44 4 Social Protection in Ghana: A Comprehensive Overview � � � � �45 4.0 Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 4.1 The Social Protection Landscape in Ghana: A Brief Historical Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 4.1.1 Phase I (1957 to 1979) � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �46 4.1.2 Phase II (1980 to 1999) � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �47 4.1.3 Phase III (2000 to 2022) � � � � � � � � � � � � � � � � � � � � � � � � � � � � �48 4.2 Overview of Some Flagship Social Protection Programmes in Ghana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50 Contents xv 4.2.1 Livelihood Empowerment Against Poverty (LEAP) Cash Transfer � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �50 4.2.2 National Health Insurance Scheme (NHIS) � � � � � � � � � � � � � � � � �51 4.2.3 Education Capitation Grant � � � � � � � � � � � � � � � � � � � � � � � � � � � �53 4.2.4 Ghana School Feeding Programme � � � � � � � � � � � � � � � � � � � � � � � �54 4.2.5 Free Senior High School Programme � � � � � � � � � � � � � � � � � � � � � � �56 4.3 Is Social Protection in Ghana Rights-based? . . . . . . . . . . . . . . . .57 4.4 Chapter Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .58 5 Research Methodology � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �59 5.0 Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59 5.1 Description of the Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . .59 5.2 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61 5.3 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .62 5.3.1 Target Population and Sample Size Determination � � � � � � � � � � � �62 5.3.2 Sampling Procedure � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �62 5.4 Data Type and Data Collection Methods . . . . . . . . . . . . . . . . . . .65 5.5 Data Entry and Data Cleaning. . . . . . . . . . . . . . . . . . . . . . . . . . . .66 5.6 Data Analysis Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .67 5.6.1 The Logistic Regression Model: Form and Assumptions � � � � � � � � �67 5.6.2 Interpretation of the Logistic Regression Model � � � � � � � � � � � � � � �73 5.6.3 Assessing the Quality (Goodness-of-fit) of the Logistic Regression Model � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �75 5.7 Empirical Model, Variable Description and Estimation Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .77 5.7.1 The Empirical Model � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �77 5.7.2 Variables Description � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �78 5.7.2.1 The Dependent Variables . . . . . . . . . . . . . . . . . . . . . . . . . 78 5.7.2.2 Main Explanatory Variables . . . . . . . . . . . . . . . . . . . . . . . 80 5.7.2.3 Control Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5.7.3 Estimation Approach � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �86 Contents xvi 5.8 Measuring Preferences: Dealing with the Challenges and Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .86 5.9 Chapter Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .88 6 Presentation and Discussion of Empirical Findings � � � � � � � � �89 6.0 Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .89 6.1 Descriptive Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .89 6.1.1 Socio-demographic Characteristics of the Sample � � � � � � � � � � � � � �89 6.1.2 Economic and Labour Market Characteristics of the Sample � � � � �92 6.1.3 Distribution of Dependent and Independent Variables � � � � � � � � � �96 6.2 Inferential Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .99 6.2.1 Preferences for Cash Transfers (LEAP) � � � � � � � � � � � � � � � � � � � �99 6.2.1.1 Empirical Model and Results . . . . . . . . . . . . . . . . . . . . . . . 99 6.2.1.2 Post Estimations Checks: Goodness-of-fit and Model Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 6.2.1.3 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 6.2.2 Preferences for Social Health Insurance (NHIS) � � � � � � � � � � � � �117 6.2.2.1 Empirical Model and Results . . . . . . . . . . . . . . . . . . . . . . 117 6.2.2.2 Post Estimations Checks: Goodness-of-fit and Model Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 6.2.2.3 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 6.3 Discussion of the Empirical Findings . . . . . . . . . . . . . . . . . . . .132 6.4 Chapter Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .137 7 General Conclusion and Policy Recommendations � � � � � � � � � 139 7.0 Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .139 7.1 Summary of Findings and Conclusion . . . . . . . . . . . . . . . . . . . .139 7.2 Policy Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .141 7.3 Limitations of the Study and Suggestions for Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143 References � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 145 Appendices � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 163 Contents xvii Questionnaire. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173 Introduction (Consent): � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �173 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173 Curriculum Vitae � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 195 Agbaam, Callistus Akachabwon . . . . . . . . . . . . . . . . . . . . . . . . . . . . .195 Contents xix List of Tables Table 5.1. Sampled Localities in the Accra Metropolitan Area. . . . . . . . .64 Table 5.2. Summary/Blocks of Explanatory Variables. . . . . . . . . . . . . . .85 Table 6.1. Logistic Regressions Results: Support for Cash Transfers (LEAP)—(I). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .101 Table 6.2. Logistic Regressions Results: Support for Cash Transfers (LEAP)—(II). . . . . . . . . . . . . . . . . . . . . . . . . . . . . .104 Table 6.3. Robustness Checks: Support for Social Cash Transfers (LEAP). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116 Table 6.4. Logistic Regressions Results: Support for Social Health Insurance (NHIS)—(I). . . . . . . . . . . . . . . . . . . . . . . .118 Table 6.5. Logistic Regressions Results: Support for Social Health Insurance (NHIS)—(II). . . . . . . . . . . . . . . . . . . . . . . .121 Table 6.6. Robustness Checks: Support for Social Health Insurance (NHIS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131 xxi List of Figures Figure 1.1. Number of Policy Areas covered in Social Protection Programmes anchored in National Legislation, 1960–2015. . . 3 Figure 5.1. Map of the Accra Metropolitan Area. . . . . . . . . . . . . . . . . . .60 Figure 6.1. Ethnic Distribution of Respondents. . . . . . . . . . . . . . . . . . . .90 Figure 6.2. Educational Attainment by Gender. . . . . . . . . . . . . . . . . . . . .92 Figure 6.3. Employment Sector by Gender. . . . . . . . . . . . . . . . . . . . . . . .93 Figure 6.4. Income Category by Employment Sector. . . . . . . . . . . . . . . .94 Figure 6.5. Distribution of Respondents by Employment Industry. . . . . . 95 Figure 6.6. Support for Cash Transfers (LEAP) by Employment Sector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .96 Figure 6.7. Support for Social Health Insurance (NHIS) by Employment Sector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .97 Figure 6.8. Mean Level of Knowledge by Employment Sector. . . . . . . .99 Figure 6.9. The Effect of Beliefs by Employment Sector. . . . . . . . . . . .107 Figure 6.10. The Effect of Institutional Trust by Employment Sector. . . . 108 Figure 6.11. The Effect of Knowledge by Employment Sector. . . . . . .109 Figure 6.12. The Effect of Knowledge by Level of Education. . . . . . .110 Figure 6.13. Index Plot for Standardized Pearson Residuals. . . . . . . . . .113 Figure 6.14. Index Plot of Pregibon’s Delta Beta Statistic. . . . . . . . . . .114 Figure 6.15. The Effect of Beliefs by Employment Sector. . . . . . . . . . .123 List of Figures xxii Figure 6.16. The Effect of Institutional Trust by Employment Sector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .124 Figure 6.17. The Effect of Knowledge by Employment Sector. . . . . . .125 Figure 6.18. The Effect of Knowledge by Level of Education. . . . . . .126 Figure 6.19. Index Plot of Standardized Pearson Residuals. . . . . . . . . .129 Figure 6.20. Index Plot of Pregibon’s Delta Beta Statistic. . . . . . . . . . .129 xxiii List of Abbreviations AMA Accra Metropolitan Area AME Average Marginal Effects AU African Union CAADP Comprehensive Africa Agricultural Development Programme CDD Center for Democratic Development CSO Civil Society Organization DMHIS District Mutual Insurance Schemes EA Enumeration Area ECG Education Capitation Grant ESS European Social Survey ERP Economic Recovery Programme FCUBE Free Compulsory Universal Basic Education FSHS Free Senior High School G-DRGs Ghana Diagnostic Related Groupings GHIPSS Ghana Interbank Payment and Settlement System GNECC Ghana National Education Campaign Coalition GPRS I Ghana Poverty Reduction Strategy I GPRS II Growth and Poverty Reduction Strategy II GSS General Social Survey ICESCR International Covenant on Economic, Social and Cultural Rights ICT Information and Communication Technology ILO International Labour Organization IMF International Monetary Fund LEAP Livelihood Empowerment Against Poverty LIPW Labour intensive Public Works 1 Introduction to the Study 6 preferences for redistribution in the European welfare states (e.g., Alesina et al., 2021; Andreoli and Olivera, 2020; Kulin and Meuleman, 2015; Olivera, 2015; Hausermann et al., 2014; Blekesaune, 2013; Jaime-Castillo, 2013; Hochtl et al., 2012; Rehm, 2009), or (ii) redistributive preferences in North American countries such as the US and Canada (e.g., Owens and Pedulla, 2014; Zilinksy, 2014; Franko et al.,2013; Alesina and Giuliano, 2009; Chong et al., 2001; Appelbaum, 2001; Fong, 2001), or (iii) a comparative analysis of both (e.g., Alesina et al., 2018; Barnes, 2015; Jaime-Castillo and Saez-Lozano, 2014; Dallinger, 2010; Kenworthy and McCall, 2008; Kenworthy and Pontusson, 2005; Alesina and Angeletos, 2005; Blekesaune and Quadagno, 2003). More so, among these contributions3, only a very few studies distinguish between preferences for different social protection policy areas (e.g., Alesina et al., 2018; Hausermann et al., 2014; Pontusson and Rueda, 2010; Kenworthy and McCall, 2007; Blinder and Krueger, 2004; Blekesaune and Quadagno, 2003). A majority of these studies generally focus on a single social protection policy area or examine public support for redistribution as a whole4. However, given that various social protection pillars entail different degrees of redistribution and risk sharing, it is possible that preferences for these policies may differ based on programme characteristics, or the heterogeneity of risk being addressed (Jordan, 2013). In addition, due to data limitations, relatively fewer contributions explicitly measure individual preferences for social protection using real or actual policy scenarios. Notable exceptions include for example, Hirvonen and Hoddinott (2021), Schuring (2014), and Boeri and Tabellini (2012). Generally, a majority of empirical studies measure individual preferences based on inequality, culture and national identity, prevailing social norms, etc. See for example Moene and Wallerstein, 2003; Kulin and Meuleman, 2015. 3 Specific reference to quantitative studies that analyse individual preferences for social protection or redistribution in both high income countries, and lowand middle-income countries. 4 A large number of studies rely on the overall measure “support for Redistribution” which according to Jason (2013) fails to capture the diversity and complexity in various programme designs. 1.2 Research Problem 7 hypothetical questions5 as contained in attitudinal surveys such as the World Values Survey (WVS), the General Social Survey (GSS) and the European Social Survey (ESS) amongst others. This approach to approximating and measuring individual preferences (i.e., based on hypothetical questions) could be very limiting since such questions tend to be general, too broad and may not inherently capture an individual’s preference for social protection as often suggested. Furthermore, studies that include factors very specific to developing countries, for example, the potential conflict of interest that arises between different groups of individuals in the formal and informal sectors, and the impact of governance challenges in formal institutions (e.g., corruption, weak accountability structures and poor enforcement mechanisms) are still extremely limited in number. Particularly, with regards to the former, so far and to the best of the researcher’s knowledge, only Berens (2015a, 2015b), and Carnes and Mares (2014) explicitly address issues related to the conflict of interest between formal and informal sector workers with respect to redistributive policy preferences in a development context. However, as highlighted in the earlier paragraphs, all three studies focus solely on countries in Latin America and the Caribbean. Nonetheless, aside these, virtually no other studies exist in this regard. With respect to the latter, with the exception of Hauk et al., 2017 and Gassmann et al., (2016) virtually no other quantitative study examines the impact of institutional quality or governance challenges on redistributive policy preferences in a development context. Studies that incorporate such factors although still few in number (e.g., Peyton, 2020; McDonald, 2020; Rothstein et al., 2012; Hetherington, 2005) are predominantly focused on analysing policy preferences in high-income countries. This study therefore attempts to address the research gaps outlined above and, in doing so, contribute to an enhanced understanding of the 5 For example in the WVS, the following question is often used to approximate preference for social protection or redistribution; “Government should take more responsibility to ensure that that everyone is provided” versus “People should take more take more responsibility to provide for themselves”. 1 Introduction to the Study 8 factors that drive public support or otherwise for social protection policies in a development context. 1.3 Overall Objective and Research Questions In view of the above, the overall objective of this study is to analyze the main factors that influence or determine public support for social protection in a lowand middle-income country context. Specifically the study seeks to answer the following research questions: • What factors determine public support for social protection? • How and to what extend do these factors differ based on the type of social protection mechanism being considered (i.e., Cash transfers versus Social health Insurance? 1.4 Organization of Chapters This main chapters of this thesis are structured as follows; Chapter one generally introduces the study. It presents the background and context within which the study is being conducted. It also presents the research problem, the study’s overall objectives, as well as the research questions guiding the study. Chapter two presents an exposition into the concept of social protection by highlighting its definition, rationale, type of mechanisms used, the actors involved in implementation, and ends with a discussion of the major global policy and legal frameworks for the development of social protection. Chapter three presents the theoretical framework for this study. Drawing on diverse strands of theoretical literature, the researcher attempts to develop a set of theoretical explanations on why individuals may support or oppose various social protection mechanisms. Specifically, the chapter provides theoretical arguments on the impact of self-interest, beliefs, institutional quality and knowledge on public support for social protection. It then ends with a set of hypothesis to be tested in the empirical part of the study. 1.4 Organization of Chapters 9 Chapter four focuses exclusively on social protection in Ghana. It highlights the social protection landscape in Ghana from a historical perspective, and as well presents a comprehensive overview of some existing social protection programmes in Ghana. The chapter ends with a short discussion on the question as to whether or not the current social protection architecture in Ghana is rights-based. Chapter five elaborates on the methodological framework for the study. It describes the study area, research design, sampling approach, data collection methods and the framework for data analysis. The chapter then concludes with a description of the empirical model and an operationalization of all the variables used in the study. Chapter six which is the empirical chapter, presents and discusses the empirical results of the study. It begins with a brief description of the socio-demographic and labour market characteristics of respondents in the sample. Thereafter, the results of the empirical analysis regarding individual preferences for the two main social protection programmes under consideration namely, the Livelihood Empowerment Against Poverty (LEAP) cash transfer programme and the National Health Insurance Scheme (NHIS) are presented. The chapter ends with an overall discussion of the study’s empirical findings with the aim of answering the research questions and hypothesis set forth earlier in the study. Chapter seven provides a general conclusion to the study. It presents a summary of the study’s main empirical findings and the overall conclusions. In addition, it reflects on the policy implications of these findings, in relation to the design and implementation of social protection programmes. The chapter ends by outlining the study’s main limitations and some suggestions for future research. 11 2 Understanding Social Protection 2.0 Chapter Overview This chapter presents a general discussion on the mechanics of social protection. It begins with a comprehensive definition of the scope and rationale for social protection. Thereafter, it discusses the different types of social protection instruments or mechanisms commonly used, as well as the various actors involved in implementing social protection globally. The chapter concludes by highlighting the major global policy and legal frameworks guiding the development of social protection. 2.1 Social Protection: Key Definitions and Rationale The concept of social protection has been widely defined by various scholars in different ways. For example, Norton et al. (2001, p. 7) define social protection as “public actions taken in response to levels of vulnerability, risk and deprivation which are deemed socially unacceptable within a given polity or society.” Barrientos et al. (2005, p. 9) posit that such measures provide an opportunity for “short-term assistance to individuals and households to cope with shocks while they are temporarily finding new economic opportunities that will rapidly allow them to improve their situation.” For Ellis et al. (2009), social protection basically entails measures that are aimed at addressing the numerous causes of poverty and vulnerability in society. Also, Deveraux and Sabates-Wheeler (2004, p. 9) provide a more elaborate and comprehensive definition of the concept of social protection as consisting of “all public and private initiatives that provide income or consumption transfers to the poor, protect the vulnerable against livelihood risks, and enhance the social status and rights of the marginalised; with the overall 2 Understanding Social Protection 12 objective of reducing the economic and social vulnerability of poor, vulnerable and marginalised groups”. Furthermore, aside the various scholarly definitions presented above, the concept of social protection has also been defined by different international institutions in ways that reflect their policy and ideological standpoints. For instance, the ILO conceptualizes social protection as a basic human right and therefore defines it as “the set of policies and programmes designed to reduce and prevent poverty and vulnerability throughout the life cycle” (ILO, 2017:1). Similarly, the World bank from a social risk management perspective also delineates the contours of social protection as entailing measures that “help individuals and societies manage risk and volatility and protect them from poverty and destitution—through instruments that improve resilience, equity, and opportunity” (World Bank, 2012:1). Likewise, for the Organization for Economic Co-operation and Development (OECD), the concept essentially denotes “policies and actions which enhance the capacity of poor and vulnerable people to escape from poverty and enable them to better manage risks and shocks” (OECD, 2009:12). According to United Nations Research Institute for Social Development (UNRISD), social protection is a core element of social policy and therefore includes measures that enable individuals to avert, cope and overcome situations that negatively affect their wellbeing (UNRISD 2018, p. 135). Lastly, the African Union (AU) also conceptualizes social protection as encompassing “a “package” of policies and programmes with the aim of reducing poverty and vulnerability of large segments of the population”6 (African Union, 2008:30). Evidently, a key feature in almost all the definitions outlined above is the emphasis on social protection as a means to addressing poverty and vulnerability among different groups of individuals within society. Hence, for those already poor, social protection is aimed at enabling them break 6 According to the African Union, this is achieved through a combination of “policies/programmes that promote efficient labor markets, reduce people’s exposure to risk, and contribute to enhancing their capacity to protect and cover themselves against lack of or loss of adequate income, and basic social services” (Africa Union, 2008:30). 2.1 Social Protection: Key Definitions and Rationale 13 out of poverty, whilst for those who are near-poor and non-poor, it essentially serves as a safety net by preventing them from falling into poverty. Furthermore, it is imperative to emphasize that social protection may not entail only “public actions” as opined by Norton et al. (2001). Rather, such measures may also encompass actions by non-state actors as highlighted by Deveraux and Sabates-Wheeler (2004). Also, contrary to the definition by Barrientos et al. (2005), social protection may as well transcend “short term assistance” to include other measures with both intermediate and long term goals. In the context of this thesis, given that the overall objective is to analyze support for government-led programmes, the researcher narrowly defines social protection to include public measures which are aimed at addressing the risk of poverty and vulnerability among individuals in society. Despite the fact that different definitions of the concept of social protection highlight different functions, some consensus seem to exist with respect to the fact that social protection functions for example as; (i) a protective mechanism, in the sense that it provides support to cushion individuals from deprivation, (ii) a preventive mechanism, given that it averts or obviates deprivation, (iii) a promotive mechanism, considering that it augments individual incomes and capabilities, and (iv) a transformative mechanism, given that it focuses on addressing structural issues pertaining to social equity such as human rights and empowerment (Deveraux and Sabates-Wheeler 2004, p. 10)7. 7 The functions listed may not be fully exhaustive given that social protection measures cover other multiple functions not explicitly captured by the conceptualization provided above (for example see Norton et al., 2001 and United Nations, 2000). Nonetheless, this conceptualization by Deveraux and Sabates-Wheeler (2004) provides a very good conceptual basis for understanding the role or functions of social protection in both developed and developing countries. 2 Understanding Social Protection 14 2.2 Types of Social Protection Instruments/Mechanisms 2.2.1 Social Assistance Social assistance is commonly defined as consisting of interventions that provide both cash and in-kind support to the extremely poor and vulnerable in society to enable them maintain a minimum living standard (Barrientos 2013, p. 25). Given their overall focus and target, social assistance programmes are usually non-contributory in nature and are mostly financed by the state through taxation (Ibid). Examples of such programmes include both conditional and non-conditional cash transfers, cash plus programmes, social pensions, food and in-kind transfers, public work programmes, graduation programmes, fee waivers and subsidies (World Bank 2018, p. 5; Carter et al. 2019, p. 13). Barrientos (2013, p. 25) argues that whereas in most developed countries social assistance programmes focus on income maintenance by providing means-tested transfers to individuals in amounts that enable them cover the poverty gap, in developing countries such transfers are usually in fixed amounts and focused on households rather than individuals. 2.2.2 Social Insurance Unlike social assistance, social insurance consist of “contributory schemes providing protection against a range of life-course and work-related contingencies” (Barrientos, 2013: 25). Such schemes are usually based on a risk pooling mechanism and typically require that individuals make regular contributions into a fund in other to be eligible to enjoy the benefits (Ibid). Social insurance programmes are usually based on mandatory or voluntary contributions by individuals. Whereas the mandatory pillar is mostly based on employment and payroll deductions, thereby targeting formal sectors workers, the voluntary contributions pillar which is mostly based on flat rate payments is often targeted at the relatively non-poor in the informal sector8 (Wagstaff 2009, p. 504). Examples of social insurance programmes include 8 Some social insurance programmes are also designed in such a manner that they exempt the extremely poor in society as well as indigent groups from paying in- 2.3 Implementing Social Protection: Actors and Roles 15 health insurance schemes, contributory old age, survivor and disability pensions, unemployment benefits, maternity or paternity benefits, sickness or injury benefits amongst others (Oxford Policy Management 2017, p. 7). Social insurance schemes are largely considered as preventive social protection mechanisms given that they primarily focus on minimizing the risk of falling into difficulty or financial hardship (Ibid). 2.2.3 Labour Market Regulations Labour market regulations refer to interventions (contributory and non-contributory) which are designed to assist individuals secure employment or protecting their existing jobs (World Bank 2018, pp. 5–6). Generally, these interventions may be categorized into two, namely, ‘passive’ labour market interventions and ‘active’ labour market interventions (Barrientos 2013, p. 26). Whereas passive labour market interventions typically focus on ensuring that the rights of workers are protected and the minimum standards of employment are guaranteed (e.g., through legislations that underpin initiatives such as minimum wages, retirement and unemployment payments), active labour market interventions tend to focus on stimulating employment via “training and skills transfer, job search and intermediation, remedial education and employment subsidies” (Ibid). More so, in view of the fact that passive labour market interventions largely focus on individuals working in the formal sector, they often tend to overlap with social insurance measures such as unemployment insurance, or even maternity and sickness benefits (Carter et al. 2019, p. 16). 2.3 Implementing Social Protection: Actors and Roles 2.3.1 The Family/Kinship and Community Networks The traditional family and kinship network as well as local community networks remain very important actors in the provision of informal social surance premiums. A typical example is the National Health Insurance Scheme in Ghana. 2 Understanding Social Protection 22 “provides a unique set of internationally-accepted standards that serve as a reference for national social security systems”. The recommendation which seeks to complement existing ILO treaties, is also deemed as completing the organization’s social security strategies (Ibid). Oritiz et al. (2016, pp. 1–2) posit that the Social Protection Floors Recommendation (No. 202) is by far the only international accord that provides a true reflection of the global consensus on the need for universal social protection systems. The Recommendation whilst clearly affirming social security as a basic human right provides guidance to national governments in setting up national social protection floors14 and implementing same using strategies that progressively ensure that as many as possible individuals are able to access and enjoy higher levels of social security in line with the existing standards of the ILO (ILO, 2012). It also enjoins countries to monitor progress on these national floors, and ensure that the design and implementation of these floors are grounded on social dialogue and the participation of all relevant stakeholders in the country (Ibid). Additionally, the World Bank Social Protection and Labour (SPL) Strategy 2012–2022 also constitutes a key policy framework for the global social protection agenda. The document which is an outcome of wide consultations with relevant social protection stakeholders (including government representatives, CSOs, private sector actors and other international agencies) basically builds on the banks previous strategy15, achievements, and practical lessons from the field. It provides a coherent framework for the extension of technical and financial support to many lowand middle-income 14 These are defined as “as set of social security guarantees that ensure, at a minimum, that all people have access to social protection at adequate benefit levels – or income security” and include amongst others, “cash transfers for children, maternity benefits, disability pensions, support for those without jobs, old age pensions as well as access to essential health care (Oritiz, et al., 2016:4). For full details on the Recommendation No. 202 please see: https://www.ilo.org/dyn/normlex/ en/f?p = NORMLEXPUB:12100:0::NO:12100:P12100_INSTRUMENT_ ID:3065524:NO 15 The World Bank’s first Social Protection and Labour Strategy 2000–2008, proposed a conceptual framework for social protection grounded on social risk management. For a review of the said strategy please see Holzmann (2003). 2.5 Chapter Conclusion 23 countries with respect to the implementation of social protection. Thematically centered on resilience, opportunity and equity, the strategy “lays out an agenda to help lowand middle-income countries build, improve and harmonize their SPL programs, to increase their capacity to respond to crises and shocks, support poverty reduction and inclusive growth, and build on the best global knowledge of what works” (World Bank 2012, p. 2). It also calls for greater collaborations and coordination among different stakeholders in the implementation of social protection given that the absence of these commonly result in programme fragmentation and consequently the exclusion of poor and vulnerable groups (Ibid). Also, the strategy underlies the World Bank’s continuous commitment to many initiatives such as the Global Partnership for Universal Social Protection, a joint collaboration between the World Bank, the ILO and other bilateral and multilateral agencies seeking to monitor progress and contribute towards the attainment of the sustainable development goal’s target on social protection. Last but not the least, the SDGs framework, a successor to the Millennium Development Goals (MDGs), probably presents a more encompassing and globally recognized policy framework for social protection. As earlier stated, SDG 1.3 specifically requires that countries implement appropriate social protection mechanisms and strive to extend coverage to previously excluded populations as well as improve benefits levels. Moreover, the framework also supports the implementation of various policy measures aimed at tackling other priority areas related to social protection such as enhancing food security (SDG2), quality healthcare (SDG3), education (SDG4), gender equality (SDG5), decent employment (SDG8), and reduced inequalities (SDG10) amongst others. Undeniably, the SDGs framework lends global credence to various efforts aimed at promoting social protection especially in developing countries. 2.5 Chapter Conclusion To sum up, given that social protection remains a widely discussed concept, this chapter through a review of the literature has attempted to provide 2 Understanding Social Protection 24 a comprehensive understanding of its underlying mechanisms. It has highlighted the main actors involved in the design and implementation of social protection, and discussed major global policy and legal frameworks that support the development and institutionalization of social protection in both developed and developing countries. In the next chapter, the researcher presents the theoretical framework for the study. 25 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 3.0 Chapter Overview This chapter provides the theoretical and conceptual underpinnings for the entire study. Based on various strands of theoretical and conceptual literature, it presents a unified and an encompassing framework for explaining why individuals may choose to either support or oppose redistributive policies, and by extension social protection. It begins by highlighting the standard self-interest model of Meltzer and Richard (1981) and proceeds to show the circumstances under which a rational actor may deviate from the predictions of the model to embracing other-regarding behavior. Next, the scope of theoretical discussion is expanded to include how other relevant factors such as the perceived level of institutional quality and knowledge of policies can also impact public support for social protection. The chapter then concludes with a set of derived hypothesis to be tested in the empirical part of the study. 3.1 Defining Preferences Strictly speaking, there is no single definition for the term “preferences” (Druckman and Lupis 2000, pp. 1–2). Rather, various scholars have defined the term differently. For example, Scherer (2005, p. 703) defines preferences more broadly as consisting of “relatively stable evaluative judgements in the sense of liking or disliking a stimulus, or preferring it or not over other objects or stimuli.” For Hausman (2012, p. 1), preferences are essentially 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 26 comparative evaluations. These comparative evaluations can either be partial, total or overall (Ibid, pp. 1–3). Preferences are said to be (i) partial when an individual considers some specific criterion in ranking between alternatives, (ii) total when an individual takes into account every relevant consideration in ranking and (iii) overall when an individual take into account most of what they consider relevant rather than a specific criterion or everything that matters to them in evaluating alternatives (Ibid). According to Hausman, although in a broad sense preferences typically refer to overall comparative evaluations, in the realm of economics, they tend to be viewed as total comparative evaluations (Hausman 2012, p. 4). Furthermore, similar to the definition by Hausman (2012), Hansson and Grüne-Yanoff (2021) also define preferences as “subjective comparative evaluations, in the form of “Agent A prefers X to Y”.” By this definition, the authors posit that preferences concern matters of value rather than objective facts given that they are attributable to individuals or a collective and entail subjective judgements on liking one object over and above another (Ibid). Altogether, they argue that a characterization of preferences as “subjective comparative evaluations” differentiates the term from other evaluative concepts which tend to focus only on evaluating a single object and excludes a subjective element (Ibid). In the context of this study, given that the researcher seeks to analyze individual preferences in the form of support or opposition for different social protection mechanisms, the study adopts the broader definition of preferences by Scherer (2005) since it inherently captures both the monadic and comparative aspects discussed above. As it will be shown, there are several factors that shape or influence individual preferences for redistributive policy measures such as social protection. A number of these factors are discussed below. 3.2 Factors Determining Preferences for Redistribution 27 3.2 Factors Determining Preferences for Redistribution 3.2.1 Self-interest Related Motives 3.2.1.1 The Standard Median Voter Model: The Effect of Income Inequality A dominant strand of economic literature on redistributive preferences proceed from the rational choice argumentation that individuals are homines economici: rational individuals act or behave intrinsically in a utility-maximizing manner driven purely by self-interest. A key and influential model in the ensuing literature on redistribution has been the standard median voter model of income redistribution proposed by Meltzer and Richard (1981). Drawing on the earlier works of Romer (1995) and Richards (1977), Meltzer and Richard (1981) set forth a simple parsimonious uni-dimensional model of income redistribution that captures the essential link between income inequality and the demand for redistribution. In this model, Meltzer and Richard argue that in a standard economic environment — where (i) the only activities of government are redistribution via lump-sum transfers and linear taxation, (ii) the real budget is balanced, and (iii) voters are assumed to be fully informed about the state of the economy,- the level of redistribution demanded is inherently determined by the utility-maximizing preferences of a decisive voter, whom under a democratic framework (majority voting rule) is said to be the voter with the median income (Meltzer and Richard, 1981). Furthermore, they contend that given higher levels of income inequality (evidenced by a right-skewed income distribution), the income of the median voter is lower than the mean income. Thus, the median voter would expect to gain from income redistribution since through progressive taxation, the burden imposed by government redistribution is relatively less than the transfers received. Consequently, the model predicts that the median voter and all other individuals with incomes below that of the median voter will support government redistribution, since they tend to become net beneficiaries of the process whilst those with incomes above the median voter 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 28 will oppose redistribution because they become net contributors and hence carry a disproportionate share of the tax burden associated with government redistribution (Ibid). Furthermore, given that the median voter has preferences which are inversely ordered by income, the larger the gap between mean and median income, the higher the demand for redistribution (Ibid). In short, the median voter model of Meltzer and Richard (1981) supports the conclusion that a positive relationship exist between income inequality and demand for redistribution. Furthermore, the model also posits that in a democratic polity, this positive demand for redistribution is expected to translate into some real generous welfare package for the poor. Since welfare policies generate some redistributive effects, net beneficiaries of such policies will more likely support them whereas net contributors are more likely to oppose. Despite the saliency of the median voter model, several scholars have questioned its inherent assumptions16. Also, the vast literature explaining redistributive preferences provides inconclusive evidence on the empirical utility of the standard median voter model,17 which clearly implies that from a self-interest perspective there may be other factors driving preferences outside income. Notable factors here include the role of risk aversion and the prospects of social mobility. According to Moene and Wallerstein (2001, 2003), uncertainty about future incomes may cause individuals to rather view redistributive programmes as insurance against future risk which might be difficult to insure privately (risk pooling measures). Thus, given the fact that demand for insurance is ordinarily assumed to rise with income, the authors predict a negative relationship between income inequality and support 16 For example Burstein (1998), governments maybe less responsive to the demands of voters contrary to the expectations of the model. Moreover, Pontusson and Rueda (2010) and Larcinese (2007) also posit that electoral turnout may as well change the position of the median voter and as a result affect the level of redistribution demanded. Other limitations of the model are discussed in the extant literature (e.g see Iversen and Goplerud, 2018 and Padovano, 2012 amongst others). 17 For example Lierse 2019; Olivera 2015; Kerr, 2014; Kenworthy and Pontusson, 2005. 3.2 Factors Determining Preferences for Redistribution 29 for redistribution, contrary to the predictions of the standard median voter model (Ibid). Also, Bénabou and Ok (2001) argue that when low income individuals perceive possibility or chances of upward mobility to be high, they may be less likely to support redistribution contrary to the theoretical postulations of the standard median voter model18. Nevertheless, albeit its shortcomings, the median voter model remains a workhorse model in contemporary political economy literature since that its hypotheses commonly provides a base for explaining redistributive policy preferences (Portmann and Stadelmann 2013, p. 1). Thus, in the ensuing, the logic of the standard median voter model19 is extended to explain individual preferences for both non-contributory or tax-financed social protection (social assistance) and contributory social protection (social insurance) by analyzing the conflict of interest that arises between individuals of different income levels (i.e., low-income versus high-income groups) in a development context. Generally, in many developing countries, the pattern of income distribution tends to be highly unequal (Simpson 2018, p. 10). As a result, two extreme income groups emerge both of whom are central to this analysis. At the head of the distribution is the majority of individuals usually with incomes below the mean (low-income group) whereas at the smaller tail end of the distribution we find the minority group of individuals with incomes usually above the mean (middle to high-income group). Since the structure of income distribution gives rise to a higher mean relative to the median, it can be expected that the individual with the median income level (median voter) will be located relatively closer to or amongst individuals in the low-income group. Thus, with regards to non-contributory or tax-financed social protection (i.e., social assistance programmes), following the logic of self-interest it can be expected that low-income individuals especially the poor will more 18 Despite the relevance of these alternative explanations, due to data limitations the researcher does not directly test them in this thesis. 19 Throughout this chapter, the standard median voter model and the Meltzer and Richard model (1981) are used interchangeably. 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 30 likely to support such programmes since they stand a better chance of benefitting from any extensions in coverage of social protection systems. This support is further reinforced by the fact that the net benefits to be derived far outweighs the net cost to be incurred via a progressive tax regime. Support is expected to be much greater, the lower an individual’s level of income is relative to the mean, since the gains from redistribution are inversely ordered with income. On the contrary, acting based on self-interest, high income individuals are expected to more likely oppose non-contributory or tax-financed social assistance programmes, as they are less likely to benefit directly from such programmes, while bearing a disproportionate share of the cost of these programmes, particularly under a progressive tax regime. In sum, it can therefore be hypothesized that: Hypothesis 1a: compared to low-income groups, high-income groups are more likely to oppose taxed-financed social assistance programmes. However, with regards to contributory programmes (social insurance), the dynamics of support may differ. Given that these programmes typically require that individuals make contribution before receiving benefits, it is reasonable to expected that low-income individuals especially the poor (who are unable to afford such contributions) will generally be indifferent, since they do not incur any direct financial cost arising from these programmes or enjoy any benefits due to exclusion. On the other hand, high-income earners will relatively be more likely to support contributory programmes since they are able to contribute and enjoy the benefits of such programmes, and also, given that contributory programmes are more likely to exclude free riders.20 Thus, it is hypothesized that: 20 Strictly speaking, this line of argument is based on the firm assumption that contributory programmes exclude the poor since they may not be able to afford the cost of premiums/contributions. However, in situations where these programme by design tend to include low income individuals and contributions into the same pool is progressively based on an individual’s income rather than for example individual risk, it is possible for the preferences of high income individuals to change towards less support since under such circumstances, high income individuals although also benefitting from the programme tend to heavily cross subsidize the poor. 3.2 Factors Determining Preferences for Redistribution 31 Hypothesis 1b: high-income individuals are more likely to support contributory social protection programmes (social insurance) since they exclude free riders. 3.2.1.2 The Effect of Labour Market Characteristics Moving further, very related to income, the structure of the labour market may also present another huge source of distributional conflict with regards to the extension of social protection. Generally, labour markets in many developing countries are often stratified into formal and informal sectors: definitions of which remain highly disputed in the literature (Berens, 2015a; Maloney, 2004). Nonetheless, in presenting the conceptual arguments, the researcher first and foremost attempts to present various definitions of the formal and informal sector, and thereafter discusses how the structure of the labour market specifically the distribution of individual income (poor and non-poor) in both formal and informal sector affects redistributive preferences from a self-interest perspective. Definitions of Formal and Informal Sectors Unlike the formal sector which is commonly defined to cover all economic activities that are officially registered, regulated and recognized by the state (Vij et al. 2017, p. 3; Weeks 1975 cited in Pratap and Quintin 2006, p. 3), definitions of what exactly constitutes the informal sector21 remains highly contested in the development literature (Turner 2020, p. 41; Gasparini and Tornarolli 2009, p. 15). Often used interchangeably or synonymously with terms such as ‘informal economy’ ‘shadow economy’ or ‘informality’ in general, characterization of the informal sector has been a subject of extensive scholarly debates for many decades now (Aguilar and Guerrero 2020, p. 280; Ulyssea 2020, 21 According to Chen (2012, p. 2) the term “informal sector” was coined by anthropologist Keith Hart during his study of low income economic activities among unskilled migrants in Accra, the capital city of Ghana in 1971. However, the term gained widespread acceptance following its usage in an ILO employment mission report on Kenya in 1972. 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 38 By beliefs, the researcher specifically refers to certain ideas, thoughts or convictions that an individual deems or considers to be true. Values on the other hand, refer to some “internalized rules that guide or motivate behavior” (Bender 2021, p. 510). Values enable individuals to make decisions regarding “what is good or bad, justified or illegitimate, worth doing or avoiding” based on the standards and principles that one is committed to (Schwartz 2012, p. 4). Therefore, the attitudes that an individual expresses towards others tend to be reflective of the kind of beliefs and values that he or she upholds or endorses32 (Arikan and Ben-Nun Bloom 2012, p. 213; Schwartz 2012, p. 16; Davidov et al. 2008, p. 241). Although, individuals may have different kinds of beliefs, in this study, emphasis is placed on beliefs that are relevant for redistribution. These beliefs which are more generally reflective of an individual attitudes towards the poor can be classified based on responsibility for the determination of outcomes: in casu, beliefs about self-responsibility versus societal responsibility for the determination of wealth and poverty (Alesina and Angeletos 2005, p. 962; Fong 2001, p. 227). Whereas beliefs about self-responsibility attribute the causes of poverty and wealth to factors that are within individual control, beliefs about societal responsibility attribute the causes of poverty and wealth to factors that are external or exogenous to the individual and for which an individual has no personal control over (Fong, 2001, p. 227). In line with the above, the researcher argues that depending on an individual’s belief regarding the causes of poverty and wealth, they may either support or oppose social protection programmes. Specifically, individuals who believe in self-responsibility and therefore attribute poverty to lack of personal effort are less likely to act in solidarity with the poor by supporting social protection programmes, as they perceive the poor to be responsible purposes of this of this study, the researcher adopts the latter position and therefore considers individual beliefs and values. systems as part of other-regarding preferences. Thus, they are discussed separately from institutions. 32 Also, individual attitudes may also be influenced by prevailing social norms. Social norms are defined as ‘ the informal rules that govern behavior in groups and societies” (Bicchieri et al., 2018). 3.2 Factors Determining Preferences for Redistribution 39 for their own predicament. To the contrary, individuals who believe in societal responsibility for the determination of outcomes may likely assume collective responsibility for the plight of the poor. As a result, they are more likely to act in solidarity with the poor by supporting social protection programmes since they perceived poverty to be caused by factors outside of personal control. Although, the present conceptual framework distinguishes between preferences for different social protection mechanisms namely, tax-financed social assistance programmes and contributory or social insurance programmes respectively, it is generally expected that the effect of beliefs will remain the same across both policy areas. Thus, the researcher hypothesizes that; Hypothesis 3a: individuals who believe that poverty is caused by external factors are more likely to support tax-financed social assistance programmes than those who attribute the causes of poverty to selfresponsibility. Similarly, Hypothesis 3b: individuals who attribute the causes of poverty to external factors are more likely to support contributory or social insurance programmes than those who attribute poverty to self-responsibility. 3.2.3 The Impact of External Institutions on Preferences for Redistribution 3.2.3.1 The Nature of Institutions Institutions have been defined in many different ways. For example, Aoki (2007, p. 7) defines institutions as ‘‘...self-sustaining, salient patterns of social interactions, as represented by meaningful rules that every agent knows and incorporated as agents’ shared beliefs about the ways how the game is to be played.” Hodgson (2006, p. 2) refers to institutions as systems of dominant social rules that guide social interactions. For North (1990, p. 3) institutions 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 40 are “the rules of the game in a society, or more formally, are the humanly devised constraints that shape human interaction.” As such, they tend to provide a structure for everyday life by reducing uncertainty and minimizing the risk of collective action problems which are inherent in human interactions (North, 1990, p. 3). Aside the specification of acceptable or prohibited conduct, institutions also entail various enforcement mechanisms which ensure that breaches or non-adherence to specific rule components are adequately sanctioned (Voigt and Engerer, 2001, p. 132). Based on their origin and type of enforcement, Kasper and Streit (1999) distinguish between internal and external institutions.33 According to the authors, whereas internal institutions refer to rules that evolve within a group in the light of experience and are privately enforced (e.g., conventions, internalized rules, customs and good manners, etc.), external institutions are rules imposed on society from above by virtue of political action, and consequently enforced by public authority through the use of force. Some examples of external institutions include constitutions, statues, by-laws, government decrees and administrative regulations (Ibid, pp. 100–110). Given the focus of this sub chapter, the researcher presents conceptual arguments on how the quality of external institutions in particular may shape individual preferences for social protection. 3.2.3.2 Institutional Quality and Support for Redistribution As previously highlighted, individuals are endowed with preferences which are either based on self-interest or other-regarding concerns. However, suffice it to say that, these preferences do not exist in a vacuum. They are embedded within an institutional context part of which are external institutions. Generally, external institutions, in casu, constitutions, decrees, statutory laws and regulations as highlighted in the preceding section generally provide the ‘rules of the game’ and as such define the constraints and opportunities for individual conduct. In this study, the researcher argues that aside their existence, the quality of external institutions matter for individual 33 North (1990) also distinguishes between formal and informal institutions. 3.2 Factors Determining Preferences for Redistribution 41 preferences. By institutional quality, the researcher simply refers to the level of effectiveness of external institutions or in other words, the extent to which the state as ‘originator’ is able to ensure compliance to formal rules and in the event of breaches enforce the stipulated sanctions. Hence, external institutions are said to be effective when they are non-discriminatory and apply to all individuals in equal measure, are transparent, reliable, and ably complimented by enforceable sanctions to deter non-adherence to rule components. Conversely, they are perceived as weak and less effective, when they tend to be discriminatory, less transparent, inconsistent and entail weak mechanisms of enforcing sanctions. In the current theoretical framework, it is argued that whereas high levels of institutional quality may tend to engender high levels of trust and confidence in the state and its inherent bureaucracy, particularly in terms of its ability to foster the rule of law and impartially sanction negative behaviors such as corruption, low levels of institutional quality tend to reduce trust and confidence in the state, especially as negative behaviors such as corruption, abuse of power and the misuse of government resources are more likely to become common under such circumstances. As a result, the study proceeds on the premise that institutional quality as reflected in the prevailing levels of trust in government, should predispose individuals to either show more or less support for social protection.34 In line with the above, this study argues that when individuals perceive the quality of external institutions to be high, they become more inclined to supporting social protection. This is because such individuals tend to trust in the ability of the state and its inherent bureaucracy to effectively implement social protection programmes and impartially sanction individuals who engage in negative behavior. However, when citizens perceive the quality of external institutions to be weak, the level of trust in the public authorities diminishes, and consequently they become less inclined to supporting social protection, since they are unable to trust in the ability of 34 Similar lines of argument concerning the effect of trust in government on welfare policies and taxation have been provided in Peyton (2020); McDonald (2020); Rothstein et al., (2012); Hetherington, (2005). 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 42 public authorities to effectively implement social protection programmes in a just and transparent manner. Moreover, regardless of type of social protection programme being considered, it can be expected that the effect of institutional quality on individual preferences would remain the same. Therefore, this study hypothesizes that: Hypothesis 4a: individuals who generally perceive the quality of external institutions to be high are more likely to support tax-financed social assistance programmes than those who perceive the quality of external institutions to be low. Equally, Hypothesis 4b: individuals who perceive the quality of external institutions to be high are more likely to support contributory social insurance programmes than those who perceive the quality of external institutions to be low. 3.2.4 Knowledge of Policies and Support for Redistribution In addition to the factors highlighted in the preceding sections, knowledge of relevant issues pertaining to specific redistributive policies may also prove important for individual decision making regarding support for social protection. By knowledge, the researcher basically refers to the level of information or awareness that an individual possess on a particular issue, and in this context on the relevant variables regarding social protection in general.35 As evident in the burgeoning literature on redistribution, the impact of knowledge or information on redistribution can be analyzed from different perspectives. Whereas some studies focus explicitly on individual’s state of knowledge or awareness concerning the specifics of a policy or programme,36 others center on how knowledge about broader societal issues such as true levels of poverty and inequality 35 For example the cost and size of the redistribution, target groups, true levels of poverty and inequality etc. 36 For example Boeri and Tabellini (2012). 3.2 Factors Determining Preferences for Redistribution 43 among others influence preferences.37 In the context of this study, the researcher adopts the former approach, by emphasizing how more or less knowledge of a specific policy area can lead to varying levels of support for social protection. The study proceeds with the assumption that individuals may have imperfect information about relevant issues pertaining to redistribution and this can in turn affect decision making and preferences (see for example, Karadja et al., 2017; Kuziemko et al., 2015; Zilensky, 2014; Boeri and Tabellini, 2012; Blinder and Krueger, 2004). Interestingly, the effect of knowledge (either more knowledge or less knowledge) on redistributive preferences tends to be bi-directional. For instance, on the one hand, whereas more knowledge concerning a particular redistributive policy may enable individuals to better understand the need or essence for the said intervention, thereby leading to support, on the other hand, it may also make more salient issues of conflict of interest and consequently serve as a basis for opposing redistribution. A similar mechanism may also hold true for individuals with less knowledge. Thus, whereas less knowledge on a particular policy or programme may result in individuals not being able to understand or appreciate the importance of the said programme, or even underestimate or misjudge the social good and the welfare effects associated with it, leading to less support, it is also possible that less knowledge of a particular redistributive policy may as well make conflicting issues less salient and therefore lead to support from individuals who would otherwise not support such policies. In line with the previous section, it is expected that the bi-directional effect of knowledge will remain the same across the different policy areas of social protection being examined. Therefore, this study hypothesizes that: Hypothesis 5a: knowledge of policies is a significant determinant of individual support for support tax-financed social assistance programmes. Hypothesis 5b: knowledge of policies is a significant determinant of individual support contributory or social insurance programmes. 37 For example, Pellicer et al. (2019), Kuziemko et al. (2015), Zilensky, (2014), etc. 3 Explaining Redistributive Policy Preferences: Towards a Theoretical Framewok 44 3.3 Chapter Conclusion In conclusion, this chapter sort to present a unified conceptual framework for explaining individual support or otherwise for redistribution, specifically for both contributory (social insurance) and non-contributory tax-financed (social assistance) social protection programmes. The various postulations derived have also been formalized into tentative hypotheses to be tested in the empirical part of this study. In the next chapter, the researcher provides a comprehensive overview of social protection in Ghana. 45 4 Social Protection in Ghana: A Comprehensive Overview 4.0 Chapter Overview This chapter provides a comprehensive discussion of the state of social protection in Ghana. It begins with a historical highlight of the social protection landscape in Ghana. Thereafter, a detailed discussion of some existing social protection programmes as well as their impact on various dimensions of human welfare is presented. In closing the chapter, the author briefly discusses whether or not social protection in Ghana can be described as rights-based. 4.1 The Social Protection Landscape in Ghana: A Brief Historical Perspective Ghana is one of the few countries in sub-Sahara Africa that has been widely recognized for its efforts to expand access to social protection especially in the areas of cash transfers, social health protection and formal sector contributory pensions. However, the progress achieved cannot be attributed to any one-off event. Rather, a sequence of policy events since independence perhaps explains the status quo. As a matter of fact, Ghana has had a long history of experimenting with various social protection or social security programmes in its post-independence era, the experiences from which may have shaped the present state of affairs with respect to social protection (i.e., path dependency). Thus, this section attempts to provide a comprehensive insight into the social protection landscape in Ghana from a historical perspective. In doing so, the researcher categorizes Ghana’s post-independence social protection efforts into three distinct phases, 4 Social Protection in Ghana: A Comprehensive Overview 46 namely, phase one (1957 to 1979), phase two (1980–1999), and Phase three (the year 2000 to 2022)38. 4.1.1 Phase I (1957 to 1979) In the period immediately after independence, the then socialist government under Ghana’s first president Dr. Kwame Nkrumah clearly pursued a set of very ambitious social welfare programmes. For example, it provided universal free universal healthcare services for all citizens under its ‘free healthcare for all’ policy financed from general taxes (Government of Ghana 2004, p. 4).39 The government also provided free universal primary education for all children of school going age and cost free education for those at the tertiary level (Abukari et al. 2015, p. 4). At the pre-tertiary education level, students from the Northern part of Ghana were absolved from paying any fees whilst those from the southern sector paid fees although very marginal (Ibid).40 Furthermore, during the 1960s, the government also designed and implemented its first contributory social security programme to provide old age income for individuals mainly in the formal sector (Asamoah and Nortey, 1987). However, as impressive as these programmes were, it was difficult to fully sustain them as they took a huge toll on the nation’s economic resources, especially following the sequence of political and economic turmoil that unfolded from the mid 1960s onwards. Consequently, this led to the gradual introduction of nominal user fees particularly in health sector in 1969 under the Hospital Fees Decree, which was subsequently amended into the Hospital Fees Act of 1971 (Nyonator and Kutzin 1999, p. 330). Badasu (2004, p. 290), believes that the introduction of user fees in the early 1970s was mainly intended to deter citizens from the unnecessary use of 38 The various phases indicated are the researchers own subjective categorizations which have been provided mainly for ease of explanation. 39 This free health care for all policy was implemented in only government or public health facilities. 40 The Northern Scholarship scheme for pre-tertiary education was established in 1961 under the Education Act, 1961. 4.1 The Social Protection Landscape in Ghana: A Brief Historical Perspective 47 healthcare services (moral hazard) rather than for revenue generation purposes considering that the amount charged as user fees at the various public health facilities were still very minimal. 4.1.2 Phase II (1980 to 1999) During the early years of phase two (1980s), the socio economic turmoil in Ghana worsened leading to the adoption of an Economic Recovery Programme (ERP) under the Structural Adjustment Programmes (SAPs) developed by the twin Bretton Wood Institutions (i.e., International Monetary Fund and the World Bank). The SAPs generally emphasized the need for countries to pursue economic liberalization polices driven largely by market forces (Kraus 1991, p. 19). As a result, the role of the state in providing social services was diminished. Under the SAPs, the government of Ghana faced with the requirement of cutting down and rationalizing public sector expenditure, resorted to the introduction of cost-sharing policies and in some cases pursued full cost recovery strategies, which resulted in citizens having to pay for essential social services such as health and education (Botchwey 1993, p. 4). For example, the “cash and carry” system was introduced through the health sector reforms in 1983 and 1985, thus requiring individuals to pay out-of-pocket for healthcare services at the point of delivery41 (Kodua et al. 2015, pp. 13–16). Similarly, cost-sharing arrangements were also introduced in the education sector necessitating copayments by parents to cater for the cost of books, boarding and other essential educational supplies (Sowa 1993 cited in Abukari et al. 2015, p. 2). Clearly, the introduction of these measures increased the financial burden associated with accessing basic social services for all Ghanaians including the poor. However, in the latter part of phase two, specifically towards the end of the 1980s into the early 1990s, given the negative consequences of the 41 Alongside the “cash and carry” system, Government also provided both full and partial exemptions for different categories of individuals such as indigents, children, healthcare workers, tuberculosis, leprosy and psychiatric patients. Also Antenatal, postnatal and immunization services were also exempted from charges (Nyonator and Kutzin, 1999). 4 Social Protection in Ghana: A Comprehensive Overview 54 p. 9). Thus, the programme assumed the form of a fee waiver by providing grants to public schools to enable them cater for tuition and other ancillary cost hitherto paid by parents, in anticipation that it will promote interest for basic education in the country (Jones et al. 2009, p. 44). After an initial pilot in about 40 deprived districts, the coverage of the capitation grant was expanded to include all public basic schools in Ghana at the commencement of the academic year in September 2006. In 2009, the government also increased the value of the grant and concurrently announced a new initiative to provide educational supplies such as textbooks and school uniforms free of charge to all school children in the poorest and most deprived districts across the country (Ibid). To date, some empirical studies have examined the impact of the education capitation grant on educational outcomes in Ghana. The evidence emerging largely points to mixed results. For instance, Akyeampong (2011) reports that although the positive impact of the grant was evident in the first two years of its implementation, such gains were gradually eroded due to other systemic challenges in the basic education system such as the lack of infrastructure. Also, whereas studies such as Osei-Fosu (2011) and Osei et al. (2009), found no significant effect of the capitation grant particularly on school enrollment and retention rates, other studies such as Padibo and Tamanja (2017) and Maikish and Gershberg (2008) provide some evidence of the positive impact of the grant with respect to increasing basic school enrollment across selected districts in Ghana. Nonetheless, almost all the studies cited above acknowledge the inherent challenges facing the ECG, especially the inadequacy of the capitation grant and the undue delays in releasing capitation payments from central government to the local districts. 4.2.4 Ghana School Feeding Programme The Ghana School Feeding Programme (GSFP) was launched in 2005 in line with the then Government of Ghana’s policy framework (i.e., GPRS II) and within the framework of the Comprehensive Africa Agricultural Development Programme (CAADP) Pillar 3, as a response to calls to step 4.2 Overview of Some Flagship Social Protection Programmes in Ghana 55 up efforts towards achieving the Millennium Development Goals (Government of Ghana, 2015b, p. 9). Under the umbrella of the New Partnership for Africa Development (NEPAD), the programme was piloted in one school drawn from each of the then ten administrative regions in Ghana (10 schools in total as at 2005). However, by August of the following year, the number of schools were rapidly scaled up to 200 in 138 districts, with plans for further expansion by the close of 2006 (Government of Ghana 2006, p. 1). According to Dunaev and Corona (2019, p. 14), as of 2017 the Ghana School Feeding Programme reached approximately 1.7 million students in 5,582 public kindergartens and basic schools across the country. The fundamental concept behind the GSFP is “to provide children in public primary schools and kindergartens with one hot nutritious meal, prepared from locally grown foodstuffs, on every school-going day” (Government of Ghana 2015b, p. 11). In the short term, the programme was expected to increase the enrollment of students, boost school attendance and retention (especially for girls), reduce hunger and malnutrition among students in deprived communities and also enhance local food production (Government of Ghana 2006, p. 21). However, in the long term, the GSFP was expected to contribute to overall poverty reduction, promote food security and generate income for the local farmers (Ibid). It is imperative to mention that the GSFP was not the first of its kind to be implemented in Ghana. Prior to its launch in 2005, various development agencies (e.g., Catholic Relief Services, Adventist Development and Relief Agency, World Food Programme, World Vision among others) had already been engaged in implementing supplementary feeding programmes albeit on a small scale in very deprived schools and communities across Ghana. Nonetheless, the GSFP launched in 2005 represented a more comprehensive and far-reaching effort by the government of Ghana to progressively provide free school meals for basic school students nationwide (Government of Ghana, 2006). Generally, some available studies on the effectiveness of the GSFP suggest that the programme is contributing immensely to increasing school enrollment and attendance especially in deprived and rural communities in 4 Social Protection in Ghana: A Comprehensive Overview 56 Ghana (e.g., Goldsmith et al., 2019; Bukari and Hajara, 2015; Oduro-Ofori and Gyapong, 2014). However, the inadequacy of the grant amount allocated per child, low quality of meals provided by caterers, frequent delays in the release of funds to caterers, and overall funding constraints, constitute some key challenges confronting the programme (Akuamoah-Boateng and Sam-Tagoe, 2018; Sulemana, et al., 2013; Essuman and Bosumtwi-Sam, 2013). 4.2.5 Free Senior High School Programme The Free Senior High School (FSHS) Programme was launched in 2017 in fulfilment of a major political campaign promise by the NPP government. The programme which is in line with the dictates of the 1992 constitution of Ghana (Article 25 1b) and the global sustainable development agenda (SDG 4) aims at providing universal access to free quality secondary education for all public senior high school students in Ghana (Government of Ghana, 2017). Through the FSHS programme the government seeks to eliminate all cost barriers to secondary education in Ghana by absorbing all approved fees including admission, tuition, uniforms, library, science development, sports, culture, entertainment, examination, utilities, Information and Communication Technology (ICT), and feeding fees among others (Government of Ghana, 2017). In addition, the programme also seeks to improve the quality of secondary education through the provision of teaching and learning materials as well as adequate staffing, and enhance equity by ensuring that 30% of placements in elite senior high schools are reserved for students from public junior high schools especially those from deprived backgrounds. In addition, the FSHS programme also aims at supporting reforms to institutionalize Technical and Vocational Education and Training (TVET) at the senior high school level in Ghana (Ministry of Education, 2021). As a matter of fact, although various forms of targeted subsidies or secondary school scholarships46 have over the years been implemented in Ghana either by government, NGOs or private organizations, the current 46 For example, the Northern Scholarship Scheme which was introduced by 1961. 4.3 Is Social Protection in Ghana Rights-based? 57 FSHS programme is the most comprehensive given that it automatically covers all students placed into public senior high schools across the country (Ministry of Education, 2021). Thus, the universal character of the current FSHS programme differentiates it from all past initiatives. Available data from the Ministry of Education suggest that the FSHS programme has since its inception impacted positively on enrolment rates in public senior high schools. Specifically, the programme is said to have increased net enrolment from 308,799 in 2016 to approximately 404,856 as of March, 2020 (Government of Ghana, 2021). However, the increase in enrolment rates has resulted in the implementation of a double track school system in many public senior high schools as a measure of containment (Ibid). Notwithstanding its successes, the FSHS programme is as well confronted with a number of challenges. According to the Public Interest and Accountability Committee (2020), most senior high schools under the programme are confronted with infrastructure challenges and inadequate educational supplies given the high enrolment rates. More so, considering the fact that the FSHS programme has no clear or dedicated funding source, financial sustainability of the programme remains a key issue of concern (Asare, 2021; GNECC, 2019). 4.3 Is Social Protection in Ghana Rights-based? The 1992 Constitution of Ghana guarantees a host of basic individual rights for all Ghanaians. Under the direct principles of state policy, the constitution mandates the state to cater for the welfare of all citizens albeit in a progressive manner. Furthermore, Ghana also has signed and ratified almost all the relevant global conventions, treaties and protocols required for social protection47, and has taken steps to operationalize these locally (Government of Ghana, 2015a). Thus, in principle, it has been argued that 47 For example, the Universal Declaration of Human Rights, the United Nations Conventions on the Rights of the Child and Persons with Disabilities, the African Union Social Policy Framework (2003), the Livingstone Declaration (2006), the Ouagadogou Declaration and Plan, The ILO Convention 202, etc. 4 Social Protection in Ghana: A Comprehensive Overview 58 “Ghana’s social protection policies and interventions are moving toward a rights-based approach to development, with social protection increasingly being viewed within official government circles as a right of every citizen” (Kaltenborn et al., 2017, p. 25). Nevertheless, some other scholars contend that, in practice there are still a number of challenges that impede against the full realization of social protection as a basic human right in Ghana. For example, Abdulai et al. (2019, p. 9–10) contend that the limited value of the LEAP grant, perceptions that the grant is charity rather than an entitlement and the weak grievance and complaints settlement mechanisms altogether negatively affect the notion that the programme is rights-based. More so, according to the ILO (2014, p. 165), a number of social protection programmes in Ghana are not directly backed by law.48 Therefore, the absence of clear legal frameworks for these programmes may tend to constrain beneficiaries from enforcing their rights or entitlements to various social protection benefits. 4.4 Chapter Conclusion This chapter provided a comprehensive overview of social protection in Ghana. It offered insights into the historical landscape of social protection development in Ghana and highlighted some major flagship programmes currently being implemented. In the next chapter, the researcher presents the methodology for the study. 48 Some exceptions include the National Health Insurance Act, 2012 (Act 852), the Persons with Disability Act, 2006 (Act, 715), National Pensions Act, 2008 (Act 766), etc. It is also important to mention that several other programmes although not directly backed by any law usually tend to be based either on obligations from international instruments ratified, the constitution of Ghana and various policy/strategy documents. For a detailed discussion on these, see for example ILO (2014) and Kaltenborn et al. (2017). 59 5 Research Methodology 5.0 Chapter Overview This chapter presents the methodological framework for the study. It begins with a brief description of the study area. Thereafter, it highlights the study’s research design, sampling approach, data collection methods and the framework for data analysis. The chapter then concludes with a description of the empirical model and the operationalization of all variables used in the study. 5.1 Description of the Study Area This study was conducted in the Accra Metropolitan Area (AMA). Located along Ghana’s southern coastal line and bordering the Gulf of Guinea, the Accra metropolis serves both as the administrative capital of the Greater Accra region and Ghana as a whole. The AMA was first established in 1898 by the erstwhile colonial administration. Since then, it has undergone a number of changes particularly in terms of its geographic size, with a number of other administrative districts constantly being carved out of it. Figure 5.1 below shows the geographic map of Accra as at the time of the study. According to the Ghana Statistical Service (2014, p. 15), the population of Accra based on the 2010 Population and Housing Census (PHC) was estimated at 1,665,086 inhabitants. Out of this figure, 51.9 percent were female with the remaining 48.1 percent being male (Ibid). Given an estimated annual population growth rate of 3.1 percent, the Accra Metropolitan Assembly estimates that the population of the metropolis as at 2018 stood at 2,036,899 inhabitants with an additional daily influx of over 2 million 5 Research Methodology 60 individuals who visit the city to undertake varied socio-economic activities.49 The population of the city continues to grow rapidly due to the prevalence of rural urban migration (Ibid, p. 20). The indigenous or traditional inhabitants of Accra are the Ga Mashie, however individuals from various ethnic groups in the country such as the Dangme, Akan, Ewe, Guan and Mole-Dagbani among others can all be found in the city. The dominant religious group in the metropolis are Christians, whereas Muslims, Traditionalist and individuals from other faith constitute religious minorities (Ghana Statistical Service, 2014, p. 20). The city has been well acknowledged for its cosmopolitan nature, considering that it houses many individuals from diverse social, economic, religious and ethnic backgrounds (Arthur 2018, p. 21). 49 Official estimates from the Accra Metropolitan Assembly. See https://ama.gov. gh/theassembly.php Figure 5.1. Map of the Accra Metropolitan Area. Source: Author’ own drawing based on data from :https://data.humdata.org/dataset/ ghana-administrativeboundaries; OpenStreet Map; Google Terrain. 5.2 Research Design 61 Given its status as the capital of Ghana, physical infrastructure in the city is relatively well developed compared to other cities in the country. Key state institutions such as the executive seat of government, parliament, government ministries and other state agencies can all be located within the metropolitan area. Generally, social infrastructure such as hospitals, clinics, schools and roads are present in the city although access to these amenities maybe limited for certain groups (for instance those living in the highly dense settlements and low class neighborhoods) due to economic and spatial inequalities (Weeks et al, 2012). With regards to education and literacy, approximately 52 per cent of the population aged 11 years and above are able to read and write in English and Ghanaian language (Ghana Statistical Service 2014, p. 32). Predominantly an urban area, the economy of the Accra Metropolis is dominated by activities in sectors such as manufacturing, wholesale and retail of goods, transportation and storage, construction, accommodation and food services amongst others (Ibid, p. 37). About 70 percent of the population aged 15 years and above are economically active with a little over 90 per cent of such persons being employed in some form of economic activity, mainly in the private sector (Ibid, p. 34). Moreover, similar to most other African cities, informality is a key feature of the economy of the Accra metropolis with an estimated 74 percent of individuals employed in the private informal sector (Ibid, p. 39). Urban poverty is also prevalent in the metropolis, with a majority of individuals particularly migrants living in slums, informal settlements and low class neighborhoods (Verutes et al 2012, p. 4). It is imperative to mention that, the AMA was purposively selected for this study primarily due to its cosmopolitan nature. More so, its centrality and relative ease of access also made it very ideal for this study given the researchers’ limited time and resources. 5.2 Research Design For this study, a cross sectional research design was adopted. Bryman and Bell (2007, p. 55) define a cross sectional research design as that which 5 Research Methodology 62 “entails the collection of data on more than one case and at a single point in time in order to collect a body of quantifiable or quantitative data in connection with two or more variables, which are then examined to detect patterns of association”. They elaborate that such a design is appropriate when a researcher’s main aim is to understand a particular phenomenon at a specific point in time rather than at multiple time periods. Thus, given the fact that the overall objective of this study is to understand individual preferences for social protection at a singular point in time, a cross sectional research design was well-suited. 5.3 Sampling 5.3.1 Target Population and Sample Size Determination The target population for this study basically included all individuals aged 18 years and above in the Accra Metropolitan Area. However, due to time constraints and the very limited budget available for study the researcher selected a sample size of 600 individuals. As highlighted by Duflo et al. (2007:3928) “sample size is often determined in large part by budget or implementation constraints.” Thus, in the context of this study, the researcher focused on balancing the constraints of time and budget against the need for precision. 5.3.2 Sampling Procedure To select the required sample, a multi-stage probability sampling procedure was utilized. The procedure adopted largely followed the sampling strategy used in the Afrobarometer surveys. The Afrobarometer survey is a well-established and widely cited public opinion survey in Africa that collects and publishes high quality data on issues related to democracy, governance, economic growth and the society. The sampling procedure in Afrobarometer emphasizes randomness at every stage of sampling and utilizes the probability proportionate to population size (PPPS) procedure to ensure that various geographic units have a proportional chance of being included in the sample (Afrobarometer Network, 2014, p. 26). 5.3 Sampling 63 To aid in the sampling process the researcher obtained a summary of the census data from the Ghana Statistical Service detailing all localities, enumeration areas and the population aged 18 years and above in the Accra Metropolitan Area. The sampling process then proceeded as follows; In Stage 1, the researcher randomly selected 10 localities from the list of all localities in the AMA50 as provided by the Ghana Statistical Service. These 10 localities then served as secondary sampling units from which primary sampling units were later drawn. In Stage 2, the primary sampling units (PSUs)51 defined as “the smallest, well-defined geographic units for which reliable population data are available” (Afrobarometer 2014, p. 28) were selected. Given that the total Sample size was 600, the plan was to collect data from 60 PSUs with 10 respondents in each. To do this, a list of all enumeration areas in each of the selected localities was obtained from the Ghana Statistical Service. Then based on a probability proportionate to population size (PPPS) procedure, the required number of PSUs for each locality was calculated and corresponding number of enumeration areas duly selected using simple random sampling. The researcher then obtained the enumeration area maps to aid in the identification of the selected PSUs. In Stage 3, sampling starting points within each enumeration area were then randomly selected. These were predominantly marked and notable physical structures in the enumeration area such churches, mosques and schools. In Stage 4, using the marked physical structures as starts points, dwelling units were identified and households randomly selected using a walk pattern with an interval of 5 or 10 households depending on the number of dwelling units or households on the walking stretch. Finally in Stage 5, which constituted the last sampling stage, an individual respondent was chosen from each selected household for the 50 As of July, 2017. 51 In Ghana, census enumeration areas (EAs) represent the smallest and well defined geographic units for which population data is available (see https://catalog.ihsn.org/index.php/catalog/3780). Thus, the EAs serve as primary sampling units (PSUs). 5 Research Methodology 70 technique is commonly used in estimating the parameters of the logistic regression model rather than other alternative techniques such as Ordinary Least Squares (OLS), because the ML procedure best maximizes the log-likelihood function of the logit model. As such, when the logit model is well specified, the ML estimator yields parameters that can be considered as “asymptotically optimal” (Kleinbaum et al 2014, p. 668). Evidently, there are a number of assumptions or requirements that need to be fulfilled for the logistic regression model to be considered unbiased, efficient and valid. Just like other types of regression analysis, logistic regression assumptions have been well discussed in the econometrics literature. The researcher provides a summary of these assumptions as discussed by Harrell, 2015; Osborne, 2015; Tabachnick and Fidell 2014; Kleinbaum et al., 2014; Dougherty, 2011; Gujarati and Porter, 2009; Hosmer and Lemeshow, 2000; Menard, 1995 amongst others. To begin with, a key assumption of logistic regression is that the independent variables should not be highly or perfected correlated with each other (Osborne 2015, p. 86; Tabachnick and Fidell 2014, p. 489). When independent variables are highly correlated with each other, the problem of multicollinearity arises (Field 2005, p. 174). According to Menard (1995, p. 65p) the presence of high and perfect multicollinearity tends to be very problematic given that it leads to an undue inflation of the standard errors and consequently results in a difficulty to attain unique parameter estimates of the logistic regression model. Given this potential negative effect, it is therefore prudent to check for multicollinearity after logistic regression. The presence of multicollinearity among independent variables can be detected using various collinearity diagnostics such as the tolerance statistic, Variance Inflation Factor (VIF), Eigen values, Condition indexes and variance proportions (Field 2005, p. 175). For the purposes of this study, the researcher employs the VIF measure to check for the presence or otherwise of multicollinearity. According to Gujurati and Porter (2009, p. 328), the VIF basically shows the extent to which the variance of a coefficient estimate is inflated due to presence of multicollinearity. As a rule of the thumb, a VIF value greater than 10 points to the problem of high multicollinearity in the 5.6 Data Analysis Framework 71 dataset, whilst values less than 10 do not call for any serious concern (Myers, 1990 cited in Field 2005, p. 260). In addition, the logistic regression model also assumes the independence of observations. This means that the individual observations in the data should not emanate from repeated measures or data that is matched (Osborne 2015, p. 86). When the assumption of independence of observations is violated, the logistic regression model may experience overdispersion or under dispersion depending on the structure of the data, and consequently increase the likelihood of Type 1 errors when testing for the effect of the independent variables (Ibid:87). Furthermore, logistic regression models also require that data should be free from outliers and highly influential observations as these can exert undue and disproportionate influence on the parameter estimates of the regression (Osborne 2015, p. 104; Tabachnick and Fidell 2014, p. 489; Menard 1995, pp. 73–79). Field (2005, p. 245) defines an outlier as a case “for which the model fits poorly” and influential observations as cases “that exert undue influence on the model”. According to Long and Freese (2001, p. 113) “not all outliers are influential”. Thus, such cases should not merely be dropped or deleted because they have very high residuals. Rather they should be examined in some more detail before any concrete action is taken (ibid). To detect the presence of outliers and highly influential cases, an analysis of residuals is usually conducted. In this study, to detect possible outliers, the standardized pearson residuals55 are calculated and examined for all cases. According to Field (2005, p. 245), residual values that are greater than +3 or –3 are potentially a cause for concern while those that are above +2.5 or –2.5 warrant some closer inspection. In addition, to identify cases that exert large influences on the regression model a statistical measure that provides summary information on the influence of each individual case on the parameter estimates in the model (known as Pregibon’s dbeta56) is also 55 Menard (1995: 72) defines the standardized pearson residuals as “the difference between the estimated probabilities divided by the binomial standard deviation of the estimated probability”. 56 Field (2009:246) posits that Pregibon’ dbeta is similar to Cook’s distance in linear regression which measures the extent to which regression coefficients 5 Research Methodology 72 calculated for all observations in the sample. According to Menard (1995, p. 79), dbeta values that are greater than 1 do possibly signal high influence. Thus, such cases warrant a closer inspection given that they tend to be problematic (Ibid). Also, in logistic regression analysis, it is commonly assumed that the model is correctly specified: meaning that the logit link function is appropriate, the model includes all relevant predictors, and contains no irrelevant variables (Osborne 2015, p. 91). As highlighted by Menard (1995, p. 58) misspecification of the logistic regression model may result in coefficients that are biased due to over or underestimation. In this study, to determine whether or not the logistic regression model is correctly specified, a linktest is conducted. Generally, a linktest is used to assess a model’s link function for the detection of specification errors. It is based on the assumption that when a model is correctly specified, it should be impossible to find any additional variables that may prove to be statistically significant in the model unless it is by chance (UCLA, n.d).57 To implement this test, two new variables, namely, _hat representing the predicted value and _hatsq representing the squared predicted value are generated. As a general rule, when a model is correctly specified, the _hat value is expected to be significant whilst the _hatsq value remains insignificant. However, when the _hatsq is significant, it indicates that the model is not correctly specified (Ibid). Furthermore, the logistic regression model also assumes a relationship that is “linear in the logit”, in other words requires that the predictor variables should be linearly related to the log odds (Menard 1995, p. 60). Also, given that the logistic regression model is estimated using maximum likelihood techniques, it typically requires a relatively large sample size to produce parameter estimates that aside “being consistent, are asymptotically efficient” (Dougherty 2011, p. 379). As such, Hosmer and Lemeshow (2000, p. 347) suggest that, as a simple rule of the thumb, the ratio of observations change when a particular case is deleted. 57 As explained in https://stats.idre.ucla.edu/stata/webbooks/logistic/chapter3/ lesson-3-logistic-regression-diagnostics/ 5.6 Data Analysis Framework 73 to predictors in a logistic regression model should be a minimum of 10 cases per each predictor in the model. Finally, notwithstanding the above considerations, it is important to note that because the logistic regression model is estimated using a maximum likelihood estimation procedure, it typically does not require distributional assumptions such as multivariate normality, homoscedasticity, autocorrelation and linearity between the dependent and independent (Harrell 2015, p. 221; Osborne 2015, p. 100). 5.6.2 Interpretation of the Logistic Regression Model Depending on the statistical software used in estimating the logistic regression model, the output or results may be presented in specific formats. However, irrespective of the output generated, there are a number of key parameters whose interpretation remain very important in understanding the results of the logistic regression model. A number of these parameters are briefly discussed below. First, a very important statistical parameter in logistic regression is the logit coefficient which is expressed in units of log odds. Commonly, the logit coefficient is interpreted as “the change in the predicted logit or the log odds for a one-unit increase in the predictor variable when holding all other predictors constant or controlling for the effects of other predictors” (Liu 2018, p. 114). In most statistical software packages, the logit coefficient is reported alongside its standard errors and confidence intervals. A positive value of the logit coefficient ordinarily indicates an increase in the predicted log odds while a negative value would as well point to a decrease in the predicted log odds. More so, the statistical significance of the logit coefficient is determined using the Wald test statistic. The Wald statistic is explained in the next section. Additionally, another critical parameter in interpreting the results of the logistic regression model is the odds ratio. The odd ratio is defined as “a ratio of two odds” that is, the odds of an event occurring in one group relative to the odds that the same event occurs in another (Liu 2018, p. 100). For Field (2005, p. 225), odds ratio provides an indication “of the change in 5 Research Methodology 74 log odds resulting from a unit change in the predictor variable”. Mathematically, it can be derived by exponentiating the logit coefficient and therefore provides the same information as the logit coefficient does although in a different way (Menard 1995, p. 49). The odds ratio has values ranging from negative infinity to positive infinity. By way of interpretation, an odds ratio value that is greater than one indicates that, the odds of success for an outcome increases with a unit increase in the predictor variable, while an odds ratio value that is less one indicates that the odds of success for an outcome decreases when the predictor variable increases by one unit (Liu 2018, p. 114). However, when the value of an odds ratio that is equal to one, it implies that the predictor variable has no effect on the odds of success for the outcome (Ibid). Finally, when interpreting the strength or magnitude of effect of coefficients in the logit model, an alternative to odds ratio is usually marginal effects. According to Bartus (2005, p. 310) marginal effects “measure the change in the expected value of y as one independent variable increases by unity while all other variables are kept constant”. For Torres-Reyna (2014, p. 8), this rate of change in probability is instantaneous when the predictor variable is continuous and discrete when the predictor variable is binary. Generally, there are 3 different types of marginal effects: Average Marginal Effects (AME), Marginal Effects at Mean (MEM) and Marginal Effects at Representative values (MER)58. According to Norton and Dowd (2018, p. 867), marginal effects are relatively easy to understand and interpret, and as well are less sensitive to model variations. Therefore, they tend to be more preferred compared to odds ratios (Ibid, p. 876)59. Similar to regression coefficients, marginal effects also range from negative infinity to positive infinity. A positive estimate generally indicates an increase in the probability of falling into a certain category while a negative estimates denotes a decrease in the said probability (Liu 2018, p. 457). 58 A detailed discussion of AME, MEM and MER is presented in Williams (2012). 59 Mood (2010) presents a detailed discussion of the advantages of using marginal effects over odd ratios and log odds. 5.6 Data Analysis Framework 75 5.6.3 Assessing the Quality (Goodness-of-fit) of the Logistic Regression Model Generally, when assessing the goodness-of-fit of a logistic regression model, a key parameter of interest is the log-likelihood statistic. According to Field (2005, p. 221), the log-likelihood statistic tends to be “analogous to the residual sum of squares in multiple regression in the sense that it is an indicator of how much unexplained information there is after a model has been fitted”. Thus, when the value of the log-likelihood statistic is large, it indicates a poor fit for the model since “the larger the value of the log-likelihood, the more unexplained observations there are” (Ibid). In logistic regression, the log-likelihood statistic is used to conduct the likelihood ratio test. The likelihood ratio test basically compares the log-likelihood statistics for two models, one of which is normally a subset of the other (Liu 2018, p. 126; Kleinbaum and Klien 2010, pp. 134–135). When the likelihood ratio chi square test statistic is significant, it implies that the variable or group of variables in the full model altogether contribute significantly in predicting the outcome compared to the reduced model. Hence, the full model fits the data better (Ibid). In addition to the likelihood ratio test, another very important test used to assess the overall quality of the logistic regression model is the HosmerLemeshow goodness-of-fit test. According to Liu (2018, p. 122) Hosmer-Lemeshow test examines whether or not there are any statistically significant differences between the observed and expected frequencies after the logistic regression model has been estimated. It is based on the idea that when a model fits the data well, the observed and predicted frequencies should match more closely (Ibid). Thus, the null hypothesis for the test indicates there is no significant difference between observed and predicted frequencies (otherwise there is goodness-of-fit) whereas the alternative hypothesis indicates opposite. As a rule of the thumb, when the p-value of the test is insignificant (> 0.05), the null hypothesis fails to be rejected, thus pointing to some goodness-of-fit and vice versa (Ibid). Furthermore, another statistical measure that is commonly used to assess the predictive efficiency or usefulness of the logistic regression model 5 Research Methodology 76 is the percentage (%) correctly classified. The % correctly classified indicates the extent to which the logistic regression model is able to accurately predict the outcome category under a given cut-off value (Tabachnick and Fidell 2014, pp. 513–514). Normally, it is expected that when a model fits the data well, the % of correctly classified cases (classification accuracy) should be substantially higher for the model than one attained by chance (Kassambara 2017, p. 145). Also, although not directly related to assessing the overall fit of a model, another important parameter in logistic regression is the wald statistic. Essentially, the wald statistic is used to evaluate the significance of individual parameters in the model (Liu 2018, p. 104). Mathematically, the wald statistic can be derived by dividing each parameter estimate by its corresponding standard error (Field 2005, p. 224). The wald statistic has a chi-square distribution and is used to test the significance of each predictor in the model as stated earlier. The null hypothesis for the wald test assumes that the coefficients of the each individual parameter in model is equal to zero while the alternative hypothesis assumes otherwise. Thus, in each case where the null hypothesis is rejected, the predictor variable is accordingly deemed to be a statistically significant predictor of the outcome variable and vice versa (Ibid). For Liu (2018: 104), a very appealing feature of the wald test is that “it can also be used to test the effect of multiple predictors simultaneously”. Finally, aside the above, there are several other statistical measures or indices that provide useful information on the goodness-of-fit of the logistic regression model. Examples of these include the McFadden’s R2, Cox and Shell R2, Nagelkerke R2, Akaike Information Criterion and the Bayesian Information Criterion. Detailed discussions of these measures are presented in Liu (2018) and Menard (2000) among others. 5.7 Empirical Model, Variable Description and Estimation Strategy 77 5.7 Empirical Model, Variable Description and Estimation Strategy 5.7.1 The Empirical Model Following the approach of Franko et al. (2013) the following econometric model is estimated; logitprSUPPORT SELF INTEREST BELIEFS I ij i i [ ]_      101 23   NNSTITUTIONALQUALITY KNOWLEDGE Xe i ikki i    4 (7) Where: SUPPORTij measures the probability that an individual supports the specific social protection policy j under consideration (i.e., Social Cash Transfers - LEAP or Social Health InsuranceNHIS). SELF_INTERESTi denotes self-interest and relates to the conflict of interest within a specific policy area (Income and labour market characteristics). BELIEFSi captures an individual’s beliefs about self versus exogenous determination of poverty and inequality. INSTITUTIONAL QUALITYi measures an individual’s level of trust in public institutions/authorities responsible for implementing social protection policies and the administration of procedural justice. KNOWLEDGEi measures an individual’s level of knowledge on the specific social protection policy area in question. Xki is a vector of other socio-demographic or contextual variables not mentioned (control variables) ei denotes the idiosyncratic error term β0 ….βk are parameters to be estimated 5 Research Methodology 78 5.7.2 Variables Description 5.7.2.1 The Dependent Variables Given the comparative nature of the study, two distinct dependent variables60 are used in the empirical analysis: support for cash transfers (i.e., LEAP) and support for social health insurance (i.e., NHIS). Support for Cash Transfers (LEAP) To capture respondent’s preferences for social cash transfers, specifically the LEAP cash transfer programme, the following question was posed during the survey; “On a scale of 1 (Strongly disagree) to 5 (Strongly agree), please tell me the extent to which you either agree or disagree with the following statement: Government should increase income taxes61 to enable it continue to and better provide income support for the poor in Ghana through social cash transfer programmes such as LEAP”. It is important to highlight that although the original answer to this question was captured on a 5 point ordinal scale, a cursory overview of respondents’ answers largely showed the clustering of answers at the two extreme ends of the continuum. Therefore, following Rehm (2005), the answers to this question were recategorized or transformed into two distinct categories; “Agree” (capturing responses that were either “Strongly Agree” or “Agree”) and “Disagree” (if respondent stated otherwise by answering “Strongly disagree”, “Disagree” and “Neither agree nor disagree”). 60 Based on varying degrees of redistribution and risk sharing. 61 Given that publicly financed social protection may come with some tax implication, the researcher included a tax stimulus (trade-off) in measuring respondent preferences for social protection. The inclusion of a tax stimulus was also aimed at reducing the chances that respondents will only answer positively due to “social desirability bias”. 5.7 Empirical Model, Variable Description and Estimation Strategy 79 Consequently, individuals with positive responses (agreed) were deemed to express support for cash transfers while those who responded negatively (disagreed) were as well deemed not to support cash transfers. The distribution of answers based on the binary operationalization of the variable is provided in section 6.1.3 of this study. Support for Social Health Insurance (NHIS) Furthermore, aside support for cash transfers as captured above, the survey also measured respondent’s preferences for Social Health Insurance, specifically support for the NHIS programme using the following question in the survey; “On a scale of 1 (Strongly disagree) to 5 (Strongly agree), please tell me the extent to which you ether agree or disagree with the following statement; Government should increase income taxes to enable it continue to and better fund the NHIS in order to improve access to quality healthcare services for all”.62 Similarly, in line with the justification provided in the preceding section, the responses to this question were also transformed or regrouped into two distinct categories; “Agree” (capturing responses that were either “Strongly Agree” or “Agree”) and “Disagree” (if respondent stated otherwise by answering “Strongly disagree”, “Disagree” and “Neither agree nor disagree”)� As previously explained, individuals who “agreed” to this question were deemed to express support for social health insurance whereas those who “disagreed” were considered not to support social health insurance. The distribution of answers based on the binary operationalization of the variable is as well provided under section 6.1.3 of this study. 62 Justification for the inclusion of a tax stimulus (trade-off) has been duly explained in the preceding section. 5 Research Methodology 86 be expected that individuals affiliated with the NPP may possibly express more support for both policies relative to those affiliated to either the NDC or other political parties, if and only if the logic of public support along the lines of political affiliation does exist. 5.7.3 Estimation Approach In estimating the empirical model, the researcher adopts a hierachical or blockwise entry approach. As such, all the relevant variables are introduced into the model in a stepwise manner. According to Field (2005), when utilizing the hierarchical approach, the general rule is to first enter known predictor variables from the extant literature, and thereafter add any other predictors that may be of interest to the study. Following this approach, the researcher estimates seven different variants of the empirical model (for each dependent variable), beginning with the base model which includes some known predictors (control variables) from existing empirical studies. Thereafter, the main variables of interest are entered into the model in a step wise manner in accordance to the sequence and logic of the theoretical framework as presented in chapter 4 of this thesis. An overview of the various groupings or blocks of variables for stepwise entry into the model is presented in the table 5.2 above. 5.8 Measuring Preferences: Dealing with the Challenges and Limitations Broadly speaking, measuring individual preferences is a very difficult and tricky task given the complexities involved in ensuring both the validity and reliability of the construct. Generally, preferences may either be stated or revealed (Segerson 2017, p. 21). Stated preferences refer to preferences inferred from individual responses to questions under hypothetical scenarios whereas revealed preferences refer to preferences inferred from observing (Ibid). Ideology may therefore not be the sole basis for joining political parties in Ghana as the case maybe in the some western democracies. 5.8 Measuring Preferences: Dealing with the Challenges and Limitations 87 and evaluating actual individual behavior (Ibid). Methodologically, a range of techniques are available for eliciting preferences with some techniques being more complex than others. For example in the literature on redistributive policy preferences, elicitation methods or techniques commonly used include survey questions, discrete choice experiments, laboratory experiments, choice ranking, etc. (for example, Hirvonen and Hoddinott, 2021; Alesina et al., 2019; Pederson and Shekha 2016; Abiiro et al. 2014; Fong, 2001; Fehr and Schmidt, 1999). In the context of this study, the researcher attempted to measure or elicit respondent stated preferences for both cash transfers and social health insurance using a hypothetical survey question with a tax trade off (see section 5.7.2). However, this approach to eliciting preferences although acceptable could possibly be challenging in relation to addressing the issue of validity (both internal and external).65 Validity refers to how accurate the construct reflects that which it is intended to measure (Heale and Twycross 2015, p. 66). For example, considering the fact that the questions that measured the dependent variables (preferences for both cash transfers and social health insurance) included a tax stimulus, there could have been a tendency for respondents to mistaken such questions as directly measuring preferences for taxation. However, given forethought, the researcher ensured that the field assistants responsible for data collection were well trained and fully understood the import of all questions in the survey instrument. Consequently, this enabled the field assistants to satisfactorily explain all questions to the understanding of respondents in other to avoid or minimize any such confusion. Also, the possibility that individual respondents would state preferences that are either biased or not reflective of their real choices when confronted with the same issue in real life situations could not be ruled. Although such an occurrence is largely beyond the control of the researcher, to minimize 65 Validity consist of both internal and external validity. Whereas internal validity reflects the extent to which “the study design, conduct and analysis answer the research questions without bias”, external validity refers to the possibility of generalization beyond the study’s context (Bender 2021, p. 510). 5 Research Methodology 88 the incidence of bias responses, the researcher and field assistants took time to explain the purpose of the survey to the understanding of all respondents before commencement of the interviews. Respondents were also politely urged to answer the questions to the best of their ability and in a candid or truthful manner66. Finally, following Busemeyer (2013), Alesina and Angeletos (2005) and Brooks and Manza (2006), individual policy preferences may possibly suffer from a feedback effect. This means that whereas aggregate policy preferences may influence or shape redistributive policies, over time redistributive policies may as well in turn influence individual preferences. However, considering that such feedback effects may be an outcome of long term experiences and typically requires data covering a time dimension (Bender 2021, p. 516), due to design and data limitations, these issues were not explored in this study. 5.9 Chapter Conclusion To wrap up, this chapter presented a comprehensive discussion of the research methodology underlying the study. The discussions therein sort to provide a context for understanding how the empirical results of the study were derived. The next section presents the results of the study. 66 Among others an alternative to the approach adopted in this study could have been a lab in the field experiment given that such methods provide a controlled environment for decision making. However, time and limited resources did not allow for a lab in the field experiment. 89 6 Presentation and Discussion of Empirical Findings 6.0 Chapter Overview This chapter presents and discusses the empirical results of the study. It begins with a brief description of the socio-demographic and labour market characteristics of respondents in the sample. Thereafter, the results of the empirical analysis regarding individual preferences for cash transfers (LEAP) and social health insurance (NHIS) are presented. In doing so, the author reports the results of the empirical model, and provides a series of post-estimation checks conducted to ensure reliability and validity of the results. The chapter ends with a discussion of the study’s findings with the aim of answering the research questions and hypotheses set forth earlier in the study. 6.1 Descriptive Analysis 6.1.1 Socio-demographic Characteristics of the Sample As stated in the preceding chapter, the empirical analysis for this study was conducted using data from a total of 596 respondents. Out of this number, 306 individuals representing 51.34% of the total sample were male whilst the remaining 290 respondents representing 48.66% were female. In general, the ages of respondents ranged between 18 years to 81 years. The mean age was approximately 38 years, depicting on the average a relatively youthful sample. With respect to the marital status of respondents, 40.44% reported being single, 48.15% were married, 3.69% were divorced and 7.72% were widowed. Thus, the majority of respondents (51.85%) were “not 6 Presentation and Discussion of Empirical Findings 90 currently married” as at the time of the survey. With regards to religious affiliation, the sample was predominantly Christian (87.75%), comprising 32.89% Catholics and 54.87% Protestants. Muslims accounted for 10.91% of the sample, while 1.34% identified with other religions or as atheist. The dominance of Christians in the sample is not surprising since the population of Ghana is predominantly Christian, despite the country’s secular constitutional framework. In terms of Ethnicity, the sample largely mirrored the cosmopolitan nature of the nation’s capital city Accra. Despite being dominated by Akans (46.31%), individuals from the other major ethnic groups were well represented in the sample. For instance, 16.61% of respondents were Ewe/ Anglo, 15.27% were Ga/Adangbe, 4.70% were Dagomba and 17.11% represented individuals from the other ethnic groups not mentioned. Figure 6.1 below provides details on the ethnicity of respondents in the sample. Figure 6.1. Ethnic Distribution of Respondents. Source: Authors’ own based on output of the analysis. 6.1 Descriptive Analysis 91 Furthermore, probably reflective of Ghana’s past efforts at expanding access to basic and secondary education for all its citizens, over 90% of respondents in the sample had attained at least some form of formal education at the time of the survey. Specifically, 30.37% of respondents indicated that they had attained some form of basic education (Primary and Junior High School), 29.70% indicated having attained secondary education and 31.21% reported having attained tertiary education. Only 8.72% reported not having attained any form of formal education. Interestingly, despite the overall good picture, a disaggregation by gender easily reveals wide gaps and huge inequalities in educational attainment between male and female respondents. While a majority of male respondents (73.2%) had altogether obtained at least some secondary and tertiary education, the majority of female respondents (52.07%) were reported to have either completed just basic education or had no formal education at all. This phenomenon which is largely reflective of the broader Ghanaian context, has been attributed to socio-economic factors such as early female marriages, which contribute to high dropout rates among girls (Ghana Statistical Service, 2014). The figure below presents a graphical representation of respondent’s educational attainment by gender. Furthermore, regarding union membership, approximately 57% of respondents reported being members of voluntary organizations including social groups, professional groups and trade unions. Among this category of respondents, females were slightly dominant (50.59%) compared to their male counterparts (49.41%). Additionally, in terms of political affiliation, individuals from the two major political parties in Ghana—the New Patriotic Party (NPP) and the National Democratic Congress (NDC)—constituted the majority (70.13%) of respondents in the sample. Specifically, 40.60% of respondents were affiliated to the NPP while 29.53% were affiliated to the NDC. The remaining 29.87% indicated their support for smaller political parties in Ghana. The dominance of individuals from the two major political parties in the sample is not very surprising given that the greater Accra region where the study was conducted, has hostorically been considered a swing region for both the NPP and NDC (Kim 2018, p. 32). 6 Presentation and Discussion of Empirical Findings 92 6.1.2 Economic and Labour Market Characteristics of the Sample Typical of a developing country context, the majority of respondents in the sample (61.91%) were located in the informal sector whilst only 38.09% were engaged in economic activities within the formal sector. Moreover, for those in the formal sector, approximately 69.6% were male while only 30.4% were female. However, in the case of the informal sector, approximately 60% were female and 40% were male. The overrepresentation of women in the informal sector is generally unsurprising, given the existing gender disparities in formal and informal sector employment in Ghana (Ghana Statistical Service 2019, pp. 77–92). The Figure below illustrates the distribution of respondents by sector of employment and gender. More so, characteristic of an urban economy, respondents were engaged in a wide range of occupational activities. Among those in the informal sector, Figure 6.2. Educational Attainment by Gender. Source: Authors’ own based on output of the analysis. 6.1 Descriptive Analysis 93 39.57% were skilled manual workers, 14.91% were unskilled manual workers, 36.59% were traders/vendors, 0.81% were farmers/fisherfolk, 1.08% indicated being a housewife/housemaker, 0.81% were supervisors/foremen, 4.07% were clerical/administrative staff of organizations and 2.17% were mid-level professionals. In contrast, among those in the formal sector, 35.68% reported being mid-level professionals, 21.15% were upper-level professionals, 25.99% were clerical/administrative staff, 11.89% were supervisors/foremen, 1.76% were skilled manual workers and 3.52% reported being unskilled manual workers. With respect to income, the self-reported net monthly income of respondents in the sample ranged between 0 and 5,000 Ghana Cedis (GH¢), with an overall average of GH¢ 685. As expected, the average monthly earnings of individuals in the formal sector was significantly higher (GH¢1,096) than those in the informal sector (GH¢ 434). Also, the average monthly Figure 6.3. Employment Sector by Gender. Source: Authors’ own based on output of the analysis. 6 Presentation and Discussion of Empirical Findings 94 income for male respondents in the sample (GH¢823) was considerably higher compared to their female counterparts (GH¢542). Furthermore, an analysis of the poverty status of respondents based on the Ghana Statistical Service upper poverty line of GH¢1,760.8 per adult equivalent per year in 2016/2017, revealed that approximately 91% of all respondents in the sample could be categorized as non-poor given their self-reported monthly incomes. More so, when analyzed accordingly to employment sector, it is interesting to note 99.1% of those in the formal sector were classified as non-poor, while only 0.9% were categorized as poor. In contrast, among those in the informal sector, 86.2% were non-poor, while 13.8% fell below the poverty line (see figure 6.4 below). Evidently, the percentage of non-poor in the sample appeared to be very high. However, this was not unexpected given that poverty in Ghana is predominantly a rural phenomenon (Ghana Statistical Service 2017, p. 14). Therefore, considering that this study was conducted in the highly urbanized district of Accra, it is likely that individuals incomes could be comparatively higher. Due to the insignificant number Figure 6.4. Income Category by Employment Sector. Source: Authors’ own based on output of the analysis. 6.1 Descriptive Analysis 95 of formal sector poor in the sample, (only 2 individuals), the researcher excluded them from the inferential analysis focused on income groups. In addition, regarding respondents’ employment status, approximately 71.81% reported being in full time employment, while 14.93% were in parttime employment. Furthermore, 3.36% indicated that they were retired from active employment, 4.19% were temporarily not working and 5.70% reported being currently unemployed. In terms of employment industry, the sample showed a mixed distribution. Specifically, 47.65% reported being self-employed, 31.38% Were being employed in the private industry, 4.36% in NGOs/Civil Society Organizations and 16.61% with State/Government institutions. The employment industry statistics are equally unsurprising considering that a majority of respondents in the sample fall within the informal sector where self-employment is highly prevalent, particularly in the Ghanaian context (Ghana Statistical service, 2019). Figure 6.5 below provides a detailed illustration of the different industries/sectors in which respondents were employed. Figure 6.5. Distribution of Respondents by Employment Industry. Source: Author’s own based on output of the analysis. 6 Presentation and Discussion of Empirical Findings 102 for cash transfers. The effect of being female, rather than male, changes from negative to positive although it still remains statistically insignificant. Interestingly, compared to the previous model, being affiliated to the NPP or NDC relative to other political parties although positive, now proves to be statistically insignificant in determining support for cash transfers. Nevertheless, all other variables in the model remain the same as explained in the previous model. In Model 3, the variable broadly capturing the potential conflict of interest between formal and informal sector is introduced. Although it has the expected negative sign, it turns out to be statistically insignificant. All other variables remain consistent with the explanations provided in Model 2. Next, in Model 4 the dummy variable “Poverty Exogenous” which serves as a proxy for an individual’s belief about self-versus-societal determination of Explanatory Variables Dependent Variable: Support for Cash Transfers (LEAP) Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Trust in Public Authorities. 2.029*** (0.504) 1.876*** (0.512) Knowledge of LEAP 0.960*** (0.352) Constant –1.002* (0.555) 2.362** (0.985) 2.098** (1.005) –0.445 (1.120) –0.984 (1.166) –0.785 (1.179) Observations 596 579 579 579 579 579 Pseudo R-squared 0.192 0.230 0.231 0.345 0.366 0.376 Likelihood Ratio 156.9*** 182.5*** 183.9*** 274*** 291.3*** 299*** (df) 910 11 12 13 14 Note: * p < 0.10, ** p < 0.05, *** p < 0.01. Robust standard errors in parentheses. Table 6.1. (Continued) 6.2 Inferential Analysis 103 poverty is added. As expected, it turns out to be positive and highly statistically significant. The variable “Formal” although still negative also becomes statistically significant at 10%. Interestingly, the effect of being “Female” changes once again from negative to positive but nevertheless remains statistically insignificant. The variable “Basic Education” also loses its statistical significance in this model. However, all other variables remain consistent with previous discussions. In Model 5, a measure of institutional quality is introduced into the model. Consistent with a priori expectations, it turns out to be positive and highly statistically significant. Also, the variable “Formal” becomes significant now at 5% whilst all other variables in the model retain the same form as explained in model 4. Finally, in Model 6, the variable capturing respondents knowledge of cash transfers “Knowledge of LEAP” is introduced into the model. It is important to note that Model 6 represents the full specification and is, therefore, our main model of interest. Evidently, as shown in table 6.1, age and level of education are the only control variables that have a statistically significant effect on support for cash transfers. Virtually all the other control variables in the model tend to be insignificant. However, in congruity to our expectations, individuals who attribute poverty to external causes, those with high levels of trust in public institutions and those with greater knowledge of LEAP tend to be positively and significantly associated with support for social cash transfers. More so, consistent with the self-interest hypothesis, higher levels of income and being in the formal sector (relative to the informal sector) negatively affects individual support for social cash transfers. Nonetheless, to further examine the self-interest argument as presented in the theoretical framework, specifically in hypothesis 2a and 2b, the researcher estimates an alternative specification of the fully specified model. Instead of using a continuous measure of income, this specification introduces three dummies (Informal poor, Informal Non-poor and Formal Non-poor) to capture the potential conflict of interest between individuals from different income groups in both formal and informal sectors. As shown in model 7 (Table 6.2 below), while being non-poor in the informal sector has the 6 Presentation and Discussion of Empirical Findings 104 expected negative sign, it turns out to be statistically insignificant compared to the base group. However, being non-poor in the formal sector has a negative and statistically significant effect on support for cash transfers compared to the informal sector poor, as expected. (Continued) Table 6.2. Logistic Regressions Results: Support for Cash Transfers (LEAP)—(II). Explanatory Variables Dependent Variable: Support for Cash Transfers (LEAP) Model 6 Model 7 Coefficient AME Coefficient AME Age 0.064*** (0.011) 0.009*** (0.001) 0.062*** (0.011) 0.009*** (0.001) Female 0.081 (0.241) 0.011 (0.033) 0.156 (0.235) 0.022 (0.033) Currently Married –0.222 (0.251) –0.030 (0.034) –0.289 (0.247) –0.040 (0.034) (ref = No Formal Education) Basic Education –0.684 (0.534) –0.102 (0.079) –0.604 (0.504) –0.092 (0.075) Secondary Education –1.267** (0.549) –0.191** (0.082) –1.277** (0.518) –0.197** (0.079) Tertiary Education –1.652*** (0.631) –0.248*** (0.096) –1.838*** (0.590) –0.283*** (0.091) (ref = Other Political Party) NPP 0.180 (0.284) 0.025 (0.039) 0.260 (0.282) 0.037 (0.040) NDC 0.272 (0.295) 0.038 (0.041) 0.387 (0.294) 0.055 (0.042) Union Member –0.333 (0.231) –0.046 (0.032) –0.333 (0.228) –0.047 (0.032) Income(In) –0.473*** (0.171) –0.065*** (0.023) Formal –0.831** (0.347) –0.119** (0.051) (ref = Informal Poor) Informal Non-Poor –0.016 (0.423) –0.002 (0.063) 6.2 Inferential Analysis 105 Furthermore, to illustrate the magnitude of effect for the variables just discussed above, the average marginal effects for model 6 and 7 are computed and presented also in table 6.2 above.70 Clearly, from model 6, it is evident that a unit increase in age increases the probability that an individual will support social cash transfers by 0.9 percentage points. With respect to level of education, compared to the base group (individuals with no formal education) attaining secondary and tertiary education reduces the predicted probability of supporting social cash transfers by approximately 19.1 and 24.8 percentage points, respectively. Ad70 The author predominantly discusses the average marginal effects for only the variables that recorded a significant value. Also AMEs are multiplied by 100% for ease of interpretation. Table 6.2. (Continued) Explanatory Variables Dependent Variable: Support for Cash Transfers (LEAP) Model 6 Model 7 Coefficient AME Coefficient AME Formal Non-Poor –1.162** (0.506) –0.172** (0.076) Poverty Exogenous 2.034*** (0.256) 0.318*** (0.037) 2.121*** (0.254) 0.338*** (0.037) Trust in Public Authorities 1.876*** (0.512) 0.259*** (0.068) 1.894*** (0.502) 0.267*** (0.068) Knowledge of LEAP 0.960*** (0.352) 0.133*** (0.048) 0.883** (0.345) 0.125*** (0.048) Constant –0.785 (1.179) –3.558*** (0.790) Observations 579 579 583 583 Pseudo R-squared 0.376 0.366 Likelihood Ratio(df) 299 (14)*** 293.1(df)*** Note: * p < 0.10, ** p < 0.05, *** p < 0.01. Robust standard errors in parentheses. AME represent average marginal effects. 6 Presentation and Discussion of Empirical Findings 106 ditionally, a one-unit increase in Income (ln) decreases the likelihood of supporting cash transfers by 6.5 percentage points. Similarly, being employed in the formal sector relative to the informal sector also decreases the likelihood of supporting cash transfers by 11.9 percentage points. Conversely, individuals who believe that poverty is caused by structural forces have an increased likelihood of supporting social cash transfers by approximately 31.8 percentage points in comparison to those who do not. Equally, a unit increase in trust in public authorities also increases the likelihood of support for cash transfers by 25.9 percentage points. The huge effect of trust in public authorities highlights the extreme importance of institutional quality in shaping citizen’s willingness to accept taxation for social protection, particularly in a developing country context. Furthermore, the results also show that a unit increase in knowledge of LEAP is associated with a 13.3 percentage point increase in the likelihood of supporting social cash transfers. Lastly, based on model 7, it is also evident that, compared to the base group (informal sector poor), individuals in the formal sector who are non-poor are 17 percentage points less likely to support social cash transfers. Moving further, to provide a much detailed understanding of the results and further illustrate the effect of our main variables on individual support for cash transfers, particularly between formal and informal sector workers, four different scenarios (predicted probabilities) are simulated71 based on model 6 and model 7. Given that the results of these simulations do not significantly differ across model specifications, the researcher presents the results based on model 6 in the sections below. However, for completeness, the simulated results for model 7 are presented in appendix 1. To start, in scenario one, the effect of beliefs about the causes of poverty (self-versussocietal determination) on support for cash transfers is simulated for individuals in the formal and informal sectors. The results of the simulation are graphically presented in figure 6.9 below. 71 The researcher focuses exclusively on the main explanatory variables as highlighted in the theoretical framework but also includes the variable “level of education” for detailed illustrative purposes. 6.2 Inferential Analysis 107 Evidently, the graph below confirms the unified effect of beliefs on support for cash transfers, as discussed in the preceding paragraphs. As shown, the attribution of poverty to external causes increases the predicted probability of supporting cash transfers from 0.17 to 0.47 for informal sector workers, and from 0.27 to 0.62 for those in the formal sector. Hence, regardless of employment sector, individuals in the sample who generally attribute the causes of poverty to external factors are significantly more likely to support cash transfers than those who attribute poverty to individual responsibility. Additionally, in scenario two, the effect of institutional trust on support for cash transfers is also simulated for individuals in both the formal and informal sectors. For illustrative purposes, the researcher specifically focuses Figure 6.9. The Effect of Beliefs by Employment Sector. Source: Authors’ own based on output of the analysis. 6 Presentation and Discussion of Empirical Findings 108 on respondents in the sample with the lowest, average and highest levels of institutional trust. The results of the simulation are presented in figure 6.10 below. Again, in line with the results presented earlier, higher levels of trust in public authorities generally increase the probability of supporting cash transfers in both groups. An individual in the informal sector with the lowest level of trust is associated with a 0.37 probability of supporting cash transfers, whereas a counterpart in the formal sector is associated with a 0.26 probability of supporting cash transfers. For those with an average level of trust, the predicted probability of supporting cash transfers is 0.49 when in the informal sector, and 0.36 when in the formal sector. Likewise, among individuals with the highest level of trust, the probability of supporting social cash transfers is 0.65 when in the informal sector and 0.53 in the formal Figure 6.10. The Effect of Institutional Trust by Employment Sector. Source: Authors’ own based on output of the analysis. 6.2 Inferential Analysis 109 sector. Conclusively, although the effect of institutional trust is present in both groups, it tends to be slightly higher in terms of magnitude for informal sector workers compared to those in the formal sector. Furthermore, in scenario three, the researcher simulates the effect of varying levels of knowledge on support for cash transfers across both formal and informal sectors. The results of the simulation are as well presented in figure 6.11 below. Graphically, the results displayed in figure 6.11 further buttress the effect of knowledge as highlighted in the preceding section. Consistent with a priori expectations, individuals with higher levels of knowledge on LEAP are associated with relatively higher probabilities of supporting cash transfers compared to those with lower levels of knowledge. The effect of knowledge although consistent in both groups, tends to be relatively greater in terms of magnitude amongst informal sector workers relative to their formal sector counterparts. Specifically, for an informal sector individual with the lowest Figure 6.11. The Effect of Knowledge by Employment Sector. Source: Authors’ own based on output of the analysis. 6 Presentation and Discussion of Empirical Findings 110 level of knowledge (Index = 0), the predicted probability of supporting cash transfers is 0.43, whereas for a counterpart in the formal sector, the predicted probability is just 0.32. Similarly, an individual with the average level of knowledge (Index = .38), is associated with a 0.48 predicted probability of supporting cash transfers when in the informal sector and only 0.36 when in the formal sector. Also, for those with the highest level of knowledge (Index = 1), the same pattern can be observed. For instance, such individuals are associated with a 0.57 predicted probability of supporting cash transfers when in the informal sector and only 0.45 when in the formal sector. Finally, in scenario four, the effect of knowledge on support for cash transfers is also simulated for individuals across different levels of educational attainment. The result of the simulation is presented in figure 6.12 below. Figure 6.12. The Effect of Knowledge by Level of Education. Source: Authors’ own based on output of the analysis. 6.2 Inferential Analysis 111 Generally, as expected, higher levels of knowledge about LEAP are associated with higher probabilities of supporting cash transfers across all levels of education. However, the magnitude of the effect tends to be slightly higher at all knowledge levels (lowest, average and highest) for individuals with no formal education compared to those with higher levels of formal education. For example, the predicted probability of supporting cash transfers for an individual with average level of knowledge (Index = .38) is 0.55 for someone with no formal education, 0.45 for someone with basic education, 0.36 for someone with secondary education and only 0.31 for someone with tertiary education. Evidently, given that higher levels of education reduce support for cash transfers as previously discussed, it is reasonable to speculate that knowledge of policies predisposes individuals to primarily act based on self-interest, since individuals with poor educational backgrounds are more likely to belong to low income categories whereas those with higher education are more likely to be economically better off and belong to middle to high income groups. Hence, it can therefore be assumed that the effect of knowledge on support for cash transfers is, to some extent, mediated by an individual’s level of education. 6.2.1.2 Post Estimations Checks: Goodness-of-fit and Model Diagnostics To ensure the statistical validity and reliability of the results presented in the section above, several post estimation tests and model diagnostics are conducted. The results of these tests are presented below. I. Goodness-of-fit Test To assess how well the estimated models fits the data, the researcher proceeds as follows. First, the log likelihood ratio test72 for each model as shown in the bottom two rows of table 6.1, are analyzed. Evidently, the chi square test statistic for the log likelihood ratio test is significant across all models. 72 The mechanics of the Log Likelihood ratio test has already been discussed in the preceding chapter. 6 Presentation and Discussion of Empirical Findings 118 Table 6.4. Logistic Regressions Results: Support for Social Health Insurance (NHIS)—(I). Explanatory Variables Dependent Variable: Support for Social Health Insurance (NHIS) Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 Age 0.027*** (0.009) 0.034*** (0.010) 0.033*** (0.010) 0.023** (0.010) 0.022** (0.011) 0.020* (0.011) Female 1.034*** (0.202) 0.943*** (0.208) 0.966*** (0.210) 1.228*** (0.230) 1.252*** (0.231) 1.238*** (0.234) Currently Married –0.345 (0.210) –0.375* (0.218) –0.379* (0.219) –0.282 (0.236) –0.275 (0.237) –0.334 (0.240) (ref = No Formal Education) Basic Education –0.266 (0.461) –0.457 (0.525) –0.465 (0.525) –0.121 (0.548) –0.123 (0.553) –0.193 (0.581) Secondary Education –0.561 (0.460) –0.603 (0.532) –0.656 (0.535) –0.207 (0.559) –0.199 (0.562) –0.345 (0.589) Tertiary Education –0.674 (0.462) –0.515 (0.561) –0.704 (0.595) –0.368 (0.622) –0.313 (0.623) –0.537 (0.648) (ref = Other Political Party) NPP 0.243 (0.223) 0.129 (0.229) 0.151 (0.230) 0.026 (0.250) 0.030 (0.252) –0.099 (0.257) NDC 0.549** (0.249) 0.391 (0.255) 0.388 (0.256) 0.249 (0.273) 0.261 (0.275) 0.184 (0.281) Union Member 0.036 (0.194) 0.084 (0.199) 0.079 (0.200) 0.115 (0.215) 0.080 (0.217) –0.108 (0.227) Income(In) –0.282** (0.142) –0.317** (0.147) –0.133 (0.156) –0.135 (0.156) –0.143 (0.158) Formal 0.277 (0.288) 0.189 (0.309) 0.087 (0.311) –0.089 (0.318) Poverty Exogenous 1.783*** (0.219) 1.689*** (0.222) 1.699*** (0.226) Trust in Public Authorities 1.157** (0.500) 1.297** (0.515) (Continued) 6.2 Inferential Analysis 119 In Model 14, the second main explanatory variable “Formal” is introduced to assess the potential conflict of interest between formal and informal sector workers with regards to support for social health insurance. Although its coefficient is positive, it fails to show any statistical significance. The variable “NDC” loses it statistical significance while all other variables in the model remain the same as discussed in the previous model. In Model 15, the variable “Poverty Exogenous” is introduced to account for an individual’s belief in self-versus-societal determination of poverty. In line with expectations, the variable records a positive and a highly statistically significant effect on support for social health insurance. However, the variables “currently married” and “lncome (ln)” lose their statistical significance, while all other variables remain unchanged. In Model 16, the variable “Trust in Public Authorities” is introduced into the model as a measure of institutional quality. As expected, it also turns out to be positive and highly statistically significant. Nonetheless, all other variables in the model remain the same as discussed in the previous model. Explanatory Variables Dependent Variable: Support for Social Health Insurance (NHIS) Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 Knowledge of NHIS 1.908*** (0.491) Constant –0.232 (0.588) 1.408 (0.994) 1.610 (1.018) –0.491 (1.102) –0.824 (1.121) –1.888 (1.188) Observations 596 579 579 579 579 579 Pseudo R-squared 0.0845 0.0943 0.0956 0.199 0.207 0.229 Likelihood Ratio 61.59*** 66.61*** 67.53*** 140.6*** 146.1*** 161.7*** (df) 910 11 12 13 14 Note: * p < 0.10, ** p < 0.05, *** p < 0.01 Robust standard errors in parentheses. Table 6.4. (Continued) 6 Presentation and Discussion of Empirical Findings 120 In the final step, a complete specification of the empirical model is presented with the inclusion of the variable measuring knowledge of NHIS (Model 17). Strikingly, the statistical significance of “Age” reduces to 10%, while “Female” remains significant even at 1%. However, aside these two variables, virtually all other control variables show no statistically significant effect on support for social health insurance. With regards to our main variables of interest, higher levels of income and being in the formal sector (relative to the informal sector) do not exhibit statistically significant effects on support for social health insurance. Nonetheless, the remaining variables—“Poverty Exogenous”, “Trust in Public Authorities” and “Knowledge of NHIS”—all have the expected positive sign and as well do prove to be statistically significant predictors of support for social health insurance, as hypothesized earlier. Furthermore, to examine in detail the effect of self-interest on support for social health insurance as highlighted specifically in hypothesis 2b, the researcher estimates an alternative specification of the fully specified model. This specification uses a set of three dummy variables to capture the potential conflict of interest between individuals of different income groups (i�e�, Informal poor, Informal Non-poor and Formal Non-poor76) in both formal and informal sectors, similar to section 6.2.1. The results of the estimation are shown in model 18 in table 6.5 below. Interestingly, although both Informal sector non-poor and Formal sector non-poor have negative coefficients, they turn out to be statistically insignificant in determining support for social health insurance. Possible reasons for the null effect of these variables are explored in detail under the discussion section. Moving further, the researcher computes and presents the average marginal effects for the relevant variables discussed above. As shown in table 6.5, a unit increase in age increases the predicted probability of supporting social health insurance by approximately 0.3 percentage points. Similarly, being female relative to male increases the likelihood of supporting social 76 As already explained, the category ‘formal sector poor” have been excluded from the empirical analyses due to the very insignificant number of cases (i.e. only 2 respondents). 6.2 Inferential Analysis 121 Table 6.5. Logistic Regressions Results: Support for Social Health Insurance (NHIS)—(II). Explanatory Variables Dependent Variable: Support for Social Health Insurance (NHIS) Model 17 Model 18 Coefficient AME Coefficient AME Age 0.020* (0.011) 0.003* (0.002) 0.021** (0.011) 0.003** (0.002) Female 1.238*** (0.234) 0.195*** (0.035) 1.294*** (0.233) 0.202*** (0.034) Currently Married –0.334 (0.240) –0.051 (0.037) –0.372 (0.239) –0.057 (0.036) (ref = No Formal Education) Basic Education –0.193 (0.581) –0.028 (0.082) –0.055 (0.555) –0.008 (0.080) Secondary Education –0.345 (0.589) –0.051 (0.084) –0.225 (0.561) –0.033 (0.081) Tertiary Education –0.537 (0.648) –0.082 (0.095) –0.527 (0.610) –0.081 (0.091) (ref = No Formal Education) NPP –0.099 (0.257) –0.015 (0.040) –0.119 (0.257) –0.018 (0.040) NDC 0.184 (0.281) 0.028 (0.042) 0.176 (0.282) 0.026 (0.042) Union Member –0.108 (0.227) –0.017 (0.035) –0.150 (0.226) –0.023 (0.034) Income(In) –0.143 (0.158) –0.022 (0.024) Formal –0.089 (0.318) –0.014 (0.049) (ref = Informal Poor) Informal Non-Poor –0.259 (0.470) –0.038 (0.067) Formal Non-Poor –0.387 (0.530) –0.058 (0.077) (Continued) 6 Presentation and Discussion of Empirical Findings 122 health insurance by approximately 20 percentage points. Furthermore, a unit increase in institutional trust also increases the likelihood that an individual will support social health insurance by about 20 percentage points while a unit increase in Knowledge of NHIS is associated with a 29.3 percentage point increase in the likelihood of supporting social health insurance. The largest effect is observed for the variable capturing an individual beliefs about causes of poverty. Clearly, the results indicate that, individuals who believe poverty is not due to individual responsibility are approximately 30 percentage points more likely to support social health insurance than those who think otherwise. This huge effect undoubtedly underscores the centrality of poverty beliefs in shaping individual preferences for social protection. To illustrate these results in detail, the researcher replicates the simulations conducted in the previous section. Similarly, in scenario 1, the effect of Table 6.5. (Continued) Explanatory Variables Dependent Variable: Support for Social Health Insurance (NHIS) Model 17 Model 18 Coefficient AME Coefficient AME Poverty Exogenous 1.699*** (0.226) 0.299*** (0.039) 1.722*** (0.224) 0.301*** (0.038) Trust in Public Authorities 1.297** (0.515) 0.199** (0.078) 1.296** (0.512) 0.198** (0.077) Knowledge of NHIS 1.908*** (0.491) 0.293*** (0.072) 2.067*** (0.492) 0.315*** (0.071) Constant –1.888 (1.188) –2.728*** (0.846) Observations 579 579 583 583 Pseudo R-squared 0.229 0.232 Likelihood Ratio (df) 161.7(14)*** 164.4(14)*** Note: * p < 0.10, ** p < 0.05, *** p < 0.01. Robust standard errors in parentheses. AME represent average marginal effects. 6.2 Inferential Analysis 123 beliefs concerning self-versus-societal determination of poverty on support for social health insurance is simulated for both formal and informal sector groups. The results of the simulation are presented in figure 6.15 below. As shown the effect of beliefs in predicting support for social health insurance appears to be nearly uniform across both formal and informal sector groups. For example, attributing poverty to societal causes rather than individual responsibility increases the probability of supporting social health insurance from 0.54 to 0.83 for individuals in the informal sector and from 0.52 to 0.82 for those in the formal sector. More so, it is also very clear that the differences in predicted probabilities between individuals holding similar beliefs in both groups are minimal, although the effect is slightly higher for those in the informal sector. Figure 6.15. The Effect of Beliefs by Employment Sector. Source: Authors’ own based on output of the analysis. 6 Presentation and Discussion of Empirical Findings 124 Again, in scenario two, the researcher replicates the simulated effect of institutional trust on support for social health insurance across both groups, purposely using respondents with the lowest, mean and highest levels of institutional trust. The results of the simulation are as well shown in figure 6.16 below. Clearly, the results of the simulation reinforces the fact that higher levels of institutional trust are associated with an increased likelihood of supporting social health insurance among respondents in the sample. However, the strength of this effect seems to slightly differ between same individuals in the formal and informal sectors. For example, the predicted probability of supporting social insurance for an individual with the lowest level of trust (Index = 0) in the formal sector is 0.60, while the predicted probability for same individual in the informal sector is 0.62. For those with an Figure 6.16. The Effect of Institutional Trust by Employment Sector. Source: Authors’ own based on output of the analysis. 6.2 Inferential Analysis 125 average level of trust (Index = .42), the predicted probability of supporting social health insurance is 0.70 in the formal sector and 0.72 in the informal sector. Likewise, for individuals with the highest level of institutional trust (Index = 1), the predicted probability of supporting social health insurance is 0.80 in the formal sector and 0.82 in the informal sector. Additionally, in scenario 3, the researcher simulates effect of knowledge about NHIS on support for social health insurance. The results are graphically presented in figure 6.17 below. A visual analysis of figure 6.17 clearly reveals that knowledge significantly increases the probability of supporting social health insurance in both formal and informal sector groups. For instance, within the informal sector, the predicted probability of supporting social health insurance is 0.44 for an individual with no knowledge of NHIS (Index = 0), 0.57 for those with average Figure 6.17. The Effect of Knowledge by Employment Sector. Source: Authors’ own based on output of the analysis. 6 Presentation and Discussion of Empirical Findings 126 knowledge of NHIS (Index = 3.8) and 0.76 for those with higher knowledge of NHIS (Index = 1). Likewise, for those in the formal sector, individuals with no knowledge of NHIS are associated with a predicted probability of 0.42, whilst those with average and high knowledge of NHIS are as well associated with predicated probabilities of 0.55 and 0.75 respectively. Thus, these results show the consistent effect of knowledge across both groups. Finally, in the last scenario, the effect of knowledge on support for social health insurance is further examined by simulating its effect across different levels of educational attainment (see figure 6.18 below). Generally, consistent with a prior expectations, higher levels of knowledge about the NHIS significantly increases the probability of supporting social health insurance regardless of an individual’s level of educational attainment. As shown, for individuals with no formal education, the probability of supporting social health insurance is 0.49 when they have no Figure 6.18. The Effect of Knowledge by Level of Education. Source: Authors’ own based on output of the analysis. 6.2 Inferential Analysis 127 knowledge whatsoever of NHIS, 0.62 when they have average knowledge of NHIS and 0.80 when they have the high knowledge of NHIS. Similarly, for individuals with tertiary education, the predicted probability of supporting social health insurance is 0.39 when they have no knowledge about NHIS, 0.53 when they have average knowledge of NHIS and 0.73 when they have high knowledge of NHIS. The same pattern is observed among individuals with only basic education as well as those with only secondary education in the sample. However, given the close clustering of all these groups, it is quite evident that the differences across educational levels for individuals with similar knowledge levels are quite marginal, except for those at the two extremes of the education ladder (no formal education and tertiary education). For instance, an individual with average knowledge of NHIS has a 0.63 predicted probability of supporting social health insurance when in the group of no formal education, 0.59 when they have some basic education, 0.56 when they have attained secondary education and 0.53 when they have attained tertiary education. It is therefore not surprising that the differences in education levels tends to be statistically insignificant in explaining individual support for social health protection. 6.2.2.2 Post Estimations Checks: Goodness-of-fit and Model Diagnostics I. Goodness-of-fit Test To evaluate the goodness-of-fit of the models estimated, the researchers undertook the following. First, the log likelihood ratio test results presented at the bottom rows of table 6.4 were analyzed. Clearly, all models exhibit a statistically significant chi-square test statistic for the log likelihood ratio test, indicating that, each estimated model provides a significantly better fit compared to the model with only the intercept. Furthermore, to evaluate the overall goodness of fit of our main model (Model 17), the Hosmer and Lemeshow’s goodness-of-fit test was conducted. The test yielded a Hosmer and Lemeshow chi-square value of 4.64 with 8 degrees of freedom and an associated p-value of 0.7951. Therefore, in line