STREAMLINED ANALYSIS WITH ADePT SOFTWARE 62258 Health Equity and Financial Protection Adam Wagstaff Marcel Bilger Zurab Sajaia Michael Lokshin Health Equity and Financial Protection STREAMLINED ANALYSIS WITH ADePT SOFTWARE Health Equity and Financial Protection Adam Wagstaff Marcel Bilger Zurab Sajaia Michael Lokshin © 2011 The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org All rights reserved 1 2 3 4 14 13 12 11 This volume is a product of the staff of the International Bank for Reconstruction and Development / The World Bank. The findings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgement on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development / The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www .copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. ISBN: 978-0-8213-8459-6 eISBN: 978-0-8213-8796-2 DOI: 10.1596/978-0-8213-8459-6 Cover photo: © Shehzad Noorani /World Bank (woman and child); © iStockphoto.com/Olga Altunina (background image) Cover design: Kim Vilov Library of Congress Cataloging-in-Publication Data has been requested. Contents Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xv Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xvii Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xix Chapter 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 PART I: Health Outcomes, Utilization, and Benefit Incidence Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Chapter 2 What the ADePT Health Outcomes Module Does . . . . . . . . . . . . . . . . .5 Measuring Inequality in Outcomes and Utilization . . . . . . . . . . . . . . . . . . . . .5 Basic Inequality Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Standardization for Demographic Factors* . . . . . . . . . . . . . . . . . . . . . . . . .7 Accounting for Inequality Aversion* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Trading Off the Average against Inequality* . . . . . . . . . . . . . . . . . . . . . . . .8 Explaining Inequalities and Measuring Inequity* . . . . . . . . . . . . . . . . . . . . . .8 v Contents Benefit Incidence Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9 Basic BIA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10 BIA under Alternative Assumptions* . . . . . . . . . . . . . . . . . . . . . . . . . . . .11 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12 Chapter 3 Data Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15 Household Identifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15 Living Standards Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15 Direct Approaches to Measuring Living Standards . . . . . . . . . . . . . . . . . .16 Indirect Approaches to Measuring Living Standards . . . . . . . . . . . . . . . .17 Health Outcome Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .17 Child Survival . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .17 Anthropometric Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18 Other Measures of Adult Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18 Health Utilization Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .20 Variables for Basic Tabulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21 Weights and Survey Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21 Determinants of Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21 Determinants of Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .22 Information on Utilization for Benefit Incidence Analysis . . . . . . . . . . . . . .22 Fees Paid to Public Providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23 NHA Aggregate Data on Subsidies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .24 Chapter 4 Example Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .27 Household Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28 Living Standards Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28 Health Outcome Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28 Health Utilization Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28 Variables for Basic Tabulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29 Weights and Survey Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29 Determinants of Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29 Determinants of Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29 Utilization Variables for Benefit Incidence Analysis . . . . . . . . . . . . . . . . . . .30 vi Contents Fees Paid to Public Providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30 NHA Aggregate Data on Subsidies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .31 Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .31 Chapter 5 How to Generate the Tables and Graphs . . . . . . . . . . . . . . . . . . . . . . .33 Main Tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .34 Determinants of Health or Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36 Benefit Incidence Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38 Chapter 6 Interpreting the Tables and Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . .41 Original Data Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .42 Basic Tabulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .43 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .43 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .43 Inequalities in Health Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .46 Concentration of Health Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .47 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .47 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48 Explaining Inequalities in Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .49 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .49 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .52 Decomposition of the Concentration Index . . . . . . . . . . . . . . . . . . . . . . . . . .54 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55 Inequalities in Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55 Explaining Inequalities in Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .57 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .57 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .57 vii Contents Use of Public Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59 Payments to Public Providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .60 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .60 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61 Health Care Subsidies: Cost Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . .62 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .62 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .64 Concentration of Public Health Services . . . . . . . . . . . . . . . . . . . . . . . . . . . .66 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .66 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .66 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .69 Chapter 7 Technical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71 Measuring Inequalities in Outcomes and Utilization . . . . . . . . . . . . . . . . . . .71 Note 1: The Concentration Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71 Note 2: The Concentration Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .72 Note 3: Sensitivity of the Concentration Index to the Living Standards Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .75 Note 4: Extended Concentration Index . . . . . . . . . . . . . . . . . . . . . . . . . . .77 Note 5: Achievement Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .78 Explaining Inequalities and Measuring Inequity . . . . . . . . . . . . . . . . . . . . . . .79 Note 6: Demographic Standardization of Health and Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .79 Note 7: Decomposition of the Concentration Index . . . . . . . . . . . . . . . . .82 Note 8: Distinguishing between Inequality and Inequity . . . . . . . . . . . . . .83 Benefit Incidence Analysis (BIA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .84 Note 9: Public Health Subsidy in Standard BIA . . . . . . . . . . . . . . . . . . . .84 Note 10: Public Health Subsidy with Proportional Cost Assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .86 Note 11: Public Health Subsidy with Linear Cost Assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .87 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .89 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .89 viii Contents PART II: Health Financing and Financial Protection . . . . . . . . . .93 Chapter 8 What the ADePT Health Financing Module Does . . . . . . . . . . . . . . . . .95 Financial Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .96 Catastrophic Health Spending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .96 Poverty and Health Spending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .98 Progressivity and Redistributive Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .98 Progressivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .99 Redistributive Effect* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .99 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .102 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .102 Chapter 9 Data Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .103 Ability to Pay (Consumption) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .103 Out-of-Pocket Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .104 Nonfood Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .105 Poverty Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .105 Prepayments for Health Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .106 NHA Data on Health Financing Mix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .107 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .108 Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .108 Chapter 10 Example Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .109 Financial Protection: Vietnam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .109 Ability to Pay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .109 Out-of-Pocket Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110 Nonfood Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110 Poverty Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110 Progressivity and Redistributive Effect: Egypt . . . . . . . . . . . . . . . . . . . . . . . .110 Ability to Pay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110 Out-of-Pocket Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .111 Prepayments for Health Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .111 NHA Data on Health Financing Mix . . . . . . . . . . . . . . . . . . . . . . . . . . . .111 Incidence Assumptions for Health Care Payments . . . . . . . . . . . . . . . . .112 ix Contents Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .113 Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .113 Chapter 11 How to Generate the Tables and Graphs . . . . . . . . . . . . . . . . . . . . . . .115 Financial Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116 Progressivity and Redistributive Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . .118 Chapter 12 Interpreting the Tables and Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . .121 Original Data Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .121 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .121 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .122 Basic Tabulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .123 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .123 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .123 Financial Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .124 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .124 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .125 Distribution-Sensitive Measures of Catastrophic Payments . . . . . . . . . . . . .127 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .127 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .128 Measures of Poverty Based on Consumption . . . . . . . . . . . . . . . . . . . . . . . . .129 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .129 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .130 Share of Household Budgets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .130 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .130 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .130 Health Payments and Household Consumption . . . . . . . . . . . . . . . . . . . . . .131 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .132 Progressivity and Redistributive Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .133 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .133 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .133 Progressivity of Health Financing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .134 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .134 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .136 x Contents Decomposition of Redistributive Effect of Health Financing . . . . . . . . . . . .137 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .137 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .138 Concentration Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .139 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .139 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .141 Distribution of Health Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .142 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .142 Interpreting the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143 Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143 Chapter 13 Technical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .145 Financial Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .145 Note 12: Measuring Incidence and Intensity of Catastrophic Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .145 Note 13: Distribution-Sensitive Measures of Catastrophic Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .147 Note 14: Threshold Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .148 Note 15: Limitations of the Catastrophic Payment Approach . . . . . . . .149 Note 16: Health Payments–Adjusted Poverty Measures . . . . . . . . . . . . .150 Note 17: Adjusting the Poverty Line . . . . . . . . . . . . . . . . . . . . . . . . . . . .152 Note 18: On the Impoverishing Effect of Health Payments . . . . . . . . . .153 Progressivity and Redistributive Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . .154 Note 19: Measuring Progressivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .154 Note 20: Progressivity of Overall Health Financing . . . . . . . . . . . . . . . .155 Note 21: Decomposing Redistributive Effect . . . . . . . . . . . . . . . . . . . . . .156 Note 22: Redistributive Effect and Economic Welfare . . . . . . . . . . . . . .158 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .158 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .158 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .161 Figures 2.1: Concentration Curve and Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 7.1: Weighting Scheme for Extended Concentration Index . . . . . . . . . . . .78 xi Contents 8.1: Health Payments Budget Share and Cumulative Percentage of Households Ranked by Decreasing Budget Share . . . . . . . . . . . . . . . . .97 8.2: Kakwani’s Progressivity Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .100 13.1: Health Payments Budget Share and Cumulative Percentage of Households Ranked by Decreasing Budget Share . . . . . . . . . . . . . . . .146 13.2: Pen’s Parade for Household Expenditure Gross and Net of Out-of-Pocket Health Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . .150 Graphs G1: Concentration Curves of Health Outcomes . . . . . . . . . . . . . . . . . . . . .48 G7a: Decomposition of the Concentration Index for Health Outcomes, Using OLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54 G3: Concentration Curves of Public Health Care Subsidies, Standard BIA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .67 G4: Concentration Curves of Public Health Care Subsidies, Proportional Cost Assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .68 G5: Concentration Curves of Public Health Care Subsidies, Linear Cost Assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .69 GF1: Health Payment Shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131 GF2: Effect of Health Payments on Pen’s Parade of the Household Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .132 GP1: Concentration Curves for Health Payments, Taxes . . . . . . . . . . . . . .140 GP2: Concentration Curves for Health Payments, Insurance, Out of Pocket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .141 GP3: Health Payment Shares by Quintiles . . . . . . . . . . . . . . . . . . . . . . . . . .142 Screenshots 5.1: Main Tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .34 5.2: Inequalities in Health or Utilization Tab . . . . . . . . . . . . . . . . . . . . . . . .36 5.3: Benefit Incidence Analysis Tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38 11.1: Financial Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116 11.2: Progressivity and Redistributive Effect . . . . . . . . . . . . . . . . . . . . . . . .118 Tables 2.1: Data Needed for Different Types of ADePT Health Outcome Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10 Original Data Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .42 xii Contents H2: Health Outcomes by Individual Characteristics . . . . . . . . . . . . . . . . . .44 H3: Health Inequality, Unstandardized . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 H6: Decomposition of the Concentration Index for Health Outcomes, Linear Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .49 H8a: Fitted Linear Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50 H8c: Elasticities, Linear Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51 H8e: Concentration Index of the Covariates . . . . . . . . . . . . . . . . . . . . . . . . .51 U3: Inequality in Health Care Utilization, Unstandardized . . . . . . . . . . . .56 U6: Decomposition of the Concentration Index for Utilization Values, Using OLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .58 S1: Utilization of Public Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59 S2: Payments to Public Providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61 S3: Health Care Subsidies, Constant Unit Cost Assumption . . . . . . . . . . .63 S4: Health Care Subsidies, Proportional Cost Assumption . . . . . . . . . . . . .63 S5: Health Care Subsidies, Constant Unit Subsidy Assumption . . . . . . . .64 9.1: Data Needed for Different Types of ADePT Health Financing Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .104 9.2: Common Incidence Assumptions for Prepayments in Progressivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .107 10.1: Financing Assumptions, Egypt Example . . . . . . . . . . . . . . . . . . . . . . .112 Original Data Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .122 1: Sources of Finance by Household Characteristics . . . . . . . . . . . . . . . .123 F1: Incidence and Intensity of Catastrophic Health Payments . . . . . . . . .125 F2: Incidence and Intensity of Catastrophic Health Payments, Using Nonfood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .126 F3: Distribution-Sensitive Catastrophic Payments Measures . . . . . . . . . . .128 F4: Distribution-Sensitive Catastrophic Payments Measures, Using Nonfood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .128 F5: Measures of Poverty Based on Consumption Gross and Net of Spending on Health Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .129 P1: Average Per Capita Health Finance . . . . . . . . . . . . . . . . . . . . . . . . . . .133 P2: Shares of Total Financing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .135 P3: Financing Budget Shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .135 P4: Decomposition of Redistributive Impact of Health Care Financing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .137 xiii Foreword World Bank researchers have a long tradition of developing and applying methods for the analysis of poverty and inequality, often working with col- laborators. And the Bank’s researchers have often tried hard to make their methods accessible to others, through “how-to” guides and training courses. In that tradition, this book is the first in a new series called Streamlined Analysis with ADePT Software. ADePT is an exciting new software tool developed by the Bank’s research department, the Development Research Group. ADePT automates the production of standardized tables and charts using a wide range of methods in distributional analysis, including some advanced methods that are technically demanding and not easily accessible to most potential users. This software makes these sophisticated methods accessible to analysts who have limited programming skills. (ADePT uses the statistical software package Stata but does not require that users know how to program in Stata, or even to have Stata installed on their comput- ers.) But we also hope that ADePT will be valuable to more technically inclined researchers too, by speeding up the production of results and by increasing their reliability and comparability. The present book provides a guide to ADePT’s two health modules: the first module covers inequality and equity in health, health care uti- lization, and subsidy incidence; the second, health financing and financial xv Foreword protection. It also provides introductions to the methods used by ADePT and a step-by-step guide to their implementation in the program. We hope you find this guide useful in your work. Please give us feedback on ADePT (see www.worldbank.org/adept) and this volume, as we wish to make them even more useful in the future. Martin Ravallion Director, Development Research Group The World Bank xvi Acknowledgments We are grateful to our peer reviewers Caryn Bredenkamp, Owen O’Donnell, and Ellen van de Poel for their excellent comments on the previous draft of the manuscript for this book. Their comments led to improvements not only in the manuscript but also in the ADePT software. Caryn and Ellen continued to provide invaluable feedback on ADePT afterwards, as did Sarah Bales and Leander Buisman. We are also grateful to the Bank’s Health, Nutrition and Population unit for financial support in the production of this book. xvii Abbreviations ADePT Automated DEC Poverty Tables BIA benefit incidence analysis BMI body mass index CHC commune health center CPI consumer price index DEC Development Economics (Vice Presidency at the World Bank) ID identification NHA National Health Account OLS ordinary least squares OOP out of pocket PPP purchasing power parity VHLSS Vietnam Household Living Standards Survey xix Chapter 1 Introduction ADePT is a software package that generates standardized tables and charts summarizing the results of distributional analyses of household survey data. Users input a Stata (or SPSS) data set, indicate which variables are which, and tell ADePT what tables and charts to produce; ADePT then outputs the results in a spreadsheet with one page for each requested table and chart. ADePT requires only limited knowledge of Stata and SPSS: users need to be able to prepare the data set, but do not need to know how to program Stata to undertake the often complex analysis that ADePT performs. ADePT frees up resources for data preparation, interpretation of results, and thinking about the policy implications of results. Users can easily assess the sensitivity of their results to the choice of assumptions and can repli- cate previous results in a straightforward way. ADePT also reduces the risk of programming errors and spurious variations in results that arise as a result of different ways of implementing methods computationally. ADePT Health is just one of several modules; other modules include Poverty, Inequality, Labor, Social Protection, and Gender. ADePT Health has two submodules: Health Outcomes and Health Financing. Together these modules cover a wealth of topics in the areas of health equity and financial protection. This manual is divided into two parts corresponding to each of these sub- modules. The following topics are covered: • Part 1, Health Outcomes: (a) measuring inequalities in outcomes and utilization (with and without standardization for need), (b) decom- posing the causes of health sector inequalities, and (c) analyzing 1 Health Equity and Financial Protection the incidence of government spending (that is, benefit incidence analysis). • Part 2, Health Financing: (a) financial protection, including cata- strophic payments and impoverishing payments, and (b) the progres- sivity and redistributive effect of health financing. Each part is divided into six chapters: • Chapters 2 and 8 explain what ADePT does in each area and pro- vide a brief introduction to the methods underlying ADePT. The methods are widely accepted in the literature and are outlined in more detail in Analyzing Health Equity Using Household Survey Data, by Owen O’Donnell, Eddy van Doorslaer, Adam Wagstaff, and Magnus Lindelow (O’Donnell and others 2008). This first sec- tion and all the other sections of this manual draw heavily on this book. • Chapters 3 and 9 explain how to prepare the data for ADePT. This is key to the successful use of ADePT, as the software has no data manipulation capability. • Chapters 4 and 10 guide users through example data sets, which are used in the worked examples in the sections that follow. • Chapters 5 and 11 show users how to generate the tables and charts that ADePT is capable of producing. Using a worked example with real data, the manual provides step-by-step instructions for using ADePT. • Chapters 6 and 12 walk users through interpretation of the tables and charts produced by ADePT. Again, this is done through a worked example using real data. • Chapters 7 and 13 contain technical notes explaining in more detail the methods used in the program. Reference O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. 2 PART I Health Outcomes, Utilization, and Benefit Incidence Analysis Chapter 2 What the ADePT Health Outcomes Module Does The Health Outcomes module of ADePT Health allows users to analyze inequalities in health, health care utilization, and health subsidies, by income or any continuous (though not necessarily cardinal) measure of living standards or socioeconomic status. In what follows, “income” is often used as shorthand for whatever measure of living standards is being used. ADePT allows analysts to see whether inequalities, in, for example, the use of health care between the poor and rich, have narrowed over time or are smaller in one country than another. Users can also analyze whether (and how far) subsidies to the health sector disproportionately benefit the better off or the poor—that is, benefit incidence analysis, or BIA. ADePT can do quite simple analysis as well as more sophisticated analy- sis. The more sophisticated features of ADePT are indicated below with an asterisk. Users not familiar with the literature may wish to focus initially on the sections without an asterisk. Except where stated, the summary in this chapter relies on O’Donnell and others (2008). Measuring Inequality in Outcomes and Utilization ADePT allows users to analyze differences in health outcomes or health care utilization across any subpopulation. However, the software’s strength lies in its ability to analyze inequalities in health outcomes and utilization by income or by some other measure of living standards. 5 Health Equity and Financial Protection: Part I Basic Inequality Analysis In addition to producing tables showing the mean values by income group (or any grouping of living standards), ADePT produces a summary inequal- ity statistic, known as the concentration index, which shows the size of inequalities in health and health care utilization between the poor and bet- ter off (see figure 2.1).1 A large absolute value indicates a high degree of inequality. The concentration index derives from the concentration curve, which is graphed by ADePT. It is obtained by ranking individuals by a measure of living standards and plotting on the x axis the cumulative per- centage of individuals ranked in ascending order of standards of living and on the y axis the cumulative percentage of total health care utilization, health or ill health, or whatever variable whose distribution is being inves- tigated. The y axis could measure, for example, the percentage of people reporting an inpatient episode. This exercise traces out the concentration curve. If hospital admissions are not related to living standards, the con- centration curve will be a straight line running from the bottom left corner to the top right corner; this is the line of equality. If the better off have Figure 2.1: Concentration Curve and Index concentration curve cumulative % of the health variable half the concentration index (in absolute terms) line of equality cumulative % of population, ranked from poorest to richest Source: Authors. 6 Chapter 2: What the ADePT Health Outcomes Module Does higher inpatient admission rates than the poor, the concentration curve will lie below the line of equality. It will lie above the line of equality in the opposite case. Twice the area between the concentration curve and the line of equality is the concentration index. By convention, it is positive when the concen- tration curve lies below the line of equality, indicating that the variable of interest is lower among the poor and has a maximum value of 1. It is neg- ative when the concentration curve lies above the line of equality, indicat- ing that the outcome variable is higher among the poor and has a minimum value of 1. Standardization for Demographic Factors* ADePT allows users to request that inequalities in health and utilization be adjusted to reflect differences across income groups in variables that are justified determinants of health or utilization.2 Utilization might be higher among the poor, for example, in part because the poor have greater medical needs, and greater medical needs translate, as policy makers hope they do, into higher levels of utilization; standardization provides a way to remove this justified inequality from the measured inequality. Similarly, health may be worse among the poor because the poor are, on average, older than the better off, and people’s health inevitably worsens with age; standardization provides a way to remove this inescapable component of health inequality.3 ADePT implements both the direct and indirect methods of standardi- zation and allows users to decide whether to include only justified influ- ences in the standardization or both justified and unjustified influences, albeit standardizing just for the former. Best practice is to include both sets of variables.4 Accounting for Inequality Aversion* The concentration index embodies a specific set of attitudes toward inequality.5 ADePT reports values of a generalized or “extended” concen- tration index with different values of an inequality-aversion parameter. The higher the value of the parameter, the greater is the degree of aver- sion to inequality. The normal concentration index has a value of 2 for the inequality-aversion parameter. 7 Health Equity and Financial Protection: Part I Trading Off the Average against Inequality* Policy makers are typically concerned not just about health sector inequali- ties but also about the level of the variable in question.6 Obviously, they would like both low inequality and better health. But at the margin they are likely to be willing to trade off one against the other, accepting a little more inequality in exchange for a dramatic improvement in the average. ADePT reports values of the health achievement index that trade off average health against inequality. Specifically, it is equal to the mean of the distribution mul- tiplied by the complement of the concentration index. It therefore reflects the average of the distribution and the concentration index. If there is no inequality, so that the concentration index is 0, the achievement index is equal to the average. If outcomes are concentrated among the poor, so that the concentration index is negative, the achievement index exceeds the average. For example, if child mortality is higher among the poor, “achieve- ment” (in this case dis-achievement!) is higher than average child mortality. ADePT also reports values of an “extended” achievement index, corre- sponding to the extended concentration index. Explaining Inequalities and Measuring Inequity* In addition to measuring inequalities in health and health care utilization across the income distribution, ADePT can also be used to explain inequal- ities in terms of inequalities in the underlying determinants.7 For example, part of the observed pro-poor inequality in utilization might be because the elderly are, on average, worse off than the nonelderly and use more services. Part of it might be due to the fact that insurance is higher among the better off and the insured use more services. ADePT allows users to see how far uti- lization inequalities are due to the concentration of the elderly among the better off rather than to the concentration of the insured among the better off. ADePT allows any number of determinants of utilization (or health) to be included and calculates the portion of inequality that is due to inequality in each determinant. Here’s how the decomposition works. Suppose the variable of interest can be expressed as a linear function of a set of determinants. Then the concentra- tion index of the variable of interest is a linear function of the concentration indexes of the determinants, where the weight on each determinant is equal to 8 Chapter 2: What the ADePT Health Outcomes Module Does the regression coefficient of the determinant in the regression of the variable of interest on the full set of determinants, times the mean of the determinant, divided by the mean of the variable of interest. So the bigger the effect of the determinant on the variable of interest, the bigger the mean of the determi- nant, and the more unequally distributed the determinant, the more the (inequality in the) determinant contributes to the inequality in the variable of interest. ADePT reports how much of the inequality in the variable of interest can be attributed to inequalities in each of the determinants. There is a link between the decomposition approach, the measurement of inequity, and the indirect standardization. Suppose we divide the deter- minants into (a) justified influences on the variable of interest (for example, health if the outcome of interest is utilization) and (b) unjustified determi- nants (for example, insurance). It turns out that the concentration index minus the combined contribution in the decomposition of the standardizing variables is equal to the concentration index for the indirectly standardized values of the variable of interest. And, in the case of utilization, the differ- ence between the concentration index for utilization and the concentration index for the indirectly standardized values of utilization is equal to one of the two widely used indexes of inequity, that is, a measure of the amount of unjustified inequality. So, in a single sweep, the decomposition provides a way not just of explaining inequality but also of measuring inequity. Actually, the decomposition approach gives analysts a good deal of flexibility in choosing what to include among the justified determinants and what to include among the unjustified determinants. For example, people get less healthy as they age, suggesting that age might be a justified influence in an analysis of health inequalities. However, the speed at which people’s health deteriorates as they age is not fixed and can be affected by policy makers. Perhaps, therefore, it ought not to be viewed as a justified or inescapable influence on health in an analysis of the causes of health inequality. The attractive feature of the decomposition is that analysts can “sit on the fence” completely and simply report the contributions to inequality coming from each of the determinants, letting readers decide where to draw the line. Benefit Incidence Analysis The final type of analysis that the Health Outcomes module of ADePT allows users to undertake is benefit incidence analysis.8 This involves 9 Health Equity and Financial Protection: Part I analyzing the distribution of government health sector subsidies across the income distribution.9 Basic BIA The problem facing analysts undertaking a BIA is that the amount the government spends providing care to a specific individual is not observed, and therefore assumptions have to be made to derive subsidies at the household level.10 The least demanding assumption—in terms of data— is that unit subsidies are constant. In this case, as table 2.1 shows, the analyst simply requires data on utilization of different types of public sec- tor health care providers (for example, health centers, outpatient care in hospitals, and inpatient care in hospitals) and the amount the govern- ment spends on each type of service. By grossing up the average amounts of utilization to the population level, ADePT estimates the total volume of utilization for each type of service. This is divided into the amount the government spends on each type of service to get the unit subsidy for each type of service. This is then assumed to be constant within a given type of service. Table 2.1: Data Needed for Different Types of ADePT Health Outcome Analysis Need National Demographic indicators Health Living Health variables and Health and other Account Fees paid Topic and standards outcome other health utilization utilization data on to public analysis indicator variable(s) determinants variable(s) determinants subsidies providers Inequalities in health No standardization ✓ ✓ Standardization and decomposition* ✓ ✓ ✓ Inequalities in utilization No standardization ✓ ✓ Standardization and decomposition* ✓ ✓ ✓ Benefit incidence analysis Constant unit sub- sidy assumption ✓ ✓ ✓ Other assumptions* ✓ ✓ ✓ ✓ Source: Authors. Note: * A more advanced and more data-demanding type of analysis. 10 Chapter 2: What the ADePT Health Outcomes Module Does BIA under Alternative Assumptions* Other assumptions also require data on the amount that different house- holds (or individuals) pay in fees for the visits to public sector providers that are recorded in the household data. The other assumptions are that unit costs are constant and proportional to the fees paid. If data are avail- able on fees paid to public providers, ADePT reports BIA estimates for these cases too. ADePT reports the average subsidy (by type of service and for all services combined) for each quintile or decile. It produces separate tables for each assumption. ADePT also reports the concentration index inequality statis- tic showing, on balance, how pro-poor or pro-rich subsidies are and graphs subsidy concentration curves for different categories of utilization. Notes 1. For further details, see technical notes 1–3 in chapter 7; O’Donnell and others (2008, chs. 7, 8). 2. For further details, see technical note 6 in chapter 7; O’Donnell and others (2008, 60–64). 3. See Gravelle (2003) and Fleurbaey and Schokkaert (2009) for discus- sions on which sources of inequality in health and health care utiliza- tion should be considered justified or fair. 4. See, for example, Gravelle (2003); van Doorslaer, Koolman, and Jones (2004); Fleurbaey and Schokkaert (2009). 5. For further details, see technical note 4 in chapter 7; O’Donnell and others (2008, ch. 9). 6. For further details, see technical note 5 in chapter 7; O’Donnell and others (2008, ch. 9). 7. For further details, see technical notes 6 and 7 in chapter 7; O’Donnell and others (2008, chs. 13, 15). 8. For further details, see technical notes 9–11 in chapter 7; O’Donnell and others (2008, ch. 14); Wagstaff (2010). 9. For further details on BIA, see technical notes 9–11 in chapter 13; O’Donnell and others (2008, ch. 14); Wagstaff (2010). For a critique of BIA, see van de Walle (1998). Empirical studies include Hammer, Nabi, and Cercone (1995); van de Walle (1995); O’Donnell and others 11 Health Equity and Financial Protection: Part I (2007). Marginal BIA tries to assess how different income groups bene- fit from an expansion of the budget (Lanjouw and Ravallion 1999). A pro-rich distribution of average benefits need not translate into a pro- rich distribution of marginal benefits, since additional spending may dis- proportionately benefit the poor rather than the rich. ADePT does not currently implement marginal BIA. However, analysts can easily repeat the same BIA on data sets from multiple years or regions within the country and see how incidence changes as budgets change. 10. Wagstaff (2010) extends the analysis in O’Donnell and others (2008). References Fleurbaey, M., and E. Schokkaert. 2009. “Unfair Inequalities in Health and Health Care.” Journal of Health Economics 28 (1): 73–90. Gravelle, H. 2003. “Measuring Income-Related Inequality in Health: Standardisation and the Partial Concentration Index.” Health Economics 12 (10): 803–19. Hammer, J., I. B. Nabi, and J. Cercone. 1995. “Distributional Effects of Social Sector Expenditures in Malaysia 1974–89.” In Public Spending and the Poor: Theory and Evidence, ed. D. van de Walle and K. Nead. Baltimore, MD: Johns Hopkins University Press. Lanjouw, P., and M. Ravallion. 1999. “Benefit Incidence, Public Spending Reforms, and the Timing of Program Capture.” World Bank Economic Review 13 (2): 257–73. O’Donnell, O., E. van Doorslaer, R. P. Rannan-Eliya, A. Somanathan, S. R. Adhikari, D. Harbianto, C. G. Garg, P. Hanvoravongchai, M. N. Huq, A. Karan, G. M. Leung, C. W. Ng, B. R. Pande, K. Tin, L. Trisnantoro, C. Vasavid, Y. Zhang, and Y. Zhao. 2007. “The Incidence of Public Spending on Healthcare: Comparative Evidence from Asia.” World Bank Economic Review 21 (1): 93–123. O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. van de Walle, D. 1995. “The Distribution of Subsidies through Public Health Services in Indonesia.” In Public Spending and the Poor: Theory and Evidence, ed. D. van de Walle and K. Nead. Baltimore, MD: Johns Hopkins University Press. 12 Chapter 2: What the ADePT Health Outcomes Module Does ———. 1998. “Assessing the Welfare Impacts of Public Spending.” World Development 26 (3): 365–79. van Doorslaer, E., X. Koolman, and A. M. Jones. 2004. “Explaining Income- Related Inequalities in Doctor Utilization in Europe.” Health Economics 13 (7): 629–47. Wagstaff, A. 2011. “Benefit Incidence Analysis: Are Government Health Expenditures More Pro-Rich Than We Think?” Health Economics, 20: n/a. DOI: 10.1002/hec.1727. 13 Chapter 3 Data Preparation ADePT has no data manipulation capability. Hence, the data need to be prepared before they are loaded into ADePT. This chapter outlines the data needed by ADePT for different types of analysis. The data required for the various analyses that ADePT can do are sum- marized in table 2.1. An alternative way of reading the table is to see what analyses are feasible given the available data. ADePT works out what tables and graphs can be produced given the data fields completed: tables and graphs that are feasible are shown in black; those that are not feasible are shown in gray. As the level of sophistication of the analysis increases, so do the data requirements. The more sophisticated analyses—hence more demanding of data—are marked with an asterisk in table 2.1. Household Identifier ADePT users must specify a household identification variable, or series of variables, that uniquely identifies the household in the data set. Living Standards Indicators ADePT analyzes the distribution of health outcomes, health care utilization, or subsidies across people with different standards of living. As table 2.1 shows, all ADePT tables and charts require a living standards measure. 15 Health Equity and Financial Protection: Part I This raises the question of how to measure living standards. One approach is to use “direct” measures, such as income, expenditure, or con- sumption. The alternative is to use an indirect or “proxy” measure, making the best use of available data. Direct Approaches to Measuring Living Standards The most direct (and popular) measures of living standards are income and consumption. Income refers to the earnings from productive activi- ties and current transfers. It comprises claims on goods and services by individuals or households. In other words, income permits people to obtain goods and services. Consumption, by contrast, refers to resources actually consumed. Although many components of consumption are measured by looking at expenditures, there are important differences between consumption and expenditure. First, expenditure excludes consumption that is not based on market transactions. Given the importance of home production in many developing countries, this can be an important distinction. Second, expenditure refers to the purchase of a particular good or service. However, the good or service may not be immediately consumed, or at least it may have lasting benefits. This is the case, for example, with consumer durables. Ideally, in this case, consumption should capture the benefits that come from the use of the good, rather than the value of the purchase itself. There is a long-standing and vigorous debate about which is the better measure of standards of living—consumption or income. For developing countries, a strong case can be made for preferring consumption over income, based on both conceptual and practical considerations. Measured income often diverges substantially from measured consumption, in part due to concep- tual differences between them—it is possible to save from income and to finance consumption from borrowing. Income data are, moreover, often of poor quality, if available at all. If consumption data are used as a measure of living standards, it is cus- tomary to divide total household consumption (or income) by the number of household members (or the number of equivalent household members) to get a more accurate measure of the household’s standard of living. The per capita adjustment is quite common in empirical work in this area. 16 Chapter 3: Data Preparation Indirect Approaches to Measuring Living Standards Many surveys do not include data on either income or consumption. Sometimes, consumption data are available, but they are not very high qual- ity. In such situations, a popular strategy is to use principal components analysis (or some other statistical method) to construct an index of “wealth” from information on household ownership of durable goods and housing characteristics. This provides a ranking variable. In other words, it is possi- ble to say whether a household is wealthier than another household, but not how much wealthier. For the analyses done by the Health Outcomes module of ADePT, this is not a limitation. (It is a limitation in the Health Financing module.) Finally, because a wealth index is not a cardinal meas- ure of living standards, ADePT users should not try to adjust the wealth index for household size. Health Outcome Variables A variety of health outcome variables can be used in ADePT to analyze inequalities in health.1 These can be grouped under (a) child survival, (b) anthropometric indicators (which apply to both children and adults), and (c) other measures of adult health. Child Survival ADePT can be used to analyze inequalities in infant- and under-five mor- tality by creating a dummy variable taking a value of 1 if the child died before his or her first (or fifth) birthday.2 To get around the censoring prob- lem (that is, some children who have not yet reached their first birthday might never do so), ADePT users might want to drop from the sample chil- dren who have not yet reached their first (or fifth) birthday. ADePT can also explain inequalities in child survival using the decomposition method. The basic output in the ADePT decomposition is based on the use of a stan- dard ordinary least squares (OLS) regression model. However, ADePT will detect if the outcome variable is binary (for example, whether the child has died) and will produce a second set of output based on the results from a pro- bit model and a linear approximation of the decomposition with marginal effects evaluated at sample means.3 17 Health Equity and Financial Protection: Part I Anthropometric Indicators Anthropometric indicators capture malnutrition.4 Raw anthropometric data on weight, age, and so forth can be turned into more meaningful indi- cators by standardizing them on a reference population. Common refer- enced variables include weight-for-age, height-for-age, underweight, body mass index (BMI), and mid-upper arm circumference. The raw data need to be converted before the data are loaded into ADePT: this can be done in Stata using the zanthro command. Anthropometric indicators are sometimes dummy variables, such as underweight or not. Sometimes they are continuous variables, such as weight-for-age, which is a z score, or the child’s percentile in the refer- ence distribution. Both dummy and continuous variables can be used in ADePT to measure malnutrition inequalities and to decompose the causes of inequality. Continuous variables such as weight-for-age and BMI lend themselves naturally to linear decomposition. Inequalities in dummy vari- ables (such as underweight) can also be decomposed. The initial output from ADePT is based on a standard OLS model, but ADePT detects whether the outcome variable is a binary variable and produces a second set of results based on the output of a probit model and a linear approximation of the decomposition with marginal effects evaluated at sample means. Other Measures of Adult Health The measurement of adult health is more complex than the measurement of child survival and the measurement of malnutrition. Adult health measures differ along several dimensions. One is whether the health data are self- perceived (the occurrence of an illness during a specific time period) or observed (blood pressure). Another is whether a measure reflects a med- ical concept of health (the presence of a chronic condition), a functional concept (impairment in ability to perform everyday activities), or a subjective concept (answers to the question, how do you rate your health?). Health variables also vary in terms of how they are measured. Some are continuous variables (the number of days off work during the past four weeks), some are binary variables (the presence of chronic illness), and some are multiple-category variables. The last cannot simply be scored 1, 2, 3, and so forth because the true scale will not necessarily be equidistant between categories. O’Donnell and others (2008) review various options for 18 Chapter 3: Data Preparation multiple-category variables. Typically these require ADePT users to assign values to the categories using one of the suggested options before loading the data into ADePT. Three points not made in O’Donnell and others (2008) are also worth making: • There has been some debate in the recent literature about whether measures of health inequality like the concentration index need “cor- recting” when the variable in question is bounded, not cardinal, or both.5 Currently, ADePT does not offer any correction. However, all corrections proposed in the literature can be made ex post in the ADePT Excel output file. • There has been some discussion about whether measures of inequali- ties in self-perceived health status are biased because the poor and better off have different cutoff points between categories of health status. For instance, at equal latent (that is, unobservable) health status, the poor might report being in very good health, whereas the better off might report being in only fair health. One approach is to anchor the cutoff points using responses to anchoring vignettes in which everyone is asked to rate the health of a hypothetical indi- vidual with specific health problems; the results (for three develop- ing countries) do not suggest that inequalities are much affected by reporting bias.6 ADePT does not allow such an adjustment to be undertaken in the program; however, users could apply the vignette anchoring methodology beforehand and load into ADePT the pre- dicted latent health scores adjusted for shifts in cutoff points. • Finally, many measures of health status capture ill health over a spe- cific period of time. For example, a classic question asks whether peo- ple were ill during the previous month. There is some evidence that the recall period chosen has differential effects on reporting by the poor and rich. A recent study finds a steeper income gradient in self- reported illness when a weekly recall period is used than when a monthly recall period is used.7 The authors argue that the poor for- get illness that becomes accepted as part of their normal life. As they put it, “Poor people are ill for large fractions of the year with ailments that are short term and, apparently, easily forgotten. Richer people, who do not forget as easily, do not suffer from acute illnesses nearly as much as the poor.” This has implications for the 19 Health Equity and Financial Protection: Part I choice of recall period when designing a survey and for the interpre- tation of results when using a survey with a longer recall period. Health Utilization Variables Health care utilization variables are required for an analysis of inequalities and inequities in health care utilization, but also for a benefit incidence analysis (BIA). Health utilization is generally easier to measure than health status, at least on the face of it. Common examples are whether or not some- one used a particular type of provider during the previous month, the num- ber of outpatient consultations during the past month, and the number of inpatient days over the previous year. Two aspects of such variables are worth noting: • The first is that these variables—like other commonly used utiliza- tion measures—are either binary or count variables. With binary variables, the issues concerning the desirability of “correcting” the concentration index are worth keeping in mind. Issues also arise in the decomposition exercise. The basic decomposition tables that ADePT produces are based on a standard OLS regression. However, ADePT checks whether the utilization measure is a binary or a count variable and, if so, also reports a further set of decomposition results based on a probit or Poisson model as appropriate, using a linear approximation of the decomposition with partial effects evaluated at the sample means.8 • The second point worth making is that utilization is always measured over a specific period of time. As with the reporting of ill health, there is some evidence that the recall period differentially affects reporting by the poor and rich, with doctor visits declining with income when measured using a weekly recall period but increasing with income when using a monthly recall period.9 This has implica- tions for the choice of recall period when designing a survey and for the interpretation of results when using a survey with a longer recall period. When ADePT users undertake a BIA, the measures of utilization should cover the utilization of different types of public facilities (and private facilities 20 Chapter 3: Data Preparation if they are subsidized through supply-side subsidies). Examples of obvious variables are the number of outpatient visits at health centers, the number of outpatient visits at hospitals, and the number of inpatient admissions. Variables for Basic Tabulations Optionally, users can specify variables beyond the living standards variable by which ADePT should tabulate health outcomes or utilization. These can be individual-level variables (age, gender, education, employment status, and an additional variable chosen from the data set) or household-level variables (urban, region, and an additional variable selected from the data set). ADePT tabulates mean values of health or utilization by categories of whichever variables are selected. Weights and Survey Settings If appropriate, weights should be specified that make the results representa- tive at the population level. This should be the household weight. ADePT will automatically multiply it by household size to make the figures repre- sentative at the population level. Determinants of Health Demographic variables and other health determinants are required if ADePT users want to standardize inequalities in health for demographic dif- ferences across quintiles of the living standards variable or to decompose inequalities in health into their causes.10 The demographic variables that are typically used in these analyses are age and gender. These are required for any standardization and decomposi- tion analysis. Other health determinants are also required for standardization and decomposition. Which other health determinants (or control variables) are included will vary according to the application, but as many as possi- ble should be included, especially those that are likely to be correlated with demographic variables. ADePT users will likely want to include at a 21 Health Equity and Financial Protection: Part I minimum the living standards indicator among these variables. It could be included as a single variable or as a vector of variables capturing quintiles of the living standards indicator. (This quintile variable needs to be con- structed before the data are loaded into ADePT.) Education (of the mother and possibly of the father too) is a common variable to include among deter- minants of child health; other obvious candidates include access to safe water and hygienic sanitation. ADePT users would do well to consult the previous literature on the determinants of the health outcome indicator(s) whose inequality is being analyzed. Determinants of Utilization Indicators of need and other determinants of utilization are required (a) if the goal is to standardize inequalities in utilization for differences in need across quintiles so as to estimate inequity or (b) if the goal is to decompose inequalities in utilization into their causes.11 The usual indicators of need include demographic variables (age and gender) and measures of health status. These are required for any standard- ization and decomposition analysis. Other (non-need) determinants of utilization are also required for stan- dardization and decomposition. Which non-need determinants (or control variables) are included will vary according to the application, but as many as possible should be included, especially those that are likely to be correlated with need variables. ADePT users will likely want to include at a minimum the living standards indicator, either as a single variable or as a vector of vari- ables capturing quintiles of the living standards indicator. (This quintile variable needs to be constructed before the data are loaded into ADePT.) Other candidates for non-need determinants of utilization include insurance status and distance to the closest health facility, among others. Information on Utilization for Benefit Incidence Analysis A BIA requires information from the household survey on the quantities used of different types of services, for example, ambulatory visits to a first- level provider, ambulatory visits to a hospital, inpatient admissions to a hos- pital, and so forth. These should all refer to the same time period. 22 Chapter 3: Data Preparation Fees Paid to Public Providers A BIA that does not rest exclusively on the assumption of a constant unit subsidy requires survey data on the amounts that patients pay to government health facilities. Ideally, this would be available by type of service, such as the amount paid out-of-pocket for each ambulatory visit to a first-level provider. ADePT is, however, able to handle the case where only total out-of-pocket payments to all public facilities combined are recorded. ADePT always assumes that the out-of-pocket payments recorded in the survey are accurate, even if, when grossed up to a national level, they do not match those recorded as income by government providers. NHA Aggregate Data on Subsidies A BIA requires National Health Account (NHA)—and potentially other— data on subsidies for each type of service for which utilization data are avail- able. This NHA table typically has the title “Health Expenditures by Health Financing Agencies and Health Activities.” It shows the amount spent by government agencies (sometimes broken down by level of government and including any social health insurance agency) on inpatient care, outpatient care, and so forth. Notes 1. See O’Donnell and others (2008) for a more comprehensive discussion. 2. For a descriptive analysis of inequalities in child survival, see, for exam- ple, Wagstaff (2000). For a decomposition of inequalities in child sur- vival, see Hosseinpoor and others (2006). 3. See O’Donnell and others (2008, 182). If marginal effects are evaluated at values other than the sample mean, different results will emerge. The OLS decomposition, by contrast, is not vulnerable to this. 4. For a descriptive analysis of inequalities in malnutrition, see, for exam- ple, Wagstaff and Watanabe (2000). For a decomposition of inequalities in malnutrition, see Wagstaff, van Doorslaer, and Watanabe (2003). 5. See Wagstaff (2005, 2009); Erreygers (2009). 6. See, for example, d’Uva, O’Donnell, and van Doorslaer (2008). 7. See Das, Hammer, and Sánchez-Páramo (2010). 23 Health Equity and Financial Protection: Part I 8. See O’Donnell and others (2008, 182). 9. See Das, Hammer, and Sánchez-Páramo (2010). 10. In principle, the standardization could be done without the nondemo- graphic determinants, but it is not considered good practice because of the risk of omitted-variable bias. 11. In principle, the standardization could be done without the nondemo- graphic determinants, but it is not considered good practice because of the risk of omitted-variable bias. References Das, J., J. Hammer, and C. Sánchez-Páramo. 2010. Remembrance of Things Past: The Impact of Recall Periods on Reported Morbidity and Health Seeking Behavior. Washington, DC: World Bank. d’Uva, T. B., O. O’Donnell, and E. van Doorslaer. 2008. “Differential Health Reporting by Education Level and Its Impact on the Measurement of Health Inequalities among Older Europeans.” International Journal of Epidemiology 37 (6): 1375–83. Erreygers, G. 2009. “Correcting the Concentration Index.” Journal of Health Economics 28 (2): 504–15. Hosseinpoor, A. R., E. van Doorslaer, N. Speybroeck, M. Naghavi, K. Mohammad, R. Majdzadeh, B. Delavar, H. Jamshidi, and J. Vega. 2006. “Decomposing Socioeconomic Inequality in Infant Mortality in Iran.” International Journal of Epidemiology 35 (5): 1211–19. O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. Wagstaff, A. 2000. “Socioeconomic Inequalities in Child Mortality: Comparisons across Nine Developing Countries.” Bulletin of the World Health Organization 78 (1): 19–29. ———. 2005. “The Bounds of the Concentration Index When the Variable of Interest Is Binary, with an Application to Immunization Inequality.” Health Economics 14 (4): 429–32. ———. 2009. “Correcting the Concentration Index: A Comment.” Journal of Health Economics 28 (2): 521–24. Wagstaff, A., E. van Doorslaer, and N. Watanabe. 2003. “On Decomposing the Causes of Health Sector Inequalities with an Application to 24 Chapter 3: Data Preparation Malnutrition Inequalities in Vietnam.” Journal of Econometrics 112 (1): 207–23. Wagstaff, A., and N. Watanabe. 2000. “Socioeconomic Inequalities in Child Malnutrition in the Developing World.” Policy Research Working Paper 2434, World Bank, Washington, DC. 25 Chapter 4 Example Data Set The data sets used in all the tables and graphs in chapters 4–6 of the man- ual were produced using the 2006 Vietnam Household Living Standards Survey (VHLSS). Vietnam implemented two broad household surveys in 1993 and 1998 and, since 2002, has implemented surveys every second year, albeit with a slimmed-down health module compared to the 1993 and 1998 surveys. The attraction of the 2006 survey is that it included an extended health module. Questions were asked on activities of daily living, self-assessed health, days lost due to illness, and other issues. The survey also asked about the use of different types of health provider (commune health centers, gener- al hospitals, private facilities), with a distinction between outpatient and inpatient care. Information was also collected on the fees paid to public and private providers and on insurance coverage—Vietnam has a social health insurance scheme that covers formal sector workers out of contri- butions, civil servants and “people of merit” out of general revenues, and the poor (and other disadvantaged groups) also out of general revenues. Consumption in the survey is measured using an extensive set of modules capturing home production of food as well as market purchases of goods and services. This chapter describes the variables used to generate the tables and graphs. ADePT produces basic descriptive statistics in an original data report, an example of which is presented in chapter 6. 27 Health Equity and Financial Protection: Part I Household Identification ADePT asks first about the household identifier in the data set. In the VHLSS, several variables are needed to uniquely identify the households. These identify the province (tinh), the district (huyen), the commune (xa), the village (diaban), and finally the family (hoso). Overall, the sample con- tains 39,071 individuals living in 9,189 households. Living Standards Indicators Living standards are measured by per capita household consumption (pcexp) measured in thousands of Vietnamese dong. This variable captures expendi- tures outside the household as well as home production of foodstuffs and the use value of consumer durables. The procedure follows previous multipur- pose surveys in Vietnam.1 Health Outcome Variables The VHLSS includes various measures of adult health status. We use nine variables of impairment in activities of daily living with respect to sight (adleyesight), hearing (adlhearing), memory (adlmemory), concentration (adl- concent), understanding (adlunderstand), walking in general (adlwalk), climb- ing stairs (adlstairs), walking 400 meters (adl400m), and climbing 10 steps (adl10steps). Other health status variables include whether the individual got sick or injured during the last four weeks (illinj4wk) and last 12 months (illinj12m), the number of days spent in bed due to health problems during the last 12 months (beddays12m), and the number of work or school days missed due to illness or injury during the last 12 months (offwkdays12m). Health Utilization Variables Health care utilization is measured by the number of outpatient visits over the previous 12 months to a general hospital (opvis_ghosp), poly- clinic (opvis_poly), commune health center (opvis_chc), village clinic (opvis_vill), private facility (opvis_privfacil), traditional healer (opvis_tradherb), and other providers (opvis_othfacil). There is also information on the 28 Chapter 4: Example Data Set number of inpatient admissions over the same period to each of the afore- mentioned providers (ipvis_ghosp,. . .,ipvis_othfacil) and on the total fees paid by the individual for each of these health care utilization categories (opexp_ghosp,. . .,ipexp_othfacil). Variables for Basic Tabulations Other individual characteristics are also available, allowing us to analyze the health outcome and utilization variables conditional on them (in tables H2 and U2). These are individual age (age), a binary variable indicating gender (male), and a six-category education level (educlev). Weights and Survey Settings The survey design is quite complex, and a detailed description can be found in the basic information document, which is available on the World Bank’s Living Standards Measurement Study website.2 In order to get nationally representative statistics, we use the weight variable (wt9), which is provided along with the data collected. Determinants of Health The examples below use age and gender as the demographic variables and per capita household consumption as the control variable. Determinants of Utilization The indicators of need include age and gender as well as the health indica- tors mentioned above. In the decompositions seeking to explain inequality, two non-need variables are included. The first is health insurance. This is measured through three binary variables indicating whether the individual is covered by health insurance for the poor (hi_poor), mandatory health insurance (hi_comp), or voluntary health insurance (hi_vol). The second non-need variable used in the decomposition analysis is government per 29 Health Equity and Financial Protection: Part I capita health expenditure. This is measured at the provincial level (ghe_cap) and has been merged into the data set from provincial government expen- diture files. Utilization Variables for Benefit Incidence Analysis For the benefit incidence analysis (BIA), the following utilization variables are used: outpatient visits to a commune health center (CHC), polyclinic, and general hospital and inpatient admissions to a general hospital. Fees Paid to Public Providers For each outpatient visit and inpatient admission, the VHLSS records the out-of-pocket payments associated with the visit. The recorded data— unlike the official data on fees kept by the ministry—include informal pay- ments and transport costs: respondents in the VHLSS are asked to include “payments for medical service and treatment,” but also “other related costs (for example, bonus for doctors, transport).” Unfortunately, there is no way to exclude transport costs from the household data. NHA Aggregate Data on Subsidies Government spending (that is, subsidies) on public facilities is in millions of dong and relates to 2005, the closest year to 2006 for which detailed National Health Account (NHA) data are available. The data are taken from NHA table 2.3 The government spending figures reflect spending by the health ministry and the central government, by health departments at the provincial and district levels, as well as by the social health insurance agency, Vietnam Social Security. The table does not break government spending down exactly by type of provider. The assumption is that spending on “traditional medicine” is all at the hospital level. Government spending on inpatient care is taken from the figures labeled “inpatient treatment” and “outpatient treatment” under the “traditional medicine” heading—that is, items HD 1.1.1 and HD 1.1.2, respectively. Spending at the CHC and poly- clinic level is taken from the heading “primary health care and school 30 Chapter 4: Example Data Set health care” under “health prevention and public health”—that is, item HD 1.2.4. This is the only item in the NHA where spending at the commune level is recorded, but it is possible that some of the polyclinic spending is recorded at a higher level. It is assumed that 75 percent of the “primary health care and school health care” spending was incurred in health facili- ties and the rest was incurred in schools. Notes 1. At the time of writing, the basic information document for the 2006 sur- vey was not available. However, the survey is similar to the 2002 and 2004 VHLSS, for which a (combined) basic information document is available at http://www.worldbank.org/lsms. 2. http://www.worldbank.org/lsms. 3. Vietnam Ministry of Health (2008, 135). See http://www.wpro.who.int/ NR/rdonlyres/69A8189E-92ED-4883-ABEC-B90CBE894522/0/ National_Health_Account.pdf. Reference Vietnam Ministry of Health. 2008. National Health Account: Implementation in Viet Nam; Period from 2000–2006. Hanoi: Statistical Publishing House. 31 Chapter 5 How to Generate the Tables and Graphs This chapter explains how to set up ADePT Health Outcomes so as to gen- erate tables and graphs. The assumption is that the data set has been pre- pared before it is loaded into ADePT. The explanation proceeds with a screen shot of ADePT, with numbers on the screenshot corresponding to the steps outlined. 33 34 Main Tab All users will need to enter information on the “main” tab as follows. Screenshot 5.1: Main Tab 1. Start up ADePT, select the Health Outcomes module, add a data set (click on the “add” button), and then label it (type a name in the box). Hint 1: If you have used ADePT before, your previous session will be reloaded. You have three options in this case: (1) Continue with the same data set, in which case just keep going. (2) Do the same type of analysis you were doing before but for a different data set with vari- ables labeled the same. In this case, simply “remove” the data set, “add” your new data set, and continue. All the boxes in which you pre- viously entered information will still include the information. If, while running, ADePT finds that your new data set does not include some of the variables, it will mark them in red in the user interface. (3) Start with a new analysis and a new data set. In this case, simply choose Project -> Reset or hit ctl-R. Hint 2: You can load several data sets into ADePT at once. The variables you want to analyze will need to exist and be similarly named in both data sets. To facilitate this, check the “enable only common variables” box; this will cause ADePT to show only the variables that appear in all the data sets you have loaded. 2. Click on the “variables” tab at the top left corner of the ADePT screen and then the “main” tab in the bottom left corner of the ADePT screen. Next provide the household identifier variable (or variables—AdePT can accommodate multiple variables in the household identification, ID, box). ADePT checks the uniqueness of identification and issues a warning message when required. You can input your household ID variables in the household ID box (and other variables in other boxes) either by dragging the variable from the list of variables at the top left of the screen or by selecting the variable(s) from the drop-down menu. Hint: You can simplify the dragging approach by typing part of the name of the variable(s) in the “search” box; ADePT will then show only those variables whose name includes the text you typed in the search box. 3. Enter the name of the living standards indicator. This is typically a measure of per capita or “equivalent” household income or consump- tion. Hint: You need to have expressed the household living standards indicator on a per capita or equivalent basis before loading the data into ADePT. 4. Indicate whether the tables should present quintiles or deciles. 5. Enter the name of the health outcome variables to be analyzed. These could be binary (for example, having a particular health condition or not), categorical (for example, self-assessed health), cardinal (for example, number of work days missed due to sickness), or continuous but not cardinal (for example, anthropometric z scores). 6. Enter the names of the health care utilization variables to be analyzed. These are typically a binary variable (for example, whether a con- sultation occurred during a specified period) or a count variable (for example, the number of consultations). 7. Optionally, enter categorical or discrete individual- or household-level variables by which you would like outcomes or utilization to be tab- ulated. ADePT invites users to enter variables capturing whether the household lives in an urban setting, the region it lives in, and the indi- vidual’s age, his or her gender, level of education, and employment status. 8. Provide household weights. If necessary, also set the appropriate survey settings by clicking on “survey settings. ” 9. If you want tabulations by the variables entered in step 7, check the boxes TH1 and TH2, TU1 and TU2, or both. 10. If you want unstandardized inequalities in health outcomes or utilization, check the boxes against TH3 (and perhaps G1) or against TU3 (and perhaps G2). These provide standardized distributions of health outcomes and utilization, respectively. 11. If you want only these tables and graphs, hit the “generate” button to start the computation and generate the selected outputs. Otherwise 35 continue to the “determinants of health” tab to enter information enabling inequalities to be explained or the “benefit incidence analysis (BIA)” tab to enter the information necessary for a BIA. 36 Determinants of Health or Utilization ADePT users wanting to explain inequalities in health or utilization should enter information on the “main” tab as outlined above before entering the necessary information on the “determinants of health or utilization” tabs. Users wanting to undertake only a BIA should proceed directly to the “benefit incidence analysis” tab after the “main” tab. Screenshot 5.2: Inequalities in Health or Utilization Tab 1. Optionally, check the box to indicate that standardizing variables for health outcomes are to be entered and enter the variable names. These might be demographic variables (such as age and gender) for which income-related health inequality is not deemed inequitable. Hint: Which variables to include here is a matter of debate; there is no right or wrong answer! 2. Optionally, check the box to indicate that control variables for health outcomes are to be entered and enter the variable names. These are factors (such as income and health insurance) for which income-related health inequality is deemed unjustified. Note that these variables need to be specified appropriately. For instance, a categorical variable such as education should be included as a set of indicator (or dummy) variables, one per education level, with the exception of a reference category. Hint: A quick way to do this is to put an “i. in ” front of the categorical variable. For example, if educn is a categorical education variable, you can force ADePT to treat it as a series of summary variables corresponding to the categories by inserting i.educn in the control variable box. 3. Optionally, check the box to indicate that standardizing variables for utilization are to be entered and enter the variable names. These are typically health care need variables (such as demographic and health status variables) for which income-related inequality in health care utilization is not deemed inequitable. Hint: The health status “need” variables might well be the same variables as the ones specified in step 9, if any, but the user needs to decide this and copy and paste the information if this is what is desired. 4. Optionally, check the box to indicate that control variables for utilization are to be entered and enter the variable names. These are factors (such as income and health insurance) for which income-related inequality in health care utilization is deemed unjustified. 5. Select which tables are to be generated. Tables H4 and H5 provide the direct and indirect standardization results for health outcomes. Tables H6 and H7 provide decompositions of inequalities in health outcomes. Table H8 provides details of these decompositions. Tables U4 and U5 provide the direct and indirect standardization results for health utilization. Tables U6 and U7 provide decompositions of inequalities in utilization. Table U8 provides details of the decompositions. 6. Specify whether the standard error of each indicator is to be produced. This is required for inference but slows down computation. . 7 Check the “frequencies” box if you want an additional page in the spreadsheet showing the frequencies (that is, number of cases) used to compute each statistic requested. 8. To produce a table or figure for a subset of cases, highlight the relevant table or graph and enter the relevant “if condition” in the “if con- dition” box. Hint: Each table or graph can have a different “if condition” assigned to it. 9. If you want all the analysis to be done for just a subset of cases, click on the “filter” tab, check the “keep observations if” box, and enter the desired condition. 10. Hit the “generate” button to start the computation and generate the selected outputs. 37 38 Benefit Incidence Analysis ADePT users wanting to undertake a benefit incidence analysis should enter information on the “main” tab as outlined above before proceeding to the “benefit incidence analysis” tab. Screenshot 5.3: Benefit Incidence Analysis Tab 1. Fill out this field when disaggregated information on out-of-pocket expenditure is not available. When this is filled out, the individual fee per health care category no longer has to be entered (that is, step 3 becomes unnecessary). 2. One health care category at a time, follow steps 2 to 5. Start by specifying the variable containing the number of health care units used by each individual (for example, number of outpatient visits in a general hospital). Hint: If you are going to do a BIA with the utilization vari- ables, make sure that these are scaled to refer to the same period. 3. Enter the variable measuring the individual fee paid for the corresponding unit of utilization. 4. Type in the total public subsidy allocated for the category of utilization in question. This is an aggregate value found in National Health Account macro data. 5. Click on the “add” button. 6. Information from steps 2–4 should appear on the list. It is possible to remove any element of the list and repeat steps 2–5 to enter infor- mation related to additional health care categories. 7. Select which tables and graphs are to be generated. Relevant ones for BIA are tables S1–S5 and graphs G3–G5. Some of the tables might not be available depending on the information provided to ADePT. 8. Specify whether the standard error of each indicator is to be produced. This is required for inference, but slows down computation. 9. Check the “frequencies” box if you want an additional page in the spreadsheet showing the frequencies (that is, number of cases) used to compute each statistic requested. 10. To produce a table or figure for a subset of cases, highlight the relevant table or graph and enter the relevant “if condition” in the “if condition” box. Hint: Each table or graph can have a different “if condition” assigned to it. 11. If you want all the analysis to be done for just a subset of cases, instead click on the “filter” tab, check the “keep observations if” box, and enter the desired condition. 12. Hit the “generate” button to start the computation and generate the selected outputs. 39 Chapter 6 Interpreting the Tables and Graphs As detailed in chapter 4, all of the tables and graphs in this chapter were produced using data from the 2006 Vietnam Household Living Standards Survey. Original Data Report Concepts The original data report gives basic statistics on the variables provided by users and shows in parentheses the field in which each variable was entered. This report is important as it makes it possible to detect many of the bold errors that may occur when preparing the data. For each vari- able, the first column shows the number of observations with a valid value (that is, a value that is not represented by a missing value in Stata). The next three columns present the mean, minimum, and maxi- mum values of each variable. Column “p1” gives the first percentile—in other words, the value that is greater than 1 percent of the values taken by the variable (and thus smaller than 99 percent of its values). Column “p50” represents the median, and “p99” represents the 99th percentile. The last column indicates the number of different values taken by each variable. 41 Health Equity and Financial Protection: Part I Original Data Report N mean min max p1 p50 p99 N_unique VHLSS 2006 tinh (Household ID) 39,071 460 101 823 101 411 823 64 huyen (Household ID) 39,071 10.7 1.0 53.0 1.0 9.0 43.0 35.0 xa (Household ID) 39,071 18.2 1.0 95.0 1.0 15.0 63.0 57 diaban (Household ID) 39,071 10.1 1.0 105.0 1.0 8.0 42.0 59.0 hoso (Household ID) 39,071 14.6 13.0 25.0 13.0 14.0 20.0 10 pcexp (Welfare aggregate) 39,071 5,481 554 153,995 1,076 4,369 21,180 9,185 wt9 (Household weights) 39,071 2,111 467 4,637 576 2,006 4,353 2,079 adlhearing (Outcome 1) 36,701 0.0 0.0 1.0 0.0 0.0 1.0 2.0 adleyesight (Outcome 2) 36,701 0.1 0.0 1.0 0.0 0.0 1.0 2.0 adlwalk (Outcome 3) 36,701 0.1 0.0 1.0 0.0 0.0 1.0 2.0 opvis_chc (Custom category 1) 39,071 0.3 0.0 80.0 0.0 0.0 6.0 30 opvis_ghosp (Custom category 2) 39,071 0.4 0.0 52.0 0.0 0.0 6.0 32.0 opvis_privfacil (Custom category 3) 39,071 0.4 0.0 80.0 0.0 0.0 8.0 34 opvis_chc (Custom category 1) 39,071 0.3 0.0 80.0 0.0 0.0 6.0 30.0 opexp_chc (Custom category 1) 39,071 7.0 0.0 3,600 0.0 0.0 150.0 182 opvis_poly (Custom category 2) 39,071 0.0 0.0 40.0 0.0 0.0 1.0 19.0 opexp_poly (Custom category 2) 39,071 2.5 0.0 22,000 0.0 0.0 10.0 83 opvis_ghosp (Custom category 3) 39,071 0.4 0.0 52.0 0.0 0.0 6.0 32.0 opexp_ghosp (Custom category 3) 39,071 64.0 0.0 105,600 0.0 0.0 1,400 425 ipadm_ghosp (Custom category 4) 39,071 0.1 0.0 13.0 0.0 0.0 2.0 12.0 ipexp_ghosp (Custom category 4) 39,071 103.7 0.0 60,000 0.0 0.0 2,500 342 age (Age) 39,071 31.0 0.0 108.0 0.0 27.0 82.0 104.0 male (Gender) 39,071 1.5 1.0 2.0 1.0 2.0 2.0 2.0 educlev (Education) 39,071 1.4 0.0 5.0 0.0 1.0 5.0 6.0 generated (Household size) 39,071 4.9 1.0 17.0 2.0 5.0 10.0 16 Source: Authors. Interpreting the Results Provided that they have been adequately prepared, the identification (ID) variables generally give the number of observations in the sample analyzed, which amounts to 39,071 in our example table. For instance, per capita expenditure (pcexp) also has 39,071 observations and thus does not contain any missing values. Mean per capita expenditure amounts to D5.5 million (pcexp is measured in thousands of Vietnamese dong) and ranges from D554,000 to D154 million. The median (D4.4 million) is smaller than the mean, which indicates the expected right-skewed distribution of living standards variables. The last column of the table usually makes it possible to identify categorical data. For instance, our three outcome variables, adlhear- ing, adleyesight, and adlwalk, appear to be binary, as they are shown to take only two different values. 42 Chapter 6: Interpreting the Tables and Graphs Basic Tabulations Concepts Compared with ADePT’s original data report, which provides basic sum- mary statistics on all variables specified, tables H1, H2, U1, and U2 provide more in-depth descriptive analysis of the health variables. Tables H1 and H2 relate to health outcomes, while tables U1 and U2 display health utilization variables. All these tables show the mean of the health variables according to household (H1 and U1) and individual (H2 and U2) characteristics. ADePT also computes standard errors, which are presented in parentheses below the corresponding means. Under the assumption of normal distribu- tion, subtracting twice the standard error from the mean, and then adding twice the standard error to the mean, yields the 95 percent confidence inter- val. This interval, which is reported in many standard Stata outputs, has a 95 percent chance of containing the true (and unknown) value of interest. Interpreting the Results All tables are read in the same way. We show here only one example, table H2, which presents the relation between health outcomes and individual charac- teristics. The health outcomes analyzed are binary variables indicating whether the individual suffers from impairment in activities of daily living with respect to hearing, seeing, and walking. Individual characteristics are a binary indicating gender, age divided into 13 classes, and education level ranging fom 0 (lowest) to 5 (highest). In general, women have more health problems than do men. For instance, the prevalence of walking impairment amounts to 4.7 percent for men and 7.3 percent for women. Given the very small standard errors, such a differ- ential is statistically significant. Table H2 also shows the expected increase in health problems with age. Age is a continuous variable that ADePT express- es as age categories in this description, but we treat age as continuous in the following analyses. Finally, the relationship between health and education is less clear-cut. However, the least educated individuals, in general, have the highest prevalence of health impairment. Health impairment does not con- tinuously decrease with education, though. For instance, the two highest education levels show a fairly high prevalence of impaired sight. 43 Health Equity and Financial Protection: Part I Table H2: Health Outcomes by Individual Characteristics adlhearing adleyesight adlwalk Gender Male 0.031 0.102 0.047 (0.0014) (0.0024) (0.0017) Female 0.035 0.125 0.073 (0.0014) (0.0026) (0.0021) Age 0–5 0.000 0.002 0.029 (0.0000) (0.0017) (0.0073) 6–14 0.005 0.017 0.005 (0.0009) (0.0017) (0.0008) 15–19 0.006 0.024 0.008 (0.0011) (0.0024) (0.0013) 20–24 0.007 0.022 0.009 (0.0015) (0.0028) (0.0016) 25–29 0.007 0.017 0.011 (0.0019) (0.0032) (0.0022) 30–34 0.005 0.016 0.010 (0.0014) (0.0028) (0.0021) 35–39 0.009 0.020 0.018 (0.0018) (0.0029) (0.0026) 40–44 0.010 0.072 0.029 (0.0019) (0.0051) (0.0032) 45–49 0.016 0.159 0.047 (0.0024) (0.0074) (0.0044) 50–54 0.034 0.235 0.068 (0.0044) (0.0097) (0.0060) 55–59 0.058 0.318 0.130 (0.0062) (0.0129) (0.0094) 60–64 0.092 0.409 0.199 (0.0097) (0.0166) (0.0139) 65 0.249 0.559 0.414 (0.0085) (0.0097) (0.0097) Education level 0 0.072 0.180 0.122 (0.0027) (0.0040) (0.0034) 1 0.022 0.085 0.041 (0.0015) (0.0030) (0.0021) 2 0.016 0.075 0.033 (0.0014) (0.0031) (0.0021) 3 0.008 0.068 0.024 (0.0017) (0.0047) (0.0030) 4 0.024 0.136 0.050 (0.0032) (0.0070) (0.0046) 5 0.014 0.146 0.035 (0.0035) (0.0108) (0.0055) Total 0.033 0.114 0.060 (0.0010) (0.0018) (0.0013) Source: Authors. 44 Chapter 6: Interpreting the Tables and Graphs Inequalities in Health Outcomes Concepts Tables H3–H5 show the distribution and concentration of health according to income or any other measure of socioeconomic status. Table H3 presents the unstandardized distribution of health, whereas the other two present standardized distributions using the direct and indirect methods. The idea of standardization is to eliminate the inequality arising from variables that are beyond the control of policy makers, notably demographic factors such as age and gender. Since standardization can be achieved in a more transparent way by means of the decomposition of the concentration index (see technical note 7 in chapter 7), we do not present tables H4 and H5 here. However, ADePT makes it possible to generate these tables; for more information, see O’Donnell and others (2008, ch. 5). The first part of the tables gives the average health status per quintile along with standard errors, which are presented in parentheses below each Table H3: Health Inequality, Unstandardized adlhearing adleyesight adlwalk Quintiles of pcexp Lowest quintile 0.0346 0.0820 0.0589 (0.0022) (0.0033) (0.0029) 2 0.0380 0.1043 0.0613 (0.0023) (0.0037) (0.0029) 3 0.0352 0.1111 0.0578 (0.0023) (0.0038) (0.0029) 4 0.0285 0.1219 0.0604 (0.0021) (0.0041) (0.0030) Highest quintile 0.0283 0.1469 0.0633 (0.0022) (0.0047) (0.0033) Total 0.0329 0.1136 0.0603 (0.0010) (0.0018) (0.0013) Standard concentration index 0.0583 0.1096 0.0147 (0.0175) (0.0090) (0.0132) Conc. index with inequality-aversion parameter 3 0.0784 0.1604 0.0157 (0.0268) (0.0130) (0.0198) Conc. index with inequality-aversion parameter 4 0.0834 0.1934 0.0165 (0.0340) (0.0159) (0.0246) Standard achievement index 0.0348 0.1012 0.0594 (0.0084) (0.0075) (0.0098) Achievement index with inequality-aversion parameter 3 0.0354 0.0954 0.0594 (0.0096) (0.0076) (0.0100) Achievement index with inequality-aversion parameter 4 0.0356 0.0916 0.0593 (0.0108) (0.0081) (0.0100) Source: Authors. 45 Health Equity and Financial Protection: Part I average. The concentration index is then displayed for three different val- ues of the inequality-aversion parameter. The concentration index is a measure of how health status is related to income. A positive value indicates that the health variable is more concentrated among richer individuals. In the case of a variable for ill health, such as impairment in activities of daily living, a positive value of the concentration index means that the rich are in worse health than the poor. A negative value indicates the opposite, whereas a concentration index not very different from 0 reflects no rela- tionship between income and health status. The standard concentration index implicitly gives the poorest individual a weight close to 2, and the weight decreases linearly to reach a value close to 0 for the richest individual. A more general measure of concentration, the extended concentration index, makes it possible to change this weighting by setting the value of an inequality-aversion parameter. When the inequality- aversion parameter equals 2, the extended concentration index equals the concentration index. When the parameter is increased (here to 3 and 4), more weight is given to the poorest individual, and this weight decreases faster than linearly to reach 0 for the richest individual (see technical note 4 in chapter 7 for more details). The last three lines of the tables display the achievement index (pre- sented in technical note 5 in chapter 7), which is a measure of average health taking health inequality into account: the greater the health inequality, the smaller is the achievement index. It can be shown that the achievement index equals average health multiplied by the factor 1 minus the extended concentration index. It is thus also sensitive to the degree of aversion to inequality, and the corresponding parameter appears in parentheses. Interpreting the Results Our example table H3 shows the income-related inequality in three health status variables. These are binary variables indicating whether the individ- ual suffers from impairment in activities of daily living with respect to hear- ing, seeing, and walking. On average, 3.46 percent of the first quintile experience hearing problems. This amounts to 8.20 and 5.89 percent for seeing and walking, respectively. Starting with the second quintile, the prevalence of hearing impairment decreases with income, but seeing impairment increases with income, and walking impairment seems largely independent of income. 46 Chapter 6: Interpreting the Tables and Graphs The table also shows the average prevalence in the whole population, which amounts to 3.29, 11.36, and 6.03 percent for limitations in hearing, see- ing, and walking, respectively. The concentration index for hearing is moderately negative ( 0.0583), which reflects the decrease in health condi- tions with income. In contrast, the concentration index for seeing impairment is clearly positive (0.1096) as a result of the increase in prevalence with income. Finally, the concentration index for walking (0.0147) is close to 0, which indicates near independence between this health condition and income. When aversion to inequality is increased, the weight of the first quintiles is increased relative to the richest quintiles. Since the first quintiles have a higher prevalence of hearing impairment, the corresponding extended con- centration index gets farther into the negatives. The opposite is true for seeing impairment, and the extended concentration index for walking impairment is barely affected by change in inequality aversion. Finally, the achievement index amounts to 0.0348, 0.1012, and 0.0594 for hearing, seeing, and walking impairment, respectively, when applying the standard concentration index weighting. These have to be compared with the nonweighted averages for the whole population. For instance, since the concentration index is positive in the case of seeing impairment, the rel- atively higher prevalence of the richest is underweighted, and, as a result, the achievement index (0.1012) indicates better health compared with average health (0.1136). When aversion to inequality is increased (that is, the last two lines of the table), the greater weight assigned to the first quintiles makes average health deteriorate slightly in the case of hearing (that is, higher average prevalence of hearing impairment). In contrast, as richer individuals suffer more from seeing impairment, higher aversion to inequality makes the achievement index improve for this condition (that is, lower average prevalence of see- ing impairment). Concentration of Health Utilization Concepts Graphs G1 and G2 present the concentration curve for each health outcome and for utilization of each health service analyzed, respectively. The curves represent the cumulative share of either health outcomes or utilization 47 Health Equity and Financial Protection: Part I Graph G1: Concentration Curves of Health Outcomes 100 cumulative % of outcome variable 80 60 40 20 0 0 20 40 60 80 100 cumulative % of population, ranked from poorest to richest adlhearing adleyesight adlwalk line of equality Source: Authors. according to the cumulative share of population, ranked by increasing con- sumption. For instance, the poorest 30 percent might suffer from 50 percent of total health conditions. These curves show how health outcomes or uti- lization vary according to consumption: the farther a curve is above the 45° line, the more the corresponding health variable is concentrated among the poorest households. When the concentration curve lies under the 45° line, the corresponding health variable is more concentrated among the richest households. Twice the area between the concentration curve and the 45° line equals the concentration index, the values of which are presented in table H3 for health outcomes and table U3 for utilization. Interpreting the Results Our example graph G1 shows the income-related inequality in three health status variables. These are binary variables indicating whether the individual 48 Chapter 6: Interpreting the Tables and Graphs suffers from impairment in activities of daily living with respect to hearing, seeing, and walking. The concentration curve for hearing impairment lies above the 45° line, which confirms that this health condition is more preva- lent among the poor. By contrast, the concentration curve for seeing impair- ment lies below the 45° line, which means that this health condition is more concentrated among the rich. The concentration curve for walking limita- tions is very close to the 45° line, which indicates very little association between this health condition and income. Explaining Inequalities in Health Concepts Tables H6 and H7 show two alternatives for the decomposition of the health concentration index by health determinant. In these tables, it is possible to distinguish inequality (the extent to which health is linked to income) from inequity (the part of inequality deemed unjustified) by separating the factors into two different groups. On the one hand, the standardizing variables are the variables whose correlation with health is deemed justifiable. Demographic factors such as age and gender are common examples of such variables. On the other hand, control variables are the factors whose corre- lation with health is deemed unfair. These variables “control” health stan- dardization in the sense that they prevent standardizing variables from picking up the effect of variables with which they are correlated. Moreover, Table H6: Decomposition of the Concentration Index for Health Outcomes, Linear Model adlhearing adleyesight adlwalk Standardizing (demographic) variables male 0.0002 0.0002 0.0009 age 0.1036 0.0802 0.0971 Subtotal 0.1038 0.0801 0.0962 Control variables pcexp 0.1015 0.0256 0.0405 Subtotal 0.1015 0.0256 0.0405 Residual 0.0606 0.0040 0.0411 Inequality (total) 0.0583 0.1096 0.0147 Inequity / Unjustified inequality 0.1621 0.0295 0.0816 Source: Authors. 49 Health Equity and Financial Protection: Part I in addition to standardization, the decomposition also makes it possible to quantify the contribution of both standardizing and control variables to overall health inequality. Tables H6 and H7 also show total income-related inequality caused by standardizing variables (the first subtotal) and by control variables (the second subtotal). When the part of income-related inequality that is not explained by the chosen determinants (that is, the residual) is added to these two subtotals, the overall inequality is obtained—line “inequality (total)” in the table. Finally, when we subtract the justifiable inequality (that is, the first subtotal) from the overall inequality, we obtain the unjustified inequality—in other words, the inequity in income-related inequality. For the decomposition the health variable needs to be regressed on standardizing and control variables. When the health variable is continu- ous, a linear model is often acceptable and it is possible to estimate the parameters by ordinary least squares (OLS). The result of this estimation is presented in table H8a, and the corresponding decomposition of the concentration index is presented in table H6. However, most health vari- ables are better estimated by a nonlinear model, as health is often measured as a binary variable (for example, having a particular health condition or not), as an ordinal variable (for example, self-assessed health), or as counts (for example, number of days spent in bed due to illness). ADePT is usu- ally able to determine the best nonlinear model to be estimated, and the result of this estimation is displayed in table H8b. The decomposition of Table H8a: Fitted Linear Model adlhearing adleyesight adlwalk coef se coef se coef se Standardizing (demographic) variables male 0.0023 0.0019 0.0056* 0.0032 0.0151*** 0.0024 age 0.0028*** 0.0001 0.0074*** 0.0001 0.0048*** 0.0001 Control variables pcexp 1.59E-06*** 2.05E-07 1.38E-06*** 4.48E-07 1.16E-06*** 3.13E-07 Intercept 0.0468*** 0.0036 0.1506*** 0.0058 0.1149*** 0.0048 Number of observations 36,701 36,701 36,701 Adjusted R2 0.09 0.21 0.16 Source: Authors. Note: *** p 0.01, ** p 0.05, * p 0.1. 50 Chapter 6: Interpreting the Tables and Graphs the concentration index corresponding to this nonlinear model is then pre- sented in table H7. Each contribution in the decomposition is the product of the sensitivity of health with respect to the corresponding determinant and the degree of income-related inequality in that determinant. Sensitivity is measured by the elasticity of health according to the determinants. These elasticities are computed with the estimated model and are presented in table H8c for the linear model and in table H8d for the nonlinear model. Finally, income- related inequality is measured with the concentration index of the determi- nants. These concentration indexes are displayed in table H8e. Since they do not rely on the estimated model, they are common to the decompositions in tables H6 and H7. Using a nonlinear model is not an ideal solution, as the decomposition itself needs to be approximated (see technical note 7 in chapter 7). A binary variable may, for instance, be estimated by OLS when one thinks that the linear probability model is acceptable. This could be the case for a well-balanced binary variable (that is, not too far from the 50–50 split) for which the probit and logit model specifications are fairly linear. The advantage is that the decomposition of the concentration index would be accurate. Thus, when dealing with a health variable whose ideal specifi- cation is nonlinear, one has to decide whether to approximate its model by a linear specification or whether to approximate the decomposition of the concentration index. The alternative consists of estimating a nonlin- ear model before using ADePT and then decomposing the inequality in Table H8c: Elasticities, Linear Model adlhearing adleyesight adlwalk Male 0.1046 0.0742 0.3792 Age 2.8138 2.1797 2.6384 Pcexp 0.2851 0.0718 0.1136 Source: Authors. Table H8e: Concentration Index of the Covariates adlhearing adleyesight adlwalk Male 0.0023 0.0023 0.0023 Age 0.0368 0.0368 0.0368 Pcexp 0.3561 0.3561 0.3561 Source: Authors. 51 Health Equity and Financial Protection: Part I the (linear) scores of the latent health variable (see Hosseinpoor and others 2006). Interpreting the Results Our example table H6 shows the decomposition of the concentration index for three health status variables. These are binary variables indicat- ing whether the individual suffers from impairment in activities of daily living with respect to hearing, seeing, and walking. Since we are reading table H6, the underlying model explaining each binary variable is linear and estimated by OLS. The interpretation below relates to hearing impair- ment only. Gender and age are used as standardizing variables, and their contribu- tion to income-related inequality in health is, respectively, 0.0002 and 0.1036. That is, hearing impairment is slightly more concentrated among the rich due to gender and significantly more so due to age. Since we have chosen to standardize health according to these variables, their total contri- bution to income-related inequality in health (0.1038) is deemed justified. This positive contribution means that if hearing problems are correlated with age and gender only, they will show a pro-rich distribution. Per capita expenditure is used as a control variable in order to avoid age and gender picking up the income effect, which would lead to overstan- dardizing health. The contribution of this factor to inequality is 0.1015, which is interesting in itself. Since it is our only control variable, total inequality due to control variables also amounts to 0.1015. Finally, income-related inequity amounts to 0.0583 0.1038 0.1621 and favors the rich in the sense that the prevalence of hearing impairment is more concentrated among poorer individuals. Overall inequality appears to be smaller (in absolute terms) than inequity, as part of it is masked by the greater need of richer individuals. Table H8a shows the estimated coefficients of the linear model along with their standard errors (se) and asterisks indicating whether the coef- ficients are statistically significant for thresholds ranging from 1 to 10 percent. On the one hand, the prevalence of both hearing and walking impairment decreases with per capita expenditure, as the estimated coefficients are nega- tive ( 1.59E-06 and 1.16E-06, respectively). On the other hand, the pos- itive coefficient of per capita expenditure in the seeing impairment 52 Chapter 6: Interpreting the Tables and Graphs model (1.38E-06) indicates that this health problem increases with income. The number of observations can change from one health variable to the next due to dropping of missing values. In this example, all three mod- els use the same sample of 36,701 observations, as do the decompositions presented in table H6. Finally, table H8a also shows the adjusted R2, which is a measure of goodness of fit. The closer the adjusted R2 is to 1, the more the model explains the variability in the health variable and, consequently, the more meaningful is the decomposition of the concentration index. The model for hearing impairment explains only 9.22 percent of the variability in this health variable. However, the models for seeing and walking impairment (21.31 and 15.59 percent, respectively) fare significantly better. Table H8c displays the elasticity of each health variable according to each standardizing and control variable. These are simply computed from the estimated models presented in table H8a and are very important, as they are part of the decompositions displayed in table H6. Elasticities, however, are only meaningful for continuous variables such as age and per capita expenditure. For instance, in the case of hearing impairment, an increase of 1 percent in age results in an increase of 2.8138 percent in the prevalence of this health problem. By contrast, when per capita expenditure is increased by 1 percent, the prevalence of hearing impairment is reduced by 0.2851 percent. The problem with binary vari- ables such as male is that they can only take the values 0 and 1, and thus a percentage increase does not make much sense. Table H8e presents the concentration indexes of the standardizing and control variables. Since they do not depend on the estimated models and the sample size here is the same for all three decompositions (that is, 36,701, see table H8a), the concentration index is the same for all three health variables. When the health variables contain missing values, drop- ping them leads to different subsamples and, ultimately, to different values of the concentration index. Males are slightly more concentrated among the poor, and this vari- able has a negative concentration index. Older individuals are most often richer individuals, and this variable has a positive concentration index. Finally, the concentration index for per capita expenditure is the Gini coefficient, as observations are ranked according to the same variable. 53 Health Equity and Financial Protection: Part I This amounts to 0.3561 and provides a measure of income inequality in the population analyzed. Decomposition of the Concentration Index Concepts Graphs G7a, G7b, G8a, and G8b all present decompositions of the concen- tration index of health variables and correspond to tables H6, H7, U6, and U7, respectively. The first two graphs relate to health outcome variables, while the last two relate to health utilization variables. Graphs G7a and G8a display decompositions of the concentration index based on a linear model explaining the health variable, whereas the decompositions shown in graphs G7b and G8b are based on a nonlinear model chosen by ADePT. These graphs offer a convenient way of showing the respective contribu- tions of the various determinants of health variables to the concentration index. When a determinant is drawn above the horizontal line, this indi- cates a positive contribution—in other words, the determinant contributes to making the concentration index of the health outcome or use variable more pro-rich. The larger the corresponding area in the graph, the more the determinant is contributing to a pro-rich distribution of the health Graph G7a: Decomposition of the Concentration Index for Health Outcomes, Using OLS 0.1000 0.0500 0.0000 –0.0500 H1 - adlhearing H2 - adleyesight –0.1000 H3 - adlwalk –0.1500 H1 H2 H3 male age pcexp residual Source: Authors. 54 Chapter 6: Interpreting the Tables and Graphs outcome or use variable. Symmetrically, when a determinant is drawn below the horizontal line, this indicates a pro-poor contribution. Finally, the con- tributions of the residuals of the health variable models are also represented on the graphs. This shows the part of the concentration index that is not explained by the determinants. Interpreting the Results Our example graph G7a shows the decomposition of the concentration index for three health status variables. These are binary variables indicating whether the individual suffers from impairment in activities of daily living with respect to hearing, seeing, and walking. Age clearly makes a considerable positive contribution to the concen- tration index for the three health variables. On the one hand, this means that age makes health impairments more frequent among richer individuals, which partly results from the joint increase in wages and health problems with age. On the other hand, gender makes such a small contribution to the concentration index that it is barely evident on the graph. Per capita expenditure makes the concentration index more pro-poor for hearing and walking impairments. In other words, income makes richer individuals healthier. However, the opposite is true for seeing impairment, which probably results from the more extensive use of this health function by richer individuals. Inequalities in Utilization Concepts Tables U3–U5 show the distribution of health care utilization by income or any other measure of socioeconomic status. They are similar to tables H3–H5, with the only difference being that they relate to health care utilization instead of health outcomes. Interpreting the Results Our example table U3 shows the income-related inequality in the utiliza- tion of commune health centers (CHCs), general hospitals, and private 55 Health Equity and Financial Protection: Part I Table U3: Inequality in Health Care Utilization, Unstandardized CHC General hospital Private facility Quintiles of pcexp Lowest quintile 0.3832 0.1578 0.1610 (0.0161) (0.0113) (0.0106) 2 0.3212 0.2447 0.3940 (0.0154) (0.0140) (0.0259) 3 0.3206 0.3053 0.4478 (0.0167) (0.0167) (0.0218) 4 0.3064 0.4865 0.5251 (0.0196) (0.0256) (0.0231) Highest quintile 0.1802 0.7295 0.7036 (0.0205) (0.0339) (0.0415) Standard concentration index 0.1167 0.3050 0.2308 (0.0166) (0.0132) (0.0153) Conc. index with inequality-aversion parameter 3 0.1592 0.4210 0.3460 (0.0247) (0.0168) (0.0181) Conc. index with inequality-aversion parameter 4 0.1832 0.4843 0.4253 (0.0302) (0.0191) (0.0192) Standard achievement index 0.3376 0.2674 0.3433 (0.0084) (0.0075) (0.0098) Achievement index with inequality-aversion parameter 3 0.3504 0.2228 0.2919 (0.0096) (0.0076) (0.0100) Achievement index with inequality-aversion parameter 4 0.3577 0.1984 0.2565 (0.0108) (0.0081) (0.0100) Source: Authors. Note: standard errors displayed in brackets. facilities. The corresponding variables measure the number of outpatient visits to these health care providers during a period of one year. In the case of CHCs, the average number of visits is 0.3832 for the first quintile, and this decreases monotonically with income to 0.1802 for the last quintile. CHC outpatient visits are thus more frequent among the poor. The opposite is true for outpatient visits to general hospitals and private facilities, which increase with income. The difference in the pattern of utilization between CHCs, on the one hand, and general hospitals and private facilities, on the other, is reflected in their corresponding concentration indexes. The concentration index of CHCs is clearly negative ( 0.1167), revealing higher utilization by poorer individuals. In contrast, the concentration indexes for visits to general hos- pitals (0.3050) and private facilities (0.2308) are clearly positive, indicating higher utilization by richer individuals. 56 Chapter 6: Interpreting the Tables and Graphs When aversion to inequality is higher, the greater weight assigned to the first quintiles renders CHC utilization even more pro-poor (that is, the concentration index becomes more negative), which indicates a greater concentration of utilization among the poor. Utilization of general hospitals and private facilities becomes even more pro-rich (that is, the concentration index increases further), which shows a greater concentration of utilization among the rich. Finally, the achievement index—in other words, the rank- weighted average utilization—of CHCs, general hospitals, and private facilities amounts to 0.3376, 0.2674, and 0.3433, respectively. With higher aversion to inequality, this quantity of utilization increases for CHCs and decreases for general hospitals and private facilities. Explaining Inequalities in Utilization Concepts Tables U6 and U7 show two versions of the decomposition of the health concentration index by determinant of health care utilization. They are sim- ilar to tables H6 and H7, with the only difference being that they relate to health care utilization instead of health status. That is why we present only example table U6 here; for more detail, refer to tables H6–H8. Interpreting the Results Our example table U6 shows the decomposition of the concentration index for utilization of CHCs, general hospitals, and private facilities. The corre- sponding variables measure the number of outpatient visits to these health care providers during a period of one year. Nine impairments in activities of daily living along with age and gender are used as standardizing variables. These variables are used to account for health care need, and their total contribution of 0.0247 (CHCs), 0.0348 (general hospitals), and 0.0278 (private facilities) to income-related inequality in health care utilization is thus not deemed inequitable. These positive contributions are mostly due to age, which is positively associated with both health care use and income. 57 Health Equity and Financial Protection: Part I Table U6: Decomposition of the Concentration Index for Utilization Values, Using OLS CHCs General hospitals Private facilities Standardizing (need) variables adleyesight 0.0065 0.0086 0.0137 adlhearing 0.0005 0.0004 0.0005 adlmemory 0.0030 0.0002 0.0006 adlconcent 0.0001 0.0009 0.0005 adlunderstand 0.0026 0.0009 0.0009 adlwalk 0.0004 0.0005 0.0005 adlstairs 0.0003 0.0001 0.0007 adl400m 0.0002 0.0012 0.0006 adl10steps 0.0000 0.0000 0.0000 male 0.0008 0.0005 0.0007 age 0.0187 0.0250 0.0136 Subtotal 0.0247 0.0348 0.0278 Control variables pcexp 0.1046 0.1177 0.1432 ghe_cap 0.0181 0.0386 0.0429 hi_poor 0.0822 0.0407 0.0487 hi_comp 0.0146 0.0600 0.0556 hi_vol 0.0232 0.0339 0.0138 Subtotal 0.1310 0.2095 0.1654 Residual 0.0104 0.0607 0.0376 Inequality (total) 0.1167 0.3050 0.2308 Inequity / Unjustified inequality 0.1414 0.2703 0.2029 Source: Authors. Per capita expenditure, government health expenditure per capita at the regional level, and three health insurance variables are used as control variables to prevent the need variables from picking up their effect. The total contribution of control factors to inequality amounts to 0.1310, 0.2095, and 0.1654 for CHCs, general hospitals, and private facilities, respectively. Total inequality can be retrieved by adding the residual to the contribu- tions of standardizing and control variables. Income-related inequity, by contrast, is equal to total inequality less the inequality justified by the inequalities in the standardizing variables: 0.1167 0.0247 0.1414 for CHCs, 0.3050 0.0348 0.2703 for general hospitals, and 0.2308 0.0278 = 0.2029 for private facilities. Finally, since need is greater for rich- er individuals with respect to all three utilization variables (that is, positive contribution of standardizing variables), inequity is less pro-rich (or more pro-poor) than inequality. 58 Chapter 6: Interpreting the Tables and Graphs Use of Public Facilities Concepts Table S1 shows how health care utilization is distributed according to per capita total expenditure (or any other welfare aggregate) for various public health services. Its first part presents the average utilization for each quin- tile of total expenditure as well as the average utilization for the whole population. The same information is then displayed again, but in relative terms—that is, the share of total health care utilization that benefits each quintile. Finally, table S1 also shows the concentration index of each public health service considered. A negative concentration index reveals that poorer individuals use more health care than the rich, whereas a posi- tive value indicates the opposite. Interpreting the Results Our example table S1 presents the utilization of four health services: out- patient visits in CHCs and polyclinics as well as outpatient visits and inpatient admissions to general hospitals. As shown in the sixth line of Table S1: Utilization of Public Facilities Outpatient Outpatient Inpatient Outpatient visits visits admissions visits (chc) (polyclinic) (hospital) (hospital) Means Lowest quintile 0.383 0.034 0.158 0.049 2 0.321 0.035 0.245 0.064 3 0.321 0.028 0.305 0.076 4 0.306 0.050 0.486 0.086 Highest quintile 0.180 0.056 0.730 0.096 Total 0.302 0.041 0.385 0.074 Shares Lowest quintile 25.3 16.8 8.2 13.2 2 21.3 17.3 12.7 17.1 3 21.2 13.6 15.9 20.4 4 20.3 24.7 25.3 23.3 Highest quintile 11.9 27.5 37.9 26.0 Concentration index 0.1167 0.1153 0.3050 0.1351 Source: Authors. 59 Health Equity and Financial Protection: Part I table S1, the most frequently used health service is outpatient visits to general hospitals, with 0.385 visit per individual on average. Polyclinics are the least frequently used, with an average of 0.041 outpatient visit, which is even below the average of inpatient admissions to general hospi- tals (0.074). There is a very striking contrast between the distribution of outpatient visits to CHCs and the use of general hospital services. The former steadily decreases with income (from an average of 0.383 visit for the first quintile to 0.180 for the last), whereas the latter increases for both inpatient and out- patient care. The relation between income and the use of outpatient ser- vices in polyclinics is less well defined, but the use of outpatient services is considerably greater for the last two quintiles. Table S1 also shows the same pattern in relative terms. For instance, the first quintile alone accounts for 25.3 percent of total outpatient visits to CHCs, and the last quintile accounts for 37.9 percent of outpatient visits to general hospitals. Finally, these patterns are also reflected in the concentration index. The concentration index of outpatient visits to CHCs is negative ( 0.1167), which provides an overall measure of the extent to which this health service is distributed in favor of the poor. The concentration index of outpatient visits in general hospitals (0.3050) is strongly positive, revealing that richer individuals use this service a lot more than the poor. Payments to Public Providers Concepts Table S2 is very similar to table S1, the only difference being that it relates to health care fees instead of utilization. It shows how health care fees are dis- tributed according to total expenditure per capita (or any other welfare aggre- gate) for various public health services. Its first part presents the average fee paid by each quintile of total expenditure as well as the average fee paid by the whole population. It then displays the same information in relative terms—that is, the share of total fees paid by each quintile. Finally, it shows the concentration index of each public health service considered. A negative concentration index reveals that poorer individuals contribute more to health care than the rich, whereas a positive value indicates the opposite. 60 Chapter 6: Interpreting the Tables and Graphs Table S2: Payments to Public Providers Outpatient Outpatient Inpatient Outpatient visits visits admissions visits (chc) (polyclinic) (hospital) (hospital) Means Lowest quintile 7.41 0.89 11.6 21.1 2 7.00 0.87 27.1 46.4 3 8.12 1.71 38.7 70.7 4 8.85 2.70 77.1 135.2 Highest quintile 4.87 8.08 202.2 261.3 Total 7.25 2.85 71.3 107.0 Shares Lowest quintile 20.4 6.2 3.3 3.9 2 19.3 6.1 7.6 8.7 3 22.4 12.0 10.8 13.2 4 24.4 18.9 21.6 25.3 Highest quintile 13.4 56.7 56.7 48.9 100.0 100.0 100.0 100.0 Concentration index 0.031 0.480 0.517 0.461 Source: Authors. Interpreting the Results Our example table S2 presents the contributions in thousands of Vietnamese dong made to four health services: outpatient visits to CHCs and polyclinics as well as outpatient visits and inpatient admissions to general hospitals. The sixth line of table S2 shows that inpatient admis- sions in general hospitals require the highest fees. On average, individuals spend D107,000 on hospital admissions. This average is computed using the whole population, not just patients. Given that in general only a small fraction of the population receives inpatient care, the average fee paid by the patients for this health service is therefore much greater. The whole population also pays D71,300, on average, for outpatient visits in general hospitals and relatively little for outpatient visits in CHCs (D7,325) and polyclinics (D2,850). There is a striking difference between the distribution of fees paid for outpatient care in CHCs and in the other health services analyzed. The former does not exhibit any well-defined pattern, but it is nonetheless clear that the richest quintile contributes considerably less than the oth- ers (D4,870). By contrast, contribution to all other health services increases steadily with income. The last quintile contributes almost 20 times more than the first to outpatient costs. Health care fees clearly 61 Health Equity and Financial Protection: Part I increase with income for outpatient visits to polyclinics, but this matters a bit less considering the relatively lower fees involved. However, if the rich pay more, it is also probably because they use more health care ser- vices. That is why this picture should be complemented by a benefit inci- dence analysis (BIA), which can be performed with ADePT by generating the tables S3–S5. Table S2 also shows the same pattern in relative terms. For instance, the last quintile alone accounts for more than half of total fees paid for outpatient visits in polyclinics and general hospitals (56.7 percent for both). This share is also very high for hospital admissions (48.9), but low for CHCs (13.4). On the one hand, the concentration index for fees paid for outpatient visits to CHCs is the only negative one ( 0.031). This shows that poorer individuals spend more than the rich on this type of health care. On the other hand, the concentration index for each of the three other types of health services amounts to approximately 0.5, which is an extremely posi- tive value. This means that richer individuals spend considerably more than the poor on these health services. Finally, when compared to health care uti- lization, the distribution of fees is significantly more pro-rich for all four health services considered. Health Care Subsidies: Cost Assumptions Concepts Tables S3–S5 show how public subsidies are distributed according to income (or any other welfare aggregate) for various health services. The difference between these tables is that the subsidy is computed applying the standard BIA constant cost assumption in table S3, using the proportional cost assumption in table S4, and applying the linear cost assumption in table S5. The choice of the method depends on whether we think that the cost incurred by the government when offering a given health service (that is, the unit cost of publicly provided care) is • the same for all individuals (table S3), • proportional to the fees paid by the individuals (table S4), or • made of a minimum fixed cost that increases at the same rate as the individual fees (table S5). 62 Chapter 6: Interpreting the Tables and Graphs Table S3: Health Care Subsidies, Constant Unit Cost Assumption Outpatient Outpatient Inpatient Outpatient visits visits admissions Total visits (chc) (polyclinics) (hospital) (hospital) subsidies Average subsidy Lowest quintile 7.68 1.97 39.93 76.76 126.34 2 5.95 2.03 56.93 89.28 154.20 3 5.37 1.39 67.86 96.36 170.99 4 4.91 2.34 105.78 92.15 205.19 Highest quintile 3.70 2.23 133.37 87.01 226.30 Total 5.52 1.99 80.77 88.31 176.60 Distribution Lowest quintile 27.8 19.8 9.9 17.4 14.3 2 21.6 20.4 14.1 20.2 17.5 3 19.5 13.9 16.8 21.8 19.4 4 17.8 23.5 26.2 20.9 23.2 Highest quintile 13.4 22.4 33.0 19.7 25.6 Total 100.0 100.0 100.0 100.0 100.0 Share in the total subsidy 3.1 1.1 45.7 50.0 100.0 Concentration index 0.1407 0.0343 0.2471 0.0222 0.1201 Source: Authors. Table S4: Health Care Subsidies, Proportional Cost Assumption Outpatient Outpatient Inpatient Outpatient visits visits admissions Total visits (chc) (polyclinics) (hospital) (hospital) subsidies Mean subsidy 1 2.01 0.08 7.85 7.83 17.78 2s 1.90 0.08 18.26 17.24 37.48 3 2.21 0.16 26.14 26.34 54.85 4 2.40 0.25 52.01 50.15 104.82 5 1.32 0.75 136.51 97.08 235.66 Total 1.97 0.27 48.15 39.72 90.11 Shares 1 20.4 6.2 3.3 3.9 3.9 2 19.3 6.1 7.6 8.7 8.3 3 22.4 12.0 10.9 13.3 12.2 4 24.4 18.9 21.6 25.2 23.3 5 13.4 56.7 56.7 48.9 52.3 Total 100.0 100.0 100.0 100.0 100.0 Share in the total subsidy 2.2 0.3 53.4 44.1 100.0 Concentration index 0.0306 0.4796 0.5174 0.4610 0.4804 Source: Authors. This choice directly affects the computation of the subsidy since it amounts to the difference between public cost of care and fees paid by the individual. The first two assumptions are extreme particular cases of the more general third assumption. According to the first assumption, individual 63 Health Equity and Financial Protection: Part I Table S5: Health Care Subsidies, Constant Unit Subsidy Assumption Outpatient Outpatient Inpatient Outpatient visits visits admissions Total visits (chc) (polyclinics) (hospital) (hospital) subsidies Mean subsidy 1 2.50 0.22 19.73 26.15 48.60 2 2.09 0.23 30.61 34.33 67.27 3 2.09 0.18 38.23 40.30 80.80 4 1.99 0.33 60.87 46.22 109.42 5 1.17 0.37 91.31 51.63 144.48 Total 1.97 0.27 48.15 39.72 90.11 Shares 1 25.3 16.8 8.2 13.2 10.8 2 21.3 17.3 12.7 17.3 14.9 3 21.2 13.6 15.9 20.3 17.9 4 20.2 24.7 25.3 23.3 24.3 5 11.9 27.5 37.9 26.0 32.1 Total 100.0 100.0 100.0 100.0 100.0 Share in the total subsidy 2.2 0.3 53.4 44.1 100.0 Concentration index 0.1167 0.1153 0.3050 0.1351 0.2203 Source: Authors. fees are not used at all when computing the public subsidy. According to the second assumption, the public subsidy is fully determined by the fees. In cases where these two assumptions appear to be too strong, the third assumption provides a convenient intermediary choice. More detail on how the individual subsidy is computed can be found in technical notes 8 to 10 in chapter 7. The first part of tables S3–S5 displays the average public subsidy received by each income quintile as well as by the whole population. The second part shows the same distribution in relative terms—namely, the share of total public subsidies received by each quintile for each health ser- vice analyzed. The share of the total public subsidy for each health service in the total public subsidy is then presented. Finally, these tables show the concentration index of the subsidies. A negative concentration index indi- cates that poorer individuals benefit more from public financing, whereas a positive value indicates that richer individuals are favored. Interpreting the Results Our example tables present the computed public subsidies (in thousands of Vietnamese dong) for outpatient visits to CHCs and polyclinics as well as outpatient visits and inpatient admissions to general hospitals. 64 Chapter 6: Interpreting the Tables and Graphs As shown in the first part of table S3, public subsidies generally decrease with income for outpatient visits in CHCs (from D7.68, on average, for the first quintile to D3.70 for the last). However, richer indi- viduals receive more public subsidies for outpatient care in public hospi- tals, with the highest quintile (133.37) receiving more than three times more than the first (39.93). Public subsidies for outpatient care in poly- clinics and inpatient care in general hospitals are only weakly positively related to income. The same pattern is shown in relative terms in the second part of table S3. For instance, the first quintile alone receives 27.8 percent of the public sub- sidies associated with outpatient visits in CHCs, while the fifth quintile receives 33 percent of subsidies associated with outpatient care in hospitals. Hospital care receives the bulk of public subsidies. This is split almost evenly between inpatient (50 percent) and outpatient (45.7 percent) care. Outpatient care in CHCs and polyclinics gets only 3.1 and 1.1 percent of total subsidies, respectively. Finally, concentration indexes show that public subsidies greatly favor the poor in the case of outpatient care in CHCs ( 0.1407) and favor the rich even more markedly in the case of outpatient care in hospitals (0.2471). Outpatient care provided in polyclinics and inpatient care in hospitals are slightly pro-rich. Overall, public subsidies are noticeably pro- rich (0.1201) due to the great share and substantial concentration index of outpatient hospital care, which more than offsets the pro-poor distribution of CHC services. Our example table S4 illustrates that when the proportional cost assumption is made, public subsidies are distributed (in relative terms) the same as health care fees (see technical note 10 in chapter 7 for more details). We thus find that while public subsidies for outpatient care in CHCs tend to decrease slightly with income, all other subsidies have a strong, positive correlation with income. The most striking increase occurs for hospital outpatient care, where the first quintile gets 3.3 percent of the total public subsidy and the fifth quintile gets 56.7 percent. Compared with the standard BIA (table S4), hospital care remains the most subsidized health service, but outpatient care (53.4 percent) receives noticeably more public funding than inpatient care (44.1 percent). This is due to the fact that more fees are collected for outpatient care, and the BIA proportional assumption allocates subsidies with respect to fees. Finally, since health care fees are substantially more concentrated among 65 Health Equity and Financial Protection: Part I the rich, all concentration indexes computed with the BIA proportional assumption are found to be more pro-rich than those computed with the standard BIA assumption. Our example table S5 illustrates that when the linear cost assumption is made, public subsidies are distributed (in relative terms) the same as health care utilization (see technical note 11 in chapter 7 for more details). This distribution always lies between the ones obtained with the standard BIA assumption (table S3) and the proportional cost assumption (table S4). It is not necessarily always the case, but in our example, the results obtained with the linear cost assumption are closer to those obtained with the standard BIA assumption. All concentration indexes are more pro-rich, and only the subsidies for outpatient care in CHCs remain distributed in favor of the poor. Overall, with a concentration index of 0.2203, total public subsidies are considerably more pro-rich than with the standard BIA assumption. Concentration of Public Health Services Concepts Graphs G3–G6 present the concentration curve of the public subsidy for each health service analyzed. Graph G3 relates to the distribution of subsi- dies computed with the standard BIA method, whereas graphs G4 and G5 describe the distributions of subsidies computed with the proportional and linear cost assumptions, respectively. Finally, graph G6 relates to the distri- bution of subsidies that were already computed in the data set. The curves represent the cumulative share of public subsidies according to the cumulative share of population, ranked by increasing consumption. For instance, the poorest 30 percent might benefit from 50 percent of total subsidies. These curves show how public subsidies vary according to consumption: the farther a curve is above the 45° line, the more the corre- sponding subsidy benefits the poorest households. For some subsidies the concentration curve might lie under the 45° line. In such cases, subsidies are more concentrated among the richest households. Interpreting the Results Example graphs G3–G5 correspond to the subsidies displayed in tables S3–S5 for outpatient care in CHCs and polyclinics and for both outpatient 66 Chapter 6: Interpreting the Tables and Graphs Graph G3: Concentration Curves of Public Health Care Subsidies, Standard BIA 100 80 cumulative % of subsidies 60 40 20 0 0 20 40 60 80 100 cumulative % of population, ranked from poorest to richest op. vis., CHCs op. vis., polyclinics op. vis., gen. hosp. inp. adm., gen. hosp. total subsidies line of equality Source: Authors. and inpatient care in general hospitals. In graph G3, the concentration curve corresponding to the subsidies granted for outpatient care in CHCs lies above the 45° line. This means that poorer households receive more subsidy than richer ones for this health service, which confirms the pro- poor concentration index (–0.1407) found in table S3. The concentra- tion curves related to outpatient care in polyclinics and inpatient care in general hospitals also lie above the 45° line, which is again consistent with the moderately pro-poor concentration indexes found for these health services. In contrast, the concentration curve for outpatient care in general hospitals lies below the 45° line, indicating, as in table S3, that richer households benefit much more from public subsidies for such care. Finally, the concentration curve of total subsidies lies slightly above the 45° line for the first quintile and farther under the line for richer 67 Health Equity and Financial Protection: Part I Graph G4: Concentration Curves of Public Health Care Subsidies, Proportional Cost Assumption 100 80 cumulative % of subsidies 60 40 20 0 0 20 40 60 80 100 cumulative % of population, ranked from poorest to richest op. vis., CHCs op. vis., polyclinics op. vis., gen. hosp. inp. adm., gen. hosp. total subsidies line of equality Source: Authors. households. Overall, the distribution of total public subsidies is thus slightly pro-rich. Consistent with table S4, graph G4 draws a very different picture of the distribution of public subsidies. In this graph, only the concentration curve for outpatient care in CHCs lies above the 45° line, whereas the concen- tration curves corresponding to all other health services lie under the line. This confirms that public subsidies computed according to the proportional cost assumption tend to be extremely pro-rich. Finally, graph G5 depicts the intermediate situation displayed in table S5, which occurs when the BIA linear cost assumption is made. 68 Chapter 6: Interpreting the Tables and Graphs Graph G5: Concentration Curves of Public Health Care Subsidies, Linear Cost Assumption 100 80 cumulative % of subsidies 60 40 20 0 0 20 40 60 80 100 cumulative % of population, ranked from poorest to richest op. vis., CHCs op. vis., polyclinics op. vis., gen. hosp. inp. adm., gen. hosp. total subsidies line of equality Source: Authors. References Hosseinpoor, A. R., E. van Doorslaer, N. Speybroeck, M. Naghavi, K. Mohammad, R. Majdzadeh, B. Delavar, H. Jamshidi, and J. Vega. 2006. “Decomposing Socioeconomic Inequality in Infant Mortality in Iran.” International Journal of Epidemiology 35 (5): 1211–19. O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. 69 Chapter 7 Technical Notes These technical notes are intended as a brief guide for users of ADePT Health Outcomes. They are drawn largely (and often with minimal changes) from O’Donnell and others (2008), which provides further information. Measuring Inequalities in Outcomes and Utilization Note 1: The Concentration Curve The concentration curve plots the cumulative percentage of the health sector variable (y axis) against the cumulative percentage of the popula- tion, ranked by living standards, beginning with the poorest and ending with the richest (x axis). In other words, it plots shares of the health sec- tor variable against shares of the living standards variable. For example, the concentration curve might show the cumulative percentage of health subsidies accruing to the poorest p percent of the population. If everyone, irrespective of his or her living standard, has exactly the same value of the health variable, the concentration curve will be a 45° line, running from the bottom left-hand corner to the top right-hand corner. This is known as the line of equality. If, by contrast, the health sector variable takes higher (lower) values among poorer people, the concentration curve will lie above (below) the line of equality. The farther the curve is above the line of equality, the more concentrated the health variable is among the poor. 71 Health Equity and Financial Protection: Part I Concentration curves for the same variable in different countries or time periods can be plotted on the same graph. Similarly, curves for different health sector variables in the same country and time period can be plotted against each other. For example, the analyst may wish to assess whether inpa- tient care is distributed more unequally than primary care. If the concentra- tion curve for one country (or time period or health service) lies everywhere above that for the other, the first curve is said to dominate the second and the ranking by degree of inequality is unambiguous.1 Alternatively, curves may cross, in which case neither distribution dominates the other. It is then still possible to compare degrees of inequality, but only by resorting to a sum- mary index of inequality, which inevitably involves the imposition of value judgments concerning the relative weight given to inequality arising at dif- ferent points in the distribution. The concentration index and the extended concentration index are such summary indexes and are described in the fol- lowing technical notes. Note 2: The Concentration Index Concentration curves (see technical note 1) can be used to identify whether socioeconomic inequality in some health sector variable exists and whether it is more pronounced at one point in time than another or in one country than another. But a concentration curve does not give a measure of the magnitude of inequality that can be compared conve- niently across many time periods, countries, regions, or whatever may be chosen for comparison. The concentration index (Kakwani 1977, 1980), which is directly related to the concentration curve, does quantify the degree of socioeconomic-related inequality in a health variable (Wagstaff, van Doorslaer, and Paci 1989; Kakwani, Wagstaff, and van Doorslaer 1997). It has been used, for example, to measure and to compare the degree of socioeconomic-related inequality in child mortality (Wagstaff 2000), child immunization (Gwatkin and others 2003), child malnutrition (Wagstaff, van Doorslaer, and Watanabe 2003), adult health (van Doorslaer and others 1997), health subsidies (O’Donnell and others 2007), and health care utilization (van Doorslaer and others 2006). Many other applications are possible. The concentration index is defined with reference to the concentration curve (see technical note 1). The concentration index is defined as twice the area between the concentration curve and the line of equality (the 45° 72 Chapter 7: Technical Notes line). If there is no socioeconomic-related inequality, the concentration index is 0. The convention is that the index takes a negative value when the curve lies above the line of equality, indicating a disproportionate concen- tration of the health variable among the poor, and a positive value when it lies below the line of equality. If the health variable is a “bad,” such as ill health, a negative value of the concentration index means that ill health is higher among the poor. Formally, the concentration index is defined as 1 C = 1 − 2∫ L h ( p ) dp. (7.1) 0 The index is bounded between –1 and 1. For a discrete living standards vari- able, it can be written as follows: n 2 1 C= N μh ∑ h r −1− N , i =1 i i (7.2) where hi is the health sector variable, mh is its mean, and ri = i/N is the frac- tional rank of individual i in the living standards distribution, with i = 1 for the poorest and i = N for the richest.2 The concentration index depends only on the relationship between the health variable and the rank of the liv- ing standards variable and not on the variation in the living standards vari- able itself. A change in the degree of income inequality need not affect the concentration index measure of income-related health inequality. The concentration index summarizes information from the concentration curve and can do so only through the imposition of value judgments about the weight given to inequality at different points in the distribution. It is possible to set alternative weighting schemes implying different judgments about atti- tudes to inequality by using the extended concentration index (see technical note 4). Inevitably, the concentration index loses some of the information that is contained in the concentration curve. The index can be 0 either because the concentration curve lies everywhere on top of the 45° line or because it crosses the line and the (weighted) areas above and below the line cancel out. It is obviously important to distinguish between such cases, and so the summary index should be examined in conjunction with the concentration curve. The sign of the concentration index indicates the direction of any rela- tionship between the health variable and position in the living standards distribution, and its magnitude reflects both the strength of the relationship and the degree of variability in the health variable. Although this is valu- able information, one may also wish to place an intuitive interpretation on 73 Health Equity and Financial Protection: Part I the value of the index. Koolman and van Doorslaer (2004) have shown that multiplying the value of the concentration index by 75 gives the percentage of the health variable that would need to be (linearly) redistributed from the richer half to the poorer half of the population (when health inequality favors the rich) to arrive at a distribution with an index value of 0. Properties of the Concentration Index The properties of the concentration index depend on the measurement characteristics of the variable of interest. Strictly, the concentration index is an appropriate measure of socioeconomic-related health (care) inequal- ity when health (care) is measured on a ratio scale with nonnegative values. The concentration index is invariant to multiplication of the health sector variable of interest by any scalar (Kakwani 1980). So, for example, if we are measuring inequality in payments for health care, it does not matter whether payments are measured in local currency or in dollars; the concentration index will be the same. Similarly, it does not matter whether health care is analyzed in terms of utilization per month or if monthly data are multiplied by 12 to give yearly figures. However, the concentration index is not invariant to any linear transformation of the variable of interest. Adding a constant to the variable will change the value of the concentration index. In many applications this does not mat- ter because there is no reason to make an additive transformation of the variable of interest. There is one important application in which this does represent a limitation, however. We are often interested in inequality in a health variable that is not measured on a ratio scale. A ratio scale has a true 0, allowing statements such as “A has twice as much X as B.” That makes sense for dollars or height. But many aspects of health cannot be measured in this way. Measurement of health inequality often relies on self-reported indicators of health. A concentration index cannot be com- puted directly from such categorical data. Although the ordinal data can be transformed into some cardinal measure and a concentration index can be computed for this (Wagstaff and van Doorslaer 1994; van Doorslaer and Jones 2003), the value of the index will depend on the transformation chosen (Erreygers 2005).3 In cross-country comparisons, even if all coun- tries adopt the same transformation, their ranking by the concentration index could be sensitive to differences in the means of health that are used in the transformation. 74 Chapter 7: Technical Notes A partial solution to this problem would be to dichotomize the categor- ical health measure. For example, one could examine how the proportion of individuals reporting poor health varies with living standards. Unfortunately, this introduces another problem. Wagstaff (2005) has demonstrated that the bounds of the concentration index for a dichoto- mous variable are not –1 and 1; instead, they depend on the mean of the variable. For large samples, the lower bound is m–1 and the upper bound is 1–m. So the feasible interval of the index shrinks as the mean rises. One should be cautious, therefore, in using the concentration index to com- pare inequality in, for example, child mortality and immunization rates across countries with substantial differences in the means of these vari- ables. An obvious response is to normalize the concentration index by dividing through by 1 minus the mean (Wagstaff 2005). If the health variable of interest takes negative as well as positive values, then its con- centration index is not bounded within the range of (–1,1). In the extreme, if the mean of the variable were 0, the concentration index would not be defined. Finally, Bleichrodt and van Doorslaer (2006) have derived the conditions that must hold for the concentration index (and related measures) to be a measure of socioeconomic-related health inequality consistent with a social welfare function. They argue that one condition—the principle of income- related health transfers—is rather restrictive. Erreygers (2006) has derived an alternative measure of socioeconomic-related health inequality that is consistent with this condition and three others argued to be desirable. Note 3: Sensitivity of the Concentration Index to the Living Standards Measure Alternative measures of living standards are often available in practice— consumption, expenditure, wealth index—and it is not always possible to establish a clear advantage of one measure over others. It is therefore important to consider whether the chosen measure of living standards influences the measured degree of socioeconomic-related inequality in the health variable of interest. When the concentration index is used as a sum- mary measure of inequality, the question is whether it is sensitive to the liv- ing standards measure. As noted, the concentration index reflects the relationship between the health variable and living standards rank. It is not influenced by the variance 75 Health Equity and Financial Protection: Part I of the living standards measure. In some circumstances, this may be considered a disadvantage. For example, it means that, for a given relationship between income and health, the concentration index cannot discriminate between the degree of income-related health inequality in one country in which income is distributed very unevenly and that in another country in which the income distribution is very equal. Yet, when one is interested in inequality at a certain place and time, it is reassuring that the differing variances of alter- native measures of living standards will not influence the concentration index. However, the concentration index may differ if the ranking of indi- viduals is inconsistent across alternative measures. Wagstaff and Watanabe (2003) demonstrate that the concentration index will differ across alternative living standards measures if the health variable is correlated with changes in an individual’s rank on moving from one measure to another. The difference between two concentration indexes C1 and C2, where the respective concentration index is calculated on the basis of a given ranking (r1i and r2i)—for example, consumption and a wealth index—can be computed by means of the following regression: 2 ⎛h ⎞ 2σ Δr ⎜ i ⎟ = α + γΔri + ε i , (7.3) ⎝μ⎠ where Δri = r1i–r2i is the reranking that results from changing the measure of socioeconomic status, and σ Δr is its variance. The ordinary least squares 2 (OLS) estimate of g provides an estimate of the difference (C1–C2). The significance of the difference between indexes can be tested by using the standard error of g.4 In practice, the choice of welfare indicator can have a large and signif- icant impact on measured socioeconomic inequalities in a health variable, but it depends on the variable examined (see, for instance, Lindelow 2006). Differences in measured inequality reflect the fact that consump- tion and the asset index measure different things or at least are different proxies for the same underlying variable of interest. But only in cases in which the difference in rankings between the measures is also correlated with the health variable of interest will the choice of indicator have an important impact on the findings. In cases in which both asset and con- sumption data are available, analysts are in a position to qualify any analy- sis of these issues by reference to parallel analysis based on alternative measures. However, data on both consumption and assets are often not 76 Chapter 7: Technical Notes available. In these cases, the potential sensitivity of the findings should be explicitly recognized. Note 4: Extended Concentration Index The regular concentration index C is equal to n 2 C= n⋅μ ∑h R i =1 i i − 1, (7.4) where n is the sample size, hi is the indicator of ill health for person i, m is the mean level of ill health, and Ri is the fractional rank in the living stan- dards distribution of the ith person (Kakwani, Wagstaff, and van Doorslaer 1997). The value judgments implicit in C are seen most easily when C is rewritten in an equivalent way as follows: n 2 C =1− n⋅μ ∑ h (1 − R ). i =1 i i (7.5) The quantity hi/nm is the share of health (or ill health) enjoyed (or suffered) by person i. This is then weighted in the summation by twice the complement of the person’s fractional rank—that is, 2(1 – Ri). So the poorest person has his or her health share weighted by a number close to 2. The weights decline in a stepwise fashion, reaching a number close to 0 for the richest person. The concentration index is simply 1 minus the sum of these weighted health shares. The extended concentration index can be written as follows: (ν −1) ν n C(ν ) = 1 − ∑ h i (1 − R i ) n ⋅ μ i =1 ν > 1. (7.6) In equation 7.6, u is the inequality-aversion parameter, which is explained below. The weight attached to the ith person’s health share, hi/nm, is now equal to u(1 – Ri)(u–1), rather than by 2(1 – Ri). When u = 2 the weight is the same as in the regular concentration index; so C(2) is the standard concen- tration index. By contrast, when u = 1 everyone’s health is weighted equally. This is the case in which the value judgment is that inequalities in health do not matter. So C(1) = 0 however unequally health is distributed across the income distribution. As u is raised above 1, the weight attached to the health of a very poor person rises, and the weight attached to the health of a person above the 55th percentile decreases. For u = 6, the weight attached to 77 Health Equity and Financial Protection: Part I the health of a person in the top two quintiles is virtually 0. When u is raised to 8, the weight attached to the health of a person in the top half of the income distribution is virtually 0 (see figure 7.1). Note 5: Achievement Index The measure of “achievement” proposed in Wagstaff (2002) reflects the average level of health and the inequality in health between the poor and the better off. It is defined as a weighted average of the health levels of the various people in the sample, in which higher weights are attached to poorer people than to better-off people. Thus, achievement might be measured by the following index: (ν − 1) 1 n I(ν ) = ∑ h iν (1 − R i ) , (7.7) n i =1 Figure 7.1: Weighting Scheme for Extended Concentration Index 4.0 3.5 3.0 2.5 weight 2.0 1.5 1.0 0.5 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 rank ν=1 ν = 1.5 ν=2 ν=3 ν=4 Source: Wagstaff 2002. 78 Chapter 7: Technical Notes which is a weighted average of health levels, in which the weights are as graphed in the preceding note and average to 1. This index can be shown to be equal to the following (Wagstaff 2002): I(v) = m[1–C(v)]. (7.8) When h is a measure of ill health (so high values of I(v) are considered bad) and C(v) < 0 (ill health is higher among the poor), inequality serves to raise the value of I(v) above the mean, making achievement worse than it would appear if one were to look at just the mean. If ill health declines monotonically with income, the greater is the degree of inequality aver- sion, and the greater is the wedge between the mean (m) and the value of the index I(v). Explaining Inequalities and Measuring Inequity Note 6: Demographic Standardization of Health and Utilization In the analysis of inequality in health and utilization, the basic aim of stan- dardization is to describe the distribution of health or utilization by living standards conditional on other factors, such as age and sex. This is referred to as the age-sex standardized distribution of health or utilization. It is inter- esting only if two conditions are satisfied: (a) the standardizing variables are correlated with the living standards, and (b) they are correlated with health or utilization. The purpose is not to build a causal, or structural, model to explain health or utilization. The analysis remains descriptive, but we sim- ply seek a more refined description of the relationship between health or uti- lization and living standards. There are two fundamentally different ways of standardizing, direct and indirect. Direct standardization provides the distribution of health or uti- lization across living standard groups that would be observed if all groups had the same age structure, for example, but had group-specific intercepts and age effects. Indirect standardization, however, “corrects” the actual dis- tribution by comparing it with the distribution that would be observed if all individuals had their own age, but the same mean age effect as the entire population. Although direct standardization requires that individuals be categorized into living standard groups, this is not necessary with the indi- rect method. 79 Health Equity and Financial Protection: Part I Both methods of standardization can be implemented through regression analysis. In each case, one can standardize for either the full or the partial correlations of the variable of interest with the standardizing variables. In the former case, only the standardizing, or confounding, variables are included in the regression analysis. In the latter case, nonconfounding vari- ables are also included, not to standardize these variables but to estimate the correlation of the confounding variables with health or utilization condi- tional on these additional variables. For example, take the case in which age is correlated with education and both are correlated with both health or uti- lization and income. If one includes only age in a health regression, then the estimated coefficient on age will reflect the joint correlations with educa- tion and, inadvertently, one would be standardizing for differences in edu- cation, in addition to age, by income. One may avoid this, if so desired, by estimating the age correlation conditional on education. Indirect Standardization The most natural way to standardize is by the indirect method, which pro- ceeds by estimating a health or utilization regression such as the following: yi = α + ∑ βj x ji + ∑ γ k z ki + ε i , (7.9) j k where yi is health or utilization; i denotes the individual; and a, b, and g are parameter vectors. The xj are confounding variables for which we want to standardize (for example, age and sex), and the zk are nonconfounding vari- ables for which we do not want to standardize but do want to control for in order to estimate partial correlations with the confounding variables. If we want to standardize for the full correlations with the confounding variables, the zk variables are left out of the regression. OLS parameter estimates ( ˆ ˆ ) α , βj ,γˆk , individual values of the confounding variables (xji), and sample – means of the nonconfounding variables (z k) are then used to obtain the pre- ˆx dicted, or “x-expected,” values of the health indicator yi : yiX = α + ∑ βj x ji + ∑ γˆk zk . ˆ ˆ ˆ (7.10) j k IS ˆ Estimates of indirectly standardized health (yi ) are then given by the dif- ference between actual and x-expected health or utilization, plus the over- – all sample mean (y ), 80 Chapter 7: Technical Notes yiIS = yi − yiX + y . ˆ ˆ (7.11) ˆ IS The distribution of yi (for example, across income) can be interpreted as the distribution of health or utilization that would be expected, irrespective of differences in the distribution of x across income. A standardized distribution of health or utilization across quintiles could be generated, for instance, by ˆ IS averaging yi within quintiles. Direct Standardization The direct method of standardization is more restrictive because it requires grouping health or utilization by categories of living standards. The regression- based variant of direct standardization proceeds by estimating, for each living standard group (g), an equation such as the following: yi = α g + ∑ βjg x ji + ∑ γ kg z ki + ε i , (7.12) j k which is a group-specific version of equation 7.9. OLS estimates of the (ˆ ˆ ) group-specific parameters α g , βjg ,γˆkg , sample means of the confounding variables (x j ), and group-specific means of the nonconfounding variables (zkg ) are then used to generate directly standardized estimates of the health or utilization variable yiDS as follows: ˆ yiDS = yg = α g + ∑ βjg x j + ∑ γˆk zkg . ˆ ˆ DS ˆ ˆ (7.13) j k This method immediately gives the standardized distribution of health or utilization across groups because there is no intragroup variation in the stan- dardized values. For grouped data, both the direct and indirect methods answer the ques- tion, what would the distribution of health or utilization across groups be if there were no correlation between health or utilization and demographics? But their means of controlling for this correlation is different. The direct method uses the demographic distribution of the population as a whole – (the x j), but the behavior of the groups (as embodied in β jg and γˆ jg). The ˆ indirect method employs the group-specific demographic characteristics – (the x jg), but the population-wide demographic effects (in β j and γˆk). The ˆ advantage of the indirect method is that it does not require any grouping and is equally feasible at the individual level. The results of the two methods 81 Health Equity and Financial Protection: Part I will differ to the extent that there is heterogeneity in the coefficients of x variables across groups because the indirect methods impose homogeneity, and the difference will depend on the grouping used in the direct method. Note 7: Decomposition of the Concentration Index For ease of exposition, we refer to any health sector variable, such as health, health care use, or health care payments, as “health” and to any (continuous) measure of socioeconomic status as “income.” Wagstaff, van Doorslaer, and Watanabe (2003) demonstrate that the health concentra- tion index can be decomposed into the contributions of individual factors to income-related health inequality, in which each contribution is the product of the sensitivity of health with respect to that factor and the degree of income-related inequality in that factor. For any linear additive regression model of individual health (yi), such as yi = α + ∑ βk x ki + ε i , (7.14) k the concentration index for y (C) can be written as follows: C = ∑ (β k xk / μ )Ck + GCε / μ , (7.15) k – where m is the mean of y, x k is the mean of xk, Ck is the concentration index for xk (defined analogously to C), and GCε is the generalized concentration index for the error term (ε). Equation 7.15 shows that C is equal to a weighted sum of the concentration indexes of the k regressors, where ⎛ x ⎞ the weight for xk is the elasticity of y with respect to xk ⎜ηk = βk k ⎟ . ⎝ μ⎠ The residual component—captured by the last term—reflects the income- related inequality in health that is not explained by systematic variation in the regressors by income, which should approach 0 for a well-specified model. The decomposition result holds for a linear model of health care. If a nonlinear model is used, then the decomposition is possible only if some lin- ear approximation to the nonlinear model is made. One possibility is to use estimates of the partial effects evaluated at the means (van Doorslaer, Koolman, and Jones 2004). That is, a linear approximation to equation 7.14 is given by 82 Chapter 7: Technical Notes yi = α m + ∑ βjm x ji + ∑ γ km z ki + u i , (7.16) j k where the β jm and γ k are the partial effects, dy/dxj and dy/dzk, of each vari- m able treated as fixed parameters and evaluated at sample means, and ui is the implied error term, which includes approximation errors. Because equation 7.16 is linearly additive, the decomposition result (Wagstaff, van Doorslaer, and Watanabe 2003) can be applied, such that the concentration index for y can be written as C = ∑ (β jm x j / μ )C j + ∑ (γ km zk /μ )Ck + GCu / μ . (7.17) j k Because the partial effects are evaluated at particular values of the vari- ables (for example, the means), this decomposition is not unique. This is the inevitable price to be paid for the linear approximation. Note 8: Distinguishing between Inequality and Inequity There typically is inequality in the utilization of health care in relation to socioeconomic characteristics, such as income. In high-income countries poorer individuals generally consume more health care resources as a result of their lower health status and greater need for health care. Obviously, such inequality in health care use cannot be interpreted as inequity. In low- income countries, the lack of health insurance and purchasing power among the poor typically means that their utilization of health care is less than that of the better off despite their greater need (Gwatkin 2003; O’Donnell and others 2007). In this case, the inequality in health care use does not fully reflect the inequity. To measure inequity, inequality in utilization of health care must be standardized for differences in need. After standardization, any residual inequality in utilization, by income for example, is interpreted as horizontal inequity, which could be pro-rich or pro-poor. Similarly, income- related inequality in health must also be standardized if we want to assess the extent of inequity that this involves. For instance, both income and health are quite naturally correlated with age. This is generally not deemed inequitable and should thus be taken out of total income-related income inequality to get a measure of horizontal inequity. As noted in technical note 7, if a health sector variable is specified as a linear function of determinants, then its concentration index can be 83 Health Equity and Financial Protection: Part I decomposed into the contribution of each determinant, computed as the product of the health variable’s elasticity with respect to the determinant and the latter’s concentration index. This makes it possible to explain socioeconomic-related inequality in health and health care utilization. In fact, the decomposition method allows horizontal inequity in utilization to be both measured and explained in a very convenient way. The concentra- tion index for need-standardized utilization is exactly equal to that which is obtained by subtracting the contributions of all need variables from the unstandardized concentration index (van Doorslaer, Koolman, and Jones 2004). Besides convenience, the advantage of this approach is that it allows the analyst to duck the potentially contentious division of determinants into need (x) and control (z) variables and so the determination of “justi- fied” and “unjustified,” or inequitable, inequality in health care utilization. The full decomposition results can be presented, and users can choose which factors to treat as x variables and which to treat as z variables. Benefit Incidence Analysis (BIA) Note 9: Public Health Subsidy in Standard BIA The service-specific public subsidy received by an individual can be expressed as follows: ski = qki ckj − fki , (7.18) where qki indicates the quantity of service k utilized by individual i, ckj rep- resents the unit cost of providing k in the region j where i resides, and fki rep- resents the amount paid for k by i (user fee). The total public subsidy received by an individual is as follows: si = ∑ γ k (qki ckj − fki ), (7.19) k where γk are scaling factors that standardize utilization recall periods across services. One might standardize on the recall period that applies for the service accounting for the greatest share of the subsidy. For example, where this is inpatient care, reported over a one-year period, then γk 1 for inpatient care and, for example, γk 13 for services reported over a four-week period. 84 Chapter 7: Technical Notes Some surveys ask the amount paid for each public health service. In this case, the public subsidy can be calculated as in equations 7.18 and 7.19. Alternatively, if the survey gives only the total amount paid for all public health services, then modify equation 7.19 to si = ∑ δ k qki ckj − fi , (7.19') k where fi is the payment for all public health care and dk is a scaling factor that standardizes the recall period for the utilization variables on the recall period that applies to the total payment variable. There is no assurance that Ski, estimated either by equation 7.19 or 7.19', will always be positive. The response to this problem is often to replace neg- ative estimates of Ski by 0 (see, for example, O’Donnell and others 2007, 96). This is not altogether satisfactory, and Wagstaff (2010) argues that the pres- ence of implied negative subsidies actually suggests that the constant cost assumption is unreasonable. Our ultimate interest is in how subsidies vary with household income. This can be measured using the concentration index, in which a positive value indicates a pro-rich distribution and a negative value indicates a pro- poor distribution (see technical note 2). Moreover, Wagstaff (2011) shows that the concentration index of subsidies to subsector k (CIsk), can be expressed in terms of the concentration indexes for utilization (CIq k) and fees (CIFk): Ck F CISk = CIqk − k CIFk , (7.20) Sk Sk where Sk and Fk are the sum of aggregate subsidies and fee revenues, respec- tively. The unit cost for subsector k (Ck) equals the sum of these two aggre- gates divided by the aggregate number of utilization units of subsector k: Sk + Fk Ck = . (7.21) ∑ i qki According to standard BIA, the concentration index of subsidies is smaller the less concentrated utilization is among the better off. But it is also smaller the more concentrated user fees are among the better off. In other words, government spending looks less pro-rich if the better off pay higher fees for a given number of units of utilization. In fact, subsidies could turn 85 Health Equity and Financial Protection: Part I out to be pro-poor if fees are sufficiently disproportionately concentrated among the better off, even if utilization is higher among the better off. Note 10: Public Health Subsidy with Proportional Cost Assumption The assumption being made in the standard approach to BIA is that each unit of utilization is associated with the same unit cost; the more fees that someone pays for a given unit of utilization, the smaller is the subsidy they receive. The reality may be quite different. It may well be that the better off pay higher fees precisely because they receive more services per unit of uti- lization; that is, they are charged according to the services they receive, not according to the number of units of utilization. In many (perhaps most) countries, user fees are explicitly linked to the quantity of services rendered, rather than being a flat rate for each unit of utilization (for example, each outpatient visit). Fee schedules also often reflect the cost of the services ren- dered; for example, higher fees are associated with more expensive drugs and tests. When fees reflect the quantity and costs of services rendered, the bet- ter off may well be paying more in fees (if they do pay more) because they get more—or more expensive—tests or drugs for a given outpatient visit or inpatient admission. An alternative to the standard BIA assumption would be that costs vary across individuals according to the fees paid (Wagstaff 2011). Expressing fees as the product of unit fees and utilization, we have Ski cki qki fki qki, (7.22) where qki indicates the quantity of service k utilized by individual i, cki rep- resents the unit cost of providing k to individual i, and fki represents the unit fees paid for k by i. As a first approximation, we could assume that unit fees and unit costs are proportionate to one another. Thus, cki αk fki, (7.23) where we expect αk to be larger than 1 given that utilization is subsidized. We have Ski αk fkiqki fki qki (αk 1)fki qki. (7.24) 86 Chapter 7: Technical Notes The fraction (αk 1) can be computed from aggregate data: Sk αk − 1 = , (7.25) Fk where Sk and Fk are the sum of aggregate subsidies and fee revenues, respec- tively. Hence, we have Sk S Ski = fki qki = k Fki , (7.26) Fk Fk so that total subsidies received by individual i are proportional to the fees paid, where the factor of proportionality is simply the ratio of subsidies to fees. Using this method, the estimated value of Ski is always nonnegative. Under the proportional cost assumption, the concentration index for subsidies is simply equal to the concentration index for fees: ClS = ClF . (7.27) k k Thus, in contrast to the standard BIA constant cost assumption, the more concentrated are fees among the better off, the greater is the pro-rich bias in the incidence of government spending. Note 11: Public Health Subsidy with Linear Cost Assumption The assumption being made in the standard approach to BIA is that each unit of utilization is associated with the same unit cost; the more fees that individuals pay for a given unit of utilization, the smaller is the subsidy they receive. The reality may be quite different. It may well be that the better off pay higher fees precisely because they receive more services per unit of utilization; that is, they are charged according to the services they receive, not according to the number of units of utilization. In many (perhaps most) countries, user fees are linked explicitly to the quantity of services rendered, rather than being a flat rate for each unit of utilization (for example, each outpatient visit). Fee schedules also often reflect the cost of the services ren- dered; for example, more expensive drugs and tests are associated with higher fees. When fees reflect the quantity and costs of services rendered, the bet- ter off may well be paying more in fees (if they do pay more) because they get more—or more expensive—tests or drugs for a given outpatient visit or inpatient admission. 87 Health Equity and Financial Protection: Part I An alternative to the standard BIA assumption would be that costs vary across individuals according to the fees paid (Wagstaff 2011). Expressing fees as the product of unit fees and utilization, we have Ski ckiqki fki qki , (7.28) where qki indicates the quantity of service k utilized by individual i, cki rep- resents the unit cost of providing k to individual i, and fki represents the unit fees paid for k by i. As a first approximation, we could assume that unit fees and unit costs are proportionate to one another. Thus, cki αk fki, (7.29) where we expect αk to be larger than 1 given that utilization is subsidized. We have Ski αk fkiqki fki qki (αk 1)fki qki. (7.30) The fraction (αk 1) can be computed from aggregate data: Sk αk − 1 = , (7.31) Fk where Sk and Fk are the sum of aggregate subsidies and fee revenues, respec- tively. Hence, we have Sk S Ski = fki qki = k Fki , (7.32) Fk Fk so that total subsidies received by individual i are proportional to the fees paid, where the factor of proportionality is simply the ratio of subsi- dies to fees. Using this method, the estimated value of Ski is always non- negative. Under the proportional cost assumption, the concentration index for subsidies is simply equal to the concentration index for fees: ClS = ClF . (7.33) k k Thus, in contrast to the standard BIA constant cost assumption, the more concentrated are fees among the better off, the greater is the pro-rich bias in the incidence of government spending. 88 Chapter 7: Technical Notes Notes 1. For an introduction to the concept of dominance, its relation to inequal- ity measurement, and the related concept of stochastic dominance, see Deaton (1997). 2. For large N, the final term in equation 7.2 approaches 0, and it is often omitted. 3. Erreygers (2005) suggests a couple of alternatives to the concentration index to deal with this problem. 4. This ignores the sampling variability of the left-hand-side estimates. References Bleichrodt, H., and E. van Doorslaer. 2006. “A Welfare Economics Foundation for Health Inequality Measurement.” Journal of Health Economics 25 (5): 945–57. Deaton, A. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore, MD: Johns Hopkins University Press. Erreygers, G. 2005. “Beyond the Health Concentration Index: An Atkinson Alternative to the Measurement of Socioeconomic Inequality of Health.” University of Antwerp, Antwerp. ———. 2006. “Correcting the Concentration Index.” University of Antwerp, Antwerp. Gwatkin, D. 2003. “Free Government Health Services: Are They the Best Way to Reach the Poor?” World Bank, Washington, DC. http://poverty .worldbank.org/files/13999_gwatkin0303.pdf. Gwatkin, D. R., S. Rustein, K. Johnson, R. Pande, and A. Wagstaff. 2003. Initial Country-Level Information about Socio-economic Differentials in Health, Nutrition, and Population. 2 vols. Washington, DC: World Bank. Kakwani, N. C. 1977. “Measurement of Tax Progressivity: An International Comparison.” Economic Journal 87 (345): 71–80. ———. 1980. Income Inequality and Poverty: Methods of Estimation and Policy Applications. New York: Oxford University Press. Kakwani, N. C., A. Wagstaff, and E. van Doorslaer. 1997. “Socioeconomic Inequalities in Health: Measurement, Computation, and Statistical Inference.” Journal of Econometrics 77 (1): 87–104. 89 Health Equity and Financial Protection: Part I Koolman, X., and E. van Doorslaer. 2004. “On the Interpretation of a Concentration Index of Inequality.” Health Economics 13 (7): 649–56. Lindelow, M. 2006. “Sometimes More Equal Than Others: How Health Inequalities Depend upon the Choice of Welfare Indicator.” Health Economics 15 (3): 263–80. O’Donnell, O., E. van Doorslaer, R. P. Rannan-Eliya, A. Somanathan, S. R. Adhikari, D. Harbianto, C. G. Garg, P. Hanvoravongchai, M. N. Huq, A. Karan, G. M. Leung, C. W. Ng, B. R. Pande, K. Tin, L. Trisnantoro, C. Vasavid, Y. Zhang, and Y. Zhao. 2007. “The Incidence of Public Spending on Healthcare: Comparative Evidence from Asia.” World Bank Economic Review 21 (1): 93–123. O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. van Doorslaer, E., and A. M. Jones. 2003. “Inequalities in Self-Reported Health: Validation of a New Approach to Measurement.” Journal of Health Economics 22 (1): 725–32. van Doorslaer, E., X. Koolman, and A. M. Jones. 2004. “Explaining Income- Related Inequalities in Doctor Utilization in Europe.” Health Economics 13 (7): 629–47. van Doorslaer, E., C. Masseria, X. Koolman, and OECD Health Equity Research Group. 2006. “Inequalities in Access to Medical Care by Income in Developed Countries.” Canadian Medical Association Journal 174 (2): 177–83. van Doorslaer, E., A. Wagstaff, H. Bleichrodt, S. Calonge, U. G. Gerdtham, M. Gerfin, J. Geurts, L. Gross, U. Hakkinen, R. E. Leu, O. O’Donnell, C. Propper, F. Puffer, M. Rodriguez, G. Sundberg, and O. Winkelhake. 1997. “Income-Related Inequalities in Health: Some International Comparisons.” Journal of Health Economics 16 (1): 93–112. Wagstaff, A. 2000. “Socioeconomic Inequalities in Child Mortality: Comparisons across Nine Developing Countries.” Bulletin of the World Health Organization 78 (1): 19–29. ———. 2002. “Inequality Aversion, Health Inequalities, and Health Achievement.” Journal of Health Economics 21 (4): 627–41. ———. 2005. “The Bounds of the Concentration Index When the Variable of Interest Is Binary, with an Application to Immunization Inequality.” Health Economics 14 (4): 429–32. 90 Chapter 7: Technical Notes ———. 2011. “Benefit Incidence Analysis: Are Government Health Expenditures More Pro-Rich Than We Think?” Health Economics, 20: n/a. DOI: 10.1002/hec.1727. Wagstaff, A., and E. van Doorslaer. 1994. “Measuring Inequalities in Health in the Presence of Multiple-Category Morbidity Indicators.” Health Economics 3 (4): 281–91. Wagstaff, A., E. van Doorslaer, and P. Paci. 1989. “Equity in the Finance and Delivery of Health Care: Some Tentative Cross-country Comparisons.” Oxford Review of Economic Policy 5 (1): 89–112. Wagstaff, A., E. van Doorslaer, and N. Watanabe. 2003. “On Decomposing the Causes of Health Sector Inequalities with an Application to Malnutrition Inequalities in Vietnam.” Journal of Econometrics 112 (1): 207–23. Wagstaff, A., and N. Watanabe. 2003. “What Difference Does the Choice of SES Make in Health Inequality Measurement?” Health Economics 12 (10): 885–90. 91 PART II Health Financing and Financial Protection Chapter 8 What the ADePT Health Financing Module Does The Health Financing module of ADePT allows users to analyze financial protection in health. Users are able to analyze the incidence of “catastrophic” out-of-pocket health spending—that is, spending exceeding a certain fraction of total household spending or just its nonfood spending. Users can also compute estimates of impoverishment due to out-of-pocket spending— that is, the effect on the estimated incidence and average depth of poverty associated with including or excluding out-of-pocket health spending from one’s measure of living standards. ADePT allows users to estimate the progressivity of all sources of health financing, including out-of-pocket spending but also private insurance, social insurance, and taxes. Users can also analyze the effects of health financing on income inequality—that is, the redistributive effect of health finance. ADePT can decompose the redistributive effect into (a) a progres- sivity component, (b) a horizontal inequity component (households with similar incomes paying different amounts for their health care), and (c) a reranking effect (households moving up or down the income distribution as a result of their health care payments). ADePT can do quite simple analysis as well as more sophisticated analysis. The more sophisticated features of ADePT are indicated with an asterisk. Users not familiar with the literature may wish to focus initially on sections without an asterisk. Except where stated, the summary in this chapter relies on O’Donnell and others (2008). 95 Health Equity and Financial Protection: Part II Financial Protection Health care finance in low-income countries is still characterized by the dominance of out-of-pocket payments and the relative lack of prepayment mechanisms, such as tax and health insurance. Households without full health insurance coverage face a risk of incurring large expenditures for medical care should they fall ill; this will affect their ability to purchase other goods and services that policy makers consider to be important, such as food and shelter. Two approaches have been used to get empirically at this idea. The first looks at payments that are catastrophic in the sense that they involve amounts of money that exceed some fraction of household consumption. The second asks whether the amount of money involved makes the differ- ence between a household being above or being below the poverty line in terms of the money it has available for things other than health care. Catastrophic Health Spending Households are classified as having out-of-pocket health spending that is catastrophic if it exceeds a certain fraction of consumption.1 The per- centage of the population being so classified is likely, of course, to depend on the threshold chosen. ADePT reports the percentage of the popula- tion experiencing catastrophic health spending for different thresholds; this is termed “the head count.” ADePT also allows users to plot a chart showing how the head count varies depending on the threshold used, as in figure 8.1. Policy makers might be concerned not just about the percentage of households exceeding the threshold but also about the amount by which they exceed it. This is analogous to the issue of poverty measurement: policy makers are concerned not just about the percentage of the population that falls below the poverty line but also about how far they fall below it. So, in addition to reporting the incidence of catastrophic health payments, ADePT Health Financing also reports the intensity of catastrophic health care payments, through a concept known as “the overshoot”: the larger the amount by which “overshooting” households exceed the threshold, the greater is the (average) overshoot. In figure 8.1, the area below both the 96 Chapter 8: What the ADePT Health Financing Module Does Figure 8.1: Health Payments Budget Share and Cumulative Percentage of Households Ranked by Decreasing Budget Share payments as share of expenditure total catastrophic overshoot O proportion H exceeding threshold 0 100 cumulative percent of population, ranked by decreasing payment fraction Source: O’Donnell and others 2008. curve and the threshold represents the total catastrophic overshoot. The “mean positive overshoot” is simply the average amount of overshoot among households overshooting the threshold. ADePT also gets at another possible concern of policy makers, namely, they might care more if the households experiencing catastrophic payments are poor than if they are rich. ADePT reports the concentration index for catastrophic payments, in which a negative value means that the house- holds reporting catastrophic payments are largely poor ones. In addition, ADePT reports a weighted head count index that weights the degree to which a household exceeds the threshold by its position in the income dis- tribution; it is the product of the head count and the complement of the concentration index. If the concentration index is negative, the weighted head count exceeds the unweighted head count. A policy maker who is averse to the poor disproportionately experiencing catastrophic payments might want to track the weighted head count rather than the unweighted one. ADePT undertakes the same exercise for the overshoot. 97 Health Equity and Financial Protection: Part II Poverty and Health Spending It is common to hear of households being impoverished as a result of large out-of-pocket spending—that is, in the absence of this large out-of-pocket spending, their living standards would have been high enough to keep them above the poverty line.2 The implicit assumption here is that out-of-pocket spending should not be seen as contributing to the household’s living stan- dards: rather, in using the money to purchase health care, the household has sacrificed what would have been purchased in the absence of the health shock that necessitated the health spending. For each household, ADePT compares aggregate consumption including out-of-pocket spending (what the household’s consumption would have been in the absence of the health shock) with aggregate consumption excluding out-of-pocket spending (what its consumption actually is). If the latter is below the poverty line and the former is above the poverty line, the household is impoverished as a result of the out-of-pocket spending. ADePT compares the poverty head count (the fraction of the population that is poor) with the two definitions of con- sumption; this allows users to get a sense of how much out-of-pocket spend- ing contributes to the poverty head count. Health spending could, of course, push a poor household even further into poverty. For each household, therefore, ADePT computes the shortfall in its consumption from the poverty line, again using two definitions of con- sumption: the first including out-of-pocket spending, the second excluding out-of-pocket spending. The poverty gap is the aggregate of all shortfalls from the poverty line. ADePT computes this for the two definitions of con- sumption to get a sense of how much the poverty gap is due to out-of-pocket spending. Progressivity and Redistributive Effect Catastrophic and impoverishing health payments speak to the issue of financial protection. An equally important issue is the fairness or equity of a country’s health financing arrangements. The amount people pay for health care through the various sources of financing—out-of-pocket pay- ments, private insurance, social insurance, and taxes—affects the amount of money they have to spend on things other than health care. Since out-of- pocket payments reflect—at least to a degree—health shocks, they are 98 Chapter 8: What the ADePT Health Financing Module Does typically seen as involuntary. So what people have to spend after paying for health care can be seen as a measure of discretionary income. The way a country finances its health care affects the distribution of discretionary income: if the financing system relies on regressive sources of financing (that is, ones that absorb a larger share of a poor household’s income than of a rich household’s), the health financing system will increase inequality in discretionary income; if the system relies on progressive sources, it will reduce inequality in discretionary income. Progressivity ADePT generates easy-to-understand tables and charts that help policy makers to see how progressive or regressive health care payments are.3 The basic approach is to see how the share of income paid toward health care varies across income groups—if it rises, health care payments are progres- sive; if it falls, they are regressive. ADePT also reports Kakwani’s progressivity index, which is based on a comparison of the income distribution (captured by the Lorenz curve) and the distribution of health care payments (captured by the payment concen- tration curve, which graphs the cumulative share of payments against the cumulative share of households, ranked in ascending order of income). If, as in figure 8.2, the distribution of payments is more unequal than the distri- bution of income (that is, the payment curve lies below the Lorenz curve), payments are progressive. They are regressive if payments are more equal than income, that is, the payment concentration curve lies above the Lorenz curve. The index is equal to twice the area between the two curves, with a positive sign in front of it if payments are progressive. It is equal to the dif- ference between the payment concentration index (twice the area between the payment concentration curve and the line of equality) and the Gini coefficient. Redistributive Effect* If health care payments are regressive, other things being equal, the distri- bution of income before health care payments is more equal than the distri- bution of income after health care payments—that is, the prepayment Gini 99 Health Equity and Financial Protection: Part II Figure 8.2: Kakwani’s Progressivity Index 100 cumulative proportion of health payments (%) 80 60 40 20 0 0 20 40 60 80 100 cumulative proportion of population (%) gross income Lorenz curve health payments concentration curve line of equality Source: Authors. coefficient is smaller than the postpayment Gini coefficient.4 By contrast, if payments are progressive, health care finance exerts an equalizing effect on income distribution—that is, the prepayment Gini coefficient is larger than the postpayment Gini coefficient. The size of the disequalizing or equalizing effect on income distribution is measured by the difference between the pre- payment and postpayment Gini coefficients. This redistributive effect depends on four things: • The progressivity of health care payments as measured by Kakwani’s index. • The share of income being absorbed by health care payments. The bigger the share, the greater is the effect on income inequality. Thus, coun- tries’ rankings by progressivity do not automatically mirror their rankings by redistributive effect. 100 Chapter 8: What the ADePT Health Financing Module Does • The degree of horizontal inequity in health care finance—that is, the degree to which households with a similar ability to pay end up spending similar amounts on health care. The source of inequity could arise in the tax system (the source of income might affect the tax rate), in social insurance contributions (members of one scheme may have a steeper contribution schedule than members of other schemes), in private insurance (risk-rated premiums may translate into the better off facing different premiums), or in out-of-pocket payments (people hit with multiple health problems at a given income level will likely end up paying more). Horizontal inequity is measured with reference to the inequality in postpayment income among groups of prepayment equals and is denoted by HI in equation 8.1. The larger the inequality in postpayment income among each group of prepayment equals, the greater is the degree of horizontal inequity. Each group’s inequality in postpayment income is weighted by the product of its population share and its postpayment income share. Horizontal inequity is always nonnegative by construction (the within-group Gini coefficients cannot be negative) and serves to reduce the equalizing effect of health care payments or amplify their disequalizing effect. In other words, horizontal inequity reduces the redistributive effect, defined as the difference between the prepay- ment and postpayment Gini coefficients. • The degree of reranking—that is, the extent to which households go up or down the distribution of discretionary income as a result of their payments for health care. The World Bank’s Voices of the Poor exer- cise recorded the case of a 26-year-old man in Lao Cai, Vietnam, who, as a result of the large health care costs necessitated by his daughter’s severe illness, moved from being the richest man in his community to being one of the poorest. Reranking is measured by the difference between the postpayment Gini coefficient and the post- payment concentration index, obtained by first ranking households by their prepayment income and then, within each group of prepay- ment income, ranking households by their postpayment income. Reranking also reduces the redistributive effect. The equation linking these concepts is, roughly speaking, RE [g/(1 g)] Kakwani HI Reranking, (8.1) 101 Health Equity and Financial Protection: Part II where RE is the Gini coefficient for prepayment income minus the Gini coefficient for postpayment income; g equals health care payments as a share of prepayment income; Kakwani is Kakwani’s progressivity index (computed on the assumption that households with similar “ability to pay” end up pay- ing the same); HI is the horizontal inequity index; and Reranking is the reranking index. ADePT reports each component of this equation for each source of financing as well as for total health care payments. Notes 1. For further details, see technical notes 12–15 in chapter 13; O’Donnell and others (2008, ch. 18). 2. For further details, see technical notes 16–18 in chapter 13; O’Donnell and others (2008, ch. 19). 3. For further details, see technical notes 19–20 in chapter 13; O’Donnell and others (2008, ch. 16). 4. For further details, see technical notes 21–22 in chapter 13; O’Donnell and others (2008, ch. 17). ADePT computes the decomposition terms using the approach outlined in van de Ven, Creedy, and Lambert (2001). References O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. van de Ven, J., J. Creedy, and P. J. Lambert. 2001. “Close Equals and Calculation of the Vertical, Horizontal, and Reranking Effects of Taxation.” Oxford Bulletin of Economics and Statistics 63 (3): 381–94. 102 Chapter 9 Data Preparation ADePT has no data manipulation capability. Hence, the data need to be prepared before they are loaded into ADePT. This chapter outlines the data needed by ADePT Health Financing for different types of analysis. The data requirements for the various analyses that ADePT Health Financing can do are summarized in table 9.1. An alternative way of read- ing the table is to see what analyses are feasible given the data available to ADePT users. ADePT works out what tables and charts can be produced given the data fields users have completed: tables and charts that are feasible are shown in black; those that are not feasible are shown in gray. In contrast to the case in the ADePT Health Outcomes module, the data requirements do not increase as the analysis becomes more sophisticated. Ability to Pay (Consumption) In a developing-country context, given the lack of organized labor markets and the high variability of incomes over time, household consumption (or at least expenditure) is generally considered to be a better measure of wel- fare and ability to pay than income. Therefore, in ADePT ability to pay is typically total household consumption. This should be gross of all payments toward health care. Out-of-pocket payments for health care may already be included in the measure of household consumption or expenditure; if not, they need to be added in before the data are loaded into ADePT. Indirect taxes are reflected in the consumption aggregate—the price that households pay 103 Health Equity and Financial Protection: Part II Table 9.1: Data Needed for Different Types of ADePT Health Financing Analysis National Health Ability to pay Prepayments Account data Topic and (typically Out-of-pocket Nonfood Poverty for health on health analysis consumption) payments consumption line care financing mix Financial protection Catastrophic payments ✓ ✓ ✓ Impoverishment ✓ ✓ ✓ Progressivity and redistributive effect Progressivity ✓ ✓ ✓ ✓ Redistributive effect* ✓ ✓ ✓ ✓ Source: Authors. Note: * A more advanced type of analysis. Prepayments include private insurance premiums, social insurance contribu- tions, and taxes. includes any indirect taxes. However, the household’s consumption aggre- gate reflects the resources it has at its disposal after the payment of direct taxes, social health insurance contributions, and private insurance premi- ums. Therefore, before the data are loaded into ADePT, the fraction of esti- mated direct tax payments that go to finance health care as well as social health insurance contributions and private insurance premiums should be added to the consumption aggregate. Out-of-Pocket Payments Out-of-pocket spending should include payments for all types of health care included in the National Health Account (NHA). This includes payments to government providers (which should include informal payments if possi- ble) as well as payments to private providers (which should include pay- ments to pharmacies). Estimates of out-of-pocket payments from survey data are potentially subject to both recall bias and small-sample bias owing to the infrequency with which some health care payments are made. Survey estimates of aggre- gate payments tend to show substantial discrepancies from production-side estimates, when the latter are available. Whether estimates of the distribu- tion, as opposed to the level, of out-of-pocket payments are biased depends on whether reporting of out-of-pocket payments is related systematically to 104 Chapter 9: Data Preparation ability to pay. Under the possibly strong assumption of no systematic misre- porting, survey data can be used to retrieve the distribution of payments, and mismeasurement of the aggregate level can be dealt with by applying a macro weight that gives the best indication of the relative contribution of out-of-pocket payments to total revenues. Nonfood Consumption In the case of low-income countries, it might be more appropriate to define catastrophic payments with respect to health payments as a share of expen- diture net of spending on basic necessities, notably food outlays. The latter has been referred to as “nondiscretionary expenditure” or “capacity to pay.” The difficulty lies in the definition of nondiscretionary expenditure. A com- mon approach is to use household expenditure net of food spending as an indicator of living standards. Poverty Line To compute poverty impacts, a poverty line needs to be established. Poverty lines are either relative or absolute. Relative lines are usually expressed as a percentage of mean or median consumption in the country in question or in a region of countries (for example, the European Union). More common in the developing world are absolute poverty lines. These define poverty in relation to an absolute amount of household expenditure per capita. Many countries have their own (absolute) poverty lines. Some are calculated by taking the cost of reaching subsistence nutritional requirements (for example, 2,100 calories a day) and then adding an allowance for nonfood expenditure (for example, adding the amount spent on nonfood by households achieving a calorific intake of 2,100 calories per person). Another approach is to use the international “dollar-a-day” poverty line developed by the World Bank but also used widely by the United Nations in tracking poverty. The dollar-a-day poverty line has, in fact, been updated recently to $1.25 a day. A second line is also used, namely, $2.00 a day. These are amounts in 2005 prices obtained using 2005 purchasing power parities (PPPs) for private consumption. For other years, the convention is 105 Health Equity and Financial Protection: Part II to deflate (or inflate) the poverty line by applying the local consumer price index (CPI). Suppose, for example, that the country’s PPP for private con- sumption in 2005 is 2.02—that is, 2.02 local currency units to the dollar. Then the $1.25-a-day poverty line is equivalent to 2.53 currency units a day in 2005. Suppose the available household data refer to 2000, and the CPI (with 2005 100) for 2000 is 78.05. Then the $1.25-a-day poverty line is equivalent to $1.97 in 2000 prices, that is, 1.25 2.02 0.7805. The 2005 PPP and CPI data are downloadable from the World Bank’s data website.1 ADePT users would do well to check the poverty calculations obtained for consumption gross of out-of-pocket payments with the World Bank’s “dol- lar-a-day” figures for the closest year available; this can be done using the Bank’s PovCal tool.2 Prepayments for Health Care Evaluation of the progressivity and the redistributive effect of health care financing requires examination of all sources of health sector funding, not simply those payments that are made exclusively for health care. So, in addi- tion to out-of-pocket payments, health insurance contributions, and ear- marked health taxes, the distributional burden of all direct and indirect taxes is relevant in cases in which, as is commonly true, some health care is financed from general government revenues. Social insurance contributions should also be considered. One source of revenue—foreign aid—is not rele- vant because the purpose of the analysis is to evaluate redistribution among the domestic population. Assuming that tax parameters have been set for the repayment of foreign loans, the distributional burden of foreign debt financing on the current generation will be captured through evaluation of the tax distribution. In summary, five main sources of health care finance should be considered: direct taxes, indirect taxes, social insurance, private insurance, and out-of-pocket payments. Survey data are unlikely to provide complete information on household tax and insurance payments. For example, income tax payments or social insurance contributions may not be explicitly identified, and payments through sales taxes almost certainly are not reported. Various approximation strategies are necessary. For example, direct tax and social insurance sched- ules can be applied to gross incomes or earnings, and the distribution of the sales tax burden can be estimated by applying product-specific tax rates to 106 Chapter 9: Data Preparation disaggregated data on the pattern of household expenditure. This type of exercise is time-consuming and even in industrial countries is hampered by the lack of relevant data (for example, the availability of net income rather than gross income data). Progressivity and redistributive effect analyses seek to determine the distribution of the real economic burden of health finance, not simply the distribution of nominal payments. So the incidence of payments—who incurs their real cost—must be established or assumed. For example, employer contributions to health insurance most likely result in lower wages received by employees. The extent to which this is true will depend on labor market conditions—in particular, the elasticities of labor demand and supply. Given that the incidence of taxes depends on market condi- tions, it cannot be determined through the application of universal rules. However, a fairly conventional set of assumptions is shown in table 9.2. NHA Data on Health Financing Mix ADePT provides users with the option of reweighting the sources of financ- ing using “macro weights”—that is, financing shares as recorded in the NHA table on mix of financing. For example, if the NHA data indicate that 20 per- cent of health expenditure is financed out of pocket, but the household data reveal that only 10 percent of the computed total comes from out-of-pocket payments, users have the option of scaling the out-of-pocket payments up (and other sources down) to mirror the NHA aggregate figures. Of course, this scales the payments of all households up and down by the same percentage, leaving the progressivity of each source unaffected; all that changes is the progressivity of the total. Table 9.2: Common Incidence Assumptions for Prepayments in Progressivity Analysis Payment toward health care Incidence Personal income and property taxes Legal taxpayer Corporate taxes Shareholder (or labor) Sales and excise taxes Consumer Employer social and private insurance contributions Employee Employee social insurance contributions Employee Individual private insurance premiums Consumer Source: O’Donnell and others 2008. 107 Health Equity and Financial Protection: Part II Notes 1. See http://databank.worldbank.org/ddp/home.do. 2. See http://iresearch.worldbank.org/PovcalNet/povcalNet.html. Reference O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. 108 Chapter 10 Example Data Sets Two data sets are used in this part of the manual to illustrate the analysis undertaken by ADePT. The analysis of financial protection uses data for Vietnam, and the analysis of progressivity and redistributive effect uses data for the Arab Republic of Egypt. Financial Protection: Vietnam The tables and graphs relating to catastrophic payments and poverty effects were produced using the 1998 Vietnam Living Standards Survey. The survey design is quite complex, and a detailed description can be found on the Living Standards Measurement Study pages on the World Bank website.1 In order to get nationally representative estimates, it is cru- cial to use the weights provided (wt9). In addition, proper inference also requires taking regional stratification into account (reg10o), as well as identifying the communes drawn as a primary sampling unit within the strata (commune). There are 5,999 households (hhw), the size of which is defined as the number of people living and eating meals together in the same dwelling (hhsize). Ability to Pay Ability to pay is measured by household yearly total consumption (exp). 109 Health Equity and Financial Protection: Part II Out-of-Pocket Payments Out-of-pocket payments are measured by medical out-of-pocket payments (net of health insurance reimbursements) during the 12 months prior to the survey (oop). Nonfood Consumption This is measured by household yearly nonfood consumption (nonfood1). Poverty Line The poverty line is calculated by taking the cost of reaching subsistence nutritional requirements (for example, 2,100 calories a day) and then adding an allowance for nonfood expenditure (for example, adding the amount spent on nonfood by households achieving a calorific intake of 2,100 calories per person). Progressivity and Redistributive Effect: Egypt All tables and graphs relating to progressivity and redistributive effect were produced with the 1997 Egypt Integrated Household Survey. In order to get nationally representative estimates, it is crucial to use the weights provided (hh_weight). In addition, proper inference also requires taking regional strat- ification into account (strata), as well as identifying the primary sampling unit (psu). Each of the 2,419 households is uniquely identified (hid), and its size is recorded (hhsize). Health care in Egypt is financed from various sources. As is common for developing countries, out-of-pocket payments contribute the greatest share of revenue, 52 percent in this case. The next biggest contribution— one-third—is from general government revenues. Social and private health insurance contribute 7 and 5.5 percent, respectively, and an ear- marked health tax on cigarette sales makes up the remaining 3 percent of revenues going toward the provision of health care. Ability to Pay Ability to pay is based on household expenditure gross of expenditures on health care whether through out-of-pocket payments or prepayments, 110 Chapter 10: Example Data Sets including taxes and social health insurance contributions. Ability to pay is defined as total household expenditure, plus tax payments that go toward financing health care, plus social health insurance contributions, all divided by the square root of household size, an often-used equivalence scale (HH_expenditure). Out-of-Pocket Payments These payments are measured by out-of-pocket medical expenses (oop) over the last 12 months. Prepayments for Health Care Prepayment variables recorded in the survey are as follows: • Direct personal taxes—that is, income, land, housing, and property taxes (direct_taxes) • Private health insurance premiums (private_insurance). Prepayment variables estimated from other survey information are as follows: • Sales and cigarette taxes approximated by applying rates to the cor- responding expenditures (cigarette_tax) • Social health insurance contributions estimated by applying contri- bution rates to earnings or incomes of covered workers or pensioners (social_ins_contributions). NHA Data on Health Financing Mix The National Health Account (NHA) shares of total health revenues in Egypt (1994–95) from various sources of finance are given in table 10.1. The table also shows which of the various sources of finance can be allocated, either directly or through estimation, using the survey data. In this example, as in most others, the main difficulty concerns allocation of the 33 percent of all health care finance that flows from general government revenues. Only direct personal and sales taxes, which account for only one-sixth of government revenues, can be allocated down to households. Nonetheless, 111 Health Equity and Financial Protection: Part II Table 10.1: Financing Assumptions, Egypt Example Share of total Method of Macro weights Finance source finance (%) allocation Case 1 Case 2 Case 3 General government revenues 32.98 Taxes 17.81 Income, capital gains, and property 0.78 Reported 0.0469 0.0552 0.0108 Corporate 4.83 Ventilated Other income, profit, and capital gains 0.62 Ventilated Domestic sales of goods and services 4.72 Estimated 0.2829 0.2825 0.0649 Import duties 3.64 Ventilated Other 3.22 Ventilated Nontax revenues 15.16 Earmarked cigarette tax 3.00 Estimated 0.0300 0.0300 0.0425 Social insurance 6.67 Estimated 0.0667 0.0667 0.0919 Private insurance 5.57 Reported 0.0557 0.0557 0.0768 Out-of-pocket payments 51.77 Reported 0.5177 0.5177 0.7132 Total 100.00 1.0000 1.0000 1.0000 Source: O’Donnell and others 2008. Note: “Ventilated” means that unallocated taxes are distributed as the weighted average of financing sources that can be allocated. it is possible to allocate down to households revenues that account for 72 percent of all health care finance. Incidence Assumptions for Health Care Payments We consider three sets of assumptions about the distribution of unallocated revenues: • Case 1. We assume that unallocated general government revenues are distributed as the weighted average of those taxes that can be allo- cated. Essentially, this involves inflating the weight given to the taxes that can be allocated. For example, the weight on domestic sales taxes is inflated from its actual value of 0.0472 of all health finance to a value of 0.2829 ( [4.72/5.5] 0.3298) to reflect the distribu- tion of unallocated revenues. • Case 2. We assume that “other income, profits, and capital gains taxes” are distributed as direct personal taxes and that import duties are distributed as sales taxes. It is assumed that the rest of the unallocated revenues are distributed as the weighted average of the allocated taxes. 112 Chapter 10: Example Data Sets • Case 3. We assume that unallocated revenues are distributed as the weighted average of all allocated payments (and not just allocated taxes). Note 1. See www.worldbank.org/lsms. Reference O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. 113 Chapter 11 How to Generate the Tables and Graphs This chapter explains how to set up ADePT Health Financing so as to gen- erate tables and graphs. The assumption is that the data set has been pre- pared before it is loaded into ADePT. The explanation proceeds with a screen shot of ADePT, with numbers on the screenshot corresponding to the steps outlined. 115 116 Financial Protection Screenshot 11.1: Financial Protection 1. Start up ADePT, select the Health Finance module, add a data set (click on the “add” button), and then label it (type a name in the box). Hint 1: If you have used ADePT before, your previous session will be reloaded. You have three options in this case. (1) Continue with the same data set, in which case just keep going. (2) Do the same type of analysis you were doing before, but for a different data set with vari- ables labeled the same. In this case, simply “remove” the data set, “add” your new data set, and continue. All the boxes in which you pre- viously entered information will still include the information. If, while running, ADePT finds that your new data set does not include some of the variables, it will mark them in red in the user interface. (3) Start with a new analysis and a new data set. In this case, simply choose Project -> Reset or hit ctl-R. Hint 2: You can load several data sets into ADePT at once. The variables you want to analyze will need to exist and be similarly named in both data sets. To facilitate this, check the “Enable only common variables” box; this will cause ADePT to show only the variables that appear in all the data sets you have loaded. 2. Click on the “variables” tab at the top left corner of the ADePT screen and provide a household size variable. 3. Provide household weights and provide the survey settings when relevant. 4. Enter the name of the household total consumption variable. This variable must be gross of all health care payments. Optionally, specify household total nonfood consumption—this is used in the catastrophic payments analysis. 5. Provide the (per capita) poverty line. It is possible to enter more than one poverty line for multiple analyses with different poverty lines; separate them by a space. 6. Indicate whether the tables should present quintiles or deciles. 7. Enter the household total out-of-pocket payments variable. 8. ADePT also produces basic tabulations of health payments according to other characteristics. Optionally, enter household-level variables in the left column of the corresponding section and various characteristics of the household head in the right column. 9. Select which tables and graphs are to be generated. Relevant figures for catastrophic payments and poverty analysis are tables TF1–TF5 and graphs GF1 and GF2. To obtain basic tabulations of the health payments, select tables T1 and T2 for household-level and household head characteristics. 10. Specify whether the standard error of each indicator is to be produced. This is required for inference but slows down computation. 11. Check the “frequencies” box if you want an additional page in the spreadsheet showing the frequencies (that is, number of cases) used to compute each statistic requested. 12. To produce a table or figure for a subset of cases, highlight the relevant table or graph and enter the relevant “if condition” in the “if con- dition” box. Hint: Each table or graph can have a different “if condition” assigned to it. 13. To produce all the analysis for just a subset of cases, instead click on the “filter” tab, check the “keep observations if” box, and enter the desired condition. 14. Hit the “generate” button to start the computation and generate the outputs. 117 118 Progressivity and Redistributive Effect Screenshot 11.2: Progressivity and Redistributive Effect 1. Start up ADePT, select the Health Finance module, add a data set (click on the “add” button), and then label it (type a name in the box). Hint 1: If you have used ADePT before, your previous session will be reloaded. You have three options in this case. (1) Continue with the same data set, in which case just keep going. (2) Do the same type of analysis you were doing before, but for a different data set with vari- ables labeled the same. In this case, simply “remove” the data set, “add” your new data set, and continue. All the boxes in which you pre- viously entered information will still include the information. If, while running, ADePT finds that your new data set does not include some of the variables, it will mark them in red in the user interface. (3) Start with a new analysis and a new data set. In this case, simply choose Project -> Reset or hit ctl-R. Hint 2: You can load several data sets into ADePT at once. The variables you want to analyze will need to exist and be similarly named in both data sets. To facilitate this, check the “enable only common variables” box; this will cause ADePT to show only the variables that appear in all the data sets you have loaded. 2. Click on the “variables” tab at the top left corner of the ADePT screen and provide a household size variable. 3. Provide household weights and provide the survey settings when relevant. 4. Enter the name of the household total consumption variable. This variable must be gross of all health care payments. Optionally, it is also possible to specify household total nonfood consumption; this is used in the catastrophic payments analysis. 5. Indicate whether the tables should present quintiles or deciles. 6. Enter the variable names for taxes, social health insurance contributions, private health insurance premiums, and out-of pocket payments. It is possible to specify more than one tax; separate them by a space. 7. Check the “use NHA weights” box if National Health Accounts (NHA) weights are to be used in aggregating across payment categories and enter NHA weights as fractions. Multiple weights can be used when there are multiple taxes, but they need to be entered in the same order as the tax variables. 8. ADePT also produces basic tabulations of health payments according to other characteristics. Optionally, enter household-level variables in the left column of the corresponding section and various characteristics of the household head in the right column. 9. Select which tables and graphs are to be generated. The relevant figures for progressivity and redistributive effect analysis are tables TP1–TP4 and graphs GP1–GP3. To obtain basic tabulations of health payments, select tables T1 and T2 for household-level and household head characteristics. 10. Specify whether the standard error of each indicator is to be produced. This is required for inference but slows down computation. 11. Check the “frequencies” box if you want an additional page in the spreadsheet showing the frequencies (that is, number of cases) used to compute each statistic requested. 12. To produce a table or figure for a subset of cases, highlight the relevant table or graph and enter the relevant “if condition” in the “if con- dition” box. Hint: Each table or graph can have a different “if condition” assigned to it. 13. To produce all the analysis for just a subset of cases, instead click on the “filter” tab, check the “keep observations if” box, and enter the desired condition. 14. Hit the “generate” button to start the computation and generate the outputs. 119 Chapter 12 Interpreting the Tables and Graphs As detailed in chapter 10, the tables and graphs for the analysis of financial protection were produced using data from the 2006 Vietnam Household Living Standards Survey, while those for the analysis of progressivity and redistributive effect were produced using data from the 1997 Integrated Household Survey for the Arab Republic of Egypt. Original Data Report Concepts The original data report gives basic statistics on the variables provided by users and shows in parentheses the field in which each variable was entered. When users enter National Health Account (NHA) weights, the sources of financing are rescaled to be consistent with their corresponding NHA aggregate. The rescaled sources of financing appear directly below the raw version of them. This report is important, as it makes it possible to detect many of the bold errors that may occur when preparing the data. For each variable, the first column shows the number of observations with a valid value (that is, those that are not represented by a missing value in Stata). The next three columns present the mean, minimum, and maxi- mum values of each variable. Column “p1” gives the first percentile or, in other words, the value that is greater than 1 percent of the values taken by 121 Health Equity and Financial Protection: Part II Original Data Report N mean min max p1 p50 p99 N_unique Egypt 1997 hhsize (Household size) 2,419 1.0 1.0 1.0 1.0 1.0 1.0 1 HH_expenditure (Total consumption) 2,419 6,886 504.6 100,956 1,503 5,567 26,334 2,418 hh_weight (Household weights) 2,419 1.0 0.8 1.5 0.8 0.9 1.5 5 direct_taxes 2,419 15.7 0.0 3,742 0.0 0.0 312 318 rescaled (Tax 1) 2,419 23.6 0.0 5,613 0.0 0.0 468 318 indirect_taxes 2,419 313 3.5 9,507 20.2 169 2,293 2,418 rescaled (Tax 2) 2,419 142 1.6 4,305 9.1 76.6 1,038 2,418 cigarette_tax 2,419 9.6 0.0 2,124 0.0 0.0 51.5 228 rescaled (Tax 3) 2,419 15.5 0.0 3,439 0.0 0.0 83.4 228 social_ins_contrib 2,419 28.4 0.0 765 0.0 6.2 186 987 rescaled (Social insurance contributions) 2,419 34.0 0.0 916 0.0 7.5 223 987 private_insurance 2,419 2.8 0.0 234 0.0 0.0 42.4 237 rescaled (Private insurance premiums) 2,419 29.2 0.0 2,474 0.0 0.0 449 237 oop 2,419 140 0.0 19,233 0.0 42.4 1,443 483 rescaled (Out-of-pocket) 2,419 274 0.0 37,675 0.0 83.1 2,827 483 region (Regions) 2,419 2.6 1.0 5.0 1.0 2.0 5.0 5 insured (Health insurance) 2,419 0.2 0.0 1.0 0.0 0.0 1.0 2 generated (Per capita consumption, gross) 2,419 6,886 505 100,955 1,504 5,567 26,336 2,418 generated (Per capita consumption, net) 2,419 6,367 400 100,712 1,381 5,191 23,376 2,418 the variable (and thus smaller than 99 percent of its values). Column “p50” represents the median, and “p99” represents the 99th percentile. The last column indicates the number of different values taken by each variable. Interpreting the Results Provided that they have been adequately prepared, the identification (ID) variables generally give the number of observations in the sample analyzed, which amounts to 2,419 in our example table. The out-of-pocket expendi- tures variable (oop) also has 2,419 observations and thus does not contain any missing values. Mean out-of-pocket expenditure amounts to LE 140 (Egyptian pounds) and ranges from LE 0 to LE 19,233. The median (LE 42.4) is smaller than the mean, which indicates the expected right-skewed distri- bution of health expenditure variables. The last column of the table usually makes it possible to identify categorical data. For instance, the variables “region” and “insured,” respectively, take five and two different values. 122 Chapter 12: Interpreting the Tables and Graphs Basic Tabulations Concepts Compared with ADePT’s original data report, which provides basic summary statistics on all variables specified, tables 1 and 2 provide more in-depth descriptive analysis of the variables related to sources of financing. These tables show the mean financing according to household (table 1) and indi- vidual (table 2) characteristics. ADePT also computes standard errors, which are presented in parentheses below their corresponding means. Under the assumption of normal distribution, subtracting twice the standard error from the mean, and then adding twice the standard error to the mean, yields the 95 percent confidence interval. This interval, which is reported in many standard Stata outputs, has a 95 percent chance of containing the true (and unknown) value of interest. Interpreting the Results Both tables are read in the same way. We show here only one example, table 1, which presents the relation between health financing and household Table 1: Sources of Finance by Household Characteristics Per capita Per capita Social Private Out-of- consumption consumption Direct Indirect Cigarette insurance insurance pocket Total net of gross tax tax tax contr. premiums payments payments payments Region 1 6,571 8.4 79.3 11.3 16.4 24.0 185.9 325.3 6,246 (324) (2.0) (6.0) (1.1) (2.3) (6.5) (30.4) (38.7) (299) 2 6,269 16.0 88.8 10.3 24.4 22.2 239.3 401.0 5,868 (234) (2.6) (6.3) (0.8) (2.2) (5.9) (30.7) (39.6) (219) 3 6,409 31.0 150.9 14.0 38.5 13.7 320.6 568.7 5,840 (380) (16.5) (19.0) (1.2) (4.6) (6.2) (48.7) (63.4) (354) 4 8,876 40.9 223.0 25.9 59.0 54.6 361.9 765.2 8,110 (646) (14.8) (24.2) (10.0) (7.2) (9.8) (118.6) (132.6) (546) 5 6,983 39.2 258.2 23.4 52.7 40.4 358.0 771.9 6,211 (546) (11.9) (40.6) (7.7) (5.8) (13.5) (91.5) (108.7) (482) Household with private insurance no 6,631 24.2 145.0 13.0 29.9 0.0 284.5 496.6 6,134 (206) (4.8) (11.6) (0.6) (2.2) (0.0) (34.2) (40.2) (180) yes 8,206 33.4 204.9 31.1 64.8 151.6 300.0 785.8 7,420 (354) (11.0) (17.3) (12.1) (4.1) (15.3) (40.3) (56.2) (320) Source: Author. Note: Numbers in parentheses are standard errors. 123 Health Equity and Financial Protection: Part II characteristics. There are two household characteristics: the region of the place of residence (five categories) and whether the household has private health insurance or not (binary variable). For instance, with LE 8,876 gross consumption per capita, the fourth region is substantially (and statistically significantly) richer than the other regions. All financing sources are also the greatest in this region, but not always statistically significantly so. Households having private health insurance are also significantly richer (LE 8,206) than the uninsured ones (LE 6,631). Financial Protection Concepts Tables F1 and F2 provide information on catastrophic health payments per quintile of gross income. The only difference between the two tables is that the former defines household ability to pay as the sum of all expenditures, whereas the latter defines it as the sum of nonfood expenditures. All meas- ures presented are based on the proportion of health payments in the household budget. The columns give different thresholds above which health payment budget shares might be deemed catastrophic. The first section of the tables displays the catastrophic payment head count (H), which represents the proportion of households with a health payment budget share greater than the given thresholds. This measure provides a simple way of assessing the incidence of catastrophic payments. The second section of the tables presents the catastrophic payment over- shoot (O), which shows the extent to which, on average, the household health payment budget share exceeds various thresholds. The overshoot is an average taken over all households in the quintile of gross income, irre- spective of their health payments. The information displayed in the last section is the mean positive overshoot (MPO), which measures the inten- sity of catastrophic payments—that is, the average excess of health pay- ment budget share of those households with catastrophic payments. Mean positive overshoot is thus the average payment excess, computed for the subsample of households with catastrophic payments in their quintile of gross income. Finally, a standard error is displayed in parentheses below each estimate. 124 Chapter 12: Interpreting the Tables and Graphs Table F1: Incidence and Intensity of Catastrophic Health Payments Threshold budget share 5% 10% 15% 25% 40% Head count (H) Lowest quintile 33.8 13.6 6.1 1.2 0.0 (2.23) (1.53) (0.96) (0.37) (0.00) 2 34.9 14.2 7.4 1.4 0.1 (1.92) (1.17) (0.93) (0.40) (0.13) 3 37.9 16.2 9.0 2.7 0.5 (1.86) (1.20) (0.87) (0.53) (0.20) 4 33.8 17.1 10.0 4.3 1.2 (1.82) (1.38) (1.01) (0.66) (0.38) Highest quintile 28.3 14.5 9.9 4.9 2.1 (1.41) (0.93) (0.82) (0.62) (0.43) Total 33.8 15.1 8.5 2.9 0.8 (1.03) (0.66) (0.47) (0.25) (0.12) Overshoot (O) Lowest quintile 1.9 0.8 0.4 0.1 0.0 (0.19) (0.12) (0.07) (0.02) (0.00) 2 2.2 1.0 0.5 0.1 0.0 (0.18) (0.13) (0.09) (0.04) (0.01) 3 2.6 1.4 0.7 0.2 0.0 (0.19) (0.14) (0.11) (0.06) (0.02) 4 2.9 1.7 1.1 0.4 0.1 (0.25) (0.20) (0.15) (0.09) (0.04) Highest quintile 3.0 2.0 1.4 0.7 0.2 (0.24) (0.21) (0.17) (0.12) (0.06) Total 2.5 1.4 0.8 0.3 0.1 (0.11) (0.08) (0.06) (0.03) (0.01) Mean positive overshoot (MPO) Lowest quintile 5.7 6.2 5.9 5.1 (0.35) (0.54) (0.75) (0.67) 2 6.2 7.1 6.3 7.2 7.6 (0.38) (0.60) (0.76) (1.95) (0.00) 3 7.0 8.5 8.4 8.4 8.2 (0.41) (0.64) (0.91) (1.58) (3.04) 4 8.7 10.1 10.9 9.8 6.5 (0.56) (0.78) (0.99) (1.43) (1.95) Highest quintile 10.6 13.8 14.1 14.5 9.8 (0.72) (1.03) (1.26) (1.51) (1.74) Total 7.5 9.2 9.6 10.5 8.5 (0.23) (0.35) (0.49) (0.75) (1.23) Source: Author. Interpreting the Results Our example table F1 shows that when the threshold is raised from 10 to 25 percent of total expenditure, the incidence of catastrophic payments in the lowest quintile of gross income falls from 13.6 to 1.2 percent, and 125 Health Equity and Financial Protection: Part II Table F2: Incidence and Intensity of Catastrophic Health Payments, Using Nonfood Threshold budget share 5% 10% 15% 25% 40% Head count (H) Lowest quintile 72.7 53.8 38.1 19.7 7.1 (2.73) (2.73) (2.26) (1.83) (1.06) 2 72.9 49.9 35.2 16.9 5.9 (1.85) (1.90) (1.92) (1.31) (0.85) 3 67.5 46.4 33.3 16.0 6.1 (1.75) (1.77) (1.69) (1.19) (0.72) 4 59.8 38.1 26.2 14.8 7.3 (2.06) (1.87) (1.75) (1.31) (0.90) Highest quintile 42.2 26.7 18.4 10.8 5.4 (1.52) (1.34) (1.13) (0.87) (0.64) Total 63.0 43.0 30.2 15.6 6.3 (1.18) (1.22) (1.06) (0.72) (0.46) Overshoot (O) Lowest quintile 10.9 7.8 5.5 2.8 0.9 (0.72) (0.62) (0.52) (0.35) (0.15) 2 9.8 6.8 4.6 2.2 0.6 (0.50) (0.44) (0.37) (0.25) (0.11) 3 9.6 6.8 4.8 2.4 0.8 (0.45) (0.39) (0.33) (0.23) (0.13) 4 8.4 6.1 4.5 2.5 1.0 (0.55) (0.48) (0.41) (0.30) (0.16) Highest quintile 6.2 4.5 3.4 2.0 0.9 (0.39) (0.34) (0.30) (0.23) (0.14) Total 9.0 6.4 4.6 2.4 0.8 (0.31) (0.26) (0.22) (0.15) (0.07) Mean positive overshoot (MPO) Lowest quintile 15.0 14.5 14.4 14.0 12.1 (0.73) (0.72) (0.81) (0.96) (1.01) 2 13.4 13.5 13.2 12.8 10.1 (0.56) (0.65) (0.74) (0.93) (1.11) 3 14.2 14.6 14.4 15.0 12.9 (0.53) (0.65) (0.70) (1.01) (1.47) 4 14.1 16.0 17.2 17.3 13.5 (0.71) (0.87) (1.02) (1.15) (1.39) Highest quintile 14.6 16.9 18.6 18.8 16.9 (0.79) (1.02) (1.21) (1.39) (1.39) Total 14.2 14.8 15.1 15.2 13.0 (0.36) (0.38) (0.43) (0.50) (0.56) Source: Author. the mean overshoot drops from 0.8 percent of expenditure to only 0.1 percent. Unlike the head count and the overshoot, the mean overshoot among those exceeding the threshold need not decline as the threshold is raised. Those in the lowest quintile of income spending more than 10 percent of total expenditure on health care spent, on average, 16.2 percent 126 Chapter 12: Interpreting the Tables and Graphs (10 percent 6.2 percent), while those spending more than 25 percent on health care spent 30.1 percent (25 percent 5.1 percent). No mean positive overshoot is displayed for the 40 percent threshold in the lowest income quintile because no catastrophic payments are observed in this case (that is, H 0). Overall, both the incidence and intensity of cata- strophic payments increase with income. The reason is that richer house- holds can spend a larger budget share on health care without having to cut spending on basic necessities. Finally, for a given threshold and gross income quintile, both the head count and the overshoot are higher, as they must be, when catastrophic pay- ments are defined with respect to health payments relative to nonfood expenditure (see table F2). Another difference is that, in table F2, the inci- dence of catastrophic payments decreases with income net of food expendi- ture. In other words, poorer households cut proportionally more nonfood expenditures to cope with health outlays. Distribution-Sensitive Measures of Catastrophic Payments Concepts Tables F3 and F4 relate catastrophic payment head count and overshoot to household income distribution. As in tables F1 and F2, each column represents a threshold above which household health payments might be considered catastrophic. The first line of these tables shows the concen- tration index of the incidence of catastrophic payments (CE). A positive value of CE indicates a greater tendency for the better off to exceed the payment threshold, whereas a negative value indicates a greater ten- dency for the worse off to exceed it. The concentration index of payment overshoot (CO) is a very similar measure. This indicates whether the average payment exceeding the threshold is greater among the better off (CO > 0) or among the poor (CO < 0). HW and HO adjust the cata- strophic payment head count (H) and overshoot (O), respectively, in order to make these measures sensitive to the distribution of income. For instance, when catastrophic payments are more frequent among the poor, HW is greater than H since, according to a social welfare perspective, such catastrophic payments are worse than if they were not at all related to income. 127 Health Equity and Financial Protection: Part II Table F3: Distribution-Sensitive Catastrophic Payments Measures Threshold budget share 5% 10% 15% 25% 40% Concentration index, C_E 0.031 0.023 0.092 0.283 0.566 (0.015) (0.022) (0.029) (0.044) (0.057) Rank-weighted head count, H_W 34.788 14.758 7.698 2.068 0.339 (1.2585) (0.7995) (0.5565) (0.2423) (0.0701) Concentration index, C_O 0.091 0.175 0.270 0.441 0.621 (0.023) (0.030) (0.036) (0.048) (0.076) Rank-weighted overshoot, O_W 2.298 1.144 0.593 0.169 0.025 (0.1162) (0.0776) (0.0499) (0.0224) (0.0067) Source: Author. Table F4: Distribution-Sensitive Catastrophic Payments Measures, Using Nonfood Threshold budget share 5% 10% 15% 25% 40% Concentration index, C_E 0.100 0.129 0.135 0.113 0.033 (0.010) (0.013) (0.016) (0.023) (0.036) Rank-weighted head count, H_W 69.336 48.539 34.339 17.405 6.543 (1.4569) (1.5202) (1.3077) (0.9355) (0.5931) Concentration index, C_O 0.104 0.098 0.083 0.045 0.036 (0.017) (0.020) (0.024) (0.032) (0.045) Rank-weighted overshoot, O_W 9.896 7.007 4.950 2.490 0.794 (0.3923) (0.3343) (0.2763) (0.1836) (0.0835) Source: Author. Interpreting the Results In the example tables F3 and F4, it is very clear that the distribution of cat- astrophic payments depends on whether health payments are expressed as a share of total expenditure or a share of nonfood expenditure. In the former case, catastrophic payments rise with total expenditure, with the only exception being the head count at the 5 percent threshold. As a result, the rank-weighted head count and overshoot are smaller than the unweighted indexes given in tables F1 and F2. But when health payments are assessed relative to nonfood expenditure, the concentration indexes are negative (with only one exception), indicating that the households with low non- food expenditures are more likely to incur catastrophic payments. As a con- sequence, the weighted indexes are larger than the unweighted indexes in tables F1 and F2. 128 Chapter 12: Interpreting the Tables and Graphs Measures of Poverty Based on Consumption Concepts Table F5 presents poverty measures corresponding to household expen- diture both gross and net of health payments. A standard error is dis- played in parentheses below each estimate. Even though caution has to be observed (see technical note 18 in chapter 13), comparison of gross and net measures is indicative of the scale of impoverishment due to health payments. The first line of the table shows the poverty head count, which represents the proportion of individuals living below the poverty line. The poverty gap gives the average deficit to reach the poverty line in the population. The normalized poverty gap is obtained by simply dividing the poverty gap by the poverty line. It is useful when making comparisons across countries with different poverty lines and currency units. Finally, the normalized mean positive poverty gap is a measure of poverty intensity—that is, the average poverty gap of the poor divided by the poverty line. Table F5: Measures of Poverty Based on Consumption Gross and Net of Spending on Health Care Gross of health Net of health payments payments Poverty line 941.8 Poverty head count 3.3 4.2 (0.52) (0.60) Poverty gap 4.9 6.6 (1.03) (1.19) Normalized poverty gap 0.5 0.7 (0.11) (0.13) Normalized mean positive poverty gap 15.8 16.8 (1.47) (1.26) Poverty line 1,883.5 Poverty head count 34.2 39.2 (1.56) (1.59) Poverty gap 160.5 192.3 (11.56) (12.47) Normalized poverty gap 8.5 10.2 (0.61) (0.66) Normalized mean positive poverty gap 24.9 26.0 (0.88) (0.87) Source: Author. 129 Health Equity and Financial Protection: Part II Interpreting the Results The first section of example table F5 uses a poverty line of D941,800 (Vietnamese dong) per year in 1998 prices. This corresponds to US$1.08 per person per day. When assessed on the basis of total household consumption, 3.3 percent of the population is estimated to be in poverty. If out-of-pocket (OOP) payments for health care are netted out of household consumption, this percentage rises to 4.2. So about 1 percent of the population is not counted as living in poverty but would be considered poor if spending on health care were discounted from household resources. This represents a sub- stantial rise of 30 percent in the estimate of poverty. The estimated poverty gap also rises almost 30 percent, from D4.9 to D6.6. The poverty gap increases from 0.5 percent of the poverty line to 0.7 percent when health payments are netted out of household consumption, but the normalized mean positive poverty gap increases only slightly (from 15.8 to 16.8). This suggests that the rise in the poverty gap is due mainly to more households being brought into poverty and not to a deepening of the poverty of the already poor. Finally, the second section of our example table uses a higher poverty line of D1,883.5. As expected, this results in a higher poverty head count and larger poverty gap. Share of Household Budgets Concepts Graph GF1 puts the household payments budget share in relation to the cumulative fraction of households ranked by decreasing value of household prepayment budget share. Two curves are represented: one with the budget share computed as out-of-pocket health expenditure divided by total expen- diture and another with the budget share expressed relative to household consumption net of food expenditure. This is a useful figure, as it shows how the catastrophic payment head count (the horizontal axis) depends on budget share threshold (the vertical axis). The flatter the curve, the more sensitive the head count is to choice of threshold. Interpreting the Results In our example, the curve using total expenditure shows that a threshold of 5 percent leads to a catastrophic payment head count of 33.8 percent. When 130 Chapter 12: Interpreting the Tables and Graphs Graph GF1: Health Payment Shares 0.8 health payments budget share 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 cumulative proportion of households ranked by decreasing health payments budget share OOP/total exp. OOP/nonfood exp. Source: Author. the threshold is raised to 10 percent, the proportion of households with cat- astrophic payments falls to 15.1 percent. If we further raise the threshold to 15 and 25 percent, the head count falls to 8.5 and 2.9 percent, respectively. These figures are displayed in table F1. Health Payments and Household Consumption Concepts Graph GF2 is a stylized version of the Pen’s parade representation of the income distribution. It charts household total consumption as a function of the cumulative proportion of households ranked in ascending order of total consumption. When health payments produce reranking in the income distribution, it is still possible to visualize the effect of health care 131 Health Equity and Financial Protection: Part II Graph GF2: Effect of Health Payments on Pen’s Parade of the Household Consumption 8 consumption as multiple of PL 6 4 2 0 0.0 0.2 0.4 0.6 0.8 1.0 cumulative proportion of population, ranked from poorest to richest pre-OOP consumption post-OOP consumption Source: Author. payments on the parade using what is referred to as a “paint drip” chart (Wagstaff and van Doorslaer 2003). The graph shows the Pen’s parade for household consumption gross of health payments. For each household, the vertical bar, or “paint drip” shows the extent to which health payments reduce consumption. If a bar crosses the poverty line, then a household is not poor on the basis of gross consumption, but is poor on the basis of net consumption. In other words, the household gets impoverished by health payments.1 Interpreting the Results Our example graph GF2 shows that health payments are largest at higher values of total consumption, but it is the households in the middle and lower half of the distribution that are brought below the poverty line by health payments. 132 Chapter 12: Interpreting the Tables and Graphs Progressivity and Redistributive Effect Concepts Table P1 shows household health financing and total consumption by quin- tile, with households ranked in ascending order of gross consumption. For each quintile, the first column displays the average household total expen- diture including health care payments, whereas the last column displays expenditure net of these. The other columns show the same information for each source of health finance along with total health financing. All financ- ing and consumption variables are expressed in terms of values per equiva- lent adult in order to take household size and economies of scale into account. The last line provides information for the whole population. Finally, a standard error is displayed in parentheses below each estimate. When NHA weights are applied, table P1 presents only the part of each financing that contributes to the health system. Interpreting the Results All example tables relating to progressivity and redistributive effect of health finance use the Egyptian data described in chapter 10. The first column in our example table P1 shows that the poorest quintile consumes, Table P1: Average Per Capita Health Finance Per capita Per capita Social Private Out-of- consumption consumption Direct Indirect Cigarette insurance insurance pocket Total net of gross tax tax tax contr. premiums payments payments payments Quintiles of per capita consumption, gross Lowest 2,725.3 3.1 38.9 9.0 15.2 9.9 102.3 178.4 2,546.9 quintile (39.13) (1.17) (2.30) (0.64) (1.43) (2.53) (13.38) (15.27) (39.60) 2 4,304.3 8.3 70.0 10.6 23.9 16.1 149.2 278.0 4,026.2 (17.52) (2.37) (2.84) (0.79) (1.72) (3.47) (13.41) (14.41) (21.41) 3 5,641.0 11.0 103.4 12.8 36.8 25.4 215.4 404.7 5,236.3 (20.42) (2.21) (3.83) (1.00) (3.56) (3.97) (20.50) (22.50) (27.98) 4 7,567.2 24.1 150.5 14.2 43.5 49.0 302.0 583.3 6,983.9 (41.63) (5.66) (5.41) (1.00) (2.81) (11.38) (36.43) (39.79) (49.89) Highest 14,516.9 83.8 423.0 36.7 66.0 54.4 669.2 1,333.0 13,183.9 quintile (419.74) (17.90) (40.18) (11.93) (5.38) (9.29) (130.12) (135.25) (381.14) Total 6,952.3 26.1 157.2 16.7 37.1 31.0 287.7 555.6 6,396.7 (196.95) (4.45) (10.83) (2.45) (2.06) (4.12) (31.34) (36.78) (174.69) Source: Author. 133 Health Equity and Financial Protection: Part II on average, LE 2,725.3 and the richest consumes LE 14,516.9. When the population is taken as a whole (last line of the table), equivalent gross con- sumption amounts to LE 6,952.3. Direct taxes appear to be borne mostly by the richest, as the first three quintiles contribute only LE 3.1, LE 8.3, and LE 11.0, on average, whereas the last two contribute LE 24.1 and LE 83.8, respectively. The average financing increases with quintile for all other sources of financing, but differences are, in general, less marked than for direct taxes. In the case of cigarette taxes, the richest quintile (LE 36.7) contributes only four times as much as the poorest one (LE 9.0). The optional NHA weights were applied in this example. The table thus dis- plays the entire sources of financing, irrespective of their final contribution to the health system. Progressivity of Health Financing Concepts Tables P2 and P3 allow us to analyze the progressivity of health financing. The first part of table P2 gives the average consumption and financing share, by quintile, with households ranked in ascending order of gross consumption. Information related to consumption gives an idea about income inequality: the greater the share of the richest quintiles, the greater is the inequality. The sources of financing show which part of the income dis- tribution bears the financing: the greater the share of the richest, the more concentrated is the financing among the rich or, put differently, the more pro-rich the financing. Table P3 is similar to table P2, with the only excep- tion being that it expresses health financing as a share of total gross con- sumption. This additional table thus shows the size of each source of financing in terms of its budget share. The second part of both tables provides measures of financing concen- tration and progressivity. The first line gives the prefinancing and postfi- nancing income inequality as measured by the Gini index. The second line shows the financing concentration index, which indicates how financing is related to income. A positive value indicates that the rich bear a greater share of the financing than the poor (that is, pro-poor financing), whereas a negative value indicates the opposite. A concentration 134 Chapter 12: Interpreting the Tables and Graphs Table P2: Shares of Total Financing Per capita Per capita Social Private Out-of- consumption consumption Direct Indirect Cigarette insurance insurance pocket Total net of gross tax tax tax contr. premiums payments payments payments Quintiles of per capita consumption, gross Lowest 7.8 2.4 4.9 10.8 8.2 6.4 7.1 6.4 8.0 quintile (0.73) (0.93) (0.72) (2.12) (1.24) (1.75) (1.33) (0.94) (0.73) 2 12.4 6.3 8.9 12.7 12.9 10.4 10.4 10.0 12.6 (0.80) (1.80) (0.81) (2.23) (1.29) (2.22) (1.40) (0.95) (0.81) 3 16.2 8.4 13.2 15.3 19.9 16.4 15.0 14.6 16.4 (0.96) (1.87) (1.31) (2.65) (1.95) (2.61) (2.14) (1.45) (0.94) 4 21.8 18.5 19.1 17.1 23.5 31.7 21.0 21.0 21.8 (0.97) (3.44) (1.24) (2.41) (1.50) (4.79) (2.87) (1.77) (0.96) Highest 41.8 64.4 53.9 44.0 35.6 35.1 46.6 48.0 41.2 quintile (2.19) (4.72) (3.18) (8.35) (2.99) (4.51) (5.71) (3.66) (2.14) Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Gini 0.3343 0.3321 coefficient (0.0127) (0.0124) Concentration 0.5843 0.4778 0.3282 0.2811 0.3332 0.3987 0.4161 Index (0.0438) (0.0271) (0.0970) (0.0209) (0.0448) (0.0550) (0.0313) Kakwani index 0.2500 0.1434 0.0061 0.0532 0.0011 0.0643 0.0818 (0.0424) (0.0192) (0.0972) (0.0183) (0.0445) (0.0515) (0.0259) Source: Author. Table P3: Financing Budget Shares Per capita Per capita Social Private Out-of- consumption consumption Direct Indirect Cigarette insurance insurance pocket Total net of gross tax tax tax contr. premiums payments payments payments Quintiles of per capita consumption, gross Lowest 100.0 0.1 1.4 0.3 0.6 0.4 3.8 6.5 93.5 quintile (0.00) (0.04) (0.08) (0.02) (0.05) (0.09) (0.49) (0.54) (0.54) 2 100.0 0.2 1.6 0.2 0.6 0.4 3.5 6.5 93.5 (0.00) (0.05) (0.06) (0.02) (0.04) (0.08) (0.31) (0.33) (0.33) 3 100.0 0.2 1.8 0.2 0.7 0.5 3.8 7.2 92.8 (0.00) (0.04) (0.07) (0.02) (0.06) (0.07) (0.36) (0.40) (0.40) 4 100.0 0.3 2.0 0.2 0.6 0.6 4.0 7.7 92.3 (0.00) (0.07) (0.07) (0.01) (0.04) (0.15) (0.48) (0.51) (0.51) Highest 100.0 0.6 2.9 0.3 0.5 0.4 4.6 9.2 90.8 quintile (0.00) (0.13) (0.22) (0.08) (0.03) (0.07) (0.88) (0.85) (0.85) Total 100.0 0.4 2.3 0.2 0.5 0.4 4.1 8.0 92.0 (0.00) (0.06) (0.12) (0.03) (0.03) (0.06) (0.42) (0.41) (0.41) Gini 0.3343 0.3321 coefficient (0.0127) (0.0124) Concentration 0.5843 0.4778 0.3282 0.2811 0.3332 0.3987 0.4161 Index (0.0438) (0.0271) (0.0970) (0.0209) (0.0448) (0.0550) (0.0313) Kakwani index 0.2500 0.1434 0.0061 0.0532 0.0011 0.0643 0.0818 (0.0424) (0.0192) (0.0972) (0.0183) (0.0445) (0.0515) (0.0259) Source: Author. 135 Health Equity and Financial Protection: Part II index not significantly different from 0 reflects no relationship between income and financing. The Kakwani index is the key information in tables P2 and P3. This index measures financing progressivity as the difference between the con- centration index and the gross consumption Gini index. A positive value reveals that financing is more concentrated among the rich than income, which indicates progressivity. A simpler way to think about progressivity is that the financing budget share (that is, financing divided by income) increases with income. Interpreting the Results The first part of our example table P2 shows that the poorest quintile con- sumes, on average, 7.8 percent of total consumption, whereas this amounts to 41.8 percent for the richest. Direct taxes appear to be borne mostly by the richest, as the first three quintiles contribute only 2.4, 6.3, and 8.4 percent, on average, whereas the last two contribute 18.5 and 64.4 percent, respec- tively. The financing share increases by quintile for all other sources of financing, but differences are, in general, less marked than for direct taxes. In the case of cigarette taxes, the richest quintile (44.0 percent) contributes only four times as much as the poorest one (10.8 percent). Table P3 shows that the largest sources of financing are out-of-pocket expenditure and indirect taxes, which, respectively, represent about 4 and 2 percent of gross consumption. All concentration indexes are positive, indicating that the better off contribute absolutely more to the financing of health care than do the poor. The concentration index is largest for direct taxes (0.5843) and smallest for social insurance contributions (0.2811), suggesting that direct taxes are the most progressive and social insurance contributions are the least so. The Kakwani indexes for both direct (0.2500) and indirect (0.1434) taxes are clearly positive, indicating progressivity. This is also the case, but a bit less marked, for out-of-pocket payments (0.0643). The Kakwani index is very close to 0 for cigarette taxes and private insurance and is moderately negative for social insurance contributions ( 0.0532), indicating regressiv- ity. The magnitude of the latter index is reduced by the near proportionality in the bottom half of the income distribution. This is difficult to see here, which is why graphs (such as graph GP2, for instance) are also necessary when assessing progressivity of financing. 136 Chapter 12: Interpreting the Tables and Graphs Decomposition of Redistributive Effect of Health Financing Concepts Table P4 provides a decomposition of the redistributive effect of health financing. As in table P2, the first part gives the average consumption and financing share by quintile, with households ranked in ascending order of gross consumption. The second part of the table starts by presenting the average financing budget share (g), that is, the average ratio of household financing to gross consumption, providing us with a measure of financing size. The second line gives the Kakwani index (Ke) of a counterfactual financing system in which horizontal inequity has been neutralized by averaging financing over gross income classes. Ke represents the progressivity that would be observed if financing were not causing any differential treatment of equals. The next four lines provide the decomposition of the redistributive effect of Aronson, Johnson, and Lambert (1994). The vertical effect, V, represents the change in income inequality that would be caused by financing in the Table P4: Decomposition of Redistributive Impact of Health Care Financing System Per capita Social Private Out-of- consumption Direct Indirect Cigarette insurance insurance pocket Total gross tax tax tax contr. premiums payments payments Quintiles of per capita gross consumption Lowest quintile 7.8 2.4 4.9 10.8 8.2 6.4 7.1 6.4 2 12.4 6.3 8.9 12.7 12.9 10.4 10.4 10.0 3 16.2 8.4 13.2 15.3 19.9 16.4 15.0 14.6 4 21.8 18.5 19.1 17.1 23.5 31.7 21.0 21.0 Highest quintile 41.8 64.4 53.9 44.0 35.6 35.1 46.6 48.0 Total100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Payments as fraction of Income (g) 1.0000 0.0037 0.0226 0.0024 0.0053 0.0045 0.0414 0.0799 Kakwani index, assuming horizontal equity (Ke) 0.2497 0.1440 0.0033 -0.0536 0.0017 0.0600 0.0795 Vertical effect (V) 9.40E-04 3.33E-03 8.00E-06 2.87E-04 7.40E-06 2.59E-03 6.91E-03 Horizontal inequality (H) 5.35E-05 6.79E-05 1.14E-05 2.37E-05 1.05E-04 1.03E-03 1.39E-03 Reranking (R) 7.57E-05 6.67E-06 6.83E-08 5.74E-06 4.21E-05 2.48E-03 3.33E-03 Total redistributive effect (RE = V - H - R) 8.10E-04 3.26E-03 1.94E-05 3.17E-04 1.55E-04 9.18E-04 2.18E-03 V / RE 1.1594 1.0229 0.4115 0.9071 0.0478 3.1638 H / RE 0.0660 0.0208 0.5849 0.0748 0.6797 0.6365 R / RE 0.0934 0.0020 0.0035 0.0181 0.2725 1.5273 Source: Author. 137 Health Equity and Financial Protection: Part II absence of differential treatment of equals. A positive effect indicates a reduction in inequality, whereas a negative value reflects an increase in inequality. H is a measure of horizontal inequity—that is, the increase in income inequality due to unequal treatment of households with the same prefinancing income. R represents the increase in income inequality due to reranking—that is, the change in the rank order of households in the income distribution that is caused by financing. The next line of the table displays the total redistributive effect (RE), which measures the overall change in income inequality resulting from financing. The total redistrib- utive effect equals the vertical effect minus horizontal inequality and reranking. Finally, since the decomposition effects are usually small, it is often easier to interpret their values relative to the total redistributive effect. This is displayed in the last three lines of the table. However, note that when V and RE have opposite signs, this ratio is misleading, and ADePT does not produce it. Interpreting the Results The first part of our example table P4 shows that the poorest quintile con- sumes, on average, 7.8 percent of total consumption, whereas the richest quintile consumes 41.8 percent. Direct taxes appear to be borne mostly by the richest: the first three quintiles contribute only 2.4, 6.3, and 8.4 percent, on average, whereas the last two contribute 18.5 and 64.4 percent, respec- tively. The financing share increases as the quintile rises for all other sources of financing, but differences are less marked than for direct taxes. In the case of cigarette taxes, the richest quintile (44.0 percent) contributes only four times as much as the poorest one (10.8). The second part of the table shows that out-of-pocket health expendi- ture is the greatest source of financing, as it represents 4.14 percent of the household budget, on average. With 2.26 percent, indirect taxes are the sec- ond largest source of financing, followed by social insurance contributions, which only amount to 0.53 percent of household gross income. In the absence of horizontal inequity, indirect taxes would have a Kakwani index of 0.1440, which indicates progressivity. Out-of-pocket payments would also be progressive, but less so (0.0600), whereas social insurance contributions would be regressive ( 0.0536). The total redistributive effect of indirect taxes shows a decrease in income inequality (3.26E-03). This is also the case for direct taxes (8.10E-04), 138 Chapter 12: Interpreting the Tables and Graphs whereas out-of-pocket health expenditure ( 9.18E-04) and social insurance contributions ( 3.17E-04) have the opposite effect. When decomposing the total redistributive effect, the decrease in income inequality due to indirect taxes is mostly a vertical effect, as the ratio V / RE is close to 1 (1.0229). Inequality is thus reduced because the rich pay more indirect taxes relative to their income. The ratio V / RE for direct taxes equals 1.1594, which means that the positive redistributive effect of direct taxes would be 16 percent greater in the absence of horizon- tal inequity (that is, H R). The negative redistributive effect (in other words, the increase in income inequality) caused by social health insurance, at 90.7 percent, is due to a negative vertical effect (V / RE 0.9071) and, at 9.3 percent, is due to horizontal inequity. Note that V / RE is much farther from 100 percent in the case of cigarette taxes and (even more so) private insurance premiums, indicating that for these sources there is con- siderable variation in the amount paid at a given level of income. Finally, since RE and V have opposite signs for out-of-pocket health expenditure, it is not possible to interpret the relative measures of the decomposition. The analysis of the corresponding absolute values shows that this financing has a strong positive vertical effect. This means that the rich contribute propor- tionally more than the poor, which is very likely due to a greater use of health care. Finally, the very strong horizontal inequity observed is likely to come from the health heterogeneity in the population. Concentration Curves Concepts Graphs GP1 and GP2 present the Lorenz curve for household total expendi- ture gross of health payments along with the concentration curve for each source of household health financing. The difference between these two graphs is that GP1 shows household taxes, whereas GP2 displays social insur- ance contributions, private insurance premiums, and out-of-pocket payments. The Lorenz curve shows the cumulative share of consumption accord- ing to the cumulative share of population ranked in ascending order of consumption. For instance, only 20 percent of total consumption might come from the poorest 30 percent of the population. This curve provides us with a visual representation of household inequality: the farther the curve is from the 45° line, the greater is the inequality. 139 Health Equity and Financial Protection: Part II Graph GP1: Concentration Curves for Health Payments, Taxes 100 80 cumulative % of payments 60 40 20 0 0 20 40 60 80 100 cumulative % of population, ranked from poorest to richest per capita consumption, gross direct tax indirect tax cigarette tax line of equality Source: Author. The concentration curves represent the cumulative share of health pay- ments according to the cumulative share of population, again ranked in ascending order of consumption. For instance, the poorest 30 percent might contribute only 10 percent to total taxes. These curves show how health financing varies according to consumption: the farther a curve is from the 45° line, the more the corresponding source of financing is borne by the richest households. For some sources of financing, the concentration curve might lie above the 45° line. In such cases, payments are more con- centrated among the poorest households. Furthermore, these graphs offer a powerful means of representing the effect of health financing on the distribution of household living standards. Indeed, whenever a concentration curve lies outside the Lorenz curve, this indicates progressivity. However, a formal test of statistical dominance is required to conclude this definitively (see O’Donnell and others 2008, ch. 7). 140 Chapter 12: Interpreting the Tables and Graphs Graph GP2: Concentration Curves for Health Payments, Insurance, Out of Pocket 100 cumulative % of payments 80 60 40 20 0 0 20 40 60 80 100 cumulative % of population, ranked from poorest to richest per capita consumption, gross social insurance contr private insurance premiums out-of-pocket payments line of equality Source: Author. Finally, it should be borne in mind that this kind of analysis does not con- sider utilization of health care. Progressivity should thus not be interpreted as the rich paying more for the same amount of health care, as this is most often not the case and not accounted for by the measures presented here. Interpreting the Results In our example graph GP1, the concentration curves for direct and indirect taxes appear to lie outside the Lorenz curve, suggesting that these are pro- gressive sources of finance. The curve for the earmarked cigarette tax appears to lie inside the Lorenz curve at lower levels of consumption but outside it at higher levels of consumption. This suggests regressivity in the first part of the consumption distribution and then progressivity for the richest households. However, the gap between the concentration and Lorenz curves is never wide. 141 Health Equity and Financial Protection: Part II Our example graph GP2 shows that the concentration curve for out-of- pocket payments lies outside the Lorenz curve, suggesting progressivity. Although the concentration curve for private insurance premiums lies below the Lorenz curve at lower consumption, the opposite is true at higher consumption. The concentration curve for social insurance contributions lies almost exactly on top of the Lorenz curve (indicating proportionality) up to the middle of the consumption distribution, but lies inside the Lorenz curve for the top half of the distribution. Social insurance premiums thus seem regressive, but only at the top of the distribution. Distribution of Health Payments Concepts Graph GP3 shows the average budget share of out-of-pocket health pay- ments (that is, health payments divided by total expenditure) by quintile of Graph GP3: Health Payment Shares by Quintiles 6 OOP as % of household expenditure 4 2 0 1 2 3 4 5 quintiles of per capita consumption, gross Source: Author. 142 Chapter 12: Interpreting the Tables and Graphs gross per capita consumption. This graph is a direct representation of the progressivity of health payments. These are progressive if their share of household consumption increases with consumption and are regressive in the opposite case. Finally, if their budget share does not vary with consump- tion, health payments are proportional to income. Interpreting the Results Our example shows that out-of-pocket payments are progressive over the first three quintiles, then stabilize, and finally become regressive for the richest quintile. Note 1. This interpretation is discussed in technical note 18 in chapter 13. References Aronson, J. R., P. Johnson, and P. J. Lambert. 1994. “Redistributive Effect and Unequal Tax Treatment.” Economic Journal 104 (423): 262–70. O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. Wagstaff, A., and E. van Doorslaer. 2003. “Catastrophe and Impoverishment in Paying for Health Care: With Applications to Vietnam 1993–98.” Health Economics 12 (11): 921–34. 143 Chapter 13 Technical Notes These technical notes are intended as a brief guide for users of ADePT Health Financing. They are drawn largely (and often with minimal changes) from O’Donnell and others (2008), which provides further information. Financial Protection Note 12: Measuring Incidence and Intensity of Catastrophic Payments Measures of the incidence and intensity of catastrophic payments can be defined analogously to those for poverty. The incidence of catastrophic pay- ments can be estimated from the fraction of a sample with health care costs as a share of total (or nonfood) expenditure exceeding the chosen thresh- old. The horizontal axis in figure 13.1 shows the cumulative fraction of households ordered by the ratio T/x from largest to smallest.1 The graph shows the fraction (H) of households with health care budget shares that exceed threshold z. This is the catastrophic payment head count. Define an indicator, E, which equals 1 if Ti/xi z and 0 otherwise. Then an estimate of the head count is given by 1 N H= ∑ Ei , N i =1 (13.1) where N is the sample size. 145 Health Equity and Financial Protection: Part II Figure 13.1: Health Payments Budget Share and Cumulative Percentage of Households Ranked by Decreasing Budget Share payments as % of expenditure total catastrophic overshoot O proportion H exceeding threshold 0 100 cumulative % of households, ranked by decreasing payment Source: O’Donnell and others 2008. This measure does not reflect the amount by which households exceed the threshold. Another measure, the catastrophic payment overshoot, captures the average degree by which payments (as a proportion of total expenditure) exceed threshold z. Define the household overshoot as Oi Ei[(Ti / xi) z]. Then the overshoot is simply the average: 1 N O= ∑ Oi . N i =1 (13.2) In figure 13.1, O is indicated by the area under the payment share curve but above the threshold level. It is clear that although H captures only the occurrence of a catastrophe, O captures the intensity of the occurrence as well. They are related through the mean positive overshoot, which is defined as follows: O MPO = . (13.3) H Because this implies that O H MPO, the catastrophic overshoot equals the fraction with catastrophic payments times the mean positive 146 Chapter 13: Technical Notes overshoot—the incidence times the intensity. Obviously, all of these can also be defined with x – f(x) as the denominator. Note 13: Distribution-Sensitive Measures of Catastrophic Payments All the measures introduced in technical note 12 are insensitive to the distri- bution of catastrophic payments. In the head count, all households exceeding the threshold are counted equally. The overshoot counts equally all dollars spent on health care in excess of the threshold, irrespective of whether they are made by the poor or by the rich. If there is diminishing marginal utility of income, the opportunity cost of health spending by the poor will be greater than that by the rich. If one wishes to place a social welfare interpretation on measures of catastrophic payments, then it might be argued that they should be weighted to reflect this differential opportunity cost. The distribution of catastrophic payments in relation to income could be measured by concentration indexes for Ei and Oi. Label these indexes CE and CO. A positive value of CE indicates a greater tendency for the bet- ter off to exceed the payment threshold; a negative value indicates that the worse off are more likely to exceed the threshold. Similarly, a positive value of CO indicates that the overshoot tends to be greater among the better off. One way of adjusting the head count and overshoot measures of catastrophic payments to take into account the distribution of payments is to multiply each measure by the complement of the respective concentration index (Wagstaff and van Doorslaer 2003). That is, the following weighted head count and overshoot measures are computed: HW H . (1 CE) (13.4) and OW O . (1 CO). (13.5) The measures imply value judgments about how catastrophic payments incurred by the poor are weighted relative to those incurred by the better off. The imposition of value judgments is unavoidable in producing any distribution-sensitive measure. In fact, it could be argued that a distribution- insensitive measure itself imposes a value judgment—catastrophic payments are weighed equally irrespective of who incurs them. The particular weight- ing scheme imposed by equation 13.4 is that the household with the lowest 147 Health Equity and Financial Protection: Part II income receives a weight of 2, and the weight declines linearly with rank in the income distributions so that the richest household receives a weight of 0. So if the poorest household incurs catastrophic payments, it is counted twice in the construction of HW, whereas if the richest household incurs cat- astrophic payments, it is not counted at all. A similar interpretation holds for equation 13.5. Obviously, different weighting schemes could be pro- posed to construct alternatives to these rank-dependent weighted head count and overshoot indexes. If those who exceed the catastrophic payments threshold tend to be poorer, the concentration index (CE) will be negative, and this will make HW greater than H. From a social welfare perspective and given the distrib- utional judgments imposed, the catastrophic payment problem is worse than it appears simply by looking at the fraction of the population exceeding the threshold because it overlooks the fact that it tends to be the poor who exceed the threshold. However, if it is the better-off individuals who tend to exceed the threshold, CE will be positive, and H will overstate the problem of the catastrophic payments as measured by HW. A similar interpretation holds for comparisons between O and OW. Note 14: Threshold Choice The value of threshold z represents the point at which the absorption of household resources by spending on health care is considered to impose a severe disruption to living standards. That is obviously a matter of judgment. Researchers should not impose their own judgment but rather should pre- sent results for a range of values of z and let readers choose where to give more weight. The value of z will depend on whether the denominator is total expenditure or nondiscretionary expenditure. Spending 10 percent of total expenditure on health care might be considered catastrophic, but 10 percent of nondiscretionary expenditure probably would not. In the literature, when total expenditure is used as the denominator, the most common threshold that has been used is 10 percent (Pradhan and Prescott 2002; Ranson 2002; Wagstaff and van Doorslaer 2003), with the rationale being that this represents an approximate threshold at which the household is forced to sacrifice other basic needs, sell productive assets, incur debt, or become impoverished (Russell 2004). Researchers from the World Health Organization have used 40 percent (Xu and others 2003) when “capacity to pay” (roughly, nonfood expenditure) is used as the denominator. 148 Chapter 13: Technical Notes Note 15: Limitations of the Catastrophic Payment Approach The idea underlying the catastrophic payments approach is that spending a large fraction of the household budget on health care must be at the expense of the consumption of other goods and services. This opportunity cost may be incurred in the short term if health care is financed by cutting back on current consumption or in the long term if it is financed through savings, the sale of assets, or credit. With cross-sectional data, it is impossible to dis- tinguish between the two. Besides this, there are other limitations of the approach. First, it identi- fies only the households that incur catastrophic medical expenditures and ignores those that cannot meet these expenses and so forgo treatment. Through the subsequent deterioration of health, such households probably suffer a greater welfare loss than those incurring catastrophic payments. Recognizing this, Pradhan and Prescott (2002) estimate exposure to, rather than incurrence of, catastrophic payments. Second, in addition to medical spending, illness shocks have catastrophic economic consequences through lost earnings. Gertler and Gruber (2002) find that in Indonesia earnings losses are more important than medical spending in disrupting household living standards following a health shock. Third, the approach depends on the measure of household resources, which can be household income or consumption measured by household expenditure. Of these, only income is not directly responsive to medical spending. That may be considered an advantage. However, the health payments-to-income ratio is not responsive to the means of financing health care, and that may be considered a disadvantage. Consider two households with the same income and health payments. Say, one household has savings and finances health care from savings, whereas the other has no savings and must cut back on current consumption to pay for health care. This difference is not reflected in the ratio of health payments to income, which is the same for both households. But the ratio of health payments to total household expenditure will be larger for the household without savings. Assuming that the opportunity cost of current consumption is greater, the “catastrophic impact” is greater for the household without savings, and, to an extent, this will be reflected if expenditure, but not income, is used as the denominator in the definition of catastrophic payments. Notwithstanding these limitations, medical spending in excess of a sub- stantial fraction of the household budget is informative of at least part of the 149 Health Equity and Financial Protection: Part II catastrophic economic consequences of illness, without fully identifying the welfare loss from lack of financing protection against health shocks. Note 16: Health Payments–Adjusted Poverty Measures Let T be per capita household out-of-pocket spending on health care, and let x be the per capita living standards proxy that is used in the standard assess- ment of poverty—household expenditure, consumption, or income. For convenience, we refer to the living standards variable as household expen- diture. Figure 13.2 provides a simple framework for examining the impact of out-of-pocket payments on the two basic measures of poverty—the head count and the poverty gap. The figure is a variant on Jan Pen’s “parade of dwarfs and a few giants” (see, for example, Cowell 1995). The two parades plot household expenditure gross and net of out-of-pocket payments on the y axis against the cumulative proportion of individuals ranked by expenditure on the x axis. For this stylized version of the graph, we assume that households Figure 13.2: Pen’s Parade for Household Expenditure Gross and Net of Out-of-Pocket Health Payments household expenditure per capita gross of OOP payment parade net of OOP payment parade poverty line (PL) C A B gross net cumulative proportion H H of population ranked by hhold. expenditure per capita Source: O’Donnell and others 2008. Note: OOP = out of pocket. 150 Chapter 13: Technical Notes keep the same rank in the distribution of gross and net out-of-pocket expen- diture. In reality, reranking will occur, as illustrated by ADePT graph F2. The point on the x axis at which a curve crosses the poverty line (PL) gives the fraction of people living in poverty. This is the poverty head count ratio (H). This measure does not capture the “depth” of poverty—that is, the amount by which poor households fall short of reaching the poverty line. A measure that does take the depth of poverty into account is the poverty gap (G), defined as the area below the poverty line but above the parade. Using household expenditure gross of out-of-pocket payments for health care, the poverty head count is Hgross and the poverty gap is equal to the area A. If out-of-pocket payments are subtracted from household expenditure before poverty is assessed, then the head count and gap must both rise—to Hnet and A B C, respectively. So Hnet Hgross is the fraction of individuals who are not counted as poor even though their household resources net of spend- ing on health care are below the poverty line. The respective underestimate of the poverty gap is B C. The poverty gap increases both because those already counted as poor appear even poorer once health payments are net- ted out of household resources (area B) and because some who were not counted as poor on the basis of gross expenditures are assessed as poor after out-of-pocket payments (area C) are taken into account. Let xi be the per capita total expenditure of household i. An estimate of the gross of health payments poverty head count ratio is ∑ sp N gross i =1 i i (13.6) H gross = , ∑ s N i =1 i where pigross = 1 if x i < PL and is 0 otherwise, si is the size of the household, and N is the number of households in the sample. Defining the gross of health payments individual-level poverty gap by g i = p i ( PL − x i ), the gross gross mean of this gap in currency units is ∑ sg N gross i =1 i i (13.7) G gross = . ∑ s N i =1 i gross The net of health payments head count is given by replacing p i with p inet = 1 if ( x i − Ti ) < PL (and 0 otherwise) in equation 13.6. The net of health payments poverty gap is given by replacing g igross in equation 13.7 with g inet = p inet ⎣PL − ( x i − Ti ) ⎦ . ⎡ ⎤ 151 Health Equity and Financial Protection: Part II When making comparisons across countries with different poverty lines and currency units, it is convenient to normalize the poverty gap on the poverty line as follows: Ggross NG gross = . (13.8) PL The net of payments normalized gap is defined analogously. The intensity of poverty alone is measured by the mean positive poverty gap, Ggross MPGgross = . (13.9) H gross In other words, the poverty gap (G) is equal to the fraction of the popula- tion who are poor (H) multiplied by the average deficit of the poor from the poverty line (MPG). The mean positive poverty gap can also be normalized on the poverty line. Note 17: Adjusting the Poverty Line It might be argued that if poverty is to be assessed on the basis of house- hold expenditure net of out-of-pocket payments for health care, then the poverty line should also be adjusted downward. This would be correct if the poverty line allowed for resources required to cover health care needs. Poverty lines that indicate resources required to cover only subsistence food needs clearly do not. Higher poverty lines may make some indirect allowance for expected health care needs, but they can never fully reflect these needs, which are inherently highly variable, both across individuals and across time. A common procedure for constructing a poverty line involves calculating expenditure required to meet subsistence nutrition requirements and the addition of an allowance for nonfood needs (Deaton 1997). More directly, the mean total expenditure of households just satisfy- ing their nutritional requirements may be used as the poverty line. Implicitly, this takes into account the expected spending on health care of those in the region of food poverty. But there is tremendous variation across households in health status and therefore in health care needs, which will not be reflected in the poverty line. This may be less of a problem in high-income countries, in which explicit income transfers exist to cover the living costs of disability. But such transfers seldom exist in low- income countries. Further, the health care needs of a given household are 152 Chapter 13: Technical Notes stochastic over time. A person falling seriously ill faces health care expenses well above the average. Meeting these expenses can easily force spending on other goods and services below the poverty threshold. So there is no reason to adjust a subsistence food poverty line, but higher poverty lines may make some implicit allowance for expected health care needs; in this case, it would make sense to adjust the poverty line downward when assessing poverty on expenditure net of health payments. One option is to adjust the poverty line downward by the mean health spending of households with total expenditure in the region of the poverty line (Wagstaff and van Doorslaer 2003). If that practice is adopted, then obvi- ously some households who spend less on health care than this average can be drawn out of poverty when it is assessed on expenditure net of health care payments. That practice is not advisable if comparisons are being made across countries or time and the standard poverty line has not been adjusted to reflect differences in mean health payments in the region with food poverty. For example, the World Bank poverty lines of $1 or $2 a day clearly do not reflect differences across countries in poor households’ exposure to health payments. Subtracting country-specific means of health spending from these amounts would result in lower poverty lines, and so less poverty, in countries that protect low-income households the least from the cost of health care. Note 18: On the Impoverishing Effect of Health Payments Under two conditions, the difference between poverty estimates derived from household resources gross and net of out-of-pocket payments for health care may be interpreted as a rough approximation of the impoverishing effect of such payments (Wagstaff and van Doorslaer 2003). These condi- tions are as follows: • Out-of-pocket payments are completely nondiscretionary. • Total household resources are fixed. Under these conditions, the difference between the two estimates would correspond to poverty due to health payments. Neither of the two conditions holds perfectly, though. A household that chooses to spend excessively on health care is not pushed into poverty by out-of-pocket pay- ments. In addition, a household may borrow, sell assets, or receive transfers 153 Health Equity and Financial Protection: Part II from friends or relatives to cover health care expenses. Household expendi- ture gross of out-of-pocket payments does not correspond to the consump- tion that would be realized in the absence of those payments. For these and other reasons, a simple comparison between poverty estimates that do and do not take into account out-of-pocket health payments cannot be interpreted as the change in poverty that would arise from some policy reform that elim- inated those payments. Nonetheless, such a comparison is indicative of the extent of the impoverishing effect of health payments. Progressivity and Redistributive Effect Note 19: Measuring Progressivity The most direct means of assessing progressivity of health payments is to examine their share of ability to pay as the latter varies. If we observe a ten- dency for this share to rise with total expenditure, this would indicate some degree of progressivity in financing. ADePT graph P3 can be used for such an analysis. A less direct means of assessing progressivity, defined in relation to departure from proportionality, is to compare shares of health payments con- tributed by proportions of the population ranked by ability to pay with their share of ability to pay—that is, to compare the concentration curve for health payments, LH(p), with the Lorenz curve for ability to pay, L(p). If payments toward health care always account for the same proportion of abil- ity to pay, then the share of health payments contributed by any group must correspond to its share of ability to pay. The concentration curve lies on top of the Lorenz curve. Under a progressive system, the share of health pay- ments contributed by the poor will be less than their share of ability to pay. The Lorenz curve dominates (lies above) the concentration curve. The opposite is true for a regressive system. ADePT Health Financing provides graphs P1 and P2 to analyze dominance. However, it is currently not possi- ble to test formally for stochastic dominance using ADePT.2 Lorenz dominance analysis is the most general way of detecting depar- tures from proportionality and identifying their location in the distribution of ability to pay. But it does not provide a measure of the magnitude of pro- gressivity, which may be useful when making comparisons across time or countries. Summary indexes of progressivity meet this deficiency but require 154 Chapter 13: Technical Notes the imposition of value judgments about the weight given to departures from proportionality at different points in the distribution (Lambert 1989). ADePT Health Financing applies the Kakwani index (Kakwani 1977), which is the most widely used summary measure of progressivity in both the tax and the health finance literatures (Wagstaff and others 1992, 1999; O’Donnell and others 2005).3 The Kakwani index is defined as twice the area between a payment con- centration curve and the Lorenz curve and is calculated as pK C G, where C is the concentration index for health payments and G is the Gini coefficient of the ability-to-pay variable.4 The value of pK ranges from 2 to 1. A negative number indicates regressivity; LH(p) lies inside L(p). A positive number indicates progressivity; LH(p) lies outside L(p). In the case of proportionality, the concentration curve lies on top of the Lorenz curve and the index is 0. The index could also be 0 if the curves were to cross and positive and negative differences between them cancel one another. Given this, it is important to use the Kakwani index, or any summary measure of progressivity, as a supplement to, and not a replace- ment for, the more general graphical analysis. Note 20: Progressivity of Overall Health Financing The progressivity of health financing in total can be measured by a weighted average of the Kakwani indexes for the sources of finance, where weights are equal to the proportion of total payments accounted for by each source. Thus, overall progressivity depends both on the progressivity of the different sources of finance and on the proportion of revenue collected from each of these sources. Ideally, the macro weights should come from the National Health Account (NHA). It is unlikely, however, that all sources of finance that are identified at the aggregate level can be allocated down to the household level from the survey data. Assumptions must be made about the distribution of sources of finance that cannot be estimated. Their dis- tributional burden may be assumed to resemble that of some other source of payment. For example, corporate taxes may be assumed to be distributed as income taxes. In this case, we say that the missing payment distribution has been allocated. Alternatively, we may simply assume that the missing payment is distributed as the weighted average of all the revenues that have been identified. We refer to this as ventilation. Best practice is to make such assumptions explicit and to conduct extensive sensitivity analysis. 155 Health Equity and Financial Protection: Part II Note 21: Decomposing Redistributive Effect One way of measuring the redistributive effect of any compulsory payment on the distribution of incomes is to compare inequality in prepayment incomes—as measured by, for instance, the Gini coefficient—with inequal- ity in postpayment incomes (Lambert 1989). The redistributive impact can be defined as the reduction in the Gini coefficient caused by the payment. Thus, RE GX GX P, (13.10) where GX and GX P are the prepayment and postpayment Gini coefficients, respectively; X denotes prepayment income or, more generally, some meas- ure of ability to pay; and P denotes the payment. Aronson, Johnson, and Lambert (1994) have shown that this difference can be written as RE V H R, (13.11) where V is vertical redistribution, H is horizontal inequity, and R is the degree of reranking. Because there are few households in any sample with exactly the same prepayment income, one needs to create artificial groups of prepayment equals, within intervals of prepayment income, to distinguish and compute the components of equation 13.11. The vertical redistribution component, which represents the redistribution that would arise if there were horizontal equity in payments, can then by defined as V GX G0, (13.12) where G0 is the between-groups Gini coefficient for postpayment income. This can be computed by replacing all postpayment incomes with their group means. V itself can be decomposed into a payment rate effect and a progressivity effect, ⎛ g ⎞ V =⎜ ⎟ KE, (13.13) ⎝ 1− g ⎠ where g is the sample average payment rate (as a proportion of income) and KE is the Kakwani index of payments that would arise if there were hori- zontal equity in health care payments. It is computed as the difference between the between-groups concentration index for payments and GX. In effect, the vertical redistribution generated by a given level of progressivity is “scaled” by the average rate g. 156 Chapter 13: Technical Notes Horizontal inequity is measured by the weighted sum of the group (j)–specific postpayment Gini coefficients, (GX − P ), where weights are given j by the product of the group’s population share and its postpayment income share (aj): H = ∑ α j GX − P . j (13.14) j Because the Gini coefficient for each group of prepayment equals is nonnega- tive, H is also nonnegative. Because it is subtracted in equation 13.11, hori- zontal inequity can only reduce redistribution, not increase it. This simply implies that any horizontal inequity will always make a postpayment distribu- tion of incomes more unequal than it would have been in its absence. Finally, R captures the extent of reranking of households that occurs in the move from the prepayment to the postpayment distribution of income. It is measured by R GX P CX P, (13.15) where CX P is a postpayment income concentration index that is obtained by first ranking households by their prepayment incomes and then, within each group of prepayment “equals,” by their postpayment income. Again R cannot be negative, because the concentration curve of postpayment income cannot lie below the Lorenz curve of postpayment income. The two curves coincide (and the two indexes are equal) if no reranking occurs. All in all, the total redistributive effect can be decomposed into four com- ponents: an average rate effect (g), the departure-from-proportionality or progressivity effect (KE), a horizontal inequity effect (H), and a reranking effect (R). Practical execution of this decomposition requires an arbitrary choice of income intervals to define “equals.” Although this choice will not affect the total H R, it will affect the relative magnitudes of H and R. In general, the larger the income intervals, the greater is the estimate of hori- zontal inequity and the smaller is the estimate of reranking (Aronson, Johnson, and Lambert 1994). That makes the distinction between H and R rather uninteresting in applications.5 More interesting are the quantifica- tion of the vertical redistribution (V), both in absolute magnitude and rel- ative to the total redistributive effect, and its separation into the average rate and progressivity effects. van Doorslaer and others (1999) decompose the redistributive effect of health finance for 12 Organisation for Economic Co-operation and Development countries. 157 Health Equity and Financial Protection: Part II Note 22: Redistributive Effect and Economic Welfare When health care payments are made voluntarily, they do not have a redis- tributive effect on economic welfare. Payments are made directly in return for a product—health care. It would not make sense to consider the welfare- reducing effect of the payments made, while ignoring the welfare-increasing effect of the health care consumption deriving from those payments. This begs the question of the extent to which out-of-pocket payments for health care should be considered voluntary. It might be argued that the moral com- pulsion to purchase vital health care for a relative is no less strong than the legal compulsion to pay taxes. But in most instances, there is discretion in the purchase of health care in response to a health problem. Notes 1. The figure is basically the cumulative density function for the reciprocal of the health payments budget share with the axes reversed. 2. For more detail on stochastic dominance, see O’Donnell and others (2008, ch. 7). 3. For other approaches to the measurement of progressivity, see Lambert (1989). 4. The concentration index is presented in great detail in the technical notes presented in chapter 7. 5. See Duclos, Jalbert, and Araar (2003) for an alternative approach that avoids this limitation and Bilger (2008) for an application of this method to health finance analysis. References Aronson, J. R., P. Johnson, and P. J. Lambert. 1994. “Redistributive Effect and Unequal Tax Treatment.” Economic Journal 104 (423): 262–70. Bilger, M. 2008. “Progressivity, Horizontal Inequality, and Reranking Caused by Health System Financing: A Decomposition Analysis for Switzerland.” Journal of Health Economics 27 (6): 1582–93. Cowell, F. A. 1995. Measuring Inequality. New York: Prentice Hall; London: Harvester Wheatsheaf. 158 Chapter 13: Technical Notes Deaton, A. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore, MD: Johns Hopkins University Press. Duclos, J.-Y., V. Jalbert, and A. Araar. 2003. “Classical Horizontal Inequity and Reranking: An Integrated Approach.” In Research on Economic Inequality: Fiscal Policy, Inequality, and Welfare, ed. J. A. Bishop and Y. Amiel, 65–100. Amsterdam, Elsevier. Gertler, P., and J. Gruber. 2002. “Insuring Consumption against Illness.” American Economic Review 92 (1): 51–76. Kakwani, N. C. 1977. “Measurement of Tax Progressivity: An International Comparison.” Economic Journal 87 (345): 71–80. Lambert, P. J. 1989. The Distribution and Redistribution of Income: A Mathematical Analysis. Cambridge, MA: Blackwell. O’Donnell, O., E. van Doorslaer, R. Rannan-Eliya, A. Somanathan, S. R. Adhikari, B. Akkazieva, D. Harbianto, C. G. Garg, P. Hanvoravongchai, A. N. Herrin, M. N. Huq, S. Ibragimova, A. Karan, S.-M. Kwon, G. M. Leung, J.-F. R. Lu, Y. Ohkusa, B. R. Pande, R. Racelis, K. Tin, L. Trisnantoro, C. Vasavid, Q. Wan, B.-M. Yang, and Y. Zhao. 2005. “Who Pays for Health Care in Asia?” EQUITAP Working Paper 1, Erasmus University, Rotterdam; Institute of Policy Studies, Colombo. O’Donnell, O., E. van Doorslaer, A. Wagstaff, and M. Lindelow. 2008. Analyzing Health Equity Using Household Survey Data: A Guide to Techniques and Their Implementation. Washington, DC: World Bank. Pradhan, M., and N. Prescott. 2002. “Social Risk Management Options for Medical Care in Indonesia.” Health Economics 11 (5): 431–46. Ranson, M. 2002. “Reduction of Catastrophic Health Care Expenditures by a Community-Based Health Insurance Scheme in Gujarat, India: Current Experiences and Challenges.” Bulletin of the World Health Organization 80 (8): 613–21. Russell, S. 2004. “The Economic Burden of Illness for Households in Developing Countries: A Review of Studies Focusing on Malaria, Tuberculosis, and Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome.” American Journal of Tropical Medicine and Hygiene 71 (2 supplement): 147–55. van Doorslaer, E., A. Wagstaff, H. van der Burg, T. Christiansen, G. Citoni, R. Di Biase, U. G. Gerdtham, M. Gerfin, L. Gross, U. Hakinnen, J. John, P. Johnson, J. Klavus, C. Lachaud, J. Lauritsen, R. Leu, B. Nolan, J. Pereira, C. Propper, F. Puffer, L. Rochaix, M. Schellhorn, G. Sundberg, 159 Health Equity and Financial Protection: Part II and O. Winkelhake. 1999. “The Redistributive Effect of Health Care Finance in Twelve OECD Countries.” Journal of Health Economics 18 (3): 291–313. Wagstaff, A., and E. van Doorslaer. 2003. “Catastrophe and Impoverishment in Paying for Health Care: With Applications to Vietnam 1993–98.” Health Economics 12 (11): 921–34. Wagstaff, A., E. van Doorslaer, S. Calonge, T. Christiansen, M. Gerfin, P. Gottschalk, R. Janssen, C. Lachaud, R. Leu, and B. Nolan. 1992. “Equity in the Finance of Health Care: Some International Comparisons.” Journal of Health Economics 11 (4): 361–88. Wagstaff, A., E. van Doorslaer, H. van der Burg, S. Calonge, T. Christiansen, G. Citoni, U. G. Gerdtham, M. Gerfin, L. Gross, U. Hakinnen, P. Johnson, J. John, J. Klavus, C. Lachaud, J. Lauritsen, R. Leu, B. Nolan, E. Peran, J. Pereira, C. Propper, F. Puffer, L. Rochaix, M. Rodriguez, M. Schellhorn, G. Sundberg, and O. Winkelhake. 1999. “Equity in the Finance of Health Care: Some Further International Comparisons.” Journal of Health Economics 18 (3): 263–90. Xu, K., D. B. Evans, K. Kawabata, R. Zeramdini, J. Klavus, and C. J. Murray. 2003. “Household Catastrophic Health Expenditure: A Multicountry Analysis.” Lancet 362 (9378): 111–17. 160 Index Figures, notes, and tables are indicated by f, n, and t, respectively. A graphs ability to pay, ADePT Health concentration curves, 139–42, Financing module 140f–141f data preparation, 103–4, 104t consumption-based poverty financial protection and, 109 measures, 129–30 progressivity and redistributive financial protection, 124–27, effect data and, 110 125t–126t table generation and interpretation, generation, 115–19 124–27, 125t–126t health payments and household achievement index, 78–79 consumption, 131–32, 132f activities of daily living, data health payments distribution, set example, 27–31 142–43, 142f ADePT Health Financing module, 95–102 household budget share, 130–31, 131f ability to pay. See ability to pay, ADePT original data report, 121–22 Health Financing module health payments. See under health budget share of household payments, payments 130–31, 131f, 134–36, 135t household identifiers catastrophic health spending. See under budget share of household payments, catastrophic health spending 130–31, 131f, 134–36, 135t consumption. See under consumption data preparation, 123–24, 123t data set preparation. See under data set identification variables, 121–22, 122t preparation income data. See under income data financial protection. See under financial living standards indicators. See under protection living standards indicators 161 Index ADePT Health Financing graphs module (continued) basic tabulations, 33–39, MPO (mean positive overshoot) 34f–35f, 43, 44t catastrophic health payments concentration index incidence and intensity decomposition, 54–55 measurements, 146–47, 146f concepts, 43 financial protection tables, health care utilization 124–27, 125t–126t concentration, 47–49, 48f NHA aggregate data. See under health care utilization National Health Account inequalities, 55–58 (NHA) aggregate data health inequalities, 45t, out-of-pocket payments. See out-of- 48–54, 49t–51t pocket payments, ADePT health outcome inequalities, 45–47 Health Financing module interpretation, 41–69 overshoot concept. See under original data report, 41, 42t overshoot concept public facilities utilization, 59–60 poverty measurements. See under public health services concentration, poverty measurements 66–68, 67f–69f quintile variables. See under public providers payments, 60–62 quintile variables results interpretation, 43, 44t tables, 115–19 subsidies, cost assumptions, 62–66, basic tabulations, 123–24, 123t 63t–64t consumption-based poverty health payments, 2, 95–102 measures, 129–30 household identifiers distribution-sensitive catastrophic data set example, 28 payments measurements, specification of, 15 127–28, 128t, 147–48 identification variables, 42, 42t financial protection, 124–27, income data. See under income data 125t–126t indirect living standards generation, 115–19 measurement, 17 health financing progressivity, living standards indicators. See under 134–36, 135t living standards indicators original data report, 121–22, 122t NHA aggregate data progressivity and redistributive benefit incidence analysis, 23 effect, 133–34, 133t data set example, 30–31 redistributive effect decomposition, poverty and health spending data, 98 137–39, 137t poverty measurements, 98 threshold choice, 148 progressivity, 98–99 ADePT Health Outcomes module, 1–2 quintile variables. See under basic properties, 5–11 quintile variables catastrophic health spending, redistributive effect, 99–102 96–97, 97f tables concentration curves. See under basic tabulations, 2, 33–39, 43, 44t concentration curves concentration index data set example, 27–31 decomposition, 54–55 data set preparation. See under concepts, 43 data set preparation health care utilization financial protection, 96 inequalities, 55–58 162 Index health inequalities, 45t, 48–54, 49t–51t health care utilization variables, 20–23 health outcome inequalities, 45–47 NHA aggregate subsidies data, 23 interpretation, 41–69 principles of, 9–12, 10t original data report, 41, 42t public health subsidies in, 84–86 public facilities utilization, 59–60 public providers fees, 23 public health services public subsidies cost assumptions, concentration, 66–69 62–66, 63t–64t, 67f public providers payments, 60–62 biased measurements, health outcomes results interpretation, 43, 44t and health care utilization subsidies, cost assumptions, 62–66, inequalities, 19–20 63t–64t binary variables, health care ADePT software, 1–2 utilization analysis, 20–21 ADePT survey settings body mass index (BMI), as anthropometric data set preparation, 21, 29 indicator, 18 prepayments for health care, health budget share of household financing data preparation, 106–7 payments, ADePT Health ADePT weight settings. See weight Financing module, 130–31, settings in ADePT 131f, 134–36, 135t adult health variables data set example, 28–31 C health outcomes analysis, 18–20 catastrophic health spending age variables ADePT Health Financing module concentration index distribution-sensitive measures, decomposition, 54–55, 54f 127–28, 128t, 147–48 health inequalities, graphic incidence and intensity representation of, 52–54 measurements, 145–47, 146f alternative assumptions, benefit incidence and intensity incidence analysis, 11 tables, 124–27, 125t–126t Analyzing Health Equity Using limitations of, 149–50 Household Survey Data, 2 ADePT Health Outcomes module, anthropometric indicators, health 96–97, 97f outcomes analysis, 18 charts, ADePT production of, 2 average health outcomes and care CHCs (commune health centers), utilization, inequality health care utilization trade off against, 8 inequalities, graphic representation of visits B to, 55–58, 56t, 58t basic tabulations variables child survival, health outcomes ADePT Health Financing measurements using, 17 module, 123–24, 123t commune health centers (CHCs), ADePT Health Outcomes module health care utilization inequalities, data set preparation, 21, 29 graphic representation of visits results and concepts, 43, 44t to, 55–58, 56t, 58t table and graph generation, concentration curves 33–39, 34f–35f ADePT Health Financing module, benefit incidence analysis 139–42, 140f–141f graphs and tables generation, budget share of household payments, 38–39, 38f–39f 130–31, 131f 163 Index concentration curves (continued) concentration curves, 139–42, ADePT Health Outcomes module 140f–141f health care utilization, graphic data preparation, 103–4, 104t interpretation, 47–49, 48f nonfood consumption, 105 inequality analysis, 6–8, 6f, 71–72 out-of-pocket payments, 104–5 outcome and utilization poverty line, 105–6 measurement, 71–72 poverty measures based on, public health care subsidies, 129–30, 129t 66–68, 67f–69f living standards indicators using, 16 progressivity measurements using, 154–55 continuous variables, anthropometric concentration index indicators, 18 average trade off against inequality, 8 cost assumptions, public subsidies benefit incidence analysis, public health data, table representation subsidies, 85–86 of, 62–66, 63t–64t of covariates, 51–54, 51t count variables, health care utilization decomposition of. See under analysis, 20–21 decomposition technique covariates, concentration index distribution-sensitive catastrophic of, 51–54, 51t payment measurement, 127–28, CPI (consumer price index), 128t, 147–48 ADePT health financing extended concentration data preparation, 105–6 index, 77–78, 78f health financing progressivity, D 134–36, 135t data set preparation health inequalities, graph and table ADePT Health Financing module interpretations, 48–54, 49t–51t basic components, 103–7 inequality analysis, 6–8, 6f original data report, 121–22, 122t inequity measurement and sample data sets, 109–13 inequality explanation, 8–9 ADePT Health Outcomes module, 1–2 properties of, 74–75 guidelines, 15–23 public facilities use, table original data report, 41, 42t interpretations of, 59–60, 59t sample data set, 27–31 public providers fees, table of summary, 15 payments, 60–62, 61t decomposition technique public subsidies cost assumptions, child survival analysis, 17 63t–64t, 65–66 concentration index sensitivity to living standards health care utilization measure, 75–77 inequalities, 57–58, 58t technical guidelines for, 72–75 linear health outcomes constant unit cost assumptions, public model, 49–50, 49t, 54–55, 54f subsidies, 62–66, 63t–64t technical guidelines, 82–83 consumer price index (CPI), ADePT health health care utilization analysis, 20–21 financing data preparation, 105–6 health determinants, 21–22 consumption. See also total household inequity measurement and inequality consumption explanation, 8–9 ADePT Health Financing module redistributive effect of health financing, catastrophic health spending impact 137–39, 138t, 156–57 on, 149–50 utilization determinants, 21 164 Index demographic variables F ADePT data set preparation, 21–22 financial protection in health inequalities, graphic ADePT Health Financing module explanation of, 49–54, 49f example data set, 109–10 inequality analysis, 7, 22 graph generation, 115, 116f–117f standardization of health and table generation, 124–27, 125t–126t utilization, 79–82 ADePT Health Outcomes module, dichotomous variables, applications, 96 concentration index, 75 catastrophic payment incidence direct measurement techniques and intensity measurements, demographic standardization of 145–47, 146f inequalities and inequity fitted linear model, graphic interpretation measures, 79–82 of health inequalities, 50–54, 50t living standards, 16 direct taxes G health care prepayments and, 106–7 gender variables health financing progressivity, concentration index decomposition, 134–36, 135t 54–55, 54f health financing redistributive health inequalities, graphic effects, 137–39, 138t representation of, 52–54 progressivity and redistributive Gini coefficient effect data and, 111 health financing progressivity, distribution-sensitive catastrophic payment 134–36, 135t measurement, ADePT Health health financing redistributive effect, Financing module, 127–28, 100–102, 137–39, 138t, 156–57 128t, 147–48 linear health inequalities models, graphic dollar-a-day poverty line representation of, 53–54 ADePT health financing data graphs. See under ADePT Health Financing preparation, 105–6 module; ADePT Health Outcomes adjustment of, 153 module dummy variables, anthropometric indicators, 18 H head count ratio, poverty measures based E on, 150f, 151–52 earnings, catastrophic health spending health achievement index, average impact on, 149–50 trade off against inequality, 8 economic welfare, redistributive health care utilization effect and, 158 benefit incidence analysis, 21–23 education indicators, ADePT data concentration, graphic representation set preparation, 22 and interpretation, 47–49, 48f Egypt, progressivity and data set example, 28–31 redistributive effect data for, demographic standardization of 110–13, 112t inequalities and inequity elasticity of health determinants, measures, 79–82 graphic interpretation of health determinants of, 22 inequalities, 51–54, 51t graph and table generation, 36–37, 36f–37f extended concentration index, graphic representation of inequalities in, 77–78, 78f 55–58, 56t, 58t 165 Index health care utilization (continued) household identifiers inequalities in, 5–8, 55–58, 56t, 58t ADePT Health Financing module public facilities use, table interpretations budget share of household of, 59–60, 59t payments, 130–31, 131f, variables in, 20–21 134–36, 135t health determinants data preparation, 123–24, 123t ADePT data set preparation, 21–22, 29 ADePT Health Outcomes module graphs and tables generation, 36–39, data set example, 28 36f–37f specification of, 15 inequalities, graphic explanation of, 49–54, 49f I health outcomes identification variables ADePT Health Outcomes module ADePT Health Financing, original applications, 1–2, 5–11 data report, 121–22, 122t benefit incidence analysis, 10–11, 10t ADePT Health Outcomes, original data set example, 28 data report, 42, 42t demographic standardization of incidence assumptions for health care inequalities and inequity payments, 112–13 measures, 79–82 income data graphic results and interpretation of ADePT Health Financing module inequalities in, 45–47, 45t catastrophic health spending by individual characteristics, 43, 44t impact on, 149–50 inequalities in, 5–8, 45–47, 45t distribution-sensitive catastrophic variables. See variables in health payment measurement, 127–28, outcomes analysis 128t, 147–48 health payments. See also redistributive effects of health financial protection financing, 137–39, 138t ADePT Health Financing module ADePT Health Outcomes module adjusted poverty measures, analysis, 5–11 150–52, 150f concentration index, 75 poverty and, 98 health care utilization inequalities, progressivity of, 98–102, 134–36, 135t graphic representation redistributive effect of, 98–102, of, 55–58, 56t, 58t 137–38, 138t health financing redistributive ADePT Health Outcomes module effect, 100–102 applications, 2, 95–102 health outcomes inequalities, graphs catastrophic health spending, 96–97, 97f and tables, 45t, 46–54, 49t financial protection and, 96 as living standards indicator, 16 impoverishing effect of, 153–54 public providers fees, table progressivity analysis. See under of payments, 60–62, 61t progressivity analysis indirect measurement techniques height-for-age variable, as anthropometric demographic standardization of indicator, 18 inequalities and inequity horizontal inequality measures, 80–82 health financing redistributive living standards, 17 effect, 100–102 indirect standardization, inequity redistributive effect of health measurement and inequality payments, 156–57 explanation, 9 166 Index indirect taxes living standards indicators health care prepayments and, 106–7 ADePT Health Financing module health financing progressivity, 134–36, 135t catastrophic health spending redistributive effects of health impact, 149–50 financing, 137–39, 138t concentration curves, inequality analysis 140–42, 140f–141f average trade off against, 8 financial protection data concentration index, 72–75 set example, 109–10 demographic standardization of health ADePT Health Outcomes module and utilization, 79–82 concentration index graphic results and interpretation, sensitivity to, 75–77 49–54, 49t–51t data set example, 28 health financing redistributive data set preparation, 15–17 effect, 101–2 health care utilization determinant, 21 health outcomes and health care quintile variables, 21–22 utilization, 6–8, 6f Lorenz curve graphic and table results and ADePT Health Outcomes concentration interpretation, 45–54, 45t curves, 139–42, 140f–141f variables measurement, 17–20 health financing, 99 inequity measurement vs., 8–9, 83–84 progressivity measurements using, 154–55 inequality aversion Lorenz dominance analysis, progressivity analysis of, 7 measurements using, 154–55 health outcomes inequalities, graphs and tables, 45t, 46–54 M inequity measurement macro weights, health financing data, 107 demographic standardization of health malnutrition, anthropometric indicators, 18 and utilization, 79–82 marginal benefit incidence analysis, 11n9 health outcomes and health care mean health spending of households, utilization inequalities, 8–9, 22 poverty line adjustment and, 153 inequality analysis vs., 8–9, 83–84 mean positive overshoot (MPO), ADePT Health Financing module K catastrophic health payments incidence Kakwani’s progressivity index and intensity measurements, ADePT Health Financing module 146–47, 146f applications, 155 financial protection tables, 124–27, health financing, 99, 100f, 134–36, 125t–126t 135t, 137–39, 138t measure of goodness fit (R2), overall health financing, 155 linear health inequalities redistributive effect of health models, graphic representation payments, 156–57 of, 50t, 53–54 measurement techniques L health outcomes analysis, 17–20 linear cost assumptions, public subsidies, living standards indicators, 16–17 63t–64t, 66, 69f, 87–88 mid-upper arm circumference, linear health outcomes model as anthropometric indicator, 18 concentration index decomposition, MPO. See mean positive overshoot 49–54, 49t–50t (MPO), ADePT Health elasticities in, 51–54, 51t Financing module 167 Index N data preparation, 104–5 National Health Account (NHA) health care prepayments and, 106–7 aggregate data health financing progressivity, ADePT Health Financing module 134–36, 135t health financing mix data, 107, 111–13 health payments distribution, original data report, 121–22, 122t 142–43, 142f out-of-pocket payments, 104–5 impoverishing effect of health ADePT Health Outcomes module payments, 153–54 benefit incidence analysis, 23 measurement of, 110 data set example, 30–31 poverty measures based overall health financing data, 155 on, 129–30, 129t non-need determinants, health care health payments adjustment utilization, 21 to, 150–52, 150f nonfood consumption progressivity and redistributive ADePT health financing data effect data on, 110–13, 112t preparation, 105 redistributive effects, 137–39, 138t measurement of, 110 overshoot concept nonlinear model, health inequalities, graph ADePT Health Financing module and table interpretation, 50–54 catastrophic health payments incidence and intensity O measurements, 145–47, 146f OLS. See ordinary least squares (OLS) distribution-sensitive catastrophic regression model payment measurement, 127–28, omitted-variable bias, nondemographic 128t, 147–48 health care utilization financial protection tables, determinants, 24n10–11 124–27, 125t–126t opportunity costs, catastrophic health ADePT Health Outcomes module, spending, 149–50 catastrophic health spending, ordinary least squares (OLS) 96–97, 97f regression model distribution-sensitive catastrophic child survival analysis, 17 payment measurement, 127–28, concentration index sensitivity to 128t, 147–48 living standards, 76–77 graphic interpretation of health P inequalities, 50–54, 50t “paint drip” chart, health payments, health care utilization analysis, 20–21 131–32, 132f health inequalities, graph and table Pen’s parade representation, income interpretation, 50–54, 50t distribution original data report poverty measurements, health ADePT Health Financing module, payments adjustment 121–22, 122t to, 150–52, 150f ADePT Health Outcomes total household consumption, module, 41, 42t 131–32, 132f out-of-pocket payments, ADePT Health per capita expenditure variables Financing module health care utilization inequalities, budget share of household payments, graphic representation, 57–58, 58t 130–31, 131f health inequalities, graphic concentration curves, 140–42, 140f–141f representation of, 52–54 168 Index poverty measurements public subsidies data ADePT Health Financing module benefit incidence analysis, 23, 84–86 consumption-based measures, cost assumptions, graphic representation, 129–30, 129t 62–66, 63t–64t health payments adjustment data set example, 30–31 to, 150–52, 150f linear cost assumption, 87–88 impoverishing effect of health proportional cost assumption, 62–66, payments, 153–54 63t–64t, 68f, 86–87 poverty line, 105–6, 110, 152–53 purchasing power parities (PPPs), ADePT Health Outcomes module, 98 ADePT health financing data PPPs (purchasing power parities), preparation, 105–6 ADePT health financing data preparation, 105–6 Q prepayments for health care quintile variables ADePT Health Financing module, ADePT Health Financing module data preparation, 106–7, 107t financial protection, 124–27, progressivity and redistributive 125t–126t effect data and, 111 health financing progressivity, redistributive effect, 156–57 134–36, 135t private insurance health payments distribution, health care prepayments and, 106–7 142–43, 142f progressivity and redistributive redistributive effects, 137–39, 138t effect data and, 111 ADePT Health Outcomes module progressivity analysis health outcomes inequalities, graphs Egypt data set example, 110–13, 112t and tables, 45–54, 45t health payments living standards indicators, 21–22 distribution, 142–43, 142f public providers fees, table of graph generation for, 115, 118f–119f, payments, 60–62, 61t 133–36, 134t–135t public subsidies cost assumptions, health financing, 98–99, 100f 62–66, 63t–64t prepayments for health care, 106–7, 107t R measurement techniques, 154–55 R2 (measure of goodness fit), linear overall health financing, 155 health inequalities models, proportional cost assumption, public graphic representation subsidies, 62–66, 63t–64t, 68f, 86–87 of, 50t, 53–54 proportionality, progressivity ranking variables, living standards measurements using, 154–55 measurement, 17 public facilities usage, table representation ratio scale, concentration index, 74–75 of, 59–60, 59t recall period public health services concentration, graphic health care utilization analysis, 20–21 interpretation and representation, health status analysis, 19–20 66–68, 67f–69f redistributive effect public providers fees economic welfare and, 158 benefit incidence analysis, 23 Egypt data set example, 110–13, 112t data set example, 30 graph generation for, 115, 118f–119f payment of, table interpretations, health financing, 98–102, 137–39, 138t 60–62, 61t prepayments for health care, 106–7, 107t 169 Index regression analysis total household consumption. See also graphic interpretation of health consumption determinants, 50–54, 50t ability to pay data and, 109 inequity measurement and inequality catastrophic health spending explanation, 8–9 impact on, 149–50 reranking concentration curves, 139–42, 140f–141f health financing redistributive consumption-based poverty effect, 101–2 measures, 129–30, 129t redistributive effect of health data preparation, 103–4, 104t payments, 157 health financing progressivity, 134–36, 135t S health payments and, 131–32, 132f sensitivity of health variables concentration index, living U standards measure, 75–77 unallocated revenues, incidence graphic interpretation of health assumptions for health care inequalities, 50–54, 51t payments and, 0 social insurance underweight variable, as health care prepayments and, 106–7 anthropometric indicator, 18 health financing progressivity, utilization variables 134–36, 135t basic properties, 20–21 progressivity and redistributive benefit incidence analysis, 21–23, 30 effect data and, 111 data set example, 28–31 redistributive effects of health determinants of, 22 financing, 137–39, 138t graphic representation of inequalities SPSS data set, ADePT software in, 55–58, 56t, 58t application, 1–2 graphs and tables generation, standard errors, linear health 36–37, 36f–37f inequalities models, graphs and tables tabs, 36–37, 36f–37f graphic representation inequalities in, 5–8, 55–58, 56t, 58t of, 50–54, 50t Stat data set, ADePT software V application, 1–2 value judgments, distribution-sensitive survey settings in ADePT catastrophic payment data set preparation, 21, 29 measurement, 147–48 prepayments for health care, variables in health outcomes analysis, health financing data 17–20 preparation, 106–7 adult health indicators, 18–20 anthropometric indicators, 18 T basic tabulations, 21 tables. See under ADePT Health child survival, 17 Financing module; ADePT concentration curve techniques, 71–72 Health Outcomes module graphic representation, 54–55, 54f taxes, progressivity and redistributive health care utilization indicators, effect data and, 111 20–21 threshold choice, ADePT Health living standards indicators, 15–17 Financing module, 148 weights and survey settings, 21 170 Index Vietnam Household Living Standards weight-for-age variable, as Survey (VHLSS) anthropometric indicator, 18 financial protection data weights settings in ADePT set example, 109–10 ADePT data set preparation, 21 health outcomes data data set example, 29 set example, 27–31 distribution-sensitive vignette anchoring methodology, catastrophic payment inequality measurements of measurement, 147–48 health status and, 19–20 welfare indicators, concentration index W sensitivity to, 76–77 wealth index, living standards World Bank data sources, measurement, 17 105–6, 109, 153 171 ECO-AUDIT Environmental Benefits Statement The World Bank is committed to preserving Saved: endangered forests and natural resources. The • 4 trees Office of the Publisher has chosen to print • 2 million Btu Health Equity and Financial Protection on of total energy recycled paper with 50 percent postconsumer • 388 lb. of net fiber in accordance with the recommended greenhouse gases standards for paper usage set by the Green • 1,752 gal. of waste Press Initiative, a nonprofit program support- water ing publishers in using fiber that is not sourced • 111 lb. of solid waste from endangered forests. For more informa- tion, visit www.greenpressinitiative.org. www.worldbank.org/adept Two key policy goals in the health sector are equity and financial protection. New methods, data, and powerful computers have led to a surge of interest in quantitative analysis that permits the monitoring of progress toward these goals, as well as comparisons across countries. ADePT is a new computer program that streamlines and automates such work, ensuring that the results are genuinely comparable and allowing them to be produced with a minimum of programming skills. This book provides a step-by-step guide to the use of ADePT for the quantitative analysis of equity and financial protection in the health sector. It also elucidates the concepts and methods used by the software and supplies more- detailed, technical explanations. The book is geared to practitioners, researchers, students, and teachers who have some knowledge of quantitative techniques and the manipulation of household data using such programs as SPSS or Stata. “During the past 20 years, an increasingly standardized set of tools have been developed to analyze equity in health outcomes and health financing. Hitherto, the application of these analytical methods has remained the province of health economists and statisticians. This book and the accompanying software democratize the conduct of such analyses, offering an easily accessible guide to equity analysis in health without requiring sophisticated data analysis skills.” Sara Bennett, Associate Professor, Department of International Health, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland, United States “As the international health community becomes increasingly focused on monitoring the impact of universal coverage initiatives, ADePT Health will help make the standard techniques more accessible to policy makers and analysts, increase the comparability of health equity and financial protection measures, and aid in generating the evidence needed to support policy.” Kara Hanson, Reader in Health System Economics, Health Policy Unit, London School of Hygiene and Tropical Medicine, United Kingdom “The ADePT software and manual make it possible for researchers without extensive statistical training to perform a range of analyses that will provide an important evidence base for introducing universal coverage reforms and for monitoring if these reforms are achieving their objectives. The ADePT initiative is an exciting and timely development that will enable researchers in low- and middle-income (as well as high-income) countries to undertake health and health system equity analyses that would previously have been lengthy and extremely resource intensive.” Di McIntyre, Professor, School of Public Health and Family Medicine, University of Cape Town, South Africa Streamlined Analysis with ADePT Software is a new series that provides academics, students, and policy practitioners with a theoretical foundation, practical guidelines, and software tools for applied analysis in various areas of economic research. ADePT Platform is a software package developed in the research department of the World Bank (see www.worldbank.org/adept). The series examines such topics as sector performance and inequality in education, the effectiveness of social transfers, labor market conditions, the effects of macroeconomic shocks on income distribution and labor market outcomes, child anthropometrics, and gender inequalities. ISBN 978-0-8213-8459-6 SKU 18459