Understanding the developing metropolis : lessons from the city study of Bogota and Cali, Colombia

UNDERSTANDING THE DEVELOPING METROPOLIS Lessonsfrom the City Study of Bogota and Cali, Colombia Rakesh Mohan A WoorId Bank Ilook     Understanding the Developing Metropolis Lessons from the City Study of Bogoti and Cali, Colombia   Understanding the Developing Metropolis Lessons from the City Study of Bogota and Cali, Colombia Rakesh Mohan Published for The lWorld Bank Oxford Universitvy Press  Oxford Universitv Press OXFORD NEW YORK TORONTO DELHI BOMBAY (.kLCUTTA MADRAS KARARCHI KUALA LUMIPLIR SINClAPORE HON(; KONG TOKYO NAIROBI DAR ES SAI-,AlAM (1APETOWN MELBOURNE AUCKlAND and associated companiies in BERLIN IBADAN C 1994 by The International Bank for Reconstruction and Development / THE WORLD BANK 1818 H Street, N.W., Washington, D.C. 20433 Publishied by Oxford University Press, Inc. 200 Madison Avenue, New York, N.Y 10016 Oxford is a registered trademark of Oxford University Press All rights reserved. No part of this publicationi may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recordinig, or otherwise, wvithout the prior permission of Oxford Universitv Press. Manufactured in the United States of America First printing April 1994 The findings, interpretations, and concilisions expressed in this study are entirely those of the author and should not be attributed in anv manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The bounidlaries, colors, denominiaiolns, and other informationi shown on any map in this volume do not imply on the part of the World Bank Giroup any judgment on the legal statuis of anly terr-itorv or the endorsemenit or acceptance of stich boundarics. Library of Congress Cataloging-in-Publication Dala Mohan, Rakesh, 1948- Understanding the developing metropolis: lessons from the city study of Bogota and Cali, Colombia / Rakesh Mohan. p. cm. Includes bibliographical references and index. ISBN 0-19-520882-X 1. Bogota (Colombia)-Economic conditions. 2. Cali (Colombia)- Economic conditions. 3. Urban policy-Developing counltries. l. Title. HC198.1B5M633 1994 333.9861'48-dc2O 93-41140 CIP  Contents Foreword xi Preface and Acknowledgments xiii 1. Introduction I City Selection 4 The Macroeconomic Setting: Were the 1970s Exceptional? 8 Overview 1 4 Notes 1 7 2. The Spatial Structure of Cities in Developing Countries 18 The Emergence of Cities 18 Growth and Decentralization 21 Citv Characteristics: Ecidence from Different Countries 26 The Consequences of Growth and Decentralization 34 Notes 37 3. Growth and the Changing Structure of Bogota and Cali 38 Growth and the Spatial Distribution of Population and Income 43 The Evolution of Land Values and Population) and Employment Densities 55 The Changing Structure of Bogota: Some Wrinkles 63 Summary 69 Notes 72 4. Poverty, Distribtition of Income, and Growth 73 Distributioni of Income in Bogota and Cali 1973-78 76 Distribution and Characteristics of the Poor 83 Summary 91 Notes 95 5. Workers and Their Earnings 96 The Benefits of Education and Experience 103 How Segmented Is the Bogota Labor Market? 108 v  vi CONTENTS Operation of the Urban Labor Market: What Have We Learned? 117 Notes 119 6. Firms and Their Location Behavior 121 Trends in the Location of Employvment 123 Patterns of Employment Location in Manufacturing 128 Factors Influencirng the Location of Manufacturing Firms 132 Modeling the Intra-Urbain Location Behavior of Manufacturing Firms 134 Trade and Service Employment Patterns 139 Implications for Location Policv 141 Notes 144 7. Shelter in a Growing City 146 HoIsillg in Bogota: The Instittional Setting 147 Bogota's Unregulated Housing Market: The Myths and Realities of Incremental Development 152 Mobility and Tenure Choice 161 Modeling Housing Demand 171 Pirala Developments and Housinig Demand: Some Lessons for Housing Programs and Policies 180 Notes 183 8. Autos, Taxis, Buses, and Busetas: The Importance of Choice in Urban Trarnsport 185 The Stipplv of Transportation Services in Bogota 189 The Regtilation of Supplv: Control, Suibsidies, and Incentives 191 The Impact of Govern inent Regulationi, Taxes, and Subsidies 198 Travel Patterns in Bogota 201 Modeling Travel Demand 215 Lessons for a Developing Metropolis 221 Appendix: Theoretical Background for Mlodeling the Choice of Transport Mode 225 Notes 232 9. Urban Government and Finances 233 Urban Government in Bogota 234 The Structure of Revenue 239 The Structure of Expenditures 255 The Public Serxice Enterprises of Bogota 259 Lessons for a Developing Metropolis: An Evaluative Summary 266 Notes 271 10. Coping with City Growth 272 Understaniding Behavior 275 Understandiig City Structure 278 Encouraging Endogenous Institutional Responses to Rapid Growth 281 Appendix: The Data 288  CONTENTS vii Bibliography 301 Index 317 Tables Chapter I 1-1 Economic Growth in Colombia, 1950-86 10 1-2 Prices and Wages in Colombia 12 Chapter 2 2-1 Population and Population Growth in Selected Latin American and U.S. Metropolitan Areas 28 2-2 Population Density in Central and Peripheral Areas of Selected Latin American and U.S. Cities 29 2-3 Population Densitv Gradients in Selected Latin American and U.S. Cities 30 2-4 Population Densitv Gradients Per Kilometer in Selected Cities Worldwide by Decade, 1880-1960 33 2-5 Populationi Densitv Gradients Per Kilometer in Selected Cities Worldwide, 1950, 1960, 1965, and 1970 34 2-6 Central City Shares of Residential Population and Employment, Selected Latin Americani Cities 35 Chapter 3 3-1 Population and Intercenisal Population Growth in Colombia 39 3-2 Gross and Per Capita Domestic Product, Colombia and Bogota 39 3-3 Average Anntial Growth in Domestic Product, Colombia and Bogota 40 34 Area. Poptilationi, Population Growth, and Density in Bogoti and Cali 41 3-5 Growth in Population and Densitv in Bogota 50 3-6 Spatial Distributioni of Monthlv lncome and Population in Bogota 54 3-7 Change in Population) Density by Ring, BogotA and Cali 56 3-8 Evolutioni of Land Values by Ring, Bogota and Cali 58 ,-9 Employment Density in BogotA 61 3-10 Distributioni and Density of Workers' Residences and W'orkplaces, Bogota 62 3-11 I.and Value and Population Densitv Gradients in BogotA and Cali, bv Sector 66 3-12 Spatial Patterni of Housinig in Bogota, 1978 67 Chapter 4 4-1 Distribution of Inconme in Bogota and Cali 78 4-2 Spatial Distribution of Income in Bogota and Cali 80 4-3 Spatial Inequality in Bogota and Cali: Individuals Ranked by tItNC(AP 82  vi i CONTENTS 4-4 Mappinig MalnTitritionl into Income Deciles in Bogota. 1978 86 415 Spatial Distribution of Malnutrition) bv Age Group in Bogoti, 1978 88 4-6 Unemployment in Bogota by Incomile Decile and Sex 92 Chapter 5 5-1 Changes in Labor Use in Colornbia, 1951-78 98 5-2 Emplovmenit Chanige in Colombia, 1973-78 98 5-3 Distribution of Workers by Sex and Education Level 10 5-4 Mean Incomc Ratios for W orkers bv Sex and( Education Level 10o Chapter 6 6-1 Employmenit bv Firm Size and Major Industry Group, 1978 124 6-2 Employment Distribution by Ring. Firm Size, and Major Industry Group, 1978 125 6-3 Employmenit Distribution in Bogota by Radial Sector. Firm Size. and Major Industry Group, 1978 126 6-4 Employment Location in Bogota, 1972 and 1978 127 6-5 Distributioll of Manufactuting Employmenit by Ring: BogotA and Cali, 1970-75 130 6-6 Changes in Manufacturing Employmenit, bv Ring: BogotS, 1970-75 131 Chapter 7 7-1 Average Data for Pirala aid .i 'orrnas ifinima5Subdivisions, Mid-1970s o58 7-2 Fiiianicinlg of Infiastructure in frafta Subdivisionis 159 7-3 Alternative Average Nominial Rates of' Return to Subdividers 161 7-4 Movinig Rates by Tenure Choice. Time Spent in Previous Residenice, andi Age of Household Head 164 7-5 Probability of Moving: Coniparison of (Oh-s and 1.0(.1 Elasticities 165 7-6 Mobility Rates in U.S. Metropolitani Areas by Selected Household Characteristics 166 7-7 Moves of Hlouseholds from One Ring to Another (Recent Movers) 167 7-8 Moves of Households from One Sector to Another (Recent Movers) 168 7-9 Tenure Choice by Household Income 169 7-10 Tenure Choice by Recencv of'Household Head's Migration 169 7-11 Elasticities for Tenure Choice Model Comparison of ots and t.oC Ir Specifications 170 7-12 Interinational Comilpar-isonis of' Owner-ship Status Elasticities 171 7-13 Character-istics of' Residences b Work Zone. Bogotul 175 7-14 Analysis of Variance: Hedonic Price Equations 176 7-15 Hotisinig Demand Equationis: Estimated Elasticities for BogotA 177 7-16 Raniges of'Hotising Demanid Elasticities fronm Various Counitries 178  CONTENTS ix Chapter 8 8-1 Transportatioin in Bogota 190 8-2 Vehicle Stock in Colombia, 1970-80 191 8-3 Vehicle Stock in Bogota, Cali, and All of Colombia, December 1977 192 8-4 Bus Subsidies in Colombian Cities, 1977 and 1978 195 8-5 Adjusted C(osts, Adjusted Income, and Capital Return for Buses, October 1980 196 8-6 Formal and Actual Governmenit Interventioni in the Transpor t Sector. (Colombia, 1980 199 8-7 Vehicle Taxation Incidenice in BogotA, 1978 200 8-8 Fuel Policv Incidenice in BogotA, 1980 201 8-9 Tranisport Tax and Subsi(ly Policy Incidenice: BogotA, 1978--S0 202 8-10 Work Trip Travel Time by Income Level: BogotA. 1972 and 1978, and Cali. 1978 205 8-11 Wor-k Trip Travel Time by Transportation Mode: Bogota, 1972 and 1978, and Cali, 1978 205 8-12 Ring of Resiclentce and Ring of Emploviyment: BogotA, 1972 and 1978 206 8-13 Average Comimiiuting Distance by Zone of Work, Tvpe of Worker. and Occupationi: Bogota. 1972 iand 1978 208 8-14 Average Nuliber of Trips Per Person and Average Number of Trips Per Hotusehold: Selected Cities 211 8-15 Trip Generationi Characteristics of All Travelers: BogotA., 1972 212 8-16 Trip) Generation Equatiorls: BogotA, 1972 213 8-17 Work Trip by Mode of Transportation: Bogota, 1972 and 1978, and Cali, 1978 217 8-18 Change in Work T'rip Modal Shares-Direct Elasticities: BogotA, 1972 and 1978, and Cali. 1978 219 8A-1 Local Registrationi Tax Incidenice: BogotA, 1980 228 8A-2 Sales Tax Inci(denice: Bogota, 1978 228 8A-3 (Customs Duti Imcideince: BogotA, 1978 229 8A-4 Incidenice of Taxes and Suibsidies on Fuiel Consumption of Plrivate Cars: Bogota, 1980 230 8A-5 Models for Choice of Work Trip Mode: Bogota, 1972 and 1978 231 Chapter 9 9-I Consolidated Per Capita Revenues: BogotA, 1961-79 240 9-2 Structtire of Coisolidated Reveniules: Bogota, 1961-79 241 9-3 Fiianicinig of L.ocal Public Expenditiure in Selected Cities by Type of Revenue 242 9-4 Distribution of l.ocal Tax Revenue in Selected Cities bv Source 246 9-5 Structure of Consolidated Total Expenditures: Bogota. 1961-79 257  x CONTENTS 9-6 Subsidies from Public Utility Pricing: Bogota, 1974 264 9A-I Per Capita Revenues of District Administration: Bogota, 1960-80 268 9A-2 Consolidated Capital Expenditures: Bogota, 1961-79 270 Appendix A-1 Households Interviewed, by Barrio 296 A-2 Distribution of Structure Types: Survey and Universe 296 A-3 General Characteristics of Surveyed Subdivisions 297 A-4 Variables in Superintendencia Bancaria Pirata Subdivision Survey, 1977 299 Figures 9-1 Total Consolidated Expenditure Per Capita of the Bogota District, 1961-79 236 9-2 Composition of the Bogota District's Real Per Capita Total Revenues, 1961-79 244 9-3 Distribution of Situado Fiscal to the Bogota District 251 9-4 Distribution of Sales Tax Revenue to the Bogota District, 1975-80 253 9-5 Collection and Distribution of Colombian Beer Tax Revenue 254 9-6 Composition of the Bogota District's Total Expenditures Per Capita 256 9-7 Composition of the Bogota District's Capital Expenditures 258 Maps 3-1 Bogota: Ring and Sector Systems Based on 1973 Comunas 44 3-2 Cali: Ring and Sector Systems 46 3-3 Bogota: Population Density by Comuna 48 3-4 Bogota: Distribution of Mean Household Income Per Capita by Comuna 52 5-1 Bogota: Distribution of Occupations by Sector 102  Foreword This book summarizes extensive empirical work on Bogota and Cali. Colom- bia, carried out under one of the first comprehensive studies of its kind in a developing country. A few similar studies have been done on cities in indus- trial countries, the first being Anatomy of a Metropolis, by Edgar M. Hoover and Raymond Vernon (Harvard University Press, 1959). Although not a model for this work, it was a source of inspirationi. Our expectation when beginninig this research was that the development patterns and behavior in Bogota and Cali would resemble those observed in industrial countries in the early decades of the twentieth century. In fact, the research revealed strong behavioral similarities of households, workers, and firms to those contemporaneously observed in industrial counltries. This is true of patterns of decentralization and movement of firms, of deter- minants of household demand for housing, of travel demand and choice of mode, and of general urban residential growth patterns. Of course, there are differences-household incomes are lower, residential densities are higher, transit use is greater, and service levels are lower in the two cities studied than in industrial countries. But it is the similarities, not the differ- ences, that have been most striking. These similarities in behavior support the view that the basic tenets of urban economic theory are applicable in large cities where household and firm decisions are determined in market settings and where passenger travel and freight transport are motorized. A basic conclusion is that cities are not chaotic collections of unpredictable activities. Their strong common patterns of behavior enable us to formulate policies for transport, housilng, urban labor markets, and local public finance and to improve the manage- ment of cities. Those of us who worked on the project are particularly indebted to Colombia for collaborating in this effort and for sharing with us its exten- sive data, intellectual resources, and institutional support. Gregory K. Ingram Admrinistrator Research Advisory Staff xi   Preface and Acknowledgments This monograph attempts to summarize the main findings of the World Bank research program known as the City Study. The study, which was under the overall direction of Gregory K. Ingram, examined five major urban sectors-housing, transportation, employment location, labor markets, and public finance-in Bogota (officially, Santa Fe de Bogota) and Cali, Colombia. The goal of the study was to increase understand- ing of these sectors in order to assess the effect of policies and projects on cities in developing countries.1 The idea of the City Study was initiated in 1975-76 by Hollis Chenery, then Vice President of the Development Policy Staff of the World Bank. It was his broad vision of development policv r esearch that enabled us to contemplate and then implement such an intensive program of research in urban economics. The research r eported in this volume was conducted over a period of about five years, 1977 through 1981, by a large team at the Urban and Regional Economics Division (headed by Douglas H. Keare) of what was then the Development Economics Department of the World Bank in Washington, in cooperation with the Area de Distribuci6n Espacial de la Poblaci6n (headed by Ramiro Car- dona) of the Corporaci6n Centro Regional cle Poblaci6n (ccRP), Bogota, ancl the Departmental and Munlicipal Planning Offices (Pla- neaci6n Departamental and Planeaci6n Munlicipal) of Cali. Participants in the study directed by Gregorv Ingram were Alan Carroll, Andrew Hamer, Valerie Kozel, Kyti Sik Lee, Johannes F. I.hin, Rakesh Mohan, Alvaro Pachon, Anna Sant'Anina, and Richard Westin of the World Bank; in Bogota, Alberto lIernandez. Rodrigo Villamizar, and Amparo de Ardila were led byJose Fernanido Pineda. About twentv-five consul- tants and research assistants based in both Bogota and Washington assisted. In addition to World Bank research funds, funding was pro- vided by Colombia's National Statistical Agency (Departamento Admi- xiii  xiv PREFACE AND ACKNOWLEDGMENTS nistrativo Nacional de Estadistica, DANE) for the 1978 Household Survey and by the Bogota Chamber of Commerce for dissemination of study results. The project also benefited from the direction of the Colombian Advisory Committee, consisting of Eduardo Aldana, Rodrigo Botero, Bernardo Gaitan, Pedro G6mez, Jorge M6ndez, Miguel Urrutia, Eduardo Wiesner, and the Resident Representative of the World Bank at the time, Ian Scott. At the World Bank, the consistent interest and encouragemenit given by Ardy Stoutjesdijk and Anthony Churchill were very helpful. In this volume, I attempt to integrate the main findings of the differ- ent segments of the project. Although I have presented the findinlgs as I have interpreted them from the work of the whole team, the primary credit for each portion of the research should go to its originators (itali- cized in the entries). LAIND MARKETs: Rakesh Mohan., Rodrigo Villamizar, Gregory K. Ingram, Ricardo Paredes, Guillermo Wiesner, M. Wilhelm Wagner. HOUSING AND RESIDENTIAL LOCATION: Gregory K Ingram, Andrew flamer, Amparo de Ardila, Alan Carroll, Jose Fernando Pineda, Rafael Stevenson, Oscar Borrero, Anna Sant'Anina. TRANSPORTATiON: Alvaro Pachon, Valerie Kozel, Richard Westin, Gregory K. Ingram, Alberto Hernandez, Jose Cifuentes, Emilio Latorre. EMPLOYMENT LOCATION: Kvu Sik Lee, YoonJoo Lee. LABOR MARKETS, POVERTY, INCOME DISTRIBUTION: Rakeshi Mohan, Gary Fields,Jorge Garcia, M. Wilhelm Wagner. PUBnIC FINANCE: Johannes F Linn, Alberto Hernandez, Jeffrey Lewis, Caroline Fawcett, Juergen Wolff, David Grevtak. OVERALL PATTERNS: Gregory K Ingram, Rakesh Mohan,Jose Fernando Pineda, Alvaro Pachon. DATA DOCUMENTATION: Yoon loo Lee. Nelson Valverde. As may be expected from the quantitative research in this volume, a legion of research assistants has contributed to the work. Thanks are due to YoonJoo Lee, Sung Yong Klang, Robert Marshall, Kathie Terrell, Wili Wagner, Jon Roseman, Nancy Hardinie, Mark Snydernmani, Leslie Kramer, and Nelson Valverde, all of whom gave much bevond the call of duty in managing the vast volumes of data processed in this study. In Colombia we also received assistance from Sonia Rodriguez, Marco Tulio Ruiz, Maria Clara de Posada, and Alejandro Vivas. We received extensive cooperation from maniy organizations in Colombia. Work in Cali was carried out with staff from the Departmetn- tal Planning Office (Planeaci6n Departamental), headed by Lacides Reves, and from the Municipal Planning Office (Planeaci6n Munici-  PRFACE AND ACKNOWLEDGMENTS xv pal), headed by Jaime Cifuentes. We collaborated with the Integrated Plan for the Development of Cali (Plan Integral para el Desarrollo de Cali, PIDECA), directed byjulian Velasco. Among the staff associated with our work in Cali were M. Sanchez, Hugo Garcia, M. V. Fuentes, J. Oso- rio, F. Fajardo, M. Santacruz, S. Prado, A. Rivera, and G. L6pez. We were also generously assisted by the staff of the Bogota District Planning Office (Departamento Administrativo de Planeaci6n Distrital), the National Centre for Building Industry Studies (Centro Nacional de Estudios de la Construcci6n, CENAC), the Social Security Institute (Insti- tuto Colombiano de Seguaros Social, Icss), and the Superintendencia Bancaria. Enrique Low and Clara Eugenia Lopez, Controllers (Contra- lores) of Bogota District, took special interest in the public finance work and gave generouslv of their staff time. We were particularly fortunate to have been able to collaborate with members of DANE, who provided us with unfailing and careful assis- tance.They agreed to incorporate our requirements into their schedule of quarterly household surveys, which resulted in the World Bank-DANE City Study Household Survey of 1978, on which many of our results are based. The high quality of this survey owed much to the diligence of its DANE supervisors, Roberto Pinilla and Maria Christina Jimnenez. The contributions of Gary Losee, Alfredo Aliaga,Jairo Arias, Alvaro Pachon, andJose Fernando Pineda to its design and implementation were nota- ble. We would like to thank the authorities of DANE for so r eadily provid- ing all the data sets used in the study. My own participation in the organization and collection of the data in the 1978 Household Survey taught me much about the practical difficulties and pitfalls involved in the collection of primary data, on which we economists are so depen- dent but which we seldom understand. For me, this work has spanned a period of what I presume will be about a third of my professional life. Douglas Keare inducted me into the World Bank in December 1976 to work in the study. Although most of the research had been completed by late 1981, the changing interests and reassignment of all participants made it very difficult to bring to final form all the work that had been done. Since 1980, 1 have returned to the Government of India twice to work on housing, urban develop- ment, and industrial development in a policymaking capacity. Greg Ingram, in his various incarnations in the 1980s, was able to find the resources to enable me to extricate myself from the government of India at various times, first to return to the World Bank in 1984, then again in 1989 and 1990 to write this volume as well as the earlier mono- graph (Mohan 1986). I appreciate his efforts in providing me these opportunities to conclude an enterprise we started jointly in January 1977. These "sabbaticals" have been invaluable in helping me keep up my research interests and in giving me respite from the pressures of nor- mal governmental work. It was also generous of Otima Bordia and A. N.  xvi PRIFACE AND ACKNOWLEDGMENTS Varma, Secretary, Industrial I)evelopment, Government of India, in 1989 and 1990, respectively, to give me time off from the ministry dur- ing difficult times of onerous work. The constant encouragement received over the years from Doug Keare and johannes linn has also been instrumental in enabling me to finish this work. In Colombia, we benefited greatly from having as our main collabora- tors Ramiro Cardona and Jose Fernando Pineda. Their war m hospital- ity, generously extended to the whole team, made everv visit to Colombia a pleasure. They contributed greatly to our knowledge and understanding of Colombia, making possible a truly collaborative project. Geri Mitchell organized the final processing of this manuscript, doing much of it herself. She was assisted byJean Ponchamni, Maria Diniatu- lac, Maria E. Sainchez, and Helen Lee. I am grateful to them for being able to do th1is in addition to their normal duties. The final production of this book has taken as long as its creation. Well-considered comments from three anonymous referees led to very useful revisions. Its readability owes a lot to the careful editing of Willa Speiser Jeanne Rosen, Kathryn Kline Dahl, and Deirdre Murphy. Our quest in this study was to find general approaches to studying the workings of large cities in developing countries and to achieve rnodes of uinder-staniding the behavior of and interactionis among the participants in the myriad activities in a developing metropolis. We feel that we achieved some success in this quest and that our findings have general relevanice to tinderstanding the growth of large cities. Our approach was to describe and analyze the various patterns of behavior that are found to be stable over time. For example, some of our work has been corrob- orated by the subsequent works of Steve Mayo and Steve Malpezzi on housing ancl by Kyu Sik Lee's continuing efforts on employment loca- tion. On other issues, our results are still arnong the few available for cities in developing countries. In writing this book, delayed though it is, I hope that large cities will begin to be better understood by profession- als and policymakers alike. I also hope that more detailed work of this kind will be forthcominlg so that we can continue to improvc our tinder- standing of the developing metropolis. My greatest debt, intellectual and otherwise, in being able to conduct this study and to bring it to a close in this book is to Gregory Ingranm. To him, my deepest gratitude. I dedicate this book to my wvife, Rasika Khanna, for allowing me to desert her at various times in the raw summer heat of Delhi in order to finish this work, which has followed me for more than a decade. Mv chil- dren, Tarini and Rasesh, who have appeared during the course of the writing of this book, will have to wait for my next effort to be eligible for dedication .  PRFACE AND ACKNOWLEDGMENTS xvii Note 1. Two monographs have been published: Rakesh Mohan, Werk, Wages, and Welfare in a Developing Afetropolis: Consequences of C;rowth in Bogota, Colombia (New York: Oxford University Press, 1986); and Ktu Sik lee, The Location of Jobs in a Deeveloping AMetropolis: Patterns of Gro70th in Bogota and Calm, Colombia (New York: Oxford University Press, 1989).   Chapter 1 Introduction The total urban population in developing countries has roughly quad- rLupled in the past four decades, from less than 300 million in 1950 to more than 1.2 billion in 1990. The rapid growth of cities is among the more striking features of the developing world. In 1950 there were only three cities in the developing world with more than 5 million people; now there are at least twenty-four such cities in developing counltries, compared with only ten in developed counltries. The word "metropolis" used to describe Western cities such as New York, Chicago, London, Paris, and Rome. Today it is as likely to bring to mind Buenos Aires, Saio Paulo, Mexico Citv, Tokyo, Manila, Seoul, Bombay, New Delhi, or Cairo. More often than not, current discussions of urban or metropolitan problems concern cities in developing couLntries. Since large metropolitan centers did not appear in what are today's developed countries until their economies had reached a comparatively advanced stage, the emergence of such cities in poor countries has fre- quently been regarded as an unusual phenomenont. A metropolis, once a source of national pride, now inspires awe if not fear. To many, large cities in developing countries conjure Up images of gigantic slums, floods of destitute migrants congregating in ramshackle shanty towns, cacophonous traffic choking the main arteries, and ineffective local gov- ernment strugglillg to proxide basic public services. Because of this widespreacl negative view and the alarm it evokes, urban policies often attempt to slow the growth of large cities rather than to manage their growth in a healthy fashion. Despite misdirected and even counterproductive policies, large cities have continued to expand and have usually prospered in the process. By the end of the century the developing world is expected to have miore than fifteen "megacities" with at least 10 million people each and m)or-e than a hundred cities with at least I million people each. United I  2 UNDERSTANDING THF DEVELOPING; METROPOLIS Nations' projections suggest that more people in the developing world will be living in urban than in rural areas by the year 2020. This is already trule in Imlost countries in Latin America, East Asia, and the Mid- dle East. South and Southeast Asia as well as Chiina are rapidly moving in the same direction, followed by Sub-Saharan Africa. A positive approach to the problems of large cities is a must if govern- ments are to meet the challenges posed by city growth. In thle preindus- trial age the concentration of people in cities was a religious, political, or military plenomenon. Today it is an economic phenomenoll, and it must bc analvzed as such. Evolving constructive metropolitan policies requires an und(ier-standing of how cities work. The need for such understanidinig led the World Bank to launch a researchl program in the 1970s knowvn as the City Study. This effort, for reasons outlined below, has focused primilarily on1 Bogota and secondarily on Cali in (Colombia. The main objective of this study has been to learn mnore about the intrinsically complex and interrelated phenomena that tindenlie the rapid growth of a city. Those of us involved in the study have sought to understand how thie components of a city's economy-its housillg. employment, labor force, land market, transportation systems, and public services-interact and behave in the context of rapid city growtlh. Our ultimiiate goal has been to bring about a more positive ori- enitationi to urban policy. Considerable research exists on1 tile old cities in Europe and North America. andI their strutctur-es are reasonably well understood. The work- ings of their housinig and( labor markets, the locational decisions of their resi(denits and employers, and the patterns of their traffic and transport are all regarded as tractable. There are stvlized facts about these cities that appear to have wide applicability. But how relevant is this knowl- ecige for cities in developing counaries? Do we need to apply different modes of understanidinig to these cities? How have their populatiolns responided to rapicd growth? What dlo project andl city planners need to know abotut the processes ancI patterns in these cities so they can con- tribute mor-e effectively to urbani development? Hlow much intervention in different markets is necessar-y for a city to accommodate rapid growth? The City Study was designedl Io illuminate these and many related quuestions. The broad theme of' the City Study is behavioral adaptation to rapid change in the context of rapidly growinig cities in developing countries. The impetus fol this work arose from the World Bank's increasinig involvement in urbani projects, beginning in the early 1970s. Along with this involvemiienit came a greater awareness of the existence of urban poverty and its concomitant problems. Whereas the manifold problems of rural poverty had been studied for a long time, little information was available on the dimensions and nattire of urban poverty. This led to the  INTRODUCTION 3 formation of an Urban Poverty Task Force at the World Bank in 1974. The Task Force assembled a group of academics working on these issues and commissioned a series of studies. These efforts did not produce a coherent program for addressing the issues of urbani poverty, but they did result in greater attentioni to urban problenms and greater apprecia- tion of the complexity and interrelatedness of urban phenomena. It also became clear that one feature of the City Studv should be a focus on urban povert. At the time this study was being launched, the notioln that everythinig in a city is interr-elated hacl taken hold, and it was widely accepted that these interrelationships were best captur ed in large-scale urban models. Researclher-s in the 1960s and early 1970s, especially in North America and the United Kingdom, had been very optimistic about the utility andc feasibility of constLructing such models and usinig themil to undler- stand how cities work and hence to plan better. Two broad modeling approaches were prevalent: behavioral or analytical models and opera- tional or planning models. A critical review of these modeling efforts was conducted as a prelude to the City Study (see Mohan 1979), ancd it led to consider-able skepticism about thle utility and practicality of large- scale models. In (lesigninig the City Study, we opted instead for detailed investigations of each area of urban decisionmnaking. It was more impor- tant, we clecided, to tinderstand and model the behavior of the various actors and markets that constitute a city than to replicate its ttjnctiolIs in a large-scale moclel. The approach we adopted was therefore 1o model cliffe rent markets, as well as the behavior of individtials. hoiseholds, firms, and the public sector within those markets. We wanted especially to examine the responises of these actors in the tace of extremilely rapid change. How, for example, (1o people react to shifting prices in the housing market? How much do people do for themselves, and how should the public sec- tor intervene? What kinds of supply responses can be expected when demand for housing burgeons? How does the public sector cope with widespread illegality in land transactionls? How do people choose their mode of transport? W'hat kind of transpor-t services appear in respolise to a variegated demand? How are inconmes determined in an urbani set- ting, and how are thev affected by changes in labor market conditiolns? How clo people clecide whether or not to participate in the labor mar- ket? These are the kinds of questions we sought to address through the behavioral moclelinig of actors and markets. We felt that urball policies could be much more effectivelv designed if there were better apprecia- tion of the underlyinig behavior that results in the phenomenia observec. The objective of the City Study was not to propose a specific program to guide city growth, but rather to develop tools (models as well as other analytical niethods) that cotld be used to estimate the spatial and eco-  4 UNDERSTANDING THE DEVELOPING METROPOI-IS nomvic effects of different kiiids of policy interventiol. The research strategy was threefold: (a) to svstematically describe the current spatial patterns of' the various economic activities that constitlte a city, along with recenit changes in the observed patterns; (h) to model behavior and to estimate parameters useful for understanding how the main tl-ban] constituents respond to rapidly changing conditions; and (c) to assess the policy imnpact of these findinigs. Tle svstematic description Would enable us to determine howv the spatial patterns of activities evolve in response to changes such as the decentralization of'jobs and residences in a growing city. These descriptions would in turni help us formulate hypotheses to he tested, andl the empirical estimations of models wotld vield the paramiieter estimates. We would also be able to observe the apparenit effects of policies that attempt to alter the spatial distributioni of activities. The magnitutde of this task, we knew, would depend criticallv on how similar the fundamenital processes of tirbani development are at differ- ent stages of economic development. If existing tools (lesignied to study cities in rich countries could be applied or adaptecl for use in poorer counJtries, thie task Would be that miuch easier. We looked for clues about the transfer-ability of these tools bv comparing the spatial patterns of cit- ies in r ich and poor countries. Ouir assessment was that the similarities outmeighed the dissimilarities (see Ingram and Carroll 1981). The research was therefore also designed to test whether familiar tools devisedI to study cities in indltistrial couLntries could be transfer-red to the stucdv of growing cities in dleveloping coUntrties. Most urban studies concentrate exclusively on issues relate(d to the supply of'inifastlrlcttre, housing, and transport services. The City Sttudy addresses these issues in cletail. But it gives equal attention to the work- ing of the labor market and the resulting income distribution, because the emphasis is on the behavior of people and firms in a situation of r apid growth. A pr-ioI knowledge of household behavior and the labor market activitv of' individuals and their employers is necessary before the housing, transport, andc infrastructuLe issues can be addressed. Map- ping the spatial distribution of income and explaining its overall trend are crticial to comprehending andl predicting the natuLre of demands for infrastrUtcttUre, housing, and transpor-t that are likely to arise in a grow- ing city. City Selection Having decided to launch a program of' research designed to improve understaniding of how cities in developing countries work, we next decided to study one city in cletail rather than to attempt to draw con-  INTRODUCTION 5 clusions from cross-country patterns. This seemed the appropriate strat- egy, given our intention to focus on understanding the intracitv patterns and behavior that characterize a rapidly growing city. Resources, both financial and human, did not permit a simultaneous, in-depth study of several cities in different countries. Once the decision had been made to focus on a single city, it became apparent that Latin America was the place to look. The urbanization experience of Latin America in the 1950s. 1960s, and early 1970s was a precursor of the fast urban- growth phenomenon common to most developing countr-ies. Maniy Latin American countries had already reached levels of urbanization that exceeded 50 percent, and there were a number of large cities from which to choose. Furthermore, the prox- imity of the region to Washington, D.C., where the World Bank's head- quarters are located, was a big advantage, since the study was expected to involve intensive work in the selected city and hence frequellt travel. Certain additional criteria led us to choose BogotA as the primary site for this study. First, the policy and planning climate in the citv and in the country had to be receptive to such a study. Ae wanted the policy- makers, plannlers, and administrators to be interested in the results and capable of absorbing them. This depended on the availability of profes- sionals working on urban issues both within and outside the govern- ment, and on1 the interest of the nationial and local governments in the management of city growth. The potential for the World Bank to make future urban-r elated investments was another considerationi. The second criterion was a favorable research climate in the selected citv and country. To carry out the study in a reasonable time, it was desirable for the city to have both a rich data base and competent indi- viduals or institutions that could participate in the work. Moreover, the potential usefulness of the study would be enhanced if the local research capability existed to utilize the results so that the knowledge could be transferred to other researchers when the study was concluded. The availabilitv of existing data bases in time series, cross-sectional, and disaggregated form was an important asset to be considered, as was the availability of relevant studies. A third criterion was the absence of an atypical development pattern that would limit the transferability of the findings. The city had to be large enough to exhibit adequate variation in land prices, travel pat- terns, density levels, and the like so that the analyses could be applied to a wide ranige of situations. Accordingly, idiosyncratic administrative cit- ies such as Brasilia were excluded. The relative stability of thc nationlal economy was also important. A cultural context commnon to many COUI1- tnes was an added asset. Given these criteria, the choice was narrowed downl to four countries: Brazil, Colombia, Mexico, and Peru. Rio de Janeir(o was eliminated  6 L!NDERSTANDING THE DEVELOPING1 METROPOLIS because of its highly untusuial geography, and Sao Paulo and Mexico City because they are far too large, complex, and unmanageable for practi- cal studv. Government interest was found to be keenest in Colombia. From discussions with government officials, consultants, and academics, it also became apparent that a concern for intracity problems was wide- spread in that country. Bogota exhibited characteristics representative of a cross-section of Latin American cities (see Ingram and Carroll 1981), which greatly enhanced the chances that the findings would be transferable to other cities. It also had a rich data base on which to build (see the appendix to this volume for details). Moreover, Colombia had a historv of policy- related analytical work on urbani issues, influeniced in part by the life- long interest of Lauchlin Curr-ie. The importanice of urbanizationi in economic development and of housing as an economic activitv had long been emphasized by Currie. In 1950 hie led the first World Bank mission to a developing country, which happened to be to Colombia. He later moved there and exerted a lasting influence on national planning. It is also worth noting that Albert Hirschman derived the crucial insights for his book, rhe Strateg- of Economic Development (1958), during a three-year stint in Colombia's National Planning Department: that the Interna- tional Labour Organisation (tt.O) sent an infltuential employmnenit rnis- sion headed by Dudley Seers to Colombia in 1967 (see iL,o 1970); and that a detailed country report on Colombia was among the early WVorld Bank country economic reports published (Avramovic and associates 1972). Of direct relevance to our study was the important urban devel- opment study of Bogota undertaken with support from the United Nations Development Programme (UNDP) in 1970-74 (known as the Phase 11 Study). This work had generated a substantial amount of bench- mark data, in particular a comprehensive household survey in 1972. At the time we began the City Study in 1977, other scholars had recently completed or were engaged in research on Bogota. Albert Berry and Miguel Urrutia had been working for years on issues con- cerning the labor market and income distribution (see Berry 1975a, 1975b; Berry and Soligo 1980; Berry and Urrutia 1976; Urrutia 1969, 1985). Gary Fields and Helena Ribe were interested in similar issues (see Fields 1975; Fields and Marulanda 1976; Fields and Schultz 1980; Ribe 1979). Richard Nelson, r. Paul Schultz, and Robert Slighton (1971) had also worked earlier on the interrelated issues of population) growth, internal migration, and labor market adjustment. Peter Amato (1968, 1970a, 1970b) had documented patterns of urban residential location a decade earlier. Georges Vernez (1973) and Rodrigo l.osada and Hernando Gomez (1976), among others, had worked on phralta developments and incremental housing development. Harold Lubell  INTRODUCTION 7 and Douglas McCallum (1978) had addressed the relation of employ- ment issues and urban development. At the World Bank, several studies on Colombia had been conducted or were in progress in the mid-197Qs: Johannes Linn was carrying out a large program of research on urban public finance (see Linn 1976a, 1976b, 1979, 1980a, 1980b); Marcelo Selowskv was working on a study on the distribution of public services (see Selowsky 1979); William Doebele and Orville Grimes had con- ducted studies on the working of the urban land market (see Doebele 1975; Doebele, Grimes, and Linn 1979); and Mariluz Cortes wvas involved in a project investigating the role of small-scale enterprises (Cortes, Berry, and Ishaq 1987). Thus a vibrant research ambience existed both in Bogota and at the World Bank that was very conducive lo initiating the City Study in Colombia. The active involvement of local researchers was essential to the suc- cess of this endeavor. Bogota was well endowed with research institu- tions and university departments that had some experience working on ti-ban issues. It was therefore easy to find researchers who could partici- pate in the study. The work was eventually based in the Corporation Central Regional de Poblacion (('CRP)-an institution engaged in popu- lation studies, ur-ban studies, and economics research. The interest that different levels of government have taken in this study is borne out by the many contributions of public institutions. The National Planning Department (Departamento Nacional de Pla- neaci6n) was active in initiatinig and facilitating the study; the National Statistical Agency (Departamento Administrativo Nacional de Estadis- tica, DANE) shared raw data from past surveys and conducted the 1978 City Study survev; and the Banking Regulatory Agency (Superintenden- cia Bancaria) provided data on pirate land developments and contrib- uted staff resources. The citv government of Bogota made staff available to help conduct the public finance portion of the study. When we were getting under way, the city government of Cali, along with the State Planning Office (Planeaci6n Departamental) of Valle, the state in which Cali is situated, expressed great interest in our work. They invited Us to broaden the study to include Cali, which we did. The City Planning Office (Planeaci6n Municipal) provided extensive staff resources to iriiplemerit the study in Cali. Finally, the World Bank had a large program in Colombia. It was therefore expected that the results of the study would find a ready audi- ence within the institution. In short, Bogota satisfied all the criteria that had been laid out for selecting a site for the City Study. Cali was added because of the local interest expressed. It proved very useful as a comparator for assessing the generality of the findings. Whereas the primary studies were under-  8 UNDERSTANDING; THE DEVELOPING METROPOLIS taken in Bogota, an attempt was made to duplicate the research wher- ever possible in Cali. Wherever significant differences exist, they have been noted. Otherwise, most general conclusions apply to both cities. The Macroeconomic Setting: Were the 1970s Exceptional? This study is being published many years after completion of the field research, which was carried out fiom 1977 to 1980. The question arises whether the results were unduly influenced by any special conditions prevailing in Colombia from 1972 to 1980, the period covered by the study. A related question is whether any developments during the 1980s give cause for reassessing the results. It is also useful to place the city- related issues addressed in this studv within the national economic coIn- text of the time. Since the study was designed to enhance our understanding of the behavioral underpinnings of how cities work, especiallv in the context of rapid growth and change, the delay in publication should not make our findings obsolete. Indeed, the economic changes that have occurred during the 1980s underscore the usefulness of taking a behav- ioral look at how a city functions rather than adopting a prescriptive approach to planning. Viewed in retrospect, the economic environment in Colombia was more buoyant in the 1970s than in most other recent decades, although not exceptionally so. In the early 1980s Colombia was on the verge of economic crisis for reasons that were in part exogenous: a drop in cof- fee prices, high world interest rates, an international recession, and the backwash of the debt crisis affecting the rest of Latin America. These externally induced economic difficulties were exacerbated by burgeon- ing domestic problems, including high fiscal deficits and a significant appreciation of the real exchange rate. Growth in employment slowed, unemployment rose, and the external trade deficit widened. Although the situation in Colombia was not nearly as strained as it was in most other Latin American countries, many of which were highly indebted, international banks, facing difficulties elsewhere on that continent, hardened credit terms for Colombia. The optimistic tone of this study, a reflection of the economic environment of the 1970s, would probably have been tempered if the period of observation had been the 1980s. Nonetheless, most of our findings are unaffected by the events of those years. The three decades prior to the 1980s marked an important demo- graphic and economic transition for Colombia. In the 1960s the rural labor force ceased to grow (Urrutia 1985) and the urban labor force expanded rapidly. People living in urban areas began to outnumber  INTRODUICTION 9 rural dwellers: they increased from 39 percent of the poptulation in 1951 to 52 percent in 1964 and 60 percent in 1973. By the late 1970s two- thirds of Colombians resided in towns and cities. Colombia, like most countries experiencing urban growth, had uLndergonie a striking demo- graphic transition in a short time. In the 1980s this process slowed down; urban residents accounted for 68 percent of the population in 1986. The urbanization process during the 1950s, 1960s, and 1970s was therefore somewhat different from that in the 1980s. That some of our findings differ from those of earlier stucl- ies (for example, 11.0 1970; Lubell and McCallum 1978) is partd because we were observing the consolidation phase of' Colombia's urban history, whereas others had observed the effervescent phase, when urbatn growth seemed excessively rapid and endless. One of the lessons drawn in this study is that understanding the growth processes in a specific city requires an appreciation of the national urban growti context. Colombia's demographic tranisitioni is also evident fiom dramiiatic changes in other social indicators duting the 1960s and 1970s. The crude birth rate (CBR) fell from 45 per thousand population in 1963 to 31 per thousanid in 1974. The CBR then declined more gradually, reach- ing 27 per thousand by 1985. Similarly, infanit mortality fell from 6.5 per thousand births in 1963 to 4.0 per thousand births in 1974 and to 3.2 per thousand births in 1985. Primary schooling became unliver-sal in the 1970s, and secondarv school enrollment increased from 17 percent of the school-age population in 1963 to 39 percent in 1974 and 50 percent in 1985. These indicators confirm the extremely rapid changes that took place in the 1960s and 1970s and the slowdown that followed in the late 1970s and early 1980s. The economic impact of these social and demographic shifts was far- reaching. The rapid growth in population, particularlv in urban popula- tion, placecl severe strains on the suppliers of urban services. The slow- ing of change toward the late 1970s made it easier for the housing supply and other urban amenities to catch tip with the backlog of urgent demand. Similarly, the impressive expansionr in the supply of education starting in the 1960s, especially for females, transformed the quality of the labor force within two decades. Some of the employment pressures durinig the 1980s may be relatedl to this expansioni, which had the effect of reducing the returns to secondary and higher education and so contributed to improvinig the distribution of income (see chap- ter 4 in this Volume; Mohan and Sabot 1988). Colombia sustained a respectable but not exceptionally high rate of economic growth in the decades following WNtorld War 11 (see table 1-1). From 1950 to 1966, the gross clomestic product (GDP) grewv an average of almost 5 percent a vear. This acceler-ated to about 6.4 percent a year  I ( UNDERSTANDING THE DEV'ELOPIN(; METROPOLIS between 1967 and 1974 but fell back to about 5.5 percent for the rest of the decade. Then came the exceptional deceleration of the 1980s. Income per capita dropped during 1980-83 for the first time in three decades. Whereas economic growth was just respectable, the increase in emplovment growth in the late 1970s was exceptional (see table 1-1 and chapter 5). The existence of relatively tight labor markets, the absence of notable segmentation, and the measurable increase in real wages, all of which wvere observed in our study, were clearly influenced by unusually high growth in employmnent during the period. In our judgmenit, however, this phenomenon was not entirely fortuitous but was in part the conse- quence of appropriate policies and sound inacroeconomic manage- ment. Indeed, Colombia has benefited from a long record of sound macro- economic management, which has usually responded successfully to changes in the economic environment. Like most other developing countries, especially those in Latin America, Colombia emphasized import-substitution industrialization policies during the 1950s and 1960s. Its periods of economic growth and stagnation have been linked to its terms of trade, especially to the movement of world coffee prices. In the late 1970s coffee prices were booming and the economy was doing well: in the early 1980s the opposite was true. The post-World War II import-substituting strategy led to the consolidation of industr-ial growth, particularly in Bogoti, which rapidly surpassed Medellin as Colombia's prime manufacturing center. The Lleras government (1966-70) introduced policy changes that helped lay the foundationi for some of the income and employment gains of the 1970s. Export pr-omotionl was pursued activelv bv introduc- ing various incentives; a crawling peg exchange rate policy was adopted to maintain a realistic real exchange rate; fiscal discipline was exercised; and a mild program for the deregulation of some import controls was put in place. This policy package was largelv continued bv the Pastrana government (1970-74), which, in addition, consciouslv introduced an Table 1-1. Economic Growth in Colombia, 1950-86 (percenlt) AV.erage ainnualgrouoth Share of capital Period (;DP Emplment Capital in G;DP 1950-57 4.9 2.5 5.3 44.1 1958-66 4.9 2.4 3.9 43.3 1967-74 6.4 2.8 4.1 42.7 1974-80 5.5 4.1 4.9 40.7 1981-86 2.7 2.8 4.2 37.8 Source: Garcia (1988).  INTR0DUcT'rI)N I I urban development thrust by promotinig constructioni activities. The result was improved utilization of capacity in the industrial scctor and a significant increase in noncoffee exports. The constructioni boom was concentrated primarily in Bogota and secondarily in Medellin and Cali. In the field of housing, the most significant development was the introduction of a new savings instrument that assured savcrs a real rate of return on their accounts through indexation. The f-unids mobilized through this instrument could then be used to finanice housing. Although there is some debate about the overall impact oin housinig of this financial innovation, it undoubtediv helped foster middle- aiid upper-middle-income housing activity (see chapter 7). Housinig and financial markets did become better linked as a result. The l.6pez administration (1974-78) deemphasized suppor-t for con- structioni and adopted an explicit urban cleconcentrationi program to direct economic activities away from the largest cities of BogotA, Medellin, Cali, and Barranquilla. Emphasis was also put on1 usirig public resources to help the poor: spending on public health and education was promoted, whereas investments in infrastructure for energy and transport were downgraded. The net result of the policy environmiient that originated in the mid- 1960s was that sectors wvith low ratios of capital to labor grew faster than other sectors, noncoffee exports increased, and educationi expan(le(I rapidly along with other- social services (see Urrutia 1985). This led to the remarkable employment boom in thie late 1970s. At about the same time, however, tihe policy package began to uLnravel. World coffee prices shot up to unipr-ecedenited levels, as did unrecorded foreign exchange carnings fromil the drug trade. The bal- ance of payrnents went into surplus. Silnce it was difficult to stelilize these foreign exchange inflows, the real exchange rate appreciated sig- nificantit, and nioncoffee expor-ts began to falter. WN'hen coffee prices crashed in the early 1980s, as showni in table 1-2, a severe balance of pav- ments crisis ensued; the currenit account deficit was 6 percent of (;)IP in 1984. The policy response was to impose stringent controls on ailmost all imports. The economic difficulties were coinpouinded by rising budget deficits, which reached 6.8 perccnt of (.,DP in 1984. These were caused in part by the bunichinig of some large infrastriucture investimienits, espe- cially in the power sector, and by the drop in coffee revenues. The eco- nomic downtur-n during 1980-84 caused significant unemploymcit. Other related developments may have contributed to the slowdown in employment growth, most notably in organized manufactur-ing. As was documented in a 1987 internal WorldI Bank report, wages in the orga- nized sector began to rise in the late 1970s. and the differential between wages in the organized and the unorganiized sectors may hiave widened. Wage agreements. which had usually beni doncluded every other year,  Table 1-2. Prices and Wages in Colombia lnlicatars 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 Intlexes Exchange rate Nominal (pesos per dollar) 20.1 22.0 23.8 27.1 31.2 35.0 36.8 39.1 42.6 47.3 54.5 64.1 78.9 100.8 142.3 196.1 243.0 299.0 Real (average) n.a. n.a. n.a. n.a. 100.0 95.4 85.7 85.5 81.7 83.5 81.6 75.6 73.6 79.9 91.4 108.5 108.2 105.6 Wages' 10)5.2 108.6 110.5 111.8 111.3 10)9.2 109.7 121.7 127.6 129.5 134.2 132.6 132.9 132.4 n.a. n.a. n.a. n.a. Coffee price index 80.9 89.3 104.8 98.6 100.0 172.6 215.7 151.() 131.2 111.4 74.9 77.0 74.2 77.2 81.7 115.2 64.7 74.8 Annutalgrowth rate (pertent) GDP deflator (1975=100) 10.8 13.0 20.2 25.4 22.8 25.5 29.1 17.1 24.1) 27.6 22.8 24.8 20.4 22.2 23.2 24.0 22.8 27.0 Exchanige rate Nominal 9.4 9.7 8.2 13.8 15.1 12.1 5.1 6.3 8.8 11.2 15.2 17.6 23.1 27.8 41.2 37.8 24.0 23.0 Real n.a. in.a. n.a. n.a. n.a. -4.6 -10.1 -0.3 -4.4 2.2 -2.3 -7.3 -2.7 8.5 14.4 18.7 -0.3 -2.4 Whlolesale piices n.a. n.a. n.a. nia. n.a. 22.9 26.7 17.6 27.8 37.5 23.5 24.6 18.0 21.2 23.0 24.4 25.2 283 n.a. Not available. a. 1970= 100. Source: Internal World Bank documilents (unipublishied 1987 and 1989 country economic memoranida); Thomas (1985).  INTRODUCTION 13 became an annual occurrence in the late 1970s. This put added pres- sure on real wages. One feature of the rising wage costs in the organized sector was the increasing share represented by employment benefits. The appreciation in the real exchange rate was exacerbated by these developments, and the growth of the manufacturing sector, especially in export-oriented industries, was adversely affected. Although this study did not find significant evidence of segmentation in the labor market (see chapter 5), some observers in the 1980s did. It is possible that cer- tain developments noted in the 1970s could have led to increasing seg- mentationi later. A specific feature of the rising unemploynment in the 1980s was high unemploymenit among educated women. As this study documents at length in chapter 5 (see also Mohan 1985), the supply of women in the labor force grew as access to education expanded in the 1960s and 1970s. Some of the rise in urban unemployment rates in the early 1980s stemmed from the increasing numbers of women desiring to work out- side the home. The travails of the early 1980s have led more recent administrations to adopt policies encouraging the use of labor. The government has also kept the real exchange rate on a par with rates of the mid-1970s, reduced the fiscal deficit to manageable levels (1.4 percent of GDP by 1987), and brought the current account back into balance by 1987. These initiatives have been helped by higher coffee prices. An attempt has also been made to shift public spending away from capital-intenisive projects. A war on1 poverty, launched by the Barco government (1987- 90), emphasized extending services to the backward regions of the country and providing health and nutrition services, education, and basic utilities to the poor. Shelter is an important component of this war on poverty. A wvelcome change in approach is the recognition that slum upgrading is a key com- ponent of housinig strategy and that communiity participation is a must in this process. Less welcome is the return to an earlier strategy of directly investing in housing through the National Housing Agency (Instituto de Credito Territorial) (see World Bank 1990). Moreover, inadequate attention is being given to rental housing as a vehicle for sat- isfying the needs of the poorest (see chapter 7). In light of C(olombia's experience in the 1980s, our analysis of shelter needs and patterns from a behavioral viewpoint, and our findings about effective policy initia- tives, remains valid (see chapter 10). A final feature of Colombia's economic environment in the 1970s is relevant to our study. Government policies after the mid-1960s did not discriminate against agriculture. Helped by the favorable terms of trade that emerged, the agricultural sector prospered (Garcia 1983; Thomas 1985). The coffee boom fiom about 1974 to 1980 also contributed, as  14 UNDERSTANDING THE DEVELOPING METROPOLIS did the prospering drug trade. Furthermore, the introduction of new coffee varieties and of new agronomic practices increased the demand for labor, and that in turn helped tighten Colombia's rural labor market in the late 1970fs. Since these policy and economic changes coincided with the demographic transition described earlier, the suipply of rural labor tightened further and real wages rose, bringing improvements in income distribuition (Urrutia 1985). These developments were reflected in the urban labor market as well, putting pressure on real wages in the late 1970s. But the deteriorationi in the general economic situation in the early 1980s, coupled with the fall in coffee prices, erased some of the gains of the 1970s. Overall, the late 1970s may be seen as a transitional period durinlg which the urbanization process in Colombia was traversing the upper bend of the sigmoid path followed bv most urbanizing countries.' In conducting this study, we observed the rapid changes in Bogota (and Cali) resulting from the rapid demographic transition that Colombia hiad gone through in the 1960s and 1970s. We also observed the fruits of macroeconomic policies in place since the mid-1960s that had stimu- lated labor using economic and urban growth. Conditions in the urbani labor market benefited as well from positive developments in the agri- cultural sector. Moreover, specific government policies adopted in the early 1970s fostered the generally high levels of urban infrastructure ser- vices noted in our study. The earlv 1980s witnessed a reversal of some of the economic gains posted in the 1970s, but the record of the late 1980s encourages renewed optimism for the future.2 Given the changinig natule and reqtiiremiients of matulilng urbaniza- tion (the stage when the urban population in a country exceeds 70 per- cent), the future growth of Bogota (and Cali) will undoubtedly exhibit different characteristics than those clocumented in our study. But many other countries are now undergoinig the phase of rapid urbanization. This volume will provide useful approaches to analyzing city growth in those countries. The improved understanding of how cities work that this study attempts to furnish should influence the adoption of more positive policies for managing large cities as they grow further. Develop- ing countries will continue to see the emergenice of large cities for a long time to come. Overview The striking result of this study is that many of the behavioral relations documented for Bogota and Cali are similar to those observed for other cities in both developed and developing countries. This suggests that cit- ies exhibit strong regularities in their patterns of development, which in  INTRODUCTION I 5 turn suggests that the transferability of basic behavioral findings is great. This conclusion is bolstered by the fact that Bogota and Cali have many characteristics typical of other large cities in the developing world. Although many of the results pertain specifically to the two cities stud- ied, the research also shows that economic forces strongly condition the course of urban development and that markets play an important role in the allocation of resources to urban areas. Public use of private incen- tives can help achieve harmonious development. Efficient city growth requires management and administration that is oriented as much to economic issues as it is to physical "bricks and mortar" issues of infra- structure. The City Study was divided into seven segments covering various aspects of urban growth and the urban economy. These include the spa- tial patterns of urban development, the operation of urban land mar- kets, employment and income distribution, the location of employment within the city, housing, transportation, and public finance and adminis- tration. This book is more or less organized around these seven seg- ments or topics.3 Chapter 2 looks at the patterns of city growth across many countries in both the developing and the developed worlds. It offers a rationale for the emergence of large cities and traces patterns of spatial develop- ment common to growing cities. The continuing decentralization of both residence and employment is a feature of most cities as they grow. Chapter 3 introduces the cities of Bogota and Cali by summarizing thei- structure in relation to the spatial distribution of population and land values. It then describes their changing spatial patterns by tracing the density and land value gradients as they evolved over time. The main characteristic of growth associated with these cities, as with other expanding cities, is the decentralization of residence and employment that has occurred. Together, chapters 4 and 5 cover the third segment of the study, on employment and income distribution. Chapter 4 includes an anatomy of the distribution of income in Bogota and Cali, with specific attention to the distribution and characteristics of poverty. The central issue con- sidered is the existence and measurement of spatial inequality-that is, the concentr.ation of the poor in certain parts of the city. The various correlates of urban poverty are also described, and the measurable reduction in poverty in the late 1970s is documented. Chapter 5 turns to the work force, which is profiled not only by the usual demographic cri- teria of age, education, and income level, but also by place of residence within the city. Estimates of the returns to education and experience are also presented in this chapter. An interesting featlLre of these estimates is how the expansion of access to education has affected the returns to higher education. The segmentation of the labor market is then exam-  16 UNDERSIANDING THE DEVELOPING METROPOLIS ined. In addition to the usual segmenting variables such as union mem- bership and the size of firms, the influence of workers' backgrounds, as proxied by their places of origin and residence, proves relevant. The interaction of labor markets with city strtucture in terms of the location ofjobs and workers' residences and the extent of workers' spatial mobil- ity emerges as the theme of this chapter. Chapter 6 reports on the changing locational patterns of employment in Bogota and Cali, with a focus on the manufacturing sector. The fac- tors influencing decisions about where to locate are identified and the process of movement described. The discussion emphasizes the role of the central city as an incubator for entrepreneurship. The findings are then generalized to provide some guidelines for public policy. Chapter 7 assesses the housing situation in Bogota and the response of housing markets to the city's rapid growth. The institutional setting is described, along with the role of government intervention. The concept of unorganized urban housing markets is afflicted with a host of myths and prejudices; the validity of these is examined in the context of the widespread unregulated housing developments in Bogota. To achieve a better understanding of the working of housing markets, the function of tenure choice and mobility as adjusting mechanisms is investigated in some detail. Similarly, the main parameters of housing demand, such as income and price elasticities, are carefully estimated. Overall, the study finds that Bogota's housing market functioned relatively well and that the supply of housing improved despite the city's rapid population growth. The sttudy points clearly to the importance of rental options in satisfying the housinlg demand of the poor. Chapter 8 takes a detailed look at the transportation sector in Bogota. The interdependence of the city's structure and its transportation pat- terns is described. The regulatory apparatus and the supply mecha- nisms of urban transport are documented, and the distributive impact of the main taxes and subsidies affecting this sector is evaluated. Bogota has a wide variety of transport modes, which work relatively well. The overall effect of regulation, taxes, and suIbsidies is progressive. The mod- eling of mode choice and other aspects of travel demand are also intro- duced. The widespread availability of transportation in Bogota at affordable flat rates, made possible in part by government subsidies, has done much to alleviate the spatial disadvantage of poor workers in Bogota. The importance of providing a range of choices in urban trans- portation systems is one of the important conclusions of this study. Chapter 9 offers a window onto the structure and performance of Bogota's urban government. The seemingly unplanned and decentral- ized approach to providing urban services has resulted in relative abun- dance. The financing pattern differs from that found in most cities.  INTRODUCTION 1 7 Borrowing to finance investments is common, and user charges service the loans. Finally, chapter 10 attempts to give an integrated view of the findings from the City Study. The main lesson seems to be that the most practical way of coping with city growth is to endogenize instittutional responses to the rapidly changing and unpredictable demands made by the city's residents. The needs of citizens of a growing city are multifaceted. Whereas overall patterns of city growth are predictable, many individual and group needs are not. lnstitutions and methods of responding to these changing situations must be developed in such a way that little central direction is necessary. The institutions themselves should be flexible enough to respond to the changing requirements of a city. Notes 1. As the level of per capita income starts growing in a developing country, the level of urbarnizationi first rises slowly, then accelerates, and finally slows down. 2. Because this study has focused on the economic sphere, the repercussions of the deteriorating social and political situation in Colombia have been ignored. 3. Almost all the findings in this study are empirically based. An appendix documenits the vast data sets that have been used in the studies that underlie this book.  Chapter 2 The Spatial Structure of Cities in Developing Countries Urbanization has been the most pervasive development phenomenon in countries whose per capita income has growrn from low to high levels. In most couLntries the emergence of large cities has been part of the urbanization process. In some of these countries, one city dominates the urbani system. In others, there are a number of large cities along with a continuum of smaller settlements, with urban activities well distributed among them. However, a notable feature of this century, and particu- larlv the latter half of this century, has been the emergence of the metropolis as a familiar place for habitation. The greatest growth of large cities, both in the size of individtual cities and in the number of large cities, has occurred in the developing countries. As late as 1960, the number of large cities (those with a population of over half a mil- lion) in developing countries was about the same as in developed coun- tries: about 100. By 1990 there were probably more than 300 such cities in developing counitries-about double the number in developed coun- tries. In 1960 such large cities accounted for a third of total urbani popU- lation in developing counltries; now they account for about half, a proportion similar to that in developed countries. As the nutmbers rise, it becomes incr-easingly importanit to uncderstand the reasons behind the emergenice of these cities, the dynamics of their growtlh, and their chaniging spatial structures. Are there similarities and patterns that can be observed and understood? The Emergence of Cities Urbaniization takes place when the predominant economic activity in a region shifts from primary production (agrictulture, mining, logging, and such) to secondary and tertiary productioni (processing and servicing 18  THE SPATIAL STRUCTURE OF CITIES IN DEVELOPING COUNTRILS 19 activities). Primary activities are characteristically land-intensive and are therefore spatially dispersed. Secondary and tertiary activities are capital- and labor-intensive and are therefore spatially concentrated. Because the elasticity of substitution between land and nonland production fac- tors in secondary and tertiary activities is greater than that in primary activities, a greater use of capital and labor per unit of land is possiblc in secondary and tertiary activities. As incomes increase with economic growth, the demand for services and nonfood goods rises, whereas the relative demand for food falls. At the same time, increases in agricultural productivity make it possible to produce more food with a smaller work force. Labor then shifts from rural areas to those where production of nonfood goods and services is concentrated. Those areas thtus become ut-ban. What acttLally constitutes an urban area? The traditional characteris- tics of an urban settlement are a population above a given size, a high density of population, and a predominance of nonagricultural activities. Why do secondary and tertiary activities concentrate in such locations. Because such activities typically exhibit economies of scale, their exist- ence r equires a spatially concentrated work force. This work force, along with its dependents, requires an increasing number of activities and people to service its needs, as do the industrial activities themselves. Population begins to snowball as more people and businesses move into the area to meet the needs of the existing enterprises. The growing demand for complementary services and products produces agglomnera- tion effects-economies achieved because of the presence of a multi- tude of activities-and lead to expansion of the economy as a whole. The combination of scale economies, agglomeration effects, transporta- tion needs, and comparatively higher nonland/land substitutioni elastic- ities in industry and services produces a concentration of people and economic activity. Thus cities emerge. Different industries exhibit different levels of economies of scale. The existence of some industries with greater degrees of scale economies leads to the emergence of large cities. The further specialization of such cities into high-level service, governmental, education, and financial sectors adds to their size. These growing cities offer increasingly varied employment possibilities for workers and their households. The concen- tration of more skilled jobs also attracts more highly skilled workers from the smaller cities and towns. This in turn generates more jobs and better communication, transportation, water, and sewerage systems, all of which make it possible for cities to grow. As an economy switches into production of more "moderni" goods and services, its cities continue to expand (Henderson 1988). Eighteenth- and nineteenth-century urbanization in European coUnI- tries was dominated by the imperatives of the Indtistrial Revoltition.  20 UNNDERSTANDINCG THE DEVELOPING METROPOLIS Most people who came to the cities did so to work in the new mills and factories. Current urbanization in developing countries is different in that the tertiary sector requires more labor than does the manufactur- ing sector. The existence and expansion of cities is therefore equally determined by the economies of scale inherent in certain industries and by the localization economies achieved by the agglomeration of differ- ent kinds of service activities in one place. Understanding the reasons for the emergence of large cities is essen- tial to understaniding their internal urban structure. The term "urban structure" refers to the kind, location, and density of activities as they are distributed across space in urban areas. The goal of this study is to understand both how this structure changes as a city grows and what determines these changes. The main constituents of a city are the people who live there, the firms that do business there, and the government. It is the residents who generate the economic activity that supports the city. Their key demands are for housing-which typically accounts for half of land use in cities-and for transportation to get to work, school, shopping, and leisure. The firms in a city employ people in Inanufacturing, trade, retailing, and other services. They decide where to locate based on the pattern of demand for their products, the technology they use in pro- duction, the location of their suppliers, and their distribution capabili- ties. The government's role is to supply infrastructure and public services and to regulate economic activity. A key area for government interventioni, whether through direct supply of services or through reg- ulationi, is that of transportation. Because people, firms, and governiments in different countries con- centrate in cities for similar reasons, and because there are common and compreherisible patterns of human behavior, cities throughout the world have quite similar structures. Given such broad similarity, the structure of any city can be studied along similar lines. Differences can be tinderstood by examining the functions of the city, the nature of eco- nomic activity found there, the characteristics of the people who live there, and the activities of the government. This chapter lays out the broad characteristics of cities, giving com- parative information for rich and poor countriies wherever possible. The next chapter describes the changes that have taken place in Bogota and Cali as their structures have changed with growth and development over the past quarter-centurv. These two cities' structural similarity to other cities in both developed and developing countries gives credence to the wider applicability of our detailed behavioral studies. The goal, then, is to look for useful ways of describing cities in a summary fashion and to identih; the phenomena that are common to all or most of them. Decentralization, which is fundamental and pervasive with city growth, is one such phenomenon.  THE SPATIAL STRUCTURE OF CITIES IN D)EVELOPING COUNTRIES 21 Growth and Decentralization The concurrent growth and decentralization of cities is nothing new. Bowden (1975, p. 78) reports thatJ. Stow described these processes at work in London in "A Survey of London, Written in the Year 1598": In 1598 London had a central business district (in the modern sense) but it had not had one for long. The London Stow describes is still predominantly medieval in that merchants, artisans, the rich and poor, storage and manufacturinig were found scattered in all wards and parishes. Great parts of the city were residential. Most people (although a declining majoritv) lived where they worked.. Yet specialized warehouses were remarked in the parishes ... an( the old district of the Italian galleymen; tenements and squatter set- tlements of aliens, primarily Flemish refugees, were concentrating in the suburbs, as well as within the city wall, many formerly clustered activities that we think of as dispersed were seen by Stow as dispersing for the first time, e.g., grocers, vintners and cooks, and mantifacultr- ing-particularly that associated with the metal and leather trades- was gravitating to the north. From a population of 50,000 in the 1540s, the city's population had probably trebled by the turn of the centUry... the crucial change is the centripelal movement of the pro- genitors of the retailers of luxury shopping goods. "Men of trades and sellers of wares in this city have often times changed their places as they have founld their best advantages." ... Most of the luxury traders and their great merchant counlterparts had moved toward the center of the city (often without moving their r esidences with them). This is a good description of the processes of concentration and decon- centration of cities, and one that is quite applicable in developing coun- tries today. Typically, activities first converge in the city center and then deconcentrate later, a process that had clearly begun in L.ondon by the sixteenthl century. Moreover, the separation of residence and workplace, while not common then, had begun. The central business district (CBD) had begun to gain specialized functions. Adna Weber, in his classic work on the structure of nineteenth-cenittury cities, said that "the most encour- aging feature of the whole situation is the tendency.., toward the devel- opment of suburban towns. The significanice of this tendency is that it denotes, not a cessation in the movement toward concentration. but a diminutioll in the inlensitv of concenitrationi" (Weber 1963, p. 458). Thliere has been a tendency to thlin-k of decentralizationi mainly in terms of the widespread growth of American suburbs in the 1950s and 1960s, which was remarkable because of the related decline in the city- center activities. I'hat was a special case of decentralizationi, perhaps typ- ical of the richest countries, but decentralizationi in general shotild be  22 UNDERSTANDINC, THE DEVELOPING METROPOLIS regarded as a wider city growth phenomenon. Mills and Song (1979, pp. 82-83) describe this process succinctly: The most important causes of urban decentralization are the growth of metropolitan areas, rising real income, and improved urban transportation. In small urban areas, the purchases of an entire urban area's population are needed to support its commercial and industrial activities, and they are therefore located centrally. As the population and real income of an urban area grow, it becomes possi- ble to support shopping and employment centers with the customers and labor force of'only part of the urban area.' Thus sub-centers of stores and workplaces appear away from the central business district. Thus fewer people are tied to the city center for jobs and shopping, and they are attracted to suburban residences because of lower land valties and correspondinlgly lower population densities. Real income growth has an additional important effect. As income rises, a family's housing demand rises and it is induced to move further from the cen- ter to take advantage of low land values. Improved transportation, whether- by public transit or automobile, has the same effect. It increases accessibility to the central business district from distant parts of the urban area, thus permitting people to take advantage of cheap suburban land for housing. Measurement Issues How is decentralization to be measured? One obvious measure is the proportion of people living or working in the CBD or in the central city. As long as the boundaries of these entities are kept constant, this mea- sure should give a good sense of concentration, if' observed at different times during the growth of a city. Such a measure can be supplemented by the proportion of people livirig or working in successive rings around the city center. The problem with such measures, however, is that boundaries of the CBD, central city, and iings are arbitrary (although often taken to be legal boundaries), and observation of concentration or deconcentration will depend on their inclusiveness. These bound- aries are often enlarged as a city grows. Moreover, the concept of the (CBD, although clear in principle, is not clear for measurement purposes. Where are its boundaries? As a city grows, the CBD typically expands. For these reasons, it is better to find methods of describing the density pat- tern of the whole city that are not so sensitive to boundary-definition issues. Given the observation of and theoretical justification for density pat- terns that decline from the city center, it is natural to describe them as being given by a functioni that shows density declining exponentially  TFIE SPATIAL STRUCTURE OF CITIES IN DEVELOPING COUNTRIES 23 from the city center: Dx = Do e-X, where D', is density at distance x from the CBD, Do is the estimated density at the center, and g is the estimated rate of decline of density. Thus g may be interpreted as the percentage decline in density per unit distance. The larger gis, the greater the rate of decline from the center and the more concentrated the city. The decline in g over time then measures the rate of decentralization of the city. And because g can be measured if there is information on the population density of different parts of a city and because it is expressed in units of density decline per kilometer or mile, its value can be com- pared across cities. Vlalues are given for a whole range of cities in the next section. A few issues concerning measurement need to be mentioned. First is the selection of the exponential functional form. Several articles (for example, Griffith 1981; Andersoni 1982; Wheaton 1982a, 1982b) have questioned the validity of the simple exponential form and suggested more complex forms for estimation. One criticism is on1 theoretical as well as econometric grounds. The simple exponential form can be derived firom economic theory assuming static equilibrium in the urban area (Mills 1967). It is argued that, in fact, urban areas grow from a his- torical process of successive accretion, and, because of the durability of structures, the urban area is not in equilibrium at any given time. The density functioni should therefore be derived firom a more realistic model of city growth. Another criticism concerns CBDs with little resi- dential population. There is a doughnut-like hole in the city center (Anderson 1982). Moving outward from the city center, the populationi density first rises and theii begins to decline toward the suburbs. Thiis is observed in some cities in developing countries as well as in most devel- oped countries (see, for example, Asabere and Owusu-Banahere 1983 for African cities). There is considerable merit in these arguments, and the functional form can clearly be refined. The elegant simplicity of the negative expo- nential is appealing, however, and although inexact, it is not inconsis- tent with some of the other formulations suggested. It also has the advantage of being the most common functional form estimated and is therefore available for comparison purposes.2 The general problem of the empty city center is often solved bv excluding city-center observa- tions from the estimation sample. The recent refinements are also use- ful in interpreting the results of the simple estimation. Chief Determinants of Decentralization It is useful to obtain some idea of what the more powerful determinants of decentralization are and which of these are expected to be especially relevant in developing countries. The most important determinant is  24 UNDERSTANDING THE DEVELOPING METROPOLIS increase in population. Historically, the decrease in transportation costs has also been very important; this was especially true during the major transition from animal-drawn vehicles to motorized vehicles and later from rail transportation to private automobiles (the latter particularly in developed countr-ies, wher e the advent of the limited-access highway fol- lowed the rise of the automobile). In developing countries many modes of transport coexist, an(d different classes of people have access to differ- ent modes at different prices (see clhapter 8). The third key determi- narnt has been rising per capita income. Again, this has been more important for developed countries in the last half-centurv. The rapid rate of city growth in developing couintries has includedl high rates of immigration; this has meant that, paradoxically, the rate of per capita income growth has been low in the largest cites even when it has been high for the country as a whole. In-migration has essentially been an equilibrating mechanism, reducing the disparities between large city/ other and between overall urban/rural income levels. Rates of income growth have therefore been high for some groups of people and corre- sponding parts of cities, and low for other groups and parts of cities. Thus, growth in population is a good explanation for residential decen- tralization in some parts of a city, whereas incoine growth may explain decentralization in another part. Rich people often live largely in one area of a city, and as the city expands and income grows, they contintte to fill spaces in the same directioni as city growth; the poor, meanwhile, live in another area of the citv aind expand in another direction. The density patterns are quite dif- ferent for these two segments of the cit. The standard residential pat- tern in cities in developed countries is one of the rich living in outer rings and the poor in inner rings. Theoretically this is because the demiianid for housinig space is highly inlcome elastic for the rich, and lower land prices at the periphery more than offset the increased trans- portation costs that result fiom living greater distances fromii the city ceenter. In cities in cleveloping countries, however, it is now quite com- mon to observe manv low-income suburban settlements, as well as a few high-income residential suburbs, on the city periphery. This pattern is documented in detail for Bogota and Cali in chapters 3 and 4. These low-income settlements are often of high densities. An examination of density gradients, g, in such cases yields the expected flattening out- but the flattening out is at a high density level everywhere. In cities in developed countries, the flatteninlg out is at the low (lensity levels that result from decreases close to the center and increases at the peripherv. It is Fascinating to speculate on the reasons for these different pat- terns. One is explicit government policy. As Asabere and Owusu- Banahere (1983) have documented for African cities, in the ptursuit of an ideal Garden City concept, it has been quite common for govern-  THE SPATIAI STRUCTURE OF CITIES IN DEVELOPING COUNTRIES 25 ments to clear congested central cities, usually populated by the poor, and move the inhabitants to distant locations, often designed as sites- and-services projects.3 The original idea that concentration was bad emanated from the dismal health conditions in European industrial cit- ies in the eighteenth and nineteenth centuries. This idea has been transferred somewhat uncritically to developinig countries. Even when people have not been relocated, it is common to design high-density res- idential complexes for poor people at the periphery, because that is where land is available and inexpensive. The second factor in high-density peripheral development has to do with the nature of housing demand and the construction technology available to the poor. As chapter 7 and other studies document, the poor spend considerable portions of their income on shelter; in fact, even for middle-income residents, housing has low income elasticity of demand. Among the poor, price elasticity of demand is also low. Given low incomiie levels, the building technology used is pretty rudinmentary, and construction is a mixture of self-help and contracting. There is evi- dence from Bogota and Cali that homes in the poorer areas have fewer floors than homes in the richer areas (see chapter 3). Low-income hous- ing is seldom more than three stories, because going higher would require steel, more cement, and formal construction methods. The extent of substitution of capital for land in low-income housing is there- fore limited. The high proportions of income spent on shelter and the low income elasticity of demand also suggest some maximum threshold for crowding. The result of these two effects is only a small density gradi- ent in poor parts of the city: the areas near the center are only slightly more densely inhabited than those on the periphery. It is typically larger households with greater space needs that are pushed toward the out- skirts to take advantage of lower land prices. In the poorer areas there is little difference in per capita space betweeni the center and the periph- ery. Transportation costs affect this pattern, too. If there is a flat fare sys- tem, the poor can afford to live on the periphery, because they spend only more time traveling-not more money. Which Came First-Decentralization of Population or Businesses? In cities in developing countries and industrial countries alike, employ- ment decentralization appears to lag behind residential decentraliza- tion. There has been relativelv earlv decentralization of large and polluting manufacturing activities in cities in developing countries. As in residential development, the Garden City philosophy has played a role in locating large-and often even small-industry in peripheral industrial parks. Again, the desire for a clean, arboreal central city has been buttressed by the availability of cheaper land at the periphery. The  26 UNDERSTANDING THE DEVELOPING METROPO.IS deindustrialized CBD in these cities then tends to specialize in retail and wholesale trade, banking and finance, and other service functions. Given the paucitv of overall demand for services because of low incomes, these functions often become more concentrated during the initial stages of urban growth. If decentralization occurs, it usually happens after the richer residents have left the central city and established com- muniities that can sustaini subcenters of employment. There is little decentralization of service functions to the poorer areas, except to meet essential daily retail needs. Overall patterns of employment decentrali- zation can therefore be quite different between manufacturing and ter- tiary sectors and between rich and poor parts of the city. Given these considerations, we can expect a continuiing trend toward employment decentr-alization in cities in developing countries when those cities exhibit continuing population and income growth and residential decentralizationl. The preceding discussion illustrates how changes in the environment and the behavior of firms, households, and individuals are connected to changes in urbani structure as measured by density gradients. The next section considers the changes in density gradients that have been obser-ved in cities in both developed and developing countries. City Characteristics: Evidence from Different Countries What are some of the broad strtictural characteristics of cities, and how have they changed over time? This section draws on research reported in Ingram and Carroll (1981); most evidence is taken from Latin Ameri- can countries so that Bogota and Cali can be seen in their appropriate context. In 1970 twenty-four Latin American cities had central-city populations of more than 600,000 or metropolitan populations of more than 1 mil- lion. North America had thirt.-six metropolitan areas with 1 million or more inhabitants. Of the large Latin American cities, comparable data were available for only thirteen. Twelve large U.S. metropolitan areas from various parts of the country were selected on the basis of compara- tive size and rale of populationi growth dur-ing 1950-70. Overall, urban population growth was much higher in the 1950s than in the 1960s in both Latin America and the United States (see table 2-1). In the Latin American cities, the average annual growth rate declined from 5.5 percent in the 1950s to 4.6 percent in the 1960s; in the U.S. cit- ies it declined much more, from about 4.1 to 2.5 percent. Although it is difficult to assemble comparable information for the period since 1970, it is apparent that these trends have continued, and urban population growth has declined further in both regions. In the United States most  THE SPATIAI STRUCTURE OF CITIES IN DEVELOPING COL'NTRIES 27 older large cities lost population (luring the 1970s while the newer cit- ies, mostly in the Southwest, continued to grow. Table 2-1 shows that except for Mexico City, the larger cities grew somewhat more slowly than the smaller cities, with this tendency being much more pro- nounced in the Ulnited States than in Latin America. Preston (1979) confirms this general pattern worldwide. This sug- gests that, for the most populous cities, the advantages of lirbani agglom- eration economies and economies of scale tend to get exhausted, and some urban diseconomies might set in. This should not be overstatecl, however, since the largest cities in developing countries continue to grow at significant rates. The main reasons for the overall decline in urbani population growth in North America and in Western Elurope are that these regionis are approaching 100 percent urbaniization levels and their total population growth has become very small-near zero percent in some countries. Within the developing world, Latin Amel-ica has the highest urbanization levels, with many countrics more than 60 percent urbaniized. A major difference between North Americani and Latin American cit- ies is that large cities in Latin America (and in developing couLntries in general) have much lower real income levels. Gross national product (GNP) per capita in 1975 was $7,1004 (in 1975 dollars) for North Arner- ica and $1,000 for Latin America (World Bank 1977). Based on growth rates reported by Kuznets (1966), North America would have had a per capita GNP of $1,000 (1975 dollars) in the 1850s. There were no cities of more than I million people theni. Hence, although some similar changes are observed over time in developed and developing countries, their levels of magnitude are often quite differenit. It is interesting to observe chaniges in the intranietropolitan popula- tion distributioni as large cities grow. Table 2-2 summarizes data for a twenty-year period for the central and peripheral areas of the ten Latin American cities with available data and the twelve U.S. cities from table 2-1. It is striking that across the cities on both continenits, both the cen- tral and peripheral densities vary by an order of magnitude: this is an indication of the definitional problems that are encotuntered when mak- ing such comparisons. W'hat is the central city? Where does it end and the peripher-y begin? In most metropolitani areas worldwide, the area considered to be the central city has a residential density of about 5,000 to 20,000 people per square kilometer (see Mohan 1980). Central densi- ties are continuinig to increase in most Latin Amnerican cities, whereas they are clearly stabilizing or decreasing in North America. Peripheral densities are rising in nearly all cities in this sample, and rising faster than central-citv densities. Thus pervasive decentralization is taking place in both North American and Latin Americani cities, with the trelnd more pronounlced in the former. The L.atin American central densities  Table 2-1. Population and Poputation Growth in Selected Latin American and U.S. Metropolitan Areas Annual gowth Tate Annualg,rowUth Tate Popfnlation (thausands) (Percent) Population (thousands) (percent) Latin American citv 195() 196() 1970 1950-60 1960-7(0 UfS. ritY 1950 1960 1970 1950-60 1960-70 iMexicoCity 3,180 5,246 8,657 5.1 5.1 NewYork 9.556 10,695 11,572 1.1 0.8 Sao Paulo 2,708 4,818 8,195 5.9 5.5 Los Angeles 4,152 6.039 7,032 3.8 1.5 BuenosAires 4,723 6,739 8,189 2.8 2.0 Chicago 5,178 6,221 6,979 1.9 1.2 Rio deJaneiroo 3,298 5,012 7,082 4.3 3.5 Philadelphia 3,671 4,343 4,818 1.7 1.0 Limab - 1,846 3,302 - 5.4 Washington, D.C. 1,508 2,077 2,861 3.3 3.3 Bogot;!' 715 1,697 2,855 6.9 5.9 Boston 2,414 2,595 2,754 0.7 0.6 Santiagod 1,509 2,170 2,820 4.6 2.7 fHouston 936 1,418 1,985 4.2 3.4 Caracas, 724 1,388 2,199 6.1 4.7 San Diego 557 1,033 1,358 6.4 2.8 X Recife 819 1,240 1,793 4.2 3.8 Miami 495 935 1,268 6.6 3.1 Belo HorizotiLe 475 888 1,606 6.5 6.1 D)enver 612 929 1,228 4.3 2.8 Guadalajara 440 851 1,455 6.8 5.5 SanJose 291 642 1,065 8.2 5.2 Monterrey 376 708 1,213 6.5 5.5 Phoeniix 332 664 968 7.2 3.8 Cali' 284 638 898 6.4 3.9 - Not available. a. Data are for 1947, 1960, and 1970. b. Data are for 1950, 1961. and 1972. c. Data are for 1951, 1964, and 1973. d. Data are for 1952, 1960, and 1970. e. Data are for 1950, 1961f, and 1971. Source: lngrTali and Car roll (1981).  Table 2-2. Population Density in Central and Peripheral Areas of Selected Latin American and U.S. Cities Area (squeare Population per square kilomneter Area (square Population per square klomneter Latin Anmcan city Location, kilometers) 1950 1960 1970 US. cily Localions kWlometeri) 1950 1960 1970 Mexico City C 138 16,225 20,558 21,074 New York C 777 10,157 10,015 10,161 P 2,192 432 1,101 2,675 P 4,758 350 612 773 Sao Paulo C 1,493b 1,380 2,287 4,005 Los Angeles C 1,326 1,675 2,130 2,390 P 6,458 79 172 343 P 9,213 210 35() 420 Buenios Aires C 200 14,952 14,872 14,897 Chicago C 578 6,275 6,140 5,825 P 3,860 473 1,025 1,418 p 9,054 170 295 400 Rio dejaneiro C 1,171 2,030 2,824 3,631 Philadelphia C 334 6,202 5,995 5,835 P 5,293 174 322 535 P 8,868 180 264 325 Bogota' C 304 2,352 5,582 9,391 Washington, D.C. (C 158 5,077 4,835 4,790 P - - - _ P 5,936 119 221 355 Recife C 290d 3,594 3,815 5,075 Boston C 119 6,735 5,860 5,387 P 1,992 148 223 367 P 2,435 662 780 867 Belo Horizonte C 335 1,053 2,070 3,686 Houston C 1,028 58() 913 1.198 P 3,335 37 58 111 P 15,250 22 31 49 Guadalajara C 188 2,204 3,940 6,383 San Diego C 552 606 1,038 1,258 1P 1,164 52 95 220 P 10,484 21 44 63 Cali' C 85 3,341 7.506 10,565 Denver C 246 1,690 2,0(08 2,092 P - - - - P 9,233 21 47 77 SanJose C 300 318 681 1,480 P 3,067 64 143 202 Plhoenix C 642 166 684 906 P 23,069 10 1 0 17 -Not availahle. a. C = center; P = peripher-N. b. 1,622 in 1959 and 1960. c. Data are for 1951, 1964, and 1973. d. 146 in 1950. Source: Inigram and Carroll (1981).  Table 2-3. Population Density Gradients in Selected Latin American and U.S. Cities Latin Anmeiran fitV ph Parnae1e, 1950 1960 1970 '.S. cty ph Pareametert 1950 1960 1970 Mexico City 1.0 /O 69,00(0 62,000 44,000 New York 0.t D0 62,000 45,000 40,000 g 0.37 0.27 (1.17 g 0.16 0.13 0.11 Sao Paulo 1.0 D, 8,400 12,000 18,000 Los Angeles 0.6 D, 4,800 5,300 5,800 g 0.14 0.13 0.12 g 0.06 (.(6 0.05 Bueinos Aires 0.6 D, 54,000 37,000 33.000 Chicago 0.5 D, 27,000 20,000 16,000 g 0.21 0.14 0.02 g 0.13 0.10 0.08 RiodeJaneiro 0.5 Do 8,700 10,000 11,000 Philadelphia 1.0 Do 20,000 16.000 14,000 g 0.(9 0.08 0.07 g ().18 0.15 0.13 Bogota 0.5 D. - 37,000 26,000 Washington, D.C. 1.0 DO 15,000 11,00(( 9,000 g - 0.25 0.12 g 0.25 0.18 0.14 Recife 0.6 DO 13,000 14,000 17,000 Boston 0.7 Do 14,000 11,000 9,300 g 0.25 0.21 0.19 g 0.16 0.13 0.12 Belo Horizonte 1.0 Do 5.000 11,000 19,000 Hotistoni 1.0 D(, 2,100 3,500 4,200 g 0.26 0.28 0.27 g 0.12 0.13 0.12 Guadalajara 1.0 D,, 14,000 28,000 39,000 San Diego 0.5 D, 2,100 3,200 3,600 g 0.45 0.46 0.41 g 0.11 0.10 0.09 Moniterrev 1.0 Do 6.200 8.500 7,40(0 Miarni 0.5 Do 8,000 6,800 7,200 g 0.32 0.27 (.19 g 0.23 0.15 (1.13 Cali (1.5 Do - 43,000 29,000 Denver 1.0 D)0 6,8(00 6,0()() 5,100 g - 0.41 (0.21 g 0.27 0.20 0.16 San Jose 0.5 D0 620 1,300 3,500 g 0.08 0.07 ().10 Phoefnix 1.0 Do 350 2.700 3,100 g (0.(08 0.16 0.14 -Not available. .Not: Density gradients were calCulated using the technique described in W1hite (1977). a. Parameters firom density = Del7X where xis the distanice in kilometers from the city centergis the density gradient per kilometer, and D. is the estimated central citv density. Saurce: Ingram and Carroll (1981).  THFE SPATIAL SlTRLICTLIRE OF CITIES IN DEVELOPING COUNTRIES 3 I are akin to older North American cities, and the newer North American cities are structurally dispersed, with low densities all over. Data in table 2-2 are based on arbitrary definitions of central and peripheral areas. Density functions are more robust than actual central- city densities as proxies for detecting city structures and for summariz- ing changes over time. As Mills and Tan (1980) report, urban popula- tion density ftictions have been estimated for almost every developed countrv in the world, and an increasing number of estimates are avail- able for developing coulitries as well. As menitioned earlier, the expo- nential density function provides a good summary indicator- of city structiure, with g measuring the percentage of decline in density per unit distance ancl D,, measuring the theoretical density at the center. The concept behind this functional form is that of a circular mono- centric citv. There has been considerable discussioll of Lhe use of other functionial forms to accounit for two problems: cities typically have more than one large employment center, and the city center typically has very little residential populationi since most land is used for economic activ- ity.5 These are useftil elaborations of the simple theory and provide for more realistic estimation, but they lose the elegant simplicitv of the exponential function. Moreover, they cannot be used for comparison purposes because most estimates that exist are for the simple exponen- tial fo m. Thus, g is a good summiiar y measur-e of decentralization. For a fixed radius x of a circular city, the smaller the g, the smaller the proportion of people living within x distance of the city center and the more decen- tralized the city. Hence, when g declines from one decade to the next, the city can be said to have decentralized. Table 2-3 displays estimates of density function parameters for the same sample of Latin American and U.S. cities. The density gradient g declined over time in most cites, incr-easing only in Guadalajara, Belo Horizonte, Houston, San Jose, and Phoenix-and in each case in only one of the two decades. The decline in the populationi density gradients in Bogota and Cali was particularly significant: both fell by almost halt between 1960 and 1970, implying a notable decentralization of popula- tion. Latin American cities generally have steeper gradients than U.S. cities, as do smaller cities compared with larger ones. Intercept (central city) densities (I),) are significantly higher in Latin American than U.S. cities; the averages are 24,000 and 10,000, respectivelv. Latin American city centers are also more highly congested in terms of residential population. WNThen we combine the comparison of intercept densities with that made for density gradients, two major points emerge. First, there is a surprising degree of similarity betwveen the density funiction parameters of large Latin American and U.S. cities. This similarity is enhaniced  32 tTND)ERSTAND)IN(I THE D)EVELOPING MFTROPOLIS when we compare the large Latin American cities to only the five older Northeastern cities (New York, Chicago, Philadelphia, Washington, and Boston). The 1970 average intercept and gradient, 17,700 and 0.116, for these five cities are similar to the 1970 averages, 24,600 and 0.12, respec- tivel1, for the large Latin American cities. It appears, too, that density function parameters follow different patterns in Latin America and the United States: As size varies, U.S. cities have fairly constant density gradi- ents and varving intercept densities, while Latin American cities have fairly constant intercept densities and varying density gradients. Cou- pling this pattern with the similarity of parameters for large cities suggests that the relatively smaller Latin American cities have larger intercept densities and steeper gradients than similar U.S. cities and therefore are much more centralized. These patterns are quite common the world over and in recent history. The regular decline in gradient over the past century (see table 24) is remarkable, and the pattern clearly transcends countries and cul- tures. The pictur-e for a larger sample (see table 2-5) is a little more var- ied, but the broad patterns continuLe to hold true. The largest cities- those with population between 5 million and 10 million-appear to have a uniformly low gradient of arounld 0.1 or less, whereas smaller cit- ies generally have much higher gradients. More densely populated coUltr ies, such as India,japan, and the Republic of Korea, tend to have higher overall densities and higher gradients in their smaller cities than Latin American coUlltinies do. Altlhough density functionL comparisons have been based on city size to some extent, the data suggest that city age, transportation technology, and per capita incomie are also important determinants of the decen- tralization observed in a city (Harrisoni and Kain 1979; White 1977). For example, density gradients vary for North American cities of different vintage, and they are similar for Latin American cities and older North American cities. Older cities developed during a period when public transit was the dominant mode, as it still is in most developing coun- tries. By the time the newer U.S. cities, like SanJose, Phoenix, Houston, and Los AnEgeles, emerged and developed, automobiles had become the dominant form of transportation. The older cities in industrial coun- tries and most cities in developing countries are therefore more central- ized and have higher densities than the newer North American cities. This also suggests that portions of developing-country cities that are newly developed or inhabited by the rich could be similar to the newer North American patterns of citv development. The level of structural similarity is quite surprising. Living conditions, per capita incomes, patterns of housing, and levels of public service infrastructure vary considerably among North America, Latin America, Europe, Africa, India, and East Asia, as do the histories and culttural  THE SPATIAl. STRLICT LRE OF CFITFS IN DEVEl. PING COUNTIRIE:S 3S Table 24. Population Density Gradients Per Kilometer in Selected Cities Worldwide by Decade, 1880-1960 Average fon- f(n/r Kingston )Year iU.S. ritiei" Chicago London Paris BRmbay (Jamalical) 1880 1.22 0.49 0.38 0.60 0.37 - 1890 1.06 0.32 - - 0.33 1.02 19(0 0.96 0.22 0.23 0.50 0.26 - 191( 0.80 0.23 - - 0.24 0.9( 1920 (.69 0.16 0.17 - 0.20 - 1930 0.63 0.13 0.17 0.50 0.17 - 1940 0.59 0.13 0.14 - 0.17 0.54 1950 0.50 0.11 0.12 0.21 (.13 - 1960 0.31 - 0.09 - 0.1( 0.33 - Not available. a. Baltimore. Milwaukee, Philadelphia, and Rochester. Source: Mills anid Tall (I 980). influiences of each of these regions. Yet decentralization of cities has beeni a pervasive phenomenon accompanying growth. Moreover, judg- ing from Nortil Amnerican and Western European cities, decentraliza- tion continutes evCeI after population growth stops. This is what attracted the most attentioni in the 1960s and 1970s, because its effects-the emp- tying out of central cities and the increasing residential separation of the rich in suburbs and the poor in the center-are stark. The phenom- enon of decetntialization observed in developinig countr-ies is then closer- to the earlier pattern for cities in developed counitries, whe n growth was accompanied by decentralization. The distinction esselitially is that in these periods D0 continies to increase or r emains constanit as g declines, whereas in later stages 1) declines as well. The documentationi for employment dlensity gracdients is less abunl- dant because of the difficulty of obtaining spatial data on employmenit, which is not typicallv reported in population censuses. However, the pat- tern of continuiing decentralization is well documented for some couIn- tries, suchl as Brazil,Japan, the Reptiblic of Korea, and the United States (for example, see, respectively, Y. J. Lee 1985; Mfills and Ohta 1976; Mills and Song 1979; and Mills 1972). Employmenit density gradients are typi- cally muich steeper than those of populationi density, which suggests thc famriliar overall pattern of inward coMInTLting for work. Employment decentralization has ustually lagged behind residential decentralization. Mantifacturing is usually the first sector to begin decenitralizing for the technological reasonis discussecl earlier, essentially con nected with the ease of substitutinig capital for land. The portion of retail trade that serves the dlaily needs of consuImlers typically decentralizes with popula- tion, but the other poi-tiotis remain centralized, along with much of wholesale trade. Office activities oftetn remaini the most centialized; the  341 UNDERSTANDING THE DEVELOPING METROPOLIS possibilities of capital/land substitution are extremely high for these activities. Because commercial activities are the most directly related to constumer buying pOW7er, their rate of decentralization is closely related to prevailinig income levels (see table 2-6). It is therefore reasonable to expect less centralizationi of these activities in rich cities and in richer sections of poor cities. Clhapter 6 documenits these patterns as they applv to Bogoti and Cali. The Consequences of Growth and Decentralization Because we can expect continued growth of large cities along with decentralization, it is necessary to examine some of the consequences of these patterns. First, we need to understandcl that although the process of ur-bani growth and decentralizationi in developing cotuntries is similar Table 2-5. Population Density Gradients Per Kilometer in Selected Cities Worldwide, 1950, 1960, 1965, and 1970 CiIN 195() 1960 1965 197(0 Large cilies New York 0.16 0.13 - 0.11 l'okvo 0.06 0.07 0.08 (.08 L.oldo.l - 0.08 - 0.07 Madras 0.24 0.24 - - Great Britain' - (.10 - (.06 West. Germani'v - 0.25 - 0.21 japan' - - - 0.48 Seou(l - 0.35 0.33 0.22 AVedium-size cities Birmingham - 0.12 - 0.04 Manchester - (.08 - 0.06 Accra - 0.52 - 0.46 Poona (India) - 1.05 - 1.07 I'tisani (Repuiblic of Korea) - - 0.26 0.13 Kiinasi ((;hana) - - 0.61 0.55 Taegu (Republic of Korea) - - 0.78 0.74 Bangalore (India) (.61 0.53 - - (Great Britainh - 0.34 - 0.25 W'est GermanvO - 0.31 - 0.31 japar" - - - 0.49 - N<it available . .1. Data are averages for cities with populations of rmiore than I million. bc Data are averages for cities with poptilations of 0.5 million to 1 million. Source: Ingiraimi and Carroll (1981) for New York; Mills and Ohta (1976) for Tokyo; Mills and Tan (1980) for Madras, Poona, and Bangalore; Njills and Song (1979) for Seotl, Pusan. and laegu; Asahbre anti Owust-Banahere (1983) fior Kmnasi:(Glickman (1979) for all others.  Table 2-6. Central City Shares of Residential Population and Employment, Selected Latin American Cities (percent) Residentialpopailation Ewonomically Professionalsl Empklrtnwnt active technical OJfice workers/ Senvice Blue-collar Year population workers salespeople Zaei ns workers CGar oa'ners vlanuflaring Commerce Service Mexicos City 195(0 73 86 81 70' _ _ _ _ _ 1960 59 72 66 49d - - - _ 1970 39 52 47 43 28 - - - Sao Paulo 1950 85 - - - - - 84 95 1960 - - - - - - 8( 92 90 1970 75 - - - - 82 71 - - BueniosAires 1950 65 - - - - - 67 - 83h 1960 46 62 55 57 - - 49c -66b, 1970 37 56 48 44 - 54 - - - Rio delaneiro 1950 76 - - - - - 77 88 1960 - - - - - - 75 86 82 1970 56 - - - - 79 76 - - Recife 1950 68 - - - - - 59 86f 1960 - - - - - - 63 85 86 1970 63 - - - - 82 63 - - Belo Hori7onte 1950 77 - - - - - 56 91 1960 - - - - - - 47 92 88 1970 81 - - - - 90 53 - - Guadalajara 1950 87 95 94 91t - 94d - _ 1960 88 96 94 9() - - 92e - - 1970 84 91 89 86 84 - 87 - - Monterrey 1950 91 94 95 92' - - - - 196() 86 89 89 85' - - - - 1970 66 78 76 71 70 - - - - -Not available. a. Service workers and blue-collar workers combined. b. (Commerce anid ser%ice combined. c. For 1964. d. For 1956. e. For 1965. Saurre: Ingram and Ca-rroll (1981).  :6 tUNDERSTANDIN(; THE DEVEIOPING; METROPOLIS to that in developed countties, the existence of different initial condi- tions and behavioral parameters ofteni makes the restilts quite varied. The emergence of large, low-income, high-density settlements on the peripheries of cities raises the questions of where jobs and transporta- tion networks should be located and how the costs of public services should be structured. For example, the establishment of a flat-fare trains- portation system would help residents overcome the disadvantage of liv- ing far from the center, but such a system would give the wrong signals about the associated resotirce costs of transportation. The provision and pricinig of other essential services, suclh as water, sewerage, and drain- age, presenit a similar problem. Residential decentralization stretches oUt the supply networks of these services, which are then expensive to provide. If the poor have little choice about where to live in the first place, however, they can hardly be charged high prices for essential ser- vices. Some of these issues are therefore difficult to analyze because of their politico-economic natutre. Another common consequence of residential decentralization is the increased spatial separationi of the rich in suburbs and the poor in the central city (see chapter 4). This separation has economic as well as social effects. Employment opportunities are likely to be more plentiful in the better-off areas, where the residents have a greater demand for goods and services. The poor then either commute long distances to work in these areas, or they seek residential land nearby. This often gives rise to squatter settlements and slums in the middle of otherwise rich neighborhoods. Because of the high land values and presstires to use the land more "productively,' the government often relocates poor resi- dents, frequently by force. The decentralization of employment and residences naturally causes a rapid conversion of agricultural land to urban uses. This is often per- ceived to be a major problem, and fears are expressed about the dwin- dling supply of agricultural land. This is, on the whole, a false concern. Even in the most (lensely populated countries, such asJapan and Korea, only a small fraction of total arable land is urbanized. The loss of agri- cultural land is, however, a legitimate concern for the farmers whose property is tirbanized. even if they are fairly compensated. When they lose their livelihood, they often do not receive appropriate training and employment in alternative occupations. The faster the growth and decentralization of a city, the more serious this problem is, as the coin- petition for land at the uLrban fringe becomes intense. The challenge for policymnakers is to ensure that land is allocated to its most economiiically efficient use and that the legitimate interests of owners and users, both old and new, are protected.  THE SPATIAL STRUCTURE OF CITIES IN DEVELOPING COUNTRIES 37 Notes 1. Compare this development with the "dispersing for the first time [ofl gro- cers, vintners and cooks" in sixteenth-century London. 2. For theoretical derivation and review of this fuinctionial form, see Mills (1967, 1972) and Mohan (1979). 3. See Berrv (1973) chapter 3 and jacobs (1961) chapter I for discussion of the Garden C ity concept. 4. All dollar amounts are U.S. dollars unless labeled otherwise. 5. For example, Griffith (1981) discusses a generalizationi of the simple expo- nenitial form to account for multicentered cities; Anderson (1982) proposes a cubic spline function to account for an empty center.  Chapter 3 Growth and the Changing Structure of Bogota and Cali Unlike many other Latin American countries-for example, Argentina, Ecuador, Peru. and Venezuela-Colombia has a well-articulated system of cities in which Bogota accounts for only about 20 percent of the total urban population. Colombia's primacy index (that is, the ratio of the populationl of the largest city to the total urbani population of the coun- try) is the lowest in Latin America except for that of Brazil (see Renaud 1981). However, Bogota's share of both population and output has increased over the past few decades (see tables 3-1 and 3-2). Within forty years Colombia was transformed from a predominantly rural country to one that is predominantly urban; its urbanization level rose from about 30 to 65 percent. During the same period Bogota grew from an intermediate-size city (bv today's standards) into a large metropolis. It had a remarkably high and sustained rate of population growth, of about 5 percent a year over almost a half a century (see table 3-1). Tables 3-2 and 3-3 give the comparable growth in economic output since 1950. The growth rate of Bogota's regional product was 7 percent per year during 1950-75, while that for Colombia was 5.2 percent. But because population grew faster in Bogota than in the country as a whole, the per capita growth rate for the city's gross product was only 0.6 percent a year, versus 2.1 percenit a year nationwide (see table 3-3). Economic opportunities in Bogota led to substantial labor mobility, and the per capita differential between Bogota and Colombia as a whole nar- rowed over the years until it was only about 1.5 by the mid-1970s. It would be even lower if price differentials were accounted for. Bogota grew faster than other Colombian cities taken as a whole over this period (and earlier-as long ago as the mid-I 800s) . and it emerged with a high concentration of people and economic activity bv the 1970s. Cali, is  GROW[rH AND THE CHANGING STRUCTURE OF BOGOTA AND CALI 39 Table 3-1. Population and Intercensal Population Growth in Colombia Population 7btal tlrban' Bogota Annual compound growth rale (percent) Yea, (thousands) (percent) (percent) Total Urban' Bogota 1938 8,702 31 3.8 n.a. n.a. n.a. 1951 11,548 39 5.7 2.2 3.9 5.2 1964 17,485 52 9.5 3.2 5.4 7.3 1973 22.500 60 12.7 2.7 3.7 5.6 1983 26.965' 65b 151' 2.0 3.0 4.0( nia. Not applicable. a. The ur ban population is the percentage living in a county seat (cabecera mvumpal). Most cotntLy seats have 1,500 people or more, but some have fewer: moreover, not all set- tlements that large are county seats. b. Data are estimates from Sokol and others (1984). c. Bogota's 1978 population was estimated from the World Bank-DANF Household Sur- vey. The 4.0 pe rcent population growth rate observed from 1973 to 1978 was assumed to continue utitil 1983. Source: 1938, 1951, 1964, and 1975 population censuses; Sokol and others (1984). too, has grown consistently (see table 3-4). This record is consistent with the idea that cities grow because they are more efficient economic pro- ducers, benefiting from agglomeration and scale economies. The average population density of Bogota shows remarkable stability in the face of rapid population growth. Peak density occurred at the end of the 1950s, the period of most rapid growth. (The area data must be regarded as somewhat dubious because the juridical definition of the area of a city is always somewhat arbitrary, as is more evident in the data for Cali. The extension of a citv's boundaries eventually does, however, reflect the area that is regarded as belonigirig to the city at any given time.) The pattern for Bogota suggests that the juridical area of the city Table 3-2. Gross and Per Capita Domestic Product, Colombia and Bogota (olombia Bogetd Bogola;'s Bogoti s (;RP pet (;RP as a captla as a (;lSP G' (;RP GRP percentage of percentage of (milleions per capila (mtilon,% per capila Colombias Cooam/nas GoP Kear Jof pesos) (pesos) ofpesos) (pesos) 6d/p per cap/la 1960 77.71 4 5,088 11,996 9,220 15.4 1.81 1965 97,968 5,455 17,208 9,591 17.6 1.76 1970 130,361 6,484 25,920 10,822 19.9 1.67 1975 176,478 7,352 37,671 11.779 21.3 1.60 ,Vote. GDP = gross domestic product. (;Rp = gross regional product. Pesos are 1970 pesos: the exchange r ate was US$1 = Col$18.40. So,rrc. Coloimbia (1977).  40 UNDERSTANDING THE DEV'ELOP[NG METROPOLIS Table 3-3. Average Annual Growth in Domestic Product, Colombia and Bogoti (percent) Colombia Bogota GDP GRP Period CDP per capita GRP per capita 1950-75 5.2 2.1 7.0 0.6 1960-75 5.6 2.5 7.9 1.7 1970-75 6.2 3.2 7.8 1.7 Note: GDP = gross domestic product. (;RP = gross regional product. Saurce: Table 3-2. tends to follow growth in population, albeit with a lag. The constant average density implies that Bogota has always been decentralizing as it has grown, in the sense that the central parts of the city have housed a successively smaller proportion of the total population. It would be interesting to see whether such a pattern holds for other cities. If so, it would go against the general notion that cities become increasingly con- gested as their populations grow. Cali's density pattern is more erratic because the juridical area has been changed in fits and starts, but there, too, density stabilized in the 1970s and is now comparable to Bogota's. Overall land development has clearly kept pace with population growth in these cities, and the city area appears to have become an endogenous variable. How does Bogota compare with other cities in the world in terms of density? Bogota falls midway between the world's densest cities, such as Bombay (140 people per hectare) and Calcutta (120 per hectare), and such North American cities as Chicago, Philadelphia, Detroit, and San Francisco, which have about 40 to 60 people per hectare. New York City (inclutding the boroughs of the Bronx, Manhattan, Queens, and Brook- lyvr) is about as dense as Bogota, with more than 100 people per hectare. There is a large degree of similarity in overall population densities in large cities in the world, and Bogota falls somewhere in the middle. How has the physical character of Bogota changed as it has grown? Its original design was the traditional rectangular grid of streets and ave- nues, with the Plaza Principal as the focus of the city. Until the early twentieth century the entire built-up area of the city occupied what is now only the central business district (CBD). The city was primarily a commercial, political, and religious center; Medellin was the industrial capital of the country. (Medellin had more manufacturing employment than Bogota until about the late 1960s. It has since become better known for more dubious activities.) High- and middle-income residents of Bogota began to move north in the second quarter of this century as the southern and wester-n parts of the city began to fill up with lower-  GROWTH AND THE CHANGING; STRUCTURE OF BOGOTA AND CALl 41 Table 34. Area, Population, Population Growth, and Density in BogotAi and Cali Avetage annul,l Dendsil Populalan grozwtlh rate (population (ilv and Year .4rea (hectares) /thowsands) (percent) per hectaret Bogogali 1560 20 - - 160() 56 - -- 1670 129 3 - 23 1720 - 20, 3.9 - 1800 - 22' 0.1 - 1850 294 30 0.6 100 1886i 610 64 2.2 105 1900 909 100 3.2 110 1928b 1 .958 235 3.1 120 1938h 2,514 330 3.5 131 1951 - 660 5.5 - 1958 8,084 1,130 8.0 14() 1964b 14,615 1,730 7.4 118 19731, 30,423 2,877 5.8 95 1978' 30,886 3,500 4.0 113 Cali 1800 50d 6' - 120 1880 l 14d 15' 1.1 127 1900 _ 24r 2.5 _ 1918 - 45' 3.6 - 1938 4001 88 3.4 255 1951 1,290d 2841 9.0 187 1958 1,850d 428' 6.0 231 1964 ',1 OOl 638h 6.3 70 1973 9',100' 930"' 4.2 1(3 1978c 9,10(0 1.10 3.4 121 - Not available. a. Data are from Lubell and McCallum (1978), table 2.1. b. Data ai-e fromil population census estimates. c. Data art from the 1978 World Bank-DANE Household Sur vey. All data for Bogota- except for the vears marked a, b, and c-are from Wiesner (1980), table 1. d. Data are from the "Plan General de Desarrolo (le Cali suX Area Metropolitana. 1970- 1985-2000,1 an internal document of the Planeaci6n Municipal dte Cali. e. Tabares ( 1 979). f Area fbr C.ali is from the Planeaci6n Municipal de (Cali. and lower-middle-income residents. In the mid-1950s Chapinero, origi- nally an outlyinig suburb north of the city, began to replace the city cen- ter as the primary commercial area. The city became elongated north to south, being bounded on the east by high mountains and on the west bv valuable agricultural land and by low-lving land that discouraged devel- opment. Only in the 1960s and 1970s did the western parts of the city begin to fill, giving the city its present almost semicircular shape.  4Y2 UNDERSTANDINC THE DEVELOPING METROPOLIS Cali, which nestles into the base of the mountains to its west, has had a somewhat similar history. Until the early part of the century the built- up area was a compact, traditional rectangular grid of streets around the central plaza (Plaza Mayor), and the citv was mainly a commercial, reli- gious, and political center for the state of Valle del Cauca. Being small, it was mixed in character until it began to expand in the mid-twentieth century, when the rich occupied the western part of the city and the poor the eastern parts. Like Bogota, Cali developed linearly north to south, with greater development taking place toward the south. The eastern part of the city consists of low-lying lands populated by the poor, who settled there in great numbers during the major land invasions of the 1950s and early 1960s. Since then, the city has filled out and has acquired a semicircular shape-almost a mirror image of Bogota, although smaller. As in other cities, it is difficult to separate the effects of institutional decisions on the spatial development of a city from those arising from individual preferences as r evealed by the market. Because much is made in this study of the particular spatial income pattern found in both cit- ies, it is interesting to note some of the main planniiing exercises that have been carried out in Bogota. The two main influences on the shape of the city were Karl Brunner in 1935 and Le Corbusier in 1951. Much of Bogota's existing zoning and planning structure can be attributed to Brunner, who organized a city planning department. The northern part of the city was zoned for low-density residential development, while the south was zoned for high-density development. Le Corbusier's main contribution was the creation of an industrial zone to the west of the citv; intentionally or fortuitously, this separated the poor south-zoned for high-density' development unider Brunner's plan-from the rich north. These basic zoning and planning regulations have had a power- ful impact on the structuire of city growth. The low-density residential development zoned for the northern part of the city has effectively seg- regated it for the rich. Similarly, the high-densitv residential zoning in the southerni part of the city' made it possible for the poor to locate there. The industrial zone in the west houses the main indutstries located in Bogota. This pattern of growth is typical of cities as they expand, except that in Bogota land is probably more effectively segregated by use-between low- and high-density residential development, and among residential, commercial, and industrial uses-than is typical for many cities in devel- oping countries. In Bogota and Cali the central cities were densely packed to begin with and could become only a little denser as the city grew. New growth, in both residenitial and employment activities, took placc largely outside the city center. Although land use planning had significarnt effects on1 land tIse segregation, it is difficult to clearly iden-  GROW'TH AND THE CHANGING STRUCTURE OF BOGOTA AND CALl 43 tify the sequence of causation. Land use regulations may essentially have followed existing land use patterns and perpetuated their continuance. The rich suburbanized from the center in successive waves toward the north; the poor replaced thern, to some extent, in the center, as is typi- cal in North American cities, but they also decentralized toward the sou th. As Bogota grew, its employmenit structur-e diversified, and now about 25 percent of its labor force is engaged in manufacturing, another 25 percent in commerce, and about 37 percent in services. This distribu- tion is typical of well-diversified cities-almost all large cities in the developing world have more than 20 percent employment in manufac- turing activities (see Mohan 1986, chap. 3)-but some manufacturing noncapital cities, such as Bombay, Calcutta, and Sao Paulo, have as much as 40 percent of their labor force in mantifacttrinig. The presenice of the national government in a capital city clearly increases the propor- tion of people employed in tertiary activities. Cali has a higher propor- tion of manufacturinig employment and a lower share in the service sector than does Bogota. It is important to look at the overall employ- ment structure in a growing city because different employment struc- tures have different implications for the spatial pattern of city growth, and, as shown later, the location patterns of various activities have changed as Bogota has grown. Specifically, manitifactuirinig has led the trend toward decentralization of employment locationi. To summarize, both Bogota and Cali have grown at very high SUS- tained rates for more than a century. Income has grown faster than pop- ulation over at least the past three or four decades, and real per capita income has therefore grown substantially. The result of growth in both population and income has been a considerable physical expansion of the city, and both population and economic activity have decentralized. The following sections document the magnitude of this decentralization and note the departures from the general pattern that might be expected from a growing, primarily monocenitric city. Growth and the Spatial Distribution of Population and Income Maps 3-1 and 3-2 show the boundary lines for rings and sectors within BogotA and Cali. They also show the comunas-smaller divisions marked by two-digit numbers-that make up each ring or sector.' Map 3-3 illus- trates the spread and densities of population in BogotA. As might be expected, the outskirts are relatively sparsely populated, whereas the center is denser; the southern part of the city is clearly more densely populated than the north. The center (CBI, romuna 31) is less dense than some surrounding areas. The densest areas are more than ten  44 UNI)ERSTANDING THE DEVELOPING METROPOLIS Map 3-1. Bogoti: Ring and Sector Systems Based on 1973 Comunas IBRD 25346 Ring System X~~~~~~~~~~~~~~~ : - - < 5 4~ ~~~~~~~~~54 EgI Rings 93l Comunas JANUARY 1994 SsU7rce: M<>han 1986, map 3-1.  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA AND CALI 45 IBRD 25347 Sector System 25 ComunasX JANUARY 1994  4tj UNDERSTANDING THE DEVELOPING NIETROPOLIS Map 3-2. Cali: Ring and Sector Systems IBRD 25343 Ring System < b tR~~~~~~E Comnunos JANUARY 1994 .S'nirc: Stohari 1986, map 43-2.  GROWTH AND THE CHANGINC; STRUICTLURE OF BOGOTA AND CALI 47 IBRD 25344 Sector System < A E ~~~~~~~~~Comunas JANUARY 1994  48 L1NDERSTANDINGr THE DEVELIOPING METROPOLIS Map 3-3. Bogoti: Population Density by Comuna IBRD 25348 1973 People per hectare 350 and over _ 300 to <350) 1EZ0 250to <300 200 to <250 150 to <200 lOO to<lS I so 771 Sto <10 I/ o as Scrur: . <ohan 1986, map 3-3.  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA AND *ALI 49 IBRD 25349 1978 People per hectore 350 and over 300 to < 350 250 to <300 200 to <250 150to <200 7}( 111100 to <150 [ <50 91 4 b j sS 54 5 241i 23 Comunos 199 JANUARY 1994  50 UNDERSTANDING THE DEVEL.OPING METROPOLIS times denser than the least dense, but some of the outlying areas in the south are about as densely populated as some of the inner ones. Thus, although exponentially declining population density functions fit Bogota and Cali well (as documented in the next section), such a sum- mary representation hides some of the actual diversity. The comunas are too disaggregated to provide a summary of popula- tion changes. There are two natural ways of dividing a city spatially for the purposes of analysis: into "rings" and into "sectors," or pie slices, as shown in maps 3-1 and 3-2. (Because Bogota and Cali are semicircular in shape, so are their rings.) These gross spatial divisions are used for analysis throughout this study. Populationi and density declined slightly in absolute terms in the center of the city over the period 1964-78 (see table 3-5). In 1964 the highest density was in the CBD (ring I and sector 1), but ring 2 (about 1-3 kilometers from the center) was the densest in 1973 even though it also lost population between the tvo census years. It is striking that ring 3 (3-6 kilometers from the center) and ring 4 had similar densities in 1973 and 1978. The fastest growth clearly occurred in rings 5 and 6. Indeed, ring 5 accommodated twice as much of the incremental population during those years as the rest of the city put together. Land prices also increased the most in ring 5. Ring 6 was Table 3-5. Growth in Population and Density in Bogoti Average annual rate of grouwth in Area, 1973 Population, 1978 population and densitY (percent) Divjision (hectares) (thousands) 1964-73 1973-78 Ring 1 398 82 -1.8 2.4 2 1,357 280 -0.2 -0.3 3 2,575 426 3.2 3.4 4 5,960 923 5.7 0.7 5 14,329 1,592 14.6 6.3 6 5,804 189 37.6 13.8 Sector 1 398 82 -1.8 2.4 2 4,357 696 5.8 7.1 3 5,313 859 8.3 2.8 4 1,914 250 7.1 -2.7 5 3,066 213 7.4 0.2 6 5,673 680 10.9 6.0 7 5,064 325 5.7 -1.9 8 4,638 388 7.1 14.5 C'ity 30,424 3,492 7.2 3.9 Note: Density for all vears has been calculated based on the 1973 area. Source: 1964 and 1973 population censuses; 1978 World Bank-n.ANr Household Survey.  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA AND CALI 51 almost uninhabited in 1964. The main part of population growth in Bogota in the period covered by table 3-5 has clearly been in the outly- ing rings. There has been some decentralization of population, partly because population in the two inner rings has declined somewhat, but mainly because of the much larger additions in the outskirts. The lower panel of table 3-5 gives the same information for the radial sectors. Sec- tor 1, the CBD, is identical to ring 1. The southern sectors (2, 3, and 4) are almost twice as densely populated as the northern sectors (6, 7, and 8). The sectors have grown somewhat uniformly, although sector 6 grew faster than the others. Not until the early 1960s did the western part of the city began to fill up; before then the city had largely been a north- south strip stretching along the mountains. Sectors 4 and 5 form the industrial corridor that Le Corbusier established in 1951. On average, the densities of rings 3 and 4 are very similar to those of sectors 3 and 4. The northern part of sector 8 is even sparser than the average. We now turn to the spatial distribution of income (see chapter 4 for a systematic discussion of this subject). Again, comunas are too disaggre- gated to provide a useful picture. It should be noted, however, that the income of the poorest comuna was only one-twelfth that of the richest. Each comuna was relatively homogeneous within itself. A glance at map 34, which illustrates the distribution of household income per capita (HINCAP) by comuna for 1973 and 1978, shows that mean income varied across the city in a remarkably regular fashion, increasing in income as one follows the map clockwise from south to north. There is no discern- ible pattern from the center to the periphery, however (see table 3-6). Only a mild pattern is discernible by rings. There is a tendency for income to rise somewhat from the center toward the periphery and then to decline; ring 5 is poorer than the rest. The differences, however, are small, and the coefficients of variation are large. For most of the rings, the proportion of households in the ring is not very different from the share of income, except for ring 5. Thus it is clear that the income distribution pattern is quite different from that in U.S. cities, where income, in general, increases as one moves from the centel to the periphery. The data partially support the idea that the poor in cities in develop- ing countries get pushed to the periphery. The pattern is more distinct when viewed by sectors. The richest sector (sector 8) has a mean HINCAP more than five times that of the poorest (sector 2). Except for sector 6, which is relatively poor, income increases as one moves from sector 2 to sector 8. The most heterogeneous zone is the CBD, with a very high coef- ficient of variation. The share of income received by the poor sectors is much less than their share of households. The ranking of the sectors did not change between 1973 and 1978. It is therefore clear that a spa- tial analysis of Bogota is of more interest when done by radial sectors  52 UNDERSTANDING THE DEVELOPING METROPOLIS Map 34. Bogota: Distribution of Mean Household Income Per Capita by Comuna IBRD 25350 1973 _ 60% 43] Comunas JANUARY 1994 Source: Mohan 1986, map 3-4.  GROWTH AND HFH CHANGING STRLI(:TL RF( OF BO(;OIA AND) CAL.[ 53 IBRD 25351 1978 40 56 i ~~~~~84 VP 2~ ~~~51 45. ~~~~~~~~~~14 25 WEE i EF4 Comunas JANUARY 1994  54 UNDERSTANDING THE DEVELOPING METROPOLIS Table 3-6. Spatial Distribution of Monthly Income and Population in Bogoti 1973 1978 Meaan Sharei of Meanr Mlean house/oLd Distibution in rome in Mlean honUsehold Distribution house/hold inromne per of house- ring or household income per oj]house- income caita holis sector income caJpita holds Dimwi.mon (1973 pesos) (1973 peso5) (percenit) (perrent) (1978 peSos) (1978 pesos) (percen7t) Ring 1 2,114 884 3.5 2.6 8.343 2,49( 3.(1 (2.10) (1.03) 2 2,976 1.016 10.8 11.5 16,047 5,570 9.8 (1.92) (1.04) 3 3.722 1,073 12.7 16.9 19.928 5,913 13.1 (1.59) (0.93) 4 3,046i 788 30.7 33.5 14,772 4,105 27.2 (1.94) (1.40) 5 2.264 533 39.1 31.7 10,818 2,277 41.6 (1.86) (1.I0) 6f 3,202 753 3.3 3.8 15,999 3,178 5.3 (1.93) (1.26) .Sector 1 2,114 884 3.5 2.6 8,343 2,490 3.0 (2.11) (1.03) 2 1,583 414 18.1 10.2 6,833 1,585 19.9 (1.44) (0.52) 3 2,002 484 24.4 17.5 9,930 2,234 24.6 (1.41) (0.86) 4 2,347 646 9.8 8.2 12,843 2,800 7.2 (1.68) (0.88) 5 2,453 700 7.7 6.8 13,893 3,067 6.1 (1.68) (1.04) 6 2,349 5,56] 17.0 14.3 12,801 2,744 19.5 (1.66) ((.98) 7 45.791 1,265 12.6 20.6 16,163 3,750 9.3 (1.61) (1.07) 8 7,895 2,258 7.0 19.7 32.804 11,354 11.1 (1.50) (o90() (Ciy 2,791 751 1(00.0 10(1 13,805 3,629 100.0 (1.89) (1.22) Nolte: Numbers in parenthieses are coeffi(-ients oF sariationi. Source: 1 973 population cenisus sample; 1978 Vsorld Bank-DANk fIlousehold Survey. than bv rings, although the rings do exihibit some interesting character- istics. Thle magnitude of differenices in incomile betweeni the diffelelit parts of the city is surpr-isinigly large for averages taken over large num- bers (see table 3-6).  GROWTH AND THtE CHAN(GING STRUCTURE OF BOGOTA AND CAL.I 55 Such differences are seldom encountered between different regions of a country. In Bogoti (as in Cali) the rich clearly concentrate in cer- tain locations. Althouigh the poor are to be found in most parts of the citv, the declining coefficient of variation between 1973 and 1978 indi- cates increasing spatial segregation of income classes. The growth of the northern part of the city, the resulting gradients for land values andl population density, the use of transportation, an(d the type of housinig all conform milch more to patterns foundc in developed countries than in the city's southiern section. One other ring characteristic should be mentioned: household size, which increases from 3.83 in the citv center to 5.3 in rinig 5 and 4.9 at the peripherv (ring 6). ILarger households require more space and hence are rnore likely to locate awvay from the center. In addition, although larger households earn higher total inconmes, the added comil- pensationi is not enougih to offset larger household size. Thus house- hold income per capita declines with increasing household size. The periphery of Bogota thierefore contains larger as well as somewhat poorer households, the household income per capita measure being a more meaningful measure of welfare than total household income. A broad pictire of the changinig structur-e of Bogota is now beginiing to emerge. Growth has entailed the densificationi of successive rings moving outward from the center-, but this densification is not unliformll betweeni different sectors of the city. The rich have continued to move northi (sectors 7 and 8) and to settle in relatively low-densityi neighbor- hoods, whereas the poor have largely moved toward the southl and southwest (sectors 2 and 3) and to some extent toward the west (sector 5). The striking element of this pattern is the relatively high constant denisity that is observed as one moves from the center toward the south- ern periphery. The poor appear to live in settlements of similar dlensi- ties whether in the center or on the outskirts. There is a clear increase in household size toward the periphery, and the densiqt of dwelliig UllitS would therefore be expected to decline outward in the poor sec- tors as well. A similar pattern can be observed in Cali. The Evolution of Land Values and Population and Employment Densities Table 3-7 expands on the information presented in table 3-5. The maxi- mum distance from the city center- is about 15 kilometers (to ring 6) in Bogota and about 10 kilometers (to ring 5) in Cali. Growth has clearly occurred by accretion in the outer rings; density in the CBD has remained at about 200 persons per hectare. Growth on the fi-inges of  Table 3-7. Change in Population Density by Ring, Bogoti and Cali BogtadC (Ai Den.iI) (populazliori per hectlare) Deatitv (population per hectare) Area (hectares) 1964 1973 1978 Area (herl(iresj 1964 197 3 1978 Ring 1 398 220 180 205 140 210 150 160 2 1.357 210 210 220 1,500 140 125 135 3 2.575 1(00 140 140 3,00() 95 135 160 4 5.960 90 150 155 3,00() 25 70 100 5 14.329 25 8( 110 1,500 25 50 70 6 5,804 1 17 32 na. na. na. na. CitV 30,424 50 95 115 9,100 70 103 121 Densitv gradient (g) 0.18 0.15 0.12 0.51 0.44 0.25 na. Not applicable. Note: All figures are r-ounded. a. City area has been kept conistant for all calCulations of city density, unlike in table 3-4. In fact, botil cities have grown duhitlg the period in question. ald many peripheral areas inicluded here were outside the 1964 city botindaries. Source: Data for Bogota are fiom 1964 anid 1973 population censiuses and 1978 'World Bank-D.xNE 1lousehold Sturey. Data for Cali were calculated front data in Iabares (1979).  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA AND CALI 57 the existing cities has accompanied densification of the inner rings. This process is somewhat different from the growth pattern observed in most U.S. cities in the early twentieth century. Thus, while Bogota and Cali have decentralized with growth in the sense that a smaller propor- tion of the total population lives within any area of constant radius, they have not yet decentralized in the manner of many cities in the United States, where central cities have actually lost populations in substantial magnitudes. The last line in table 3-7 gives the measured density gradients from 1964 to 1978 for the two cities. The unit of observation is the barrio, or neighborhood.2 These density gradients were estimated for Bogota from about 300 observations in 1964 and about 450 in 1973 and 1978 and for Cali from about 130 observations in 1964 and about 200 in 1973 and 1978. These data are more disaggregated than those usually avail- able for such calculations. The gradients reported for different cities in chapter 2, for example, are based on far fewer data points in each case. As noted in chapter 2, g declines as expected for both cities and is higher for Cali, the smaller city, than for Bogoti. For purposes of com- parison, we may recall that, around 1970, gwas 0.11 for New York, 0.08 for Chicago, 0.08 for Tokyo, 0.07 for London, 0.22 for Seoul, 0.17 for NMexico City, 0.12 for Sio Paulo, and 0.12 for Bogota. Among smaller cit- ies like Cali, it was 0.19 for Monterrey, 0.41 for Guadalajara (Mexico), 0.27 for Belo Horizonte, 0.19 for Recife (Brazil), 0.25 for medium-size cities in Great Britain, and 0.31 for similar cities in Germany. Thus Bogota and Cali have population density gradients that are typical for cities of their income levels, size, and available transport systems. The gradients have declined over time with increases in population and income, much as would be expected. Both cities exhibit substantial decentralizationi with growth but with successive densificationi of inner rings, unlike many North American cities. Simple urban economic theory (Mills 1972) leads us to expect that the land value and density patterns should be broadly consistent. The competition for land for its highest economic use gives land its value. Hligher density implies higher relative use, and, conversely, higher land value gives people an incentive to use land more intensively. In a full} developed urban model, land values and population densities would both be endogenous variables that would be expected to move in a con- sistent fashion. For most cities it is typically difficult to obtain good information on land values, especially comparable information over a long period of time. It is also difficult to separate the value of land from that of the structures built on it. We were fortunate in obtaining a unique data set of about 6,000 transactions in Bogota covering the period 1955-78 from Guillernmo Wiesner of Wiesner and Cia. Ltda., a long-established Bogota  Table 3-8. Evolution of Land Values by Ring, Bogota and Cali Bogoti Cali Averagre A verage Average ailcual .4ve7age annual distancefrom growu1th mrte, distarcefrom growth rate, the centfr land values (1978 pesos per square meter) 1964-78 the center Land values (1978 peyos persquare mter) 1963-79 (kilometers) 1963--65 1972-74 1975-77 (percent) (kilomneters) 1963 1974 1979 (percent) Ritng I 0 4,250 3,900 3,100 -2.3 0 5,900 4,600 6,400 0.6 2 2.2 1,850 1,660 1,550 -1.3 1.8 1,100 1,100 2,400 5.6 3 3.8 1,350 1,350 1,32() -0.2 3.4 520 480 1,030 4.9 4 6.5 870 1,080 1,130 1.9 5.4 380 410 960 6.6 5 9.8 570 800 850 2.9 6.9 150 370 810 12.0 6 15.4 370 700 730 4.9 n.a. n.a. n.a. n.a. n.a. Gradienta (h) n.a. 0.16 0.08 0.07 n.a. n.a. 0.55 0.51 0.25 n.a. n.a. Not applicable. Vote: 1978 exchange rate: US$1 =Col$38. a. Lanid value gradienits were calculated as = - Veh', wher e V, is the price of land x from the center, V, is the theoretical value at the center, and h is the gra- dient. Source: For Bogoti. Villamizar (1981). For Cali, Velasco and Mier (198(0).  GROWTH AND) THE CHANGING STRLUCTL'RE. OF BOGOTA ANI) CAL.I 59 real estate firm. Because all transactions observed were for land that was not built on, we did not have to artificially separate the values of the land and the structures. (For detailed analysis of land values in BogotA, see Villamizar 1981; Mohan and Villarnizar 1982; and WAagner 1984.) The data set for Cali was obtained from the Cali Planning Office. Table 3-8, which shows the pattern of the evolving land price surface for Bogota ancl Cali by rings, is remarkable for the highly regular pat- tern that emerges and is consistent with expectations. The prices are given in constant 1978 Colombian pesos because the earlier prices were inflated by the consumer price index. For both Bogota and Cali. land prices decrease from the center toward the peripherv in a regular fash- ion; central land values remain broadly constant in real terms, whereas considerable and consistent increases are recorded towardl the periph- ery. This is in keeping with the evolution of density patterns shown in table 3-7. The constancy of real land prices in the central areas of Bogot.i dur- ing the 1960s and 1970s is remarkable. A word of caution is necessary regarding the significanit decline in the C:RD (ring I ). Wagner (1984) has demonstrated the existence of some biases in the measulement of cen- tral land values. As may be expectcd, less and less vacant land is available in centr-al areas over time; the ntimiber of vacant-land t-ansactions there- fore falls, and larger lots became eveni scarcer. In the CBD larger lots fetched higher unit prices than smaller lots, but since the number of smaller-lot tranisactiolIs was higher, the unlweighted averaging process biased the average CBDt land values downwarcl. Hence, the fall in real urbani land values in the Bogota CBI) duLinig the 1960s and 1970s is par- tially illusory. I.and values may be regar-ded as having bcen constant over the period. It is also important to note some findings that go against the conven- tional wisdom. It is widely thoughit that la.cl values have been skyrocket- ing in cities in developing countr-ies and that they will do so inclefillitely. In fact, in BogotJi between 1955 and 1977 they grew at only about 3 per- cenlt a year il real terms, a rate not far friom either the per capita GN' girowth rate for the cotintrv or the return on any alternative asset. The absolute land prices were not very different for Cali. a finding consistent with those of Mills and Song (1979). They' foutnd that, in the Republic of Korea, land values in larger and smaller cities were not ver-v different, and the rate of incr-ease was marginally higher among cities categorizedl as second-largest than among the largest cities. The main point is that cities grow in a regular fashion and that the evidence from BogotA and Cali is quite consistent with the theoretical expectations outlined in chapter 2. These regularities are, moreover, results of the r egularitics in hulimaln behavior that are investigated in the rest of this study. This evolving patterin of land value gradients conifir ms  60 tUNDERSTANDING THE DEVELOPING METROPOLIS that land values have essentially been determined by the economic com- petition for access to space according to its best use. Distance from the city center emerges as the best explanation for variance in land value aroulid the cit,. It is therefore the access characteristics of each plot of landl that are the principal deter-minanits of its value. Other variables, sucIh as amenities, availability of inifrastr ucture, neighborhood charac- teristics, and zoninig, are less significant explanators of land value vari- ance in BogotA. The next section shows that although distance fromii the CBD is the prime determiiiianit of land values, other factors, such as access to subsidiary shopping centers and main transport arteries, are also significanit in explaining the ridges and values encotintered in the land value sur-face of Bogota. There is little evidence that the land mar- ket has not been working competitively in Bogota. Uhy have real urban land prices in Bogota and Cali risen at moderate rates? First, (lurinig the period tinder consideration the supply of houls- ing gener-ally matched the rapid gr(i)wth in demand (see chapter 7). The rapid land development that took place in the barrios pirata.s (extralegal developimienits) essentially kept tip the stipply of developed or semidevel- oped land in response to constanitly rising demand. Second, explicit governmenit policv aimed at rapid investment in urban inifrastr-uctur-e helped provide at least a minimiial level of services to the emergilig new developments-both legal and extralegal. Thircl, the availability of dif- ferenit sets of instruments for investimient in Colombia discouiraged over- crowding in real estate investment. Fourthi, the competition between developers, who had relatively free entr-v into the housiig and land( mar- kets, helped keep the returins to landc dlevelopment. withinl moderate lim- its, which molre or less matchedl national income growth per capita over- this period. Comparing the densitv and land valuie gradients in tables 3-7 and 3-8 provides ftirtlher evidence: Bogrod (:cjli Land 7 value Den.sity Lanid vabir DentlNi 1964 -0.15 -(.18 -0.51 -0.51 1 973 -0.08 -0.15 -0.25 -0.44 1978 -0.07 -0.12 -0.23 -0.25 According to standard urbani economic theory, we wotIld expect the patternl of land valtie gradients to he similar to density gradients. Fur- therniore. their levels are consistently lowver than the density gradients, as is often theorized (Mills and Song 1979). A comparisoni of estimated 1-) (lanid value at the center of the city) with actual V( reveals that our estimates are consistently lower than the actual values. This implies that the gradient of the curve shouldl, in fact, be mtich steeper at the center  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA ANt) CAlA 61 of the city than we have estimated. Because of the high concentr-atioln of economic activity at the center, one can expect larnd values to be deter- mined much more by employment density than by residential density. The latter falls rapidly from the center, and it is quite plausible that land values will exhibit a similar decline. Estimated V10 is therefore likely to be lower than this central peak. The estimated densities were consistently higher than the actual densities for the CBD, whereas estimated land val- ties were consistently lower than real land values. What is to be expected is that residential densities will increase somewehat fi-om the CBD anid thenl decline, whereas land values should show a rapid decline from the citv center at first and then a slower decrease. L.and values are expected to be extremely high at the city center because of the concentration of economic activity there. It was sug- gested in the last chapter that employment densities are expected to be higher in the center and that they typically have a steeper gradient. Table 3-9 vividly illustrates this for Bogota and Cali. Estimates of employ- ment density gradienlts are more difficult to come by, than estimates of population density because employmetnt data are not usually available bv location. Thie gradients for Bogota in 1978 appear to be similar to those calculated by Mills (1972) for a sample of L.S. cities in 1963. The table shows that manufactturing has decentralized most, as expected, fol- lowed by commerce and serxices. Finanicial activities are heavily concen- trated in the center and might have become more so dur ing the 1970s. Table 3-9. Employment Density in Bogoti (jobs per he tare) Mlan u- Item All facrlunng Commwrce 'inance .Srvices jobs per lectare, 1978 Ring 1 425 43 97 102 128 Ring 2 158 28 36 21 54 Ring 3 77 24 14 4.4 26 Ring 4 42 12 8.( 1.8 15 Ring 5 21 5.6 4.8 0.4 6.0 Ring 6 7.0 1.1 O.fi 0.1 3.0 Employment density gradients Bogota. 1978 0.30 0.21 0.30 0.56 0.29 Bogota, 1972 0.33 0.33 0.32 0.44 0.35 Cali, 1978 0.72 0.71 1.0( - 0.64 Average for selected U.S. cities, 1963 0.26 0.27 0.35 - 0.33 - Not available. Sourre: Colomibian data were calcuilated from K. S. Lee (1989), 1972 Phase 11 Household Survey, and 1978 World Bank-DANE Household Suirvey. U..S. data are from Mills (1972).  Table 3-10. Distribution and Density of Workers' Residences and Workplaces, Bogoti Ditrnbilbion ol rwrkers 1972' 1978b DensitY of Workers BY residence BY voirkpluce BY residence By wvorkplare 1972 1978 1972 1978 Division Area (herares) (percent) (percent) (perceit) (percenit) BY residence per hectare BY zworkplace per hectare Ring 1 398 1.7 25.7 2.4 14.5 37 73 507 420 2 1,357 10.9 15.0 9.7 18.3 70 85 87 155 3 2,575 13.1 16.3 13.3 17.0 44 61 50 76 4 5,960 34.5 20.7 26.0 21.2 50 52 27 41 5 14,330 37.6 20.4 43.0 25.5 23 36 11 20 6 5,804 2.2 1.8 5.5 3.6 3 11 2 7 Sector 1 398 1.7 25.7 2.4 14.5 37 73 507 420 2 4,357 18.3 7.6 19.1 8.5 36 52 14 22 3 5,313 25.1 13.4 23.4 13.4 41 52 20 29 4 1,914 9.5 9. 3 6.5 9.2 43 40 38 55 5 3,065 7.1 10.9 6.3 12.6 20 24 28 47 6 5.673 1 6.3 9.8 17.8 12.4 25 37 14 25 7 5,065 13.3 9.9 10.1 10.2 23 2 4 15 23 8 4,638 8.8 13.4 14.3 19.M) 10 37 23 47 City 30.423 100.0 100.0 100.0 10(.(0 28 39 26 38 Note The su-rvey inc luded 877.000 workers in 1972 (incIlidi ng 786,00/) w hose worIkplaces were identified) an d 1.185 000 workers in 1978 (including 1,150,00() whose workplaces were identified). Informatioun on the workplaces of sonie workers WaS Unavailable. a. Data are fiom Pachoin (1979)) table 5. Primary data for workers bs residenice was from the 1972 Phase 11 Household Survey househiold file, and bv work- place 1rom the 1972 Phase 11 Household Survey person file. h. Data are fiom 1978 World Bank-DANE Household Survey.  GROWTH AND THIE (CHANGIN(; STRI!(:TURE OF BOGOTA AND CM.A h:l Employnient density gradienits are muLch steeper than population den- sitv gradients. Cali is still very centralized. It really has only one major employmeint concentration-the center. Because we do not have reliable (lata for ear- lier vears, it is difficult to say whetlher employment has decentralized in Cali vet, except in marn ufactuLing. The employment pattern is examined in more detail itl chapter 6. Here we merely neecl to establish that the emplovymient structule of Bogota and Cali is consistent with the expected patterni of growth and decentralization. Bogot.i's CBD had a net loss of jobs in the 1970s, prima- IrilY in manufactUring. SoIle Of the CBD funictions shifted to rinlg 2, in the nor-thernl sector, which is more easily accessible to the rich living in sector 8. The Changing Structure of Bogota: Some Wrinkles So far we have treate(d the city in a relatively simple manner; The mea- surenmenit of each gradlienit assumiles that the citv is symmetric around the city ceniter and is essenitially monocentric. This is justified in the main because both BogotA and Cali are roughly semicircular, though con- strainied by moLunitainis on one sidle, with the city center roLughly at the geograplhic center of the semicircle in each case. Thie income andi popu- lation distributionis are not, however, svmmnletric. In BogotA the nor-thernl sectors (7 and( 8) were characterized as particularly richi anld the souith- ern sectors (2 and 3) as poor. In Cali the picture is more mixedl, but broadly, the western sectors (2, 6, and 7) are riche r than the eastern sec- tors (3, 4, an( 5). In generalh jobs exceed the numihber of resident work- ers in the ricIh sectors, and the converse is true in the poorer sectors (see table 3-10). As might be expected. population density is higher in the poor than in the rich sectors. This suggests that the smooth patterlls depicted in the precedinig discussioni hide considerable diversitv and that it would be useful to look at the density and land( value patterns il dtifferenit sectors to observe the extenit to which earlier genieralities remaini true. These patterns are partictila-ly importanit for houisirng and transport decisiois. For example, the denisity inifo-mationi presentecd in tables 3-9 and 3-10 shows that emplovmenit dlensities are greater than resideential densiiies in the inner rings but less in the ouiter r ings. Although the commiiutinig pattern would be predominantly inward, the decentralizationi of both employmenlt and residence suggests that a decr-easinlg proportion of trips for work would be made toward the cen- ter as tirne goes on. A more interestinig residential and work patterni is shown in the bot- tom part of table 3-10. Most striking is the job deficit in pooI sector 2 (in  14 UNDERSTAND[NG THE DFVELOPING METROPOLIS the south) and the job surplus in rich sector 8 (in the north), which leads one to expect conisiderable crosstown commuting in addition to radial commutiig. This chariging pattern points to the need to increase the number of circumfer-ential routes so that north-south crosstown traf- fic can avoid inner-city congestioni. This is possible onlv if appropriate investimienits and changes are made in the prevailing road structure. It is therefore importanit to observe and uniderstanidl trends in transportationi patterins in order to design workable transportatioti routes and rnake appropriate policy decisions about public investment in infrastructure. One anomalyv may be observed in table 310: In contrast to tihe gen- eral pattern of decentralization, there seems to have been anl increase in the density of populationi of worker-s in Bogota's (:BD between 1972 and 1978. Table 3-7 also shows some iicrease in populationi denisity in the CBD (from 180 persons per hectare to 220 persons per hectare). This is partly explained bv the rapid change in the character of the CBD's resi- d(ent poptilationi dIurinig the 1970s. Many families moved to ouiter loca- tions and were replaced bv single-member households (single workers). The housing str-ucture also changed from apartments to rooming houses (popularly known as inquilinatos in Bogoa). Single-member households, which are characteristically (liffictilt to cover in household suLrveys, were probablv undercotiuited in the 1972 sur vev. Repeat visits in the 1978 World Bank-DANE survey ensuredl better coverage. The sub- stantial shift over time towaid single-worker- residences in central cities is quite cojiniiion; it is well knowin in Nor tih America and EuLropean cities and has also begun to appear in the central cities in some developing coutnitries. UJntil now the distance friom the city center hias been emphasized as the kev variable affecting the access characteristics-that is, the proxim- ity of land parcels to economic oppoirtunities-of a particular location and, therefore, the value of land. T he hvpothesis is that large or dense agglomiier-ation.s of people are instrumental in increasing these eco- nonic opportuniities, and this is the reason for the clustering of popula- tion near the city center. Thus a concenitration of economic activity in the center produces relatively high populationi densities anid corre- spondinigly high land valLies, both of which decline with distance. The observation that the rich live in some parts of the city and the poor live in other areas leacis to the revision of some of these ideas. That more jobs are located in rich sectors means that those sectors are eco- nomically more attractive and firms have a greater tendency to locate there. The lower population densities of those areas are more than enhanced by the purchiasing power of the resident population. Access characteristics conifer-red by proximitv to the city center or by short dis- tance from workers' resid(ences are he ightened by the greater purchlas- ing power of richer households. Thtus the access characteristics of  GROWTHJ ANI) THE CHANGING SI UCTtURF OF BOGOTA AND (.AI I 65 locations within close proximity to richer areas of the city arie improved, and land prices are bid up. There is more competition for retail space, tor example, in these areas. The establishment of shoppinig centers and other activities then attracts greater employment, contribtitinig fur thetr to the positive access characteristics of these locations. T'he prodtict of' populationi and mean incomile is probably not a goo(n measure of purchasinig power because the reqtuirements of a large inim- ber of poor people do not aggr-egate: each househlold has meager demands, so poor sectors can support only a limited number of cco- nomic activities-small retail activitv for subsistence daily needis but lit- tle else. Thuis, in Bogoti, much other retail and wholesale aclivitv continuies to be concenitrated in the center, and more people live in the poor southerin part of the city thani there are jobs there. Conversely, the r ich neighbor-hoocds in the north have a mtich highier clemand for goods and services than tileir population densities wotildl suggest. Moreover. office and other professionial activity has tendled to move northi in the same direction as the rich. Thus the access char-acter-istics ot' these neighborhoods have improved fturther andI are not adeqtiately mea- sured by populationi densities. In addition, because the infiastrulcture (for example, roads, lighting, water supply, and sewerage) is better in the richier areas, the quality of life and the intrlinsic characteristics of' the sites are better as well. All these factors combinie to produce land valies that are higher in the r ich areas than in other neighborhoods no farther fiom the citv center. The estimates of populationi dlensity andl lanid value gradienits for each sector in Bogota and Cali (table 3-11) are striking. The declininig expo- nential funictioni is still a good approximation of lancl valtues for each sector. The populationi densities do not do as well; the estimated density gradients are not significaitly dlifferent from zer-o in a tinmber of sec- tors. Indeed, in Cali sectors 4 and 5 exhibit mildly positive gradients. whereas in Bogota the density gradienits are low or insignificant for 'ec- tors 2, 3, 5, and 6. Wliat these sectors have in commoni is relatively low mean iicome. The lancl valie graclienlts. however, are not significanitly diffei-erit fi-om those in othier sectors. To udiclerstanid thlese phenomenia. we need to delve furtiher into the role of land valutes and their effect on urbani structure. Wheni land val- tues are high, capital is substittited for land, and the result is the co0n- struction of' taller buildinigs. We can therefore expect to observe, on average, taller buildings in city centers and in zones in which land prices are high. As land prices increase, single-family homes arc replaced by multifamly dwellings (apartmen ts), and residlential densities rise. Althouigh residential denisities rise per unit of land area, living space per personi does not necessarily decrease in wealthv areas. These options, however, are not open to the poor.  Table 3-11. Land Value and Population Density Gradients in Bogoti and Cali, by Sector Bogotd ( ali Household Household mean inotzoe Learnd Valueh Poptulation density meant itncome Land vialued Poepulalion densiqy indexa g R2 h Y indexa ' g R2 h R2 Sector 1 l6 in.a. n.a. n.a. n.a. 163 n.a. n.a. n.a. n.a. 2 52 -0.15 0.54 -0.11 0.06 212 -0.42 0.69 -0.13* 0.0o 3 74 -0.01 0.09 -0.02* 0.01 84 -0.45 0.73 -0.10* 0.0 1 96 1 1 -0.12 0.09 58 -0.42 0.69 +0.13 0.11 5 103 -(.08 0.71 40.05* 0.03 82 -0.55 0.77 +0.07* 0.02 6 97 -0.10 0.71 -0.05* 0.02 125 -0.21 0.41 -0.26 0.16 7 122 -0.08 0.72 -0.16 0.50 219 -0.86 0.59 -0.13* 0.05 8 236 -0.07 0.55 -0.14 0.32 n.a. n.a. n.a. n.a. n.a. n.a. Not applicable. Note: All coefficients are significant at the 0.01 level except those marked with an asterisk. a. Percentacge of meani household incomile ior the city. b. Data are for 1975-78. c. Data are for 1978. d. Data are for 1979. .Siource:Velasco and Mier (1980); Tabares (1979); Paclhoni (1979); Villainsizar (1981); City Stmdy barrio file.  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA AND CAlA 67 The rates of price increase were no higher in the rich areas than in the poor ones (Villamizar 1981). In Bogoti the rate of increase (adjusted for inflation) in the rich sector (sector 8) was about 2.5 per- cent a year between 1955 and 1977; it was about 4 to 7 percent a year in the poorer sectors (2, 3, and 6). The price levels, however, were consis- tently lower in the poor areas. These data indicate that although each land parcel is nonsubstitutive to some extent, a citvwide land market is functioning. Whereas land prices in poor areas continued to be lower than in richer areas, a catch-up phenomenon was observed: prices of land parcels equidistant from the city center were not too dissimilar. The natural result of this phenomenon is that while the rich substitute for land with capital, the poor substitute for land by crowding. Much of the housing in Bogota still has fewer than two floors. The number of floors declines rapidly from the C.BD, and the northern part of the city, the richer sector 8, has taller buildings than other areas (table 3-12). The average number of floors declines systematically by ring, as does the average age of dwellings, and the proportion of single- family homes increases with distance from the center. These patternts are very much according to expectations: capital is being substituted for land in the shape of taller buildings in the inner rings; the city has grown by accretion at its edges, and therefore the outer rings have newer houses; and apartment buildings or semidetached houses are replacing single-family houses as prices increase near the CBD. Land Table 3-12. Spatial Pattern of Housing in Bogota, 1978 A verage age of Average diweli ng Ml7ean haousehold Single-family dweledng una Auerage number space perpemxon Division income index units (percent) (wears) of floors (squoare mtelers) Ri nsg 1 62 39 16 7.1 14 2 116 57 21 3.5 23 3 124 74 1 6 2.8 30 4 112 86 12 1.7 23 5 82 95 9 1.8 18 6 122 100 8 1.5 26 Seclor 1 61 39 16 7.1 14 2 53 96 13 1.4 12 3 74 91 10 1.8 19 4 96 91 1 1 1.9 25 5 103 72 18 3.4 20 6 97 92 10 2.1 21 7 122 84 1 7 1.9 30 8 236 59 1 0 2.9 45 City 100 85 12 2.1 2 1 Soeurce: Mohan and X'illamizar (1982).  68i UNDERSTANDING THE DEVELOPING METRROPOLIS prices are performinig their function well, and the housing market seems to be responding as expected. There is no clear pattern of average dwelling-unit space per person except that it is low in the CBD. We would expect that living space per personi wotild be greater in the outer rings because people would be trading space for higher transport costs. If we now look at the sectoral pattern, it is clear that the poorest sectors (2 and 3) in the south have much less living space per person than the rich northern sectors. Thus the poor are substituting crowding for land, and the rich are substitut- ing capital. (Chapter 7 analyzes the hotIsing market in detail. One of the important findings reported there is that housing demand is inelastic among low-income residents, and they spend a high proportion of their incomes on hotlsinig.) Crowding in sector 2 is about the same as in the CBD; lower land prices at the periphery have not led to more space per person. It would seem that the poor have limited choices and that there is a limit to substitution by crowding. The percentage of single-family units increases consistenitly toward the peripherv, while the average nulmber of floors per structture declines. The population density in the poorer peripheral residential areas is not lower than in the center, but the density of dwelling units and households is. .arger households can take advantage of lower land prices at the periphery and live in larger dwelling unlits to have about as much per-personi living space as smaller hotiseholds nearer thie city center. We now begin to understand why the land value gradients hold tip even when thie cities are disaggregated into sectors, whereas the density gradienits do not. The way the land market funictionis results in land val- ues that are not too different at similar distances from the center. The rich sectors hiave highel- land values because of better employment opportunities, as well as better neighborhoods and inifrast-uclttre. Becauise lan(d values are relatively regular, the poor have no choice but to substituite for landc by crowding. Since it is the larger houscholds that locate on the periphery, they still have to live at higlh densities to coin- pensate for the land prices, which are similar to land prices in the rich subuirbs. They cannlot buyv more space by substituting capital for land becatise housinig would be too expensive. For example, the average number- of floors is 2.9 in sector 8 and only 1 .4 to 1 .8 in sectors 2 and 3; average dwelling unit space per personi is lowest in sector 2 and is also low in peripheral ring 5 (see table 3-12). It would have been instructive to see how these indicators had changed since the early 1970s, btit com- parable clata were n ot readily available. In any event, we observe high poptilationi clensities on the periphery of some parts of the citv and con- sequently there is no measurable density gradient in those sectors. The rich sectors still have a density gradient, and we can therefore observe gradients for the city as a whole. Nonexistent density gradients in sorne sectors of the city are consistent %4ith relatively strong land value gradi-  GROWTH AND THE CHANGING STRUCTlJRE OF BOGOTA AND CALI 69 ents. This suggests a need for caution in interpreting similarities between citywide population density and land value gradients. One other important aspect of land value patterns merits further discussion. As a large city grows, it acquires many new competing com- mercial centers that begin to rival the old CBD. These alternative (or additional) economic centers, in turn, are strong motivating forces for the decentralization of residential population. We examined this process by looking at the land value peaks along key urban corridors in Bogota (see Villamizar 1981 for details of the pat- tern). The detailed analysis along ridges of the land value surface reveals small hills, in accordance with the access characteristics that go with the higher levels of economic activity in the developing subcenters of a rapidly growing city. Because of such developments, the gradient of land prices decreased as the city grew. The relative importance of the (:BD declined, and secondary gradients developed around the sub- centers. In general, as a city grows, the smooth land price surface cen- tered around the CBD develops wrinkles in the form of ridges, valleys, and small hills, as observed around Bogota's new subcenters. Mills and Song (1979) found that in Korean cities commercial land values were always higher than residential land values in the (:BD and in the rest of the city at points equidistant fiom the center. The evidence we have found is consistent with their findinigs. Indeed, at equal dis- tances from the CBD, the proportion of area covered by commercial activity in any neighborhood is a good predictor of land values in that area. These results are consistent with our expectations about access characteristics of neighborhood and with the observed higher land val- ues in the richer areas of the city. Commercial activity locates in the rich areas of the city. The purpose of this section has been to illustrate the complexity that is usually hidden behind regularities in city structure. We have shown, however, that these complexities are also the results of economic behav- ior on the part of households and firms-specifically the pattern of housing demand by households and the location decisions of firms-in combinationi with governmental decisions regarding zoning and infra- structure. Physical characteristics such as building heights and residen- tial and employment densities are appreciated better when behavior is understood; conversely, behavior is affected by the prevailing structuire. Thus, it is important to study both the overall structure of a city and the behavior of its elements. Summary The growth of Bogota and Cali has been characterized by continuing decentralization of populatiotn and, to some extent, of employment.  70 UNDERSTANDING THE DEVELOPING METROPOlIS Although cities in developed countries have experienced a similar phe- nomenon, in that the concentration of population and employment in city centers has declined, there are some significant differences. Lower suburban population densities in North American cities are mainly attributable to the much larger size of residential lots in the sutburbs rel- ative to those in the central city. In Bogota and Cali residential lot size increases only moderately with distance from the city center. Although both the rich and poor have dispersed their residences, neither group has developed the large-lot, low-density suburbs typical of developed countries. The poor have moved wherever there is available space, and thev continue to live in high densities at the periphery' as well. In BogotA and( Cali some peripheral densities are as high as central densities. It is important to unlderstand how these highi peripheral densities are consistent with a land( value strtzctture that confor-ms to the expectations of a regular decline from the city center. The functioninig of the land market is such that land values at similar distances from the city center are quite similar, although they are slightly higher in the richer areas because of the higher level of economic activity there. Nevertheless, the poor have to pay rotghliV equivalent prices, and hence they substitute for land by crowding. Larger poor households, in particular, have to locate in crowded conditions at the periphery. The poor cannot substi- tute more capital for land because that would mnake the housin)g too expelnsive. The high level and growtlh of land prices, often attributed to undesir- able speculation and monopoly ownership, are usually regarded as unlwarranted and hienice as villains in city growth. Impressions are often based on faulty observation or inadequate analysis, however, as the anal- ysis of a uniique data set spannling the past quarter century of BogotA's development suggests. A careful analysis of this data set indicates that the pattern of land values is consistent with the evolving density pattern of residence and employment, with land values remaining nearly con- stant in the CBD and rising at the periphery. Land prices follow a regular pattern: they are highest within the CBD and decline as distance from the center incre ases. They are determinied much more by standard accessibility considerations thani by availability and quality of infrastruc- ture. Lot size also increases price. Large lots in central locations are r are and are priced hiigher per unit of area than smaller lots similarly located. By contrast, large lots on the city periphery are lower in price per Unlit of area than smaller lots similarly located. I.and use on the periphery' is not as intensive, and large parcels are commoni, so large parcels of land are offered for sale at a discount. There was no explosive increase in real land prices; overall land val- ties in Bogota grew in step with the city's economic activity as measured by this real product. This implies that the aggregate renlts from land  GROWTH AND THE CHANGING STRUCTURE OF BOGOTA AND ClAtl 71 remained a roughly constanit proportioni of Bogota's product and sug- gests that land has a unitary elasticity of substitution with other factors of urban produiction. The indications in Bogota and Cali were that land, on average, had returns quite comparable to and not higher than returins to other assets. The land market in these cities should therefore be judged as operating quite efficiently and as giving appropriate signals for the allocation of resotirces. Increases in land prices are much more the result of increasing opportunlity costs resuIlting from growth and agglomerationi than of the market imperfection or "rampant specula- tionI" that is ustially thotight to characterize the operation of urbaln land markets. These observations have a nutmber of significant implications for pol- icy approaches to expanding cities. First, the economic forces resulting fromii household and firm behavior are powerful in pushing city struc- ture toward r elatively well-tiuiderstood and predictable changes as a city's population grows. This suggests that planninig for land-use zoning, infrastructurc, transport, and other areas of urban development would do well to conform to these general trends rather than fight thern, as planining authorities ofteni do. sometimes unwittingly and sometimes intentionally. Second, fast-rising land prices at the periphery shot-ld not be a matter of concern; they corn e part and parcel with city growth, and artificial measures that attempt to stem suchi growth in prices will sel- domn succeed. Third, given the reqtuire ments of' an expanding urban population, the utmost muist be done to let the supply of' developed urbani land expand in responise to demand; this is the sur-est way of keeping land prices roughly consistent with the growth of the rcst of the econlomy. The decentr alizationi of employment in Bogota and Cali has been led by the decentralizationi of' mantifacturing employment, as has happened in cities in developed counitr-ies. Yet the CBDs continue to attract service employment anid to retainl large retail and wholesale trade activity. In Bogota the result has been that overall C:BD employment has remained almost conistant in absolute terms, although its share has been falling- to about one-sixth of all citv employmenit by the late 1970s. The center has become more specialized in its service and retail functions. The analysis indicates that decentralizatioll can be expected to continue because the tniderlyinig causes-increases in real income, urbani popula- tion growlth and transport improvemilenits-are also expected to con- tinue. This impor-tant feature of rapicd city growth is of'ten neglected in the planniniig of uribanl services; for example, transport investments are ofien based on forecasts of risinig rather thani constant employment in a city's CBD. The existinig radially oriented system, with transport primarily focused toward the city center, should be supplemenlted with circumfer- enltial services.  72 UNDERSTANDING THE DEVELOPING METROPOLIS Although decentralization has many benefits, it also has accompany- ing costs. It requires more extensive utility networks and therefore addi- tional investment in public infrastructure. The per-household costs of manv of these services are inversely related to density. Because the net residential densities of most new developments in Bogoti and Cali con- tinue to be high, infrastructure costs per household are not as high as they might otherwise be. But the location of many poor households near and on the periphery raises the costs of infrastructure extension, and these costs could be difficult to recover from the poor through user charges. Notes 1. The size of the comunas varies significantly, the smallest being about 174 hectares and the largest twenty times that size at 3,680 hectares. The population ranges from about 25,000 to 400,000. See the appendix for a description of the main data sets used in the study and more details about the spatial classification adopted. 2. The barrio, or neighborhood, is the smallest observation unit in the censuis. Because it is much smaller than a comuina, many more observations can be used to estimate the density gradient functions.  III Chapter 4 Poverty, Distribution of Income, and Growth In studVying the growth of a couintr-V it is comm>on to look at the chang- ing distributioni ot incomiie to assess the clistributioll of the gains restilt- ing from growth)1 'Ae study the distribution of incomiie in Bogot.i in a similar conitext but with one addled dimensioni-thiat of spatial dlistribu- tion withlinl the city. This is crucial for understanding thie spatial patterni of houisilig, tranispotatioll, an d public services and for projecting the future demand for thiese services. One problemii in studyring the spatial distributioll of income is tihat thiert are few esial)lishedl patterns for mak- ing comparisonis and no r obtist theory for making predictions. The onlly theory that exists (for example, Minth 1969) hias grown otit of patterns obser-ved in cities in developed coulitnles and depend ls on two crticial asstinptiois. Thle first is iliat the incomile elasticity for the demand foir lhousing (n/,) he greater than 1; the second is that the elasticity of the mar-ginial cost of n-ansport (as distance increases) withi respect to income be small relatiVe to f11. lln ler these assullmptions. in a inonocenitric city in eqtuilibritim, one woul]d expect higher-income houIseholds to live farther from the city center than lower-incomne hiotiselolds do. There is increasilig evidence thiat the incomile elasticitv for houisinig deinanid is less than 1 (see chapter 7). Hence, althotigh the thieor-v explains the existing patternis of income distributioni in cities in clevelopedl counitries, its methodological basis can be called into question. Moreover, in situationis where many tranis- port modes exist and tr-aflic is slow, the second assuilmptioll mav also be n1ot -uie for cities in poor coulitries. Chapter 3 established that there is little evidence of svstematic variationi of income bv distance from the citv center- but that the pattern is rnuch more distinict bv radial sector. There is eviclence of sonie segregation in residence by incomie, whichi also affects the distributioll of employmeit. This chapter is therefor-e neces- sarily exploratory andl primiarily descriptive, concenitratilig on e stablish- 7'  74 tUNDERSTANDIN(; THE DEVELOPINC. METROPOlIS ing the existing patterns in Bogota and Cali and exploring useful methods of measurenmenit and description. It is (lifficult to generalize from these finclings, but it is hoped that similar measurement for other cities w-ill, over timne, establish recognizable patterns that are also expli- cable t'rom a theoretical point of view. Recognition of the patterns is necessary for understanding how Bogota functionis. There is considerable concer-ni about the existence of "excessive" pov- ertv in rapidlv growing cities. Sometimes this is expressed as concern about the growth and characteristics of the informal sector (see, for example, Mazulimdlar 1976; Sethuramani 1974; Gilbert and GLugler 1982; and WXebb 1977) and sometimes as concerin about the location of the poor in unidesirable parts of cities and the interior coniditions unider which they live. Much of the literature Supports the beliet that the qual- ity ot' life for the poor has worsened in large cities in developing coun- tries. There is also a wicdespr-eacl impression that "unemployment, underemployment and misemployment add up to a massive problem in third world cities related to dimensions of ineqtiality" (Gilbert and Gugler 1982, p. 70). Much normative evaluation has characterized the work that maniv of the lpoor do (particularly in the service sector, as (lomnestic servants, street vendors, and shoeshine bo's, for example) as socially wasteful o1r unproduictive. The other most commoll concern is with thle housing conditions of' the poor. E:arlier, sluLims and squatter r esettlemenits were regarded as evidence of a cultuIre of poverty; the new view, nIow wi(dely accepted, usually sees the incremental housing typical of manyv poor settlements as evidence ot hope, innovationi, anid opportu- nitv. It is therefore illuminating to assess the actual existence of poverty il cities like Bogota and Cali and to confront impressions and beliefs with empirical evidenice. This chapter illustrates some of' the diftficulties of even identifyinig the poool; clespite a plethora of'data for these two cities. It then examinles the correlates ofl poverty (given plausible ranges of measuremiient errors). T'he interest is as much in measurement techniques and methods of analysis as in the descr-iption of the situation in Bogota and Cali per se. T'he chaniges in income distributioni ancl povertv in Bogota during the 1970s are better appreciated with some uLiderstandinig of the overall economic conditionIs in Colombia during this period ancl in the eco- nomnic preconiditions of recent decades. The late 1940s and early 1950s in Colombia saw rapidl industrialization and creation of' new industries unider condlitionis of substanitial protection. It has been argued that nonagricultural income distributioni probably worsenied fronm the mid- l930s until the early 1950s while this industrialization was taking place (Berry and Uriutia 1976). Manv of the protectecl indutstries that emerged were relatively capital-intensive and( not labor-intenisive. Given the skewed and highlv concenitratecd ownership of capital, such industri-  POVERTY, DISTRIBUTION OF INC(OME, AND GROWTH 75 alization could be expected to lead to a lower share of labor earnings. Because rural-urban migration was also very high during this period, providing almost "unlimited supplies of labor" at the low end, it was possible to keep blue-collar wages down. Conversely, this burst of indus- trialization created a sudden demand for highly skilled and educated workers, whose real earnings rose rapidly The whole period from the 1930s to the mid-1950s could be described as involving major structutral changes in the Colombiani economy-in particular, a shift from a predominantlv agricultural rural economy to one based on urban industries and services. These structural changes contintued apace until the mid-1960s, but the latter part of the period, frorm the mid-1950s to the mid-1960s, witnessed dramatic increases in blue-collar wages. Berry and UIrrutia hypothesized that during this period the degree of pi-otected import substitution may have decreased, and small-scale competitive enterprises may have become more success- ful. The expansioni of primary educatioin in the 1950s also contributed to higher average levels of workers' skills. The period between the mid- 1950s and mid-1960s was one of slower economic growth and was in many ways a period of consolidation. Surprisingly, lower-skilled workers' incomes rose rapidly, even though ruial-urbanl migration continuied at high rates. The overall income distribution in Colombia has remained hiighly unequal and is probably among the most inieqtual in the world. The share of the bottom 40 percenit has remained aroundicl I0 percenit of total income, whereas the share of the top 5 percent has been about one- third.2 This is a little worse than in some comparable middle-income countries. In Malaysia, for exanmple, the shar-e of the bottom 40 percent was estimated to be 12 to 18 percent and that of the top 5 percent was a little less than 30 percent (Anand 1983). The Gini coefficient for Colombia is consistenitly estimated to be above 0.5. For comparison, in developed couLtties the share of the bottomii 40 percent of households is typically estimated to be about 15 to 20 percent and that of the top 20 percernt, between 35 and 45 percent of total income, with ;ini coefficients of between 0.3 andl 0.4. OUI estimates, which are only for Bogota and Cali between 1973 and 1978. suggest some improvement in the late 1970s. The poverty estimates corroborate this tendency, so there probably was some improvement in income distribution in Colombia as a whole. If such improvement did occur, it would be consistent with a number of other developments in the Colombian economy during this period. With declines in fertility rates and the progressivelv smaller proportion of people left in the rural areas, the rates of rural-uri-bani migration and of urbanization have slowed. Moreover, there is evidence that the wages of rural unskilled workers rose significantlv in the late 1970s (ULr--titia  76 UNDF:RSTANDING THE: DEVELOPING METROI'OLIS 1985). The Colombian agricultural economy benefitted greatly from relatively high coffee prices throughout the 1970s and particularly fi-om 1975, wheni a booni began thiat lasted until almost the end of the decade. Overall growth in agriculture was also consistently hIigh over this decade, with an average anntual growthi rate of 5 percent (see Sokol and other-s 1984). Thomas (1985) has forcefully argued that these changes were not entirely fortuitous but rather resulted from good macroeconiomilic management along with progressive policies that sup- ported agricuiltural development. Ihe net result was that the demo- graphiic trailsitioln takinlg place in Colombia in the 1970s, coupled withi robust growth in the agriculture sector, proclucedI a relatively tight r ural labor market. This in turn applied pressure oni the urbani labor market, and real wages rose at the same time as uLbani employmelnt expanded substanitially. Corr-espondingly, it appears tilat the noniagr-icultural sectors were also following a labor-usinig strategy that lecl to a i-ate of growth of emplov- mnent that exceedled output growth in the 1970s (Sokol andc othe rs 1984). Again, apart from fortuitous circurmstances, this success can be attributed to successftil fiscal and u-ade policies that promoted labor- UsinIg manitifactur-ed exports. The extensive expansioni of uLbaln public SerVices inlVolinig the public sector also contribtItecl significaltly to the expansioni of demand for labor. A malor result of the expanded demanicd foir labor was thie increased participation of unskilled womeni in the labor force, whichi increased the houiseholdl income of poorel hlouse- hIoldCs (see Mohani 1986, UIrrtia 1985, and chapter 5 of this book for *ietails). Distribution of Income in Bogota and Cali, 1973-78 An assessment of inlcomite distributiOnl is typically plagued by measure- ment problems. Most censuses in developing counitries do not iicilide an incomile question, and those that do merit considerable skepticism about the veracity of responses. Household surveys are usually regarded as monre accur-ate because they are more carefulyv done, with smaller samples. Rural income surveys presenlt greater problems than urban surveys because it is difficult to evaluate both outputs and inputs. In uribani su-vevs, wage earniligs are easier to evaliate than income from property. Thie maini sources of data that were used in this study were thie 1973 populatiion census for Bogota and Cali and household surveys in 1975, 1977, and 1978 conducted by DANE, the Colombiani nationial statis- tical agency. Tle 1978 survty was conclucted especially for the City Study, and the incomie questioni was more detailed than in otlier sur- vevs. The 1973 census seemed to have covered only about 5() percent of  POVER-FY, DISTRIBUTION OF INCOMt, AND) GROW [ft 77 total personal income, whereas the 1978 survey may have covered as much as 90 percernt and the other sur-veys perhaps 60 to 65 percent. Although it is difficult to assess accurately the extent of undercover- age for each income group ancl for different years, oui overall judgmtent is that coverage is probably' better at the low end than at the top end of' the income (listribuitiol. The evidence was conflictinig, but the earnings of the least-skilled occupationial categories recorded in thc 1973 census were broaclv compar-able with othier data onl the labor earnings of these categories of workers. Female par ticipation in the labor force, however, may have been unidercounited in the 1973 census, contributing to some underestimationi of hiousehold incomiles. Overall, although thler was substanltial unidercoverage of'incoine in the 1973 population census, the dislrzbulion of income may be regarded as reliable and comparable xvith other clata sets (see Mohiani 1984 and Mohan, Wagner, and Garcia 1981) lHenice, despite problems with the comparability of' differenlt dlata sets over time, it is possible to conclude from the evidenice that the (listribui- tion of income improved marginally over the late 1970s in Bogota and Cali. These details illustrate typical issues that are encountered in esti- mating income distributioll trends in just two cities that are daLa-rich in comparison with most others in developing countries. Table 4-1, which gives the decile shares of income for 1973 and 1978 for Bogota and Cali ancl the valties of' two indexes of inequialilv-the Ginii coefficient and the Theil index4-reflects the strikinig inequality that exists in C'olombia. The income share of the bottom 40 percenlt of householcis is less thani half the share received by the top 5 pelceIt. Even in 1978, wheni the income coverage was greater than 90 percent, it appeared that there was substanitial unider-cover-age of incomiie ftromil cap- ital. Since sulch income is likely to accrue to the richer houselholds, the actual distributioni would be eveni worse than that apparent fiom this table. The results are similar for Bogotli and Cali andc for the dif'ferent inlconmei measures and rankinigs.? The overall improvemenit in distributioni between 1973 and 1978 is evident. These results were also consistent for differenit incomile con- cepts, different ranking proceduires, and interimlediate years (1975 and 1977) for both cities,C implving a high degree of confidlenice in these restilts. One general restilt obtained but not repolrted here was that all the indexes tisecl yield highel inequality of' househlolds when ranked by HINCAP. This implies that even though households of' larger size tend to have higher total hotiselhold incorime, their per capita houisehold inicome tends to be lower. In sumimary, the levels of overall inequality foundl in Bogota and Cali are not substantially different f'rom earlier estimates for ulrban Colombia, although a slight tendenicy towaid improvement is observed in the late 1970s. The distributioni of incomile is slightly worse in BogotA thani in Cali.  78 UNl)DERSIAND)IN(G THF IlEVELOPIN(; MElTROPOLtS Table 4-1. Distribution of Income in Bogoti and Cali )peicentage shares of tota] incomyne) B&go1a (.li Categorv 197 3 1978 1973 1978 Moo seholds 7rnked by hmoseh(ld incotne Bottorm 2(0 percent 3.3 4.( 3.4 4.1 Bottom 4() percenit 9.8 11.9 1 0.4 1 2.4 Tol) 20 pecent 62.5 55.5 60.0 54.) Top 5 percent 30.2 25.1 31.7 25.0 Gini coefficient 0.565 0.507 0. 53 0.487 Theil in(lex 0.610 0.458 0.601 0.429 Individualt rainked b1, hausehold incenne per. capila Bottom 20 peiceit 3.3 4.0 3.4 4.2 Bottom 40 perceIt 9).9 11.5 10.5 12.4 Top 20 petrcernt 62.6 38.0 60.5 55.5 Top -percent 30.8 29.0 31.4 27.3 Gini coefficieitit 0.568 0.522 0.554 0.500 Thleil inidcx 0.649 0.508 0.632 0.470 Scmore: 1973 po)ullatio n censu s: 1978 Wcoild Bank-[)\NlN Fliousehol(i Silr ,ev. We now addlress the issue of the distribution of hIcom11e across space in Bogotai and Cali. What arc the spatial patterns? Hlow do We measure themil svsternaticalvxS Can we trace trenids over timne? Boith Bogotui. and Cali exhibit striking patterns of incoine segregationi by space. The ratio between meani incomes in different zotnes of the city is as miuch as I to 6. with the means taken over relatively large zones. This kind of spatial dis- parity in incomes is tuniusuial even among the regions of' a countrv, but because there is little intformationi of' this type for other cities, it is cliffi- cult to sa' what the nor-mii wouild be for spatial distribuitioni withinl a city. Discussioln of the extent of spatial inequality within a city and compari- son witli others are diffictilt because there are n(o naittiral units of'aialyv- sis. UInlike the world, wlichi can be divided into cotintries, or a country, whicil can be divided into states or regions. there are no natural divi- sions within a city. In the LUniteci States much cdisciission1 focuIses on the contr asting chara(teristics of ce ntr-al cities and sibtb-hs, btit even in this case there is no natural definiitioni of' what constittites a centr-al city. Any division of a citv is arbitrary. In this stttdv we contiitte to uise the two systems of zonificatioti intro- duced in chapter 3-that is., riigs and sectors. Muich utrban econoimic theory for a nionocentric citv wotild stiggest s'steimatic differences iy  POVERTY, DISTRIBUTION OF INCOME, ANI) (;ROW)UTH 7) (listanice fiomi) the city center-that is, by rings-but thlere is also a respectable traclitiol), or-iginiatinig with Homer Hovt (1939, 1966), that divides the city into pie slices (sectors) for studyinlg the distribution of income. Hoyt traced the historical developmenit of a large number of North Americin cities. He concluded that income grotips tended to locate in one section of the city and that as the city developed. the rich continlued to locate in the same dlirectioni as that in whicih the citv was expaniliig. Peter Amato (1968) establislhed a similar patterin for Bogota: he focttsedl on the elite and shiowecd that they have tended to locate pre- dominantly in the noirthern part of the city and have continied to m<ove in that clirection. In later articles Amnato (1970a, I9701)) notecd patterils in QuLitc), Lima. and Santiago that were broadly similar to those in Bogota, althotigih not always as distinct. In this exploratory section we suggest the use of a more systematic methocd of inequality measur-emiteint to assess the extent of spatial ine- qtialitv and its trenid over time. As indlicatedl in table 4-2, in 1978 mealn hotisehiold inconie per capita in the poorest sector (2) in Bogota was abotit one-fifth of that in the richest sector (8). In (ali the poorest sec- tor (4) hiad about one-qtuarter the mean hotiselioldi income of thie rich- est (2). Some anomiialies may be note(d in the data for Cali. 'The data lor tihe (CBD sector in 1978 seems to have suffered from some sample errol's: populationi is shown to have fallen from about 4.3 percent of total popu- lation in 1973 to only 1 1 percent in 1978, whereas meani houselhold income as a proportioni of the overall mean increased froml about 1.03 to 1.78. This is not plausible. The data for Cali do suggest. however, that the deceiiti-alizationi process for ricih houselholds had just begun in the 1970s. Many rich hoUseholds contilued to live within or near the cen- ti-al city. The movemenit of households from the nich sector- 2. located near the central city, to sector 6, a developing stubulbhan residenitial area. was just starting. OuL- observationis in the late 1970s wer-e made over this period of rapid transformation. Cali, being a muchi smaller citv thiani Bogota, still had a long way to go in terms of growth an(d decenti'aliza- tioi. althoughi the process of subtjrbaniizationi of tlhe rich had begun. Overall, however, the patterns for BogotA andl Cali wer-e similar in terms of'segregationi of'tlhe rich and poor. Holow nicih is hiddeni behind the meanis in table 4-2?) Are there large variances aroulnd these means, so that the means do not necessarily imply clifterinig characteristics in the sectors? The measureilelm t of ine- quality is essentially the measur-em-Dent of varianc(e in a distribution. Whiat we now seek to do is to clecoliupose this variance i nto two parts: tile vari- ance between the means of incomile in rings, or sectors, or any other grouping of people and the varianice in incormes between people within a group. The Tlheil index cani be decomposed in this way. Total ineqLial- ity can be clecom posed in to wvithi n-group (or withinl-sector, within-ring)  Table 4-2. Spatial Distribution of Income in Bogota and Cali Bogota Cali 1973 1978 1973 1978 Share of sertor in Share of sector in Share of sector in Share of sect o in cily population city population city population ditY pofrulation Sertor (percent) Mean Hr\'cAP' (percent) Mean HInCAP' (percent) Mean HIN(,AP' (percen t) Alean HIM.-AP' 1 ^2.3 1.07 2.4 0.80 4.3 1.03 1.1 1.78 2 18.2 0.57 20.6 0.50 4.2 3.68 4.8 2.36 3 26.0 (.68 25.2 0.73 19.5 0.89 15.2 0.88 4 9.3 (.90 7.2 0.98 21.2 0.69 17.8 0.64 5 7.3 0.98 6.2 1.03 35.1 0.68 42.2 0.73 6 18.2 0.80 20.0 0.91 11.1 1.61 12.8 1.42 7 12.0 1.71 9.1 1.44 4.5 1.37 6.0 2.13 8 6.6 2.87 9.1 2.60 n.a. n.a. n.a. n.a. Cit"v 100.0 697" 100.0 1,012) 100.0 594b 100.0 2,425b ni.a. Not applicable. a. Mean household mo,ithlv incomile per capita taken across individuals in the sector as a multiple of overall mean HINEAP. b. Mean household monthiv income pet capita taken across individuials in Bogoti or Cali, expressed in current pesos. Consumer price index (1970 - 100): 1973 = 150; 1978 = 400. Stnirc: 1973 population census; 1978 World Bank-DANE hlousehold Survey.  POVERTY, DISTRIBUTION OF IN(:OME, AND GROWTH 81 contribtition and between-group (or between-sector, between-ring) con- tributionis. The latter can be regarded as an index of spatial inequality in incomes.' The Theil inequality indexes in table 4-3 show the level of inequality within each sector of the two cities. In general, the Theil index for each sector is lower tlhani for the city as a whole; that is, inequality within each sector is less than overall inequality (see table 4-3). The lowest inequality is found within the poor sectors (sectors 2 and 3 in Bogota and sector 4 in Cali). Similarly, in 1978 the richest sectors (sector 8 in Bogota and sector 2 in Cali) also have low levels of inequalitv. This is a clear indica- tion that the rich and pooI parts of both cities became more homoge- neous over the 1970s. The degree of spatial income segregation increased during that period, and the rich and poor areas of the city became more recognizable as suchI. The lower part of table 4-3 presents measures of the betweeni-grotip contribution of the sectors and rings. The decompositioni of inequality permitted a calculationi of this between-group measure, which has been interpreted as a measure of the spatial inequalitv in income within a city. The spatial contributioll of groupings by sector is substantial-about 20 to 25 percent of total inequality. Because each sector is distinct and interinally homogeneous, spatial inequality between sectors is high in both cities. A low spatial contribution (4-5 percent) by rings confirms that the level of ineqLialitv within rings is not very different fi-omil overall income inequality within the whole city (see Mohan 1986). Each ring is about as heterogeneouis as the whole citv. For rorntnas, shown in the last row of the table, spatial inequiality is between 32 and 40 percent, show- ing that these smaller spatial units are even more distinct in character. These systematic measures confirm once again that there is no dis- cernible pattern in household income according to distance from the central city in either Bogota or Cali-as is often observed in cities in developed countries. Income groups are significantly segregated by sec- tor. Like households have been gravitating toward each other in differ- ent parts of the city. The metho(dology used here suggests a systematic measuire of spatial inequality that can be used to make similar measure- ments in other cities. It is difficult to judge the magnitude of such spatial inequalitv. Is 20- 25 percent spatial inequality low or high? I have seen no similar indexes for other cities. Anand's (1984) work on Malaysia, however, provides some estimates for comparison. He foulid interstate and rural-urban contributions to inequality in Malaysia to be 9.1 and 13.7, respectively, accordinig to the Theil index. By these standards, Bogota and Cali must be judged to be highly segregated spatially by income. Moreover, this inequality is increasing, despite evidence of some overall decrease in income inequiality. If spatial segregationi of income groups has long-  Table 4-3. Spatial Inequality in Bogota and Cali: Individuals Ranked by HINCAP Bogoli CIali Ina,me 1973 1978 Income 1973 1978 rankingl ranking, Item (197,Y) 7heil index Rankingh Theil index Rankingb (197,8) Theil index Ranking" Theil index Ranking0 Sec/or 1 3 0.635 8 0.458 8 5 0.573 5 0.549 7 2 1 0.429 3 0.286 1 7 0.368 2 0.257 2 3 2 0.390 1 0.309 2 3 0.429 3 0.315 3 4 5 0.407 2 0.362 4 1 0.328 1 0.251 1 5 6 0.553 5 0.418 5 2 0.435 4 0.337 4 6 4 0.514 4 0.426 6 4 0.741 6 0.495 6 7 7 (.572 6 0.444 7 6 0.781 7 0.465 5 8 8 0.602 7 0.357 3 n.a. n.a. n.a. n.a. n.a. (Citly 0.649 0.508 0.632 0.470 Intergroup contibution to inequalily (percent) Sectors 21.6 26.9 22.7 22.5 Rings - 6.0 - 4.0 (Comunas 32.6 37.8 37.3 40.6 n.a. Not applicable. - Not available. Vote: HINcLXP = houisehold income per capita. a. Ranked in ascending order of meani HiNcAP. h. Ranked in ascending order of inequality. Soturce: Mohan (1986). tables 4-9. 4-10, and 4-11.  POVERTY. DISTRIBUTION OF INCOME. AND (;ROW(TH 8'$ term deleterious effects on the efficient and harmonious working of a city, the trend in Bogota is a cause for concern. This kind of segregation can lead to increasing distance between the different social groups of a city and impede social mobility. As an example, the problem of educational segregation has received great attention in the United States. One solution attempted there has been to bus children to school across neighborhoods, particularly to provide better opportunities to children coming from spatially dis- advantaged neighlborhoods. Busing caused considerable social tension. If significanit spatial differences start emerging in cities in developing countries, social problems similar to those experienced in developed countries may well arise. For example, spatial separation, by redticinig the personal contacts that are so important in transmitting infornmation about the labor market, could contribute to labor market segmentation. If similar studies are carried out in many cities, it may also be possible to identify the reasons behind higher and lower levels of spatial differ- entiation in cities and their effects on labor markets and social welfare as a whole. For example, are such phenomena caused by highi rates of growth where similar people congr-egate near each other to begin with? Over time does fur ther sorting take place and the city become less a col- lection of differenitiated neighibor-hoods and mor-e like an organic whole? Or are such tendenicies exacerbated as the city stabilizes. Clearly, whatever the answers to these larger questions, the spatial dis- tribution of income in Bogota and Cali does not conform to expecta- tions from the theory for largely monocentric cities, despite regular- and expected patterns of population densities and land valtues. It is there- fore importanit to study cities in greater spatial detail even if thiey con- form to overall recognizable patterns. Distribution and Characteristics of the Poor To identify the correlates and composition of poverty in Bogota and Cali, we have to address the problem of how to define and measure pov- erty. Despite the high quality and large quanitity of data at hand, it is pos- sible to identify only the people who must, in all probability, be poor. And it is difficult to identify all the people who are poor, except within rather large ranges of estimates. It is mnuch easier to identify relative pov- erty, defined merely as the bottom x percent of the population-say, 30 percent. We adopt a mixture of these approaches. A lower bound is first placed on the proportion of people who must be poor. We can then trace the movtmenit of this proportion over time, compute a range of estimates of others who might be poor, and, finally, collate the charac-  RI UNDFRSTANI)ING THE DEVELOPING METROPOLIS teristics of the bottom 30 percent, who roughly coincide with those who have been identified as poor. The notion of absolute poverty has been linked with that of malnutri- tioIi. BecaLuse food is the first necessity of household consumption, if people are malnourished, they can scarcely be consuming other things. Estimates of malniutritioi and poverty have therefor-e been made on this basis for countries (for example. Ohja 1970 and Dandekar and Rath 1971 tOr India) and for the world as a whole (for example, Reutliniger and Selowskv 1976). In these studies, estimates of minimum expendi- tures on food required for adequate nutrition were compared with estimates of actual consumption by income or consumption groups. Although nutr-ition involves the intake of various elemenits such as calo- ries, Aitaminls, proteins, and minerals, it is generally agreed that if' con- sumption of calories (Ihat is, energy) is adequate, other elements are also usually adequately consunied (see. for example, Sukhatme 1978). This, if we know the caloric content of foods, the usual food basket foi different income groups, the prices of foods, the expenditure on food by different types of families. and the share of food in total consump- tion or total income, we can relate the likely level of nutrition to income levels. Thle key problem in uSillg such a procedtire is that each individual's norm is different; some individuals, for example, need only 1,800 calo- ries a day, while others need 2,500. The minimum caloric consumption is arounld the usual requirement levels. A small percentage change in the minimuim caloric consumption used as the basis for estimating the number of malnourished people can lead to substantial differences in the estimate of malnourished people. For Colombia, the Instituto Colombiano de Bienestar Familiar (Colombian Institute for Family Wel- fare) has established 1,970 calories as the average minimum for ade- quate daily caloric intake, taking into accounlt Colombian conclitionis and the age and sex distributioni of the population. Using the same data sour-ces as earlier (the 1973 populatioll census and the 1978 World Bank-DANE Househol(d Survey), we attempted to calculate the proportion of people in Bogota and Cali who can in all probability be classified as malnourished and therefore poor. Because we had only incomne data from these sources, we used other information to translate the income level of each household into likelv calories con- sumed per person, taking into account the age and sex composition of the household and its likelv consumption pattern. Even if it is assumed that anyonie consuming less than 80 percent of reqtiired mean calories must be malnourished, different methods gave Us a likely range of 30 to 50 percenit malh-ntrition in Bogota and Cali in 1973 and about 15 to 25 percent in 1978. One problematic issue was the undercoverage of incomes in 1973 alluded to in the last section. Anothcr issue is that low-  POVERTY. DISTRIBITTION OF INCOMF. AND (GROWTH 85 income people typically eat foods that are calorie-rich; that is, the price of each caloric is lower for many cheap food items typically consiumled by the poor. Taking thiat inito account, we conclided that otir estimate of the share of the populationi that must be malnouirished could be nar- rowedl downi to about 25 to 30 percent in 1973 and 12 to 15 percent in 1978. These are the people whose incomes do not allow for adequate nutrition. There mav well be additional numbers of people who have adeqtiate incomes but suffer firom manititritioni fbr other reasons. Clearly, identifying poverty is not a simple task even with detailed data. Table 4-4 provides a quanititative ide a of the kind of misclassifica- tion that can result if we take, say, the bottomil 30 percent of the incomne distributioni as poor. The first row gives the breakdown, by decile, of people who constime less thian 80 percent of reqtuired calor-ies (0.8R). Virtuallv all the malinotirished, so defined, fall within the bottom 3() per- cent. The next three rows give the percenitage in each decile that falls into each nutritional range-less thani 0.8R, between O0Rand 1.OR, and more than 1.LOR If we defined all the bottom 30 percent as poor, we wotild misclassify as poor as many as 32 percent in the secon d decile and 58 percent in the third decile who are not necessar-ily malnour-ished (with 80 percenit of requir-emenits as the norm). Conversely, it is onl, in the top five deciles that we can be sur-e there are no maliourished or poor. As the table shows, household size declines monotonically with income, suggestinig that the malinouLishied are predoininiantly in large families with high dependency ratios.8 It is useful to compar-e the minimum wage ( 1,738 pesos in 1978) wsith the average incomies presented in the last two rowss of the table. Because the imajo rit of the second decile and part of the third decile are mal- nourished, an intermediate lhousthold incormie level in the range of 4,000 to 5,000 pesos (1978) can be utilized as necessary to attain ITini- mumit adequate nutrition. It is clear, tilen, that for adequate nutrition those households need roughly 2.5 workers per houiseholdl at millimuLm1 wages. In fact, thev characteristically have higher dependency ratios; that is, they have fewer than 2.5 worker-s per household. It would seem that the minimiiumit wage does not provide even a small family with enough income for the barest minimnum of nutrition. The monthly wages of fully employed people in typical low-income occupations- constr-uction workers, cooks, waiters, and so on-were betA-een 2.0 arid 2.5 times the miniimum wage in 1973 and 1978. Moreover, monthly wages hiad growin at an average of 8 to 9 percent a year between these *gears. It Would he dlifficult, therefore, to argue that mininuim wages were too high durinig that period or that they contribilted to protecting the forimial sector. In conclusion, our r esults are mixed. First, wc must be wary of malnu- tritioni or poverty estimates reached( with the use of simple income cut-  Table 44. Mapping Malnutrition into Income Deciles in BogotA, 1978 Decile" them 1 2 3 4 5 6 / 8 9 10 Total Percenitagc of O.8Rin decile 46 32 20 1 1 0 0 0 0 0 100 Percentage of decile in nutritionial category 0.8R 100 68 42 1.7 2.5 0 0 0 0 21 0.8Rto l.OR 0 28 35 40 16 3 0 0 0 0 12 L.(R+ 0 4 23 58 82 97 100 100 100 1()( 67 Total 100 1(0 100 100 100 1(10 100 100 100 1(0 100 Percentage of'populationi in decile 9.8 10.0 9.8 9.7 10.0 10.4 9.8 10.1 10.3 10.3 100 Average househiold size (personis) 6.1 5.5 5.4 5.3 5.2 4.8 4.6 4.3 4.1 3.5 4.8 Average monthly household inconie per capita (1978 pesos) 440 720 925 1,200 1,470 1,818 2.30() 3,110 4,780 12,640 2,980 Average monthly household income (1978 pesos) 2,580 3,980 5,050 6,340 7,680 8,710 10,675 13,400 19,670 51,790 18,140 Note: O.8R riefers to people who consumiie less than 80 per-cent of the minitiiim nutrition requiremenits. I OR r efers to people who consume less than the min- imum requiremenits. [.OR+ refers to those who meet or exceed the minimum rlequiremelits. a. The population (3.5 millioni) is ranked from poorest decile (1) to richest (10) accordinig to household income per capita. b. The minimum monthly wage in 1978 was 1,738 pesos. Source. Mhan (1986), table 5-7.  POV'FRTY, DISTRIBUTION OF INCOME. AND GROWTH 87 off levels. Second, one can get a sense of the likely ranges of malnutrition and poverty through rather detailed work. For 1978, for example, there is almost zero probability that any malnourished people are in the top five deciles, a small probability that they are in the fourth and fifth deciles, and a high probability that they are in the bottom 30 percent. Taking the bottom 30 percent in this case leaves out the few who are in higher deciles but does include some who are not necessarily malnour- ished in the bottom three. Third, there is a high likelihood that the def- initelv malnourished and poor decreased from about 25 to 30 percent in 1973 to 12 to 15 percent of the population in Bogoti and Cali in 1978. Having found reasonable estimates of people who must be poor, we now need to identify their demographic. spatial, and economic charac- teristics. Tlhe concenltration of the pooI in particular identifiable groups according to each of these dimensions is notable. The incidence of mal- nutrition is strikingly high among children betwveen the ages of 5 and 14 in large households of more than six members, a high proportioni of which are located on the peripher-y of the two cities and in the particu- larlv poor sectors (in the south in Bogota and the east in Cali). (For Bogota, see table 4-5; the situation in Cali is similar.) An index of between 140 and 150 for childr-en ages 5 to 14 in sector 2 means that the children are 1.4 to 1.5 times more likely to be malnourished than the nmean of about 28 to 35 percent of this group. Note that this mean is itself much hiigher than the overall mean of 21.1 percenit. A few other features stand out from this table. Although, on average, the old are less likely to be malnourished and poor, those on the periph- ery (ring 5) and in the southern part of Bogota (sector 2) are excep- tions and disproportionately so. The children in these areas are not particularly hard hit, but the adults are worse off compared with the city average. Since malnutrition is accentuated by bad sanitation, poor hygiene, disease, and crowding, the purely income-based estimates of malnutrition are more likely to be understated for the pooI rieighbor- hoods. There, even people with somewhat higher- incomes who can afford more food could suffer from malnutrition because of chronic stomach disorders and other neighborhood effects.9 We can now bring together a number of results that are important for understanding the structure of Bogota. We have pointed out in chapter 3 that densities in the poorer parts of Bogota and Cali are relatively uni- form and do not decline with distance, even though land values do. We now find that many of the households on the distant periphery are large, poor households with high dependency ratios; that is, households have responded to lower land prices and located at the periphery; but because of higher household size, poptilation densito continties to be higlh in these areas. It would seem that poor households wvith more older people incapable of working tend to descend furthe r into poverty.  Table 4-5. Spatial Distribution of Malnutrition by Age Group in Bogota, 1978 Percentage malnmiLlished 'Age group bidex of mia/nurition' in age grollp Ring 1 2 3 4 5 6 0-4 160 92 54 77 117 101 19.7 5-9 28 76 77 81 120 112 28.4 10-14 126 82 76 86 112 98 34.1 15-54 77 65 45 84 133 9 2 17.7 55+ 131 97 51 68 140 88 15.4 Total 101 73 55 82 127 98 21.1 Sector 1 2 3 4 5 6 8 8 0-4 160 147 94 54 81 103 75 40 19.7 5-9 28 144 98 73 55 96 70 77 28.4 10-14 126 147 95 88 6i7 94 70 58 34.1 15-54 77 156 97 85 71 120 57 29 17.7 55+ 131 188 108 83 141 84 51 23 15.4 Fotal 101 155 98 80 72 14 61 38 21.1 a. Tlhe index of Tialnutrition is calculated as: Ai = Percentave 0.8R in age group i in sector Percentage 0.8R in age group i where 0.8R refers to people who ConsuLie less tiant 80 percent of the minifmljnu nut itional requiirenilents set bs the Institiuto Colombiano de Bienestar Famiiiliar. Source: Mohan (1986). table 5-12.  POVERIY, DISTRIBLITION OF INCOME, AN)D GROWTH 89 This tendency is exacerbated for income earners with relatively flat age- earnings profiles. As is well known (and documented for Bogota in chapter 5), people who are poorly educated and unskilled do ncot ear-n more with experience. As these workers get older and enter the stage of household formation, tlhey have to support more people on the same inicome. Children of the unskilled are therefore disproportionately poor on the whole. The situation is exacerbated when old people are present in the household. It is interesting to observe how this process works itself out and how spatial effects can be identified in terms of residenice patterns. Although economic growth in the mid- to late 1970s was quite effective in reducing the number of people who have no poten- tial of achieving minimally adequate nutrition, the concentration of the poor in particular areas remains striking. The concentration does, however, allow easier targeting of antipoverty, nitrition, and educaCtion programs. Having establishecl the demographic and spatial dimensions of pov- erty, we now need to look in a little more detail at who the workers are who happen to be poor. For this purpose, poor workers are identified as being in the bottom 30 percent of households ranked bv HINCAP. The very fact of working does much to alleviate poverty. Onilv about 20 percent of male workers and 15 percent of female workers fall into the bottom 30 percent. Mfost women are secondary workers (that is, not household heads). A household is better off if there is more thani one income eariner. As might be expected, almost half of ial male workers with no education and about one-third with a primary education are classified as poor. An analysis of poverty by age an-d educationl groups confirms that childreni suffer disproportioniately from poverty. Amonig both males and females, workers in the 35-44 age group are mor-e likelv to be poor. Thus, whereas about 20 percenlt of all male workers ar e clas- sified as poor, more than one-third of the 35-44 age group is so classi- fied. Most of these are workers with little or no education, but about 15 percent have a secondarv-level education. Workers with little eduication have relativelv flat age-income profiles (see chapter 5), and it is these workers who appear to fall into poverty during the period in their lives when their households are expanding and their children are going to sclhool. This highlights the real economic hardship unclergonie by the poor in educating their children into their tcens rather than sending thein out to work. Of those in the 12-14 age group who work (and the sample of this group is not large), very few are founid in poverty. Those who wor-k clearly contribute to houschold income and help lift the household out of poverty. Poor workers are concentrated among a few occupations and corre- sponding industries. Blue-collar occupations accoun1t for about 70 to 75 percent of poor workers but only 55 to 57 percent of all workers. Produc-  4(! UNDERSTANDINO THE DEVELOPING METROPOLIS tion, constructiorn, and transport workers are particularly likely to be poor; along with service workers, these groups make up about 70 per- cent of all poor male workers. Among women, it is notable that maids (domestics), who account for about 20 percent of the female labor force, are not particularly poor. Almost one-third of poor female workers are service workers (excluding maids). Among the production workers, the traditional industries of textiles and footwear, lumber and wood prod- uicts, an(l printing and publishing account for the majority of the poor workers. Female participation in productioni activities is mainly in the textile and footwear industries, which do not seem to pay women well. Do poor workers work long hotlrs or are they short of work? Poor men are strikingly overrepresented among workers who work long hoturs; about 20 percenit of poor, male workers are in this category. Only about 8 percent of the poor wvork fewer than forty hours a week, and about 55 to 60 percent work the "normal" forty- to forty-eight-hour week. Sales, construction, and transport workers wvork the longest hours; among the poor, only professional and technical workers work relatively short hoturs. The storv is different for female workers; they appear to be overrepresented among those working shorter hours. (It is mainlv sales workers and maidls who work very long hours.) It seems that many poor, female, part-time worker-s wotuld like more work but find it difficult to get or else cannot work longer hour s because of responsibili- ties at home. Overall, it appears that pool male workers are not lacking for work. There is little overt unideremployment; indeed, if anythinig, the poor are overworked but are locked into low-productivity occupatioas. implying a need for traininig programiis to improve people's skills. The sales work- ers many actually be under-employed-that is. idle for a substanitial part of their workinig hour-s-but they comprise only about 25 percenit of all poor workers. Almost half of the poor. overworked men are transport or construction workers. It is unlikely that they are idle during much of their woi-kinig day. On the other hand, a substanitial portion of the women are either genuliniely underemployed (willing and able to work more hours) or uLnable to work longer hours. Poverty will therefore not be alleviatecd by mere expansion of employment opportunities. Poor womein need greater employment opportunities, but they could also benefit from day-care progr-anms-which themselves generate emplov- ment for the providers. It is often said that recent migrants earn less than others and have to take very low-paying jobs while waiting for better-paid opportunities. The evidence is mixed. In Bogota in 1975, 22 percent of the most recent arrivals earned less than the minimum wage, compared with about 38 percent of all workers. Among those who had been in Bogota eleven and twenty years, however, as many as 20 percent earned less than mini-  l'OVER-TY, DISTRIBUJTION OF INCOME. ANI) (GROwTrH 911 mum wage. It appears that low wages are spread fairly evenly among workers of all vintages and that experience in the city does not guaran- tee a highelr wage. On the whole, it is clear that recent migrants have no monopoly on poverty; indeed, 60 to 70 percen)t of all workers earninig below minimumil wages have been in Bogota for more than ten years. We also examined in detail who the unemployed are. Because we have information about currenit income only, it is difficult to distinguish between people who are poor because they are unemployed and those who are Ul mploed because they are poor. This problem is partially addressed by looking at primary and secondary workers separate1y and examining in detail the incidenec of unemployment in each decile. Even if wc regard all unlemilployed primary workers as temporarily out of work, the unemployment rate in the bottom 30 percent woul( still be disproportionately hiigh. We therefore conclude that the probability of unemploymnenit is higher among thc chroniically poor (see table 4-6). Our coniectures are further supportecd by the particularly low participa- tion rates in the bottomil deciles. which containi a substantial proportion of people who are not working. Male participationi rates rise and thien flatten out. We emerge withi a distinctive profile of the unlemilployed. In the bot- tom decile-that is, among the verv poorest-only about 4(0 to 50 per- cent of males and 15 percent of feImales over 12 years of age participate in the labor force compared with the overall participation rate of about 70 percent for men ancd about 30 per-cent for women. Among males (over age 12) in the bottom decile who dol) participate in the labor force. almost 30 percent were otit of work. Conse(qtuenitly, onilv about 35 per- cent of all males and about 8 to 10 percent of all females over age 12 in the bottom decile have jobs. Of thc remainidler, some are discotiraged workers and others are incapacitated by sickness or injury. Generating employmerit and upgrading skills are therefore only partial solitions for poverty removal. Many of the poor are probably unemployable, and their poverty must be dealt with directly tilrotigh welfare measur es. Summary The 1970s marked a turning poilnt in Colombia's economic history. Pru- dent macroeconomic and trade policies, along with the successful sup- port of agricultural growth durinig the late 1960s and the early to mid- 1970s, had helped create a growth-oriented, labor-using environiniienit. Colombia's uniprecedenited expansion of emplovmenit during the 1970s brought about a palpable tighteninig of the labor market. This process was aided by the boom in intcrilational coffee prices dtiring the late 1970s. In addition, this was a per iod of demographic transitioni anic the  Table 4-6. Unemployment in Bogota by Income Decile and Sex Income deoile& Total Sex and year 1 2 3 4 5 6 7 8 9 10 All (thonisands) Number in the Unentmp/vinent ?-ate (pfercent) labarforfce Male 1975 30.8 11.4 7.3 8.7 8.0 5.2 3.8 2.5 2.6 0.4 6.9 624 1977 30.2 12.2 6.2 6.5 6.6 5.5 3.9 4.4 2.7 0.5 6.2 782 1978 19.8 10.2 6.1 5.6 4.2 2.7 2.0 2.9 0.7 (.6 4.7 753h Female 1975 29.2 12.1 12.7 9.6 7.9 7.8 6.1 3.7 1.9 1.1 6.4 337 1977 16.3 11.1 9.4 9.5 7.7 6.4 5.8 4.2 1.7 2.1 5.8 469 1978 6.2 7.3 5.1 8.9 7.3 4.3 2.5 1.9 1.7 0.4 4.3 463b Nuzmber l)j.tburiution off the anempilved (pe..rrent) utieottiployed Male 1975 23.8 15.5 11.2 14.2 12.4 8.1 6.6 3.7 4.0 0.6 100.( 43 1977 23.2 15.0 9.() 10.7 1(.3 10.2 7.1 8.1 5.0 1.5 10(.0 49 1978 30.9 16.2 10.3 11.2 9.4 6.0 4.7 7.9 1.8 1.6 100.() 35 Fernale 1975 13.0 11.8 13.0 12.4 11.2 12.4 10.6 8.1 4.3 3.1 100.( 21 1977 15.1 10.0 11.(0 10.7 11.1 10.5 11.0 9.7 5.1 6.0 10(.0 28 1978 21.9 16.7 9.5 13.6 13.8 7.9 5.5 5.0 5.2 1.0 100.0 20 Note: Percentages may not add to I00 because of rounding. a. Ranked fiom poorest (1) to richest (10), based oin monthly houisehold income per capita. h. The cstimated population of BogotA was lower than expected in 1978. The estimated labor force in 1978 is also low becatise the estimates for 1975 and 1977 use tihe original r)-NE expansion factof s. Source: 1975 DANE Special Bogota Household Survey; 1977 DANr llouschold Surve%: 1978 World Bank-DANE Household Surivev.  POVERTY, DISTRIBUTION OF INCOME, AND GROWTH 93 achievement of an urbanization level of more than 60 percent. The result was a slowing of urbanization that was somewhat unexpected after almost fifty years of rapid urban growth. The tightening of the labor market led to measurable growth in household income, and poverty decreased in Bogota and Cali during the 1970s. According to the data at hand, the distribution of income in the two cities seems to have improved somewhat between 1973 and 1978. The quality of the data is not robust enough for this to be stated conclusively, but such a tendency would be consistent with the other changes in the national economv. Despite this improvement, inequality in the two cities remained very high: the Gini coefficient remained between 0.50 and 0.55. The spatial separation of rich and poor appears to have increased during the period under study, despite the overall improvement in income distribution. The spatial separation is more pronounced by radial sectors than by distance from the city center, the norm in cities in developed countries. In this study, we have placed great emphasis on understanding the spatial distribution of households according to their income levels and other characteristics. This is important as a prelude to comprehending the changes that take place in patterns of housing demand, transporta- tioIn, and demand for infrastructure as a city grows. Whereas some of the specific patterns found, particularly the high degree of spatial segre- gation of the rich and poor, may be peculiar to Bogota and Cali, our findings illustrate the importance of understanding these patterns in order to make policy regarding infrastructure and other urban services. Spatial disparities in an urban area can increase segmentation in the labor market by exacerbating differences in educational background, family background, and other determinants of earnings. The deficien- cies in household envir-onment and school quality that a poor child is subject to can be worsened by such segregation. If there is a large con- centration of poorly educated or illiterate people in specific areas of the city, it becomes much more difficult to operate good schools in those areas. It is difficult to find good teachers, the student peer group con- sists of equally disadvantaged children, and the quality of education suf- fers. Similarly, the peer group of poor workers or of the chronically unemployed usually comprises other poor workers and those who are chronically unemployed. The diminution of aspirations and contacts decreases the worker's chances of moving out of the low-income or unemployment trap. The r icher areas of a city are typically better served by public utilities than poorer areas. Richer residents are simply more effective in voicing their demand for urbani public services. If an area has a mix of both rich  94 UNDERSTANDING TH1E DEVEl OPING MEFROPOLIS and poor residents, the poor benefit from the better sanitationi, water supply, roads, schools, medical services, and the like that may exist because of the influence of the rich. If. however, the poor live in large, separate tracts, as in the southerni part of Bogota and the eastern part of Cali, it becomes easier for hard-pressed authorities to neglect them. In Bogota andl Cali the severity of the potential spatial disadvantage has been mitigated by the operation of decentralized housing markets and private land development. The decentralized nature of the urban public service agencies has also helped make t,hese agencies more responsive to demands, including the demands of the poor, and the service deficits in the poor areas have not been as large as they might have been. There is some indication that the land-uise zoning policies have con- tribtited to the prevailing spatial development pattern. Resicdential- density zoning often consciously separates lower-density areas from high-denisitv areas. Although there are market forces that tend to sepa- rate the rich and the poor through the operation of the land and hous- ing markets, care must be taken in land-use zoning not to exacerbate such natural tendencies, but instead to promote more mixed land use. Although we have indicated in this chapter the difficulties of measur- ing the extent of poverty in a city, we have also demonstrated the possi- bility of arriving at some robust estimates. T'he areas of Bogota and Cali that have high malnutrition anid poverty levels are characterized by large, low-income families with high dependency ratios becatise of the presence of many children. Workers who have low lifetime earnings sink into greater poverty as their households expand. In cities in developing countries large numbers of workers with low education levels have flat age-earnings profiles. The expansion of their families dur-ing their mid- dle age pushes them into poverty, and hence an atypically large propor- tion of children are found in the malnotirished or poverty grotip. This finding provides powerful support for the idea that child-directed nutri- tion and health programs are effective weapons against poverty. If pov- ertv is concentrated spatially, it becomes easier to target such programs, bitt knowledge of city structure is then essential. Another important finding has been that many people who are fully employed are poor: their low productivity, not unemployment, is the cause of poverty. We also found that unemployment is more likely to occur among poor households. The removal of core poverty therefore involves basic welfare measures: mere increases in employment, although essential, will not be sufficient. investmenits in human capi- tal-both of adtilts and of those who are potential workers-would enhance productivity and therefore earnings (see chapter 5).  POVERTY, DISTRIBUTION OF INC OME, AND GROWTH 95 Notes I. See, for example, Urrutia (1985) and Aniand (1983) for excellent studies of Colombia and Malavsia, respectivelv. 9. Estimates for 1964 are available in Selowsky (1979) and Berrn and lUrrutia 1976i); see Berrv anid Soligo (1980) for 1974 estimates. 3. Mor-risoni (1984) claimed that this shair-e is similar for capitalist and socialist countr ies. 4. See Ananid (1983) and Mohani (1986i), chapter 4. for an exposition on the derivation of the two measures. Both indexes vari between 0 anid 1, with 0 beiig equivalent to complete equaility. The Theil index maxv be interpreted as measur- ing the departur-e of an incomne distribuitioni from an even distribuition. 5. A word about these differ-enlt ineasires is in order here. The most commilon distributioni used has been thiat of' househiolds ranked by houselhold income. It has been argued by Datta alid Meermanl (1980) and Aniind (1983) that this dis- tribution does not provide a good indication of differences in the level of living in the poptilation. Our general concer-n is with the welfare of individuals rather than of households. Moreover. hotisehoilds varx bv size as well as by age and sex compositioni Hence, for true comparisons, adjustments should be made to accounit for these variations. The best soltitioni would be to make adjutstments bv deriving "adtilt equivalenit" measures of constiilption. Practically speaking, this is difficult; the next-best strategy is to use HIN APt' to ranik indixiduals (or house- holds) for welfare comilparisoils. As Datta and Meermart also totind. although the ranikings of par-ticillar indixiduals change substiaintially when ranked by HIN- (,AP as opposed to household incomiie, the overall result does not. The particular rankings are clearly important wheni the aimi is to find out the character-istics of the poor or the rich, but the distuibutioln is not too different. 6. Reported in detail in Mohani (1984), whicih also reports estimates of other ineqiualitv measures. 7. See Anand (1983) and Mohan (1984, 1986) foi a (derivation of the decom- position of the Theil index. The wvithin-group contr-ibLtioll is essentially a weighted average of the indexes of inequality wvithin each grotip, and the between-group contribtition is a measure of the differences in the means, appropriately weighted. In the Theil index the weight used is the group share in incomie. It can be showil that this decomposition is additive. 8. t'he dependenicy ratio is thie ratio of the dependenit populattioni (defined ias un(ler 15 or over 64 years old) to tire working-age populationi (ages 15-64). 9. This has beeen confirimied to be the case from dletailed clinical utitrition studies in Cali assoc iated withl those repor-ted in McKav and others (1978).  Chapter 5 Workers and Their Earnings In this chapter we describe the structure of employment and the associ- ated pattern of labor earnings. We also report on the determinalnts of these earnings in order to under-stand the changes in income that have already been documetnted. And we briefly discuss segmentation in the labor market in terms of the identification of an informal sector. One verv interesting feature of the labor market in the late 1970s was the increasing participation of women in the labor force; this is also ana- lyzed. Altlhough estimations were conductecd for both Bogoti and Cali, for ease of presentation we provide the results for Bogota only. The findings for Cali were generally consistent with those for Bogota. Only the main results are presented here, becatuse details of model structure and estimation procedures ai-e available in Mohan (1986, chaps. 6, 7, 8, and 9). The overwhelminlg message that emerges is that high urban unem- ployment, long-term employmenit problems, high poverty levels, and segmented labor markets are not inherent in a situationi of rapid urban gr-owthi. An appropriate economic environment bolstered with a judi- cious mix of policies can lead to positive results, as occurred in Bogota and in Cali in the late 1970s. For backgrlound it is important to be aware of the overall changes in the employmnent structur-e that have taken place in Colombia over the past fifty years. We documented earlier that Colombia went through an intensive phase of ulrbanizationi over this period, going from a level of about 30 percent to 65 percent. As happenis in maniy countries, the change in employment structure followed that in sectoral output, with some lag. Hence, as is evident fi-om the data below showing each sec- tor's percentage share in total employmenit, a dramatic change in the structure of employment OCcuIrred in the relatively short period after the mid-1960s. 946  WORKERS AND THEIR EARNINGS 97 Secthn 1925 1938 1951 1964 1973 1978 Primary 67.1 67.1 57.3 50.6 41.9 35.2 Secondary 15.7 17.8 16.3 17.7 20.6 22.4 TerLiary 17.2 15.1 20.4 31.7 37.5 42.4 The increase in the share of the secondary and tertiary sectors, which are primarily urbani, is remarkable over the 1973-78 period. The expan- sion of education (see table 5-1) that occurred during the 1950s and 1960s increased the quality of the labor force very rapidly. Emplovment change, however, was more marked in the late 1970s (see table 5-2). In absolute terms it is clear that the tertiary sector was the mnost important component of this growth in employment, but the rates of growth were also high for manufacturing and construction-between 7 and 9 per- cent a year, higher than output growth in all these sectors. The output growth in agriculture was higher than employment growth over the same period, thereby implying significant improvements in agricultural labor productivity. Hence the predominanitly urbani sectors have grown in a labor-using maniner. This is contrary to the usual scenario in devel- oping countries, where the urban sectors are regarded as capital-usilng and agriculture as labor-using. Stcucttl-al change in productioll usiallv takes place faster than the change in employment struclure: output growth in manufacturillg usually occurs much faster than employment growth, and it is the agricultutral sector that tends to absorb the expancl- ing labor force in the absence of employment opportunities in the mani- ufacturin-g and tertiary sectors. This is the distinctive phenomenon that occurred in Colombia in the late 1970s, which makes this study of the Bogota labor market and earninigs determiinationi particularly interest- ing as an example of rapid population growth in a city accompanied by employnment and income growth. There was a remarkable downturni in unenmploymenit rates in Coloin- bia in the late 1970s, implying a consi(derable tighteniing in the labor mnarket: participation rates rose until the mnid-1970s because of the mas- sive expansion in education butt then began to stabilize. The last COIlIuIm of table 5-1, labeled "Other," includes mainly housewives: their increased participation in market work betwveen 1973 and 1978 is clear. The expansion of edlucationi has applied equally to women, and this has begun to manifest itself in their greater inclinationi to participate in market work. Given its large share of the total urban population and its role as capital city, the course of Bogota in the 1970s was generally that of urban Colombia. It had more than its share of the employmenlt boom: its unemployment rates were consistently below those of otlher cities. The constellation of economic policies after 1966 tended to promote exports, which are usually more labor-using. The governmenit was also  98 UJNDERSTANDING THE DEVELOPING METROPOLIS Table 5-1. Changes in Labor Use in Colombia, 1951-78 'Ibtal population age Percentage of total 15 and over Year (thoausands) in sehool Employed Unemployed Other, 1951 6,910 2.4 54.1 1.7 41.8 1964 9,528 5.3 50.2 3.4 41.1 1973 12,449 9.4 45.5 5.4 39.7 1978 14,698 10.3 52.1 3.1 34.5 Average annual rate of change (percent) 1951-64 2.4 8.5 1.9 8.1 2.4 1964-73 3.0 7.8 1.9 8.4 2.6 1973-78 3.4 5.4 6.2 -7.7 0.5 a. Nfainlsi housewives. Source: 151, 1964, 1973 population censuises; 1978 IDANE National Household Survey (Junre): Sokol and others (1984). more oriented toward a pattern of public expenditures that helped the poor: for example, expenditures in public health and education were emphasized. The improvement in the availability of public services is also quite evident in Bogota durinlg the 1960s and 1970s (see chapter 9). All this combined to produce increased demand for labor in urban Colombia and, specifically, in Bogota. We observed a marked increase in the participation of working-age women in the tightened labor market of Bogota in the late 1970s. Because the participation rate of prime-age men (ages 25-55) in the labor market is usually close to 100 percent in any case, the increase in the labor force has to come from people not traditionally in the labor force-prime-age women, young men, and young women. Because of the expansion of educational opportunities, however, young men and young womeni defer their entry into the labor force and opt to stay in Table 5-2. Employment Change in Colombia, 1973-78 Thousands Thousands Net change, Percentage Annual rate of vorkers, oj workers, 1973-78 oftotal of change .Sector 1973 1978 (thousands) change (percent) Agriculture 2,3(05 2,629 324 16 2.7 Mining 68 69 1 0 0.0 Manufacturing 912 1,357 445 22 8.3 Construction 255 360 105 6 7.1 Tertiary 2,124 3,249 1,125 56 8.9 Total 5,664 7,664 2,000 100 6.2 Source: Sokol and others (1984).  WORKERS AND THEIR EARNIN(;S 99 school longer. This tends to have a lagged effect in that people with higher education are more likely to participate in the labor force. The participation rates of women with higher education increased from about 43 to 53 percent between 1973 and 1978, compared with partici- pation rates of about 36 percent and 28 percent of women with primary and secondary education, respectively, in 1978 (having changed from 30 percent and 31 percent, respectively, in 1973). The increasing trend in participation was at the low and high ends of the education spectrum during this time of labor market tightening. Some of the improvement in poverty and income distribution indexes is attributable to the increased utilization of female unskilled labor dur- ing the boom vears of the late 1970s in Bogota. The participation rate of females in the bottom 30 percent as classified by household income per capita (HINCAP) increased from about 17 to 34 percent in the short time between 1975 and 1978. A detailed analysis of the participation behavior of women showed that the own-wage elasticity of all married wvomen was between 0.6 and 0.8, whereas that of primnc-age nonmarried women was much lower-about 0.15 (see Mohan 1986, chap. 7). The latter group included widows and never-married women. The analysis indicates that women in cities in developinig counitries can be expected to participate in market work in much greater numbers in the future. Both a secular increase in education levels, which raises the opportunity cost of not workinig, and the pure wage response under conditions of a tightening labor market will motivate the rise. The education level of the work force changed along with its makeup (see table 5-3). By the late 1970s the Bogota work force was almost fully literate. Education was more advanced as well as more widespread. In 1973 about 40 percent of the labor force was educated tip to at least the secondary level in Bogota; just five years later this propoltion had increased to about 55 percent. The expansioin of higher educatioll was quite dramatic, especially for the women. With this vastly increased sup- ply of workers with higher education, the earnings differentials between higher and primary education have declined-which is another expla- nation of the overall improvement in income distribution (see table 5-4). Education expanded at very high rates in Colombia right through the 1960s and 1970s. In 1960 primarv school enrollment was about 75 per- cent; by the mid-1970s it was 100 percent. Enrollmenit in secondary schools increased even more dramatically, from abotit 12 percent of the relevant age cohort in 1960 to about 40 percent bv 1975 and just under 50 percent by 1980. Enrollment in higher edticationi increased from 2 percent in 1960 to 9 percent in 1975 and 12 percent by 1980. The speed of expansion was particularly marked between about 1965 and 1975. The private sector has played a substanitial role in the expansion of edu- cation in Colombia. At the primary level, almost 20 percent of enroll-  1(0 UNDERSTANDING THE DEVELOPIN(; MFTROPOLIS Table 5-3. Distribution of Workers by Sex and Education Level (percent) 1973 1978 E.ducation level Males Fenales All Males Females All Nonte 5.1 9.2 6.5 2.3 5.7 3.6 Primai y 55.0 53.9 54.7 42.0 42.3 42.1 Secondarv 31.1 32.5 31.6 38.8 37.6 38.4 Ifigher 8.8 4.3 7.3 16.8 14.4 15.9 lotal 1(00.( 100.0 100.() 100.( 10(0.( 100.( Total numilher of workers 488,000 249,000 738,000 732.000 454,000 1,186.000 lotal nmbl)er in sample 45,080 23,041 68,121 3,078 1,914 4,992 Vole: Percentages in 1973 do not total 100 becauise of some omiiitted categories. Soura': 1973 population census sample; I 978 World Bank-rXF: Houscehold Survey. ments were in private schools; this rose to ahout 54 percent at the secondary level and remained fairly steady at 47 percent at the higher- edtication level in the late 1960s (see Jallade 1974). Durinig the 1960s the private sector increased its share at the primary and higher levels, whereas the public sector expanded much faster at the secondary level. Hence, the expansion of the Colombian education system was due as much to private initiative as to specific governmental initiatives. Total government expenditures onl education increased from less than 2 percent of GDP inl 1960 to about 3 percent by 1970. As might be Table 54. Mean Income Ratios for Workers by Sex and Education Level 1 97,3 1978 E.ducation level Males' Females Feimale/male Males F emales' Female/male None 0.66 0.66 0.51 0.69 (.80 0.75 Primlalry 1.00 1.00 0.51 1.00 1.00 0.65 (1,175) (595) na. (4,903) (3,176) na. Secondary 2.11 2.62 0.63 0.159 1.81 0.74 Higher 6.75 6.22 0.47 4.82 3.10 0.42 T'otal 1.83 1.72 0.48 1.87 1.59 0.55 Total nutmber oftworkers 488,000 249,000 n a. 732,000 454,000 n.a. ii.a. Not applicable. a. Ratio of mearn earnings, with those of worker-s with primary education used as a base. Nole: Ntimbers in pairentheses are actual earnings (in 1973 or 1978 pesos) of workers with primary etducadon. Sourme: 1973 popltilationi censuis sample; 1978 World Bank-A.kNF Household Survey.  WORKERS AND THEIR EARNINGS 101 expected, private-sector schools were more prominent in the larger cit- ies: government expenditures on primary and secondary educationi were more directed to smaller towns and the rural areas. Education finiancing was quite complex; it included substantial subvention from the federal government to state (departmental) and local governments and to other decentralized agencies. As will be seen in chapter 9, much of Colombia's developmental public expenditure is rotied through decentralized, relatively autonomotis agencies, and this *sas tuue in edui- cation as in other areas. The period under study immediately ftollowed this very substanitial educational expansion. The participation of the private sector indicates that the public perceived high returns to e ducatio>n. Moreover, employ- ment expansion in the 1970s was siuch that therc was adequate demand for this educated labor. The proportion of professional and technical workers increased con- siderably between 1973 and 1978 among both men and women. In oth- er respects, the distribution of workers among occupationis was quitc similar from onc vear to the other. Among males, production workers accounted for about one-thircl of the labor force, constructioni and transport workers for about 16 percent, professional administrative and technical for another 15 percent, and otlier service for the remaining third. Among females, domestic servants constituted the largest group, accounting for one-fifth of all women workers; professionial administr-a- tive arid technical workers were about II percent, office workers 19 per- cent, and production workers about 15 percent, with other services rnaking up the remaining third. The increase in the professional and technical workers was for both men and women. W"e found that overall mean nominal earnings increased by more than 400 percent betwceen 1973 and 1978, exceeding the inflation rate of 250 to 300 percent over that period. It is quite striking that, consistent with earlier observations of the decline in earnings differentials between higher and less-educated workers, the rate of increase in the less-skilled occupationis was higher than that for the professionals, wlho on average barely kept pace with in- flation. The larger increases occurred in lower-paid fields like procluc- tioIn, construction, and transport. Production wor-kers were among the lowest paid. Among women, even domestic servanlts were better remu- nerated than the production workers, if the servants' income in kind is also accounted for. Domestic servants typically live in with their em- ployers and get free food and some clothes in addition to their mone- tary wages. As might be expected from the spatial distribution of income described in the chapter 4, the lower-paid, less-skilled workers live pre- dominantly in the southern part of Bogota, and the professionals and administrators live in the north (see map 5-1). Sectors 2 in the south  102 UNDERSTANDING THE DEVElPIN(G METROPOL.IS Map 5-1. Bogota: Distribution of Occupations by Sector IBRD 25345 \> ) ~administra ors r Pomucn aoundai e , w~~~ruto o kr Scarce: lerks rnd, ap tyists, 5-!. | uction, co o - Comuna boundaries >j E l Sectors JANUARY 1994 .Sauxrr., Mohian 1986, map 5-1.  WORKERS AND THEIR EARNING,S 103 and 6 in the west are quite similar in their worker composition, although workers in sector 2 are poorer on average. The starkest con- trast is between sector-s 2 and 8, the poorest and richest sectors. Sixty percent of the workers who live in sector 2 have only primary education or less; in contrast, more than 45 percent of the workers in sector 8 have higher education. This picture has not changed over the years except for a slightly increased tendency for workers with similar occupations to cluster together-a pattern consistent with the increasing spatial ine- quality based on incomes. I)espite these spatial differences in terms of income and occupation we find no exidence of similar differences in participation rates. People in the poorer areas are not less likely to work. In summary, we observe that the recorded increases in income in Bogota have resulte(d from changes in both the demand and the supply sides of the labor market. On the demand side, the continuing overall changes in the structur-e from a predominantly agricultural economy to an industry- and service-oriented one has resulted in continuinig ex- pansion of employment in these sectors. These sectors' expansion was labor-using in the late 1970s. with cmployment expansion being greater than output expansion. On the supply side, these changes have been matched by continuing migration from rural areas. increased participa- tion of women in the labor force, and the improvement in educational quality of the labor force. We therefore observed overall increases in labor- earnings. There was little perceptible change in the structure of employment within Bogota, with industry- or production-related employ- ment accounting for just under one-third of total employment through the 1970s, although there was a clear shift toward more highly skilled oc- cupations. Workers lived whele they might be expected to, given spatial inconme distribution, with less-skilled workers concentrating in the south and west of Bogota and the professional and administrative workers in the north. Because the earnings distribution in urban areas is often said to result from the segmentationi of labor markets into the formal and informal sectors, this issue is emphasized in the rest of this chapter. We estimate earnings functions to be the predominant analytical tool used to distin- guish the detcrminants of earnings and to test whether similar people earn different incomes in different sectors, however defined. Overall, we find little evidence of segmentation in the tight labor market condi- tions of Bogota in the late 1970s. We also explore the carnings differ- ences that persist across space in Bogota. The Benefits of Education and Experience There has been considerable controversy over the inter-pretationi of returns to educationi as measured by differences in earnings, and much  104 UNDERSTANDING THE DEVELOPING METROPOLIS effort has gone into estimating these returns. Psacharapoulos (1980) summarized the results for different countries. Among the more accessi- ble works on the economic benefits of education for urban workers are Anand (1983) and Mazumdar (1981) on Malaysia and Bourguignon (1983) and Fields and Schultz (1980) on Colombia. Most estimates are lower for developed countries than for developing countries. This is partly a measurement issue, because the variance in education levels is typically lower in developed countries and it becomes more important to measure the quality of schooling. The private returns to schooling range between 12 and 18 percent for developing countries and between 7 and 10 percent for middle- and high-income countries. The highest rates of return seem to be found in Latin America. The typical pattern in other r egions is for primary education to bring the highest marginal returns and secondary education the lowest. Colombia conforms to the broad Latin American pattern, but the returns for higher education exceed those for primary and secondary education. Most of these returtns are calculated with the assumption of no tuition costs during the time of schooling and are therefore somewhat overestimated. We esti- mate the returnis to schooling in Bogota in some detail in order to understand thie measured pattern of income distribution and to follow the changes that have been observed. *e build on the human-capital model of labor earnings as developed by Schultz (1961), Becker (1964), and Mincer (1974).' One of the key criticisms of the traditional human capital earnings function (equation 5-3 in note 1) attempting to estimate the returns to education has been the omission of a measure of "ability"-that is, a measure of v. It is argued that the productixity of schooling depends on the level of initial ability, and ability itself is affected by family and other background. The earnings function should therefore, at a minimum, contain measures of ability as well as background variables. The data sets we used for Bogota were rich but did not have anv direct measures of ability, family background, or schooling quality. We did, however, have somewhat comparable data sets for four different years between 1973 and 1978, so some time trends could be observed. The only background variables that were available, however, were the loca- tion of currenit residence, location of previous residence, and place of birth. In view of the distinct characteristics observed for different parts of the city, we regard the location of residence as a good background proxy. We also use place of birth, categorized by size of town or rural community, as another proxy. These variables may be argued to be good proxies for family background and schooling quality because children typically go to neighborhood schools. The hypothesis is that quality and intensity of schooling might vary positively with city size and the overall  WORKERS AND THEIR EARNIN(;S 105 quality of neighborhood. If these are good proxies, we correct at least partiallv for the omitted variables bias, and the resulting estimated rates of return to schooling may be relied on. So far, we have neglected on-the-job training as a component of human capital. This is a straightforward extension, and we add the years of experience as another explanatory variable. To test for segmentation we can also add different variables measuring characteristics of employ- ment. The final equations that we estimate are then of the form:2 y= f (schooling, experience, region of origin, current location of residence, characteristics of employment). Trhe overwhelming result of oui- estimations is that the simple human capital model, using merely sclhooling and on-thejob-training as expla- nations of earnings, works well in Bogota and explains as much as 50 percent of the variance in earnings. WN'e paid particular attention to a couple of issues. The major expan- sion in education has been documented and commented on. It lhas been suggested that this may have led to a decrease in the returns to ed- ucation. The availability of data sets for four different years enabled Us to test for this hypothesis between 1973 and 1978. We have also corn- mented on the particular expansion of higher education. It was therefore of interest to measure the rates of return to education distinguished by different levels: primary, secondary, and higher. Another important issue was screening. It is alleged that part of the returns to education) are actually returns to certification, rather than to the knowledge gained from an additional year of schooling: employers are willing to pay a pre- mium to someone with a high school diploma or a completed college degree but not to others with similar years of schooling who lack the certification. The schooling variable was introduced into the regressions in two ways. First was the conventional schooling variable-the years of school- ing completed by the worker. The second was a "splined" variable.3 The years of schooling were broken down into different chunks for primary, higher, and postgraduate education. In addition, separate dummy vari- ables were used for those completing primary, secondary, or higher edui- cation in order to test for the certification premium. Both specifications were used for all the years in order to trace changes. The overall rate of return estimated for the marginal year of educa- tioII was between 12 and 14 percent for 1978. It had declined from between 17 and 18 percent in 1973. This decline occurred across differ- ent specifications and can therefore be stated confidently. Moreover, Bourguignon (1983) has collected results from various studies and  106 UNDERSTANDING THE DEVELOPING METROPOLIS added his own estimates to obtain a profile of the returns to education in Bogota from the mid-1960s to mid-1970s and also showed a consistent decline. The return to higher education was found to be much greater than that to secondary education, which in itself was greater than that of pri- mary education; and the estimated returns to each level of schooling declined over time. Because about half' of total secondary and college enrollment was in private schools with significant tuition costs, however, the estimated rates of return to higher and secondar y schooling as cal- culated from earnings functions are then overstated, given the assump- tion of low tuition costs. Nonetheless, the magnitudes of these returns are probably higher than those of other investments, and continue to be so despite the declines documented. The splined specification mentioned earlier permits us to measure the certification bonus. The additional bonus for high school gradua- tion was about 20 percent and for college graduationi about 25 percent. That is, people who graduate can be expected to earn 20-25 percent more than their counterparts who attended as many years of' school but did not graduate. There was n1o measurable bonus for completing pri- mary education. These estimates imply that the economic benefits of completing education can be quite high, and people have high incen- tives to receive certification. The low bonus for primary school comple- tion suggests that the Bogota labor market has become quite sophisticated, with a greater demand for more highly skilled, better- educated labor. It is also possible that completing and receiving certifi- cation for a specific education level provides more information on1 ability and should not be interpreted merely as certification. Certifica- tion in itself implies the ability to carry out a task to completion: a trait that is valuable in most employment situations. We also tested the declining rieturns for education by making separate estimates for different age cohorts: 15-24, 25-34, 35-44, 45-54, and 55- 64. As might be expected, the mean years of schooling increase with each younger cohort-except for the youngest, many of whom are still in school. The returns to educationi were the highest for the 35-44 age group. This is consistent with our other results on high mean returns to education earlier in the 1970s. There are two processes at work. Initially, as the proportion of people with higher and secondary education increases, the variance in skill levels also increases, and those with higher education are rewarded for their scarce skills. The average return to education rises as a result of "educational deepening." As this deepening continues, the relative shares change, the return to higher education starts declininig, and the composition effect becomes pre- dominant. The age cohort restilts suggest that the 25-34 age group is  WORKERS AND THEIR EARNIN(;S 107 not getting as high returns to education as the preceding cohort (see Mohan and Sabot 1988). We now have a much better understanding of the income distribution in Bogota and Cali. A portion of the high inequality observed can be explained by the larger returns to higher education and certification, which had a greater scarcity value in previous years. Given these high returns, public policy and private initiative have responded by expand- ing the supply of education. It is also clear that people have opted to attend school longer and enter the labor force increasing ages later. We can expect these trends to contintue until the returns to education decline further and become comparable to returns to other invest- ments. How does experience affect earnings? It is argued in the literature that as people work they add to their human capital as they do with schooling but at a lower rate, because schooling is usually full time and on-the-job training (OJT) is only part time. While working. a person partly consumes, or gets returns from, his previous schooling and par- tially invests in learning by doing. Some of the segmentation literature suggests the existence of an inter-nial labor market fbr firms. It is argued that only firm-specific experience leads to improvements in a worker's performance. There is also an information asymmetry: firms may not be able to evaluate a person's previous experience as well as they can evalu- ate time spent with the tirm. If this were true, labor mobility would decrease. It would also be the case that workers who join better firms to begin with would be expected to earn more than people of otherwise equivalent training and ability. We would then expect higher returns to firm-specific experience, measured as time elapsed since leaving school. Another specific measure of experience is the number of years spent in the same occupation. It may be expected that the returns to occupation- specific experience would be higher than those to generalized work experience. The data enabled us to test for these hypotheses because the 1978 World Bank-DANE survey asked respondents how long they had worked for their current firm and how long they had worked in their curr-ent occupation. It was surprising to find that job mobility has been quite high in Bogota. Although the mean value of the generalized experience variable (EXPER)4 was about 20, the mean value of years spent with the current firm (YRSFIRM) was 5.6, and in the current occupation (YRSOC, CUP), 8.5. Workers thus appear to be quite mobile between jobs as well as occupations-a finding contrary to the general impression that urban labor markets in developing countries are rigid. The standard specification of the experience variable is to enter it as a quadratic in the earnings function along uith education.5 The quad-  108 IINDERSTANDING THE DEVELOPING METROPOLIS ratic term was expected to have a negative coefficient reflecting the decline in the rcturns to each marginal year of experience after some point. The marginal returns to an additional year of experience, on average, were 4.6, 2.8, and 1.0 percent after ten, twenty, and thirty years of work experience. The earnings for those with primary education peak early, at the age of about forty-five, for those with secondary educa- tion at abotut fifty-two, and for those with higher education at about fifty- seven. Each of the three different specitications of the experience variable was cntetred in a similar way. The results were: EXPER YRSOcCCUP . RSFIRM '21 0.063 0.040 0.044 P22 -0.0009 -0.0006 -0.0008 where NRsocCuP is the numlilber of years spent in the same occupation and YRSFIRNI is the number of ylears spent with the same firm. These coefficients indicate that, contrary to expectations, firm-specific experi- ence and occupation-specific experience are not valued more highly than generalized work experience. The marginial contribution of an additional year of YRSOCCUP and VRSFIRM was in each case only about 2.8 percent after ten years, which is much less than the 4.6 percent esti- mated for EXPE R. There is little evidence of strong firm-specific inter-nial labor markets. In developing countries this is said to exist particularly in large firms in the formal or protected sector, which reportedly pay much higher wages than) other firms. This hypothesis is not borne out for the labor market in Bogota or for our estimation for Cali. Having established the rclatively regular nature of the determination of earnings in Bogota in relation to the usual human capital variables of education and lengtlh of work experience, we now investigate further some of thc other influiences on earnings: worker backgrounds on the supply side and firms' characteristics on the demand side. How Segmented is the Bogota Labor Market? Much analysis of labor markets is motivated by a desire to understand what the determinants of existing inequalities are. In the preceding sec- tion we found that differences in education and work experience alone explain about half the variation in labor earnings. We now investigate differenit kinds of labor market imperfections that could result in differ- ent levels of earnings for otherwise similar people.6 The task for analysis then becomes measurement of "similarity" and of differences in earn- ings after this simiilarity has been accounted for. Somc of the confusion  WORKERS AND ITHEIR EARNINGS 19 in the discussion of labor market segmentation arises from the possible relationship between what might be termed the human capital market and the physical capital market. Given an unequal physical capital distri- bution, to the extent that acquisition of human capital is related to the initial income and wealth distribution, the resulting human capital dis- tribution would be correlated with the initial asset and income distribu- tion. Without deliberate policy intervention this situatioll could perpetuate itself indefinitely. The main traditional intervention to break the link between the asset and human capital distributions has been the provision of free education to all or the provision of tuition fee corices- sions linked to means tests. We have alreadv found that the massive expansion of education in Colombia has been the result of both private initiative and explicit government policv and that there is at least some evidence that this expansion inight have served to reduce the scarcity returns to higher education and thereby to reduce the inequality in earnings. This inassive expansion also means that a larger proportion of people now have easier access to higher levels of educationi. The Measurement of Labor Markel .Segnentation Considerable methodological progress has taken place in the discussion of segmentation in recent years. Early argumenits wvere based merely on the observation of heterogeneity in the labor market. Differences in means in earnings between different groups wvere seen as evidence of segmentation. Now much of the work is devoted to identifying differ- ences after accounting for similarities between human capital variables. Thus it is argued that differences arising fiom differences in human capital endowments are in some sense justified, or at least to be expected. Other differences are generally attributed to inarket imper- fections, and their removal would be argued to lead to greater efficiency. An econonmetric identification problem remains even in this approach, which is followed here. If there is a protected sector that, for whatever reasons, pays higher wages, it could make its selection criteria such that the more highly educated people would be found in these higher-paying jobs, even if tile jobs do not actually require skills resulting from higher education. An earnings functioni wouldl then attribute thesc higher earnings to hunanl capital variables rather thani to segmentation result- ing from protectioni. However, the existe1nce of such a situatioln is a little difficult to argue persuasively for private profit-seeking firms. It may be a little more defensible for public sector jobs, which often use educa- tional achievements as screening criteria. Because the marginal contri- bution of public sector jobs is mor-e difficult to measure, it is possible that the official requir-ements could be higher than necessary. Using similar arguments it is also possible to adduce sucil a situation for larger  I10 tUNDERSIAND)IN(; THEF DEVElOPING M ETR(O)POI.IS firms, which also have much greater indirect office activity in support of their production activities. In this section we investigate the issue of mar-ket imper-fections fromii both the supply and the demand sides. The main issue from the supply side is that the measurement of vears of schooling masks differences in abilitv, in family background, and in schooling quality. The other key supply-side influenice in earninlgs is hypothesized to be the existence of UnlioIns. From the demand side, the issue is usually posed as the exist- ence of a formal or protected sector in urban labor marlkets, where somc workers earni more than other workers because of various kinds of restrictive practices. Sucih practices can result from the governmenit set- ting a minimum wage, which has the effect of limitinig employinent and keeping wages in the "legal" or "formal" sector higher than the rest. Similar effects can be caused by governmenit-legislated social security paymenits, which employers mav be requir-ed toi make. It is commoni to suggest that these characteristics are highly correlated andl exist mostly in larger firms, which are ofteni owvned by foreign enterprises. The size of a firm is theni used as the indicator var-iable separating the formal and infornal sector. All of governmenit is usually assigned to the formal sec- tor. The basic idea is to identify intervening or segmeniting variables that serve to restrict mobility within the labor market and can then help to explain unequal earnings between people who would otherwise be regarded as equivalenit. In this study we attempt to identify the formal sector by variables such as firm size, the existence of formal social secu- rity systems and contracts in Firms, and the type of industry or activity a firm is cngaged in. The overall finding is that the supply-side variables have a slightly greater influence on earnings than clo the demand-side effects. The protected sector is simply diffictilt to find in BogotA. It is possible, how- ever, that these results are partly due to the boom conditions that pre- vailed in Colombia during the period of this study. The Influence of Warkers' Background.s on Farnings There are two ways to measurie the effects of background. Wc can either hypothesize that people from more advantaged backgrounids have a higher return to schooling, or that the effect of their background is manifested as a premiumn on earnings. In the first case the sample would needl to be stratified by the backgrounid variable, and we wouildl test for differences in the returins to schooling-that is, differences in the edu- cation coefficient. In the second case a dumimiv variable for backgrcound would be used in the estimation test for the existence of a prermium. WeU have utilized both methods, even thougih the first method causes some econometric problems. If the dependent variable-that is. carnings-is highly corirelated with the stratifying variable, the restiltinig coefficients  WORKERS AND THEIR EARNINGS HII would be highly biased. In this case, if people of a specific background are particularly poor, the estimated returns to their education would be biased and on the low side. This procedure must be used judiciotisly, and conclusions drawn carefully. The region-of-origin variable used was the place of birth, distin- guished by size of settlement: rural, small town (less than 100,000 popu- lation), large town (between 100,000 and I million population), city (more than I million population) but excluding Bogoti), and Bogotd. Because our interest is in environmental effects on quality of schooling, Bogota is also regarded as the place of origin for people who migrated to Bogota before the age of ten and presumably received the bulk of their schooling there. In addition to the possible difference in quality of education, it is reasonable to hypothesize that people who grow tip in metropolitan areas are exposed to more varied influences and a wider variety of information, with the result that their schooling may be formed into human capital stock more efficiently. The schools in larger cities are also likely to have better teachers. It is also often suggested that migrants are at a disadvantage in competing with natives in the urban labor market. These background variables did not affect the educationi coefficients, however. There seems to be little correlation between place of origin and returins to schooling. The dummy variables for region of origin, on the other hand, were all substanitially significanit, the base comparison being with workers fi-om rural backgrounds: I'ercei tage Plae oJ onrgin in Bogai C( efficien t Bogot,t 46 0.10 to 0.1I Large citv 3 0.15 to 0.18 L.arge town 6 0.19 to, 0.23 Small town 26 0.08 to 0.Io Rural (base) 19 0 Both the Bogota natives and migrants from other urbani areas appear to be 10 to 20 percent better off than rural migrants, but there is little to differentiate Bogota natives from migrants from other urbanl areas. If anything. migrants from other large towns are better off than native- born Bogotanos-and this may be a r esult of self-selection. The estimate of returns to education from the stratified samples of workers produced similar results. There was little difference in the returns to education for Bogota natives, workers in big cities, and other urban migrants, but the rural migrants did get significantly lower returns. Given the consistency between the two methods, it is reasonable to conclude that, on average, there is no evidence of differences in the quality of education between small and large towns and Bogoti, but the rural folk probably do suffer from poor schooling and other negative environimenital influences.  112 UNDERSTANDING THE DEVELOPING METROPOLIS Using the location of residence within Bogota as a variable in earn- ings functions estimations is one of the mor-e controversial aspects of this study. Is it reasonable to hypothesize that within a city the location of residence "causes" a different rate of return to education, or that it influences earnings in other ways? Is the labor market in fact segmented in this manner? The argument is that current location is acting as a proxy for unmeastired variables. One can expect the quality of school- ing to differ by location within the city. To measure only the years of schooling is to ignore differences in quality; however, the location of res- idence could be a proxy for schooling quality. Parents' education and income, which largely determine where the family lives, then affect the children's schooling by virtue of location. Residence in a low-income neighborhood can also affect a person's network of contacts for obtain- ing a high-income job. Because the lower-income neighborhoods show rather flat age-earnings profiles, one can also expect that the demon- stration effect dampens residents' expectations and aspirations.7 This can have a cumulative effect on earnings, similar to the discouraged- worker hypothesis for nonparticipators. Finally, location of residence is also a proxy for social class, which probably has significant effects on earnings. The larger question is whether this pattern is peculiar to Bogota or whether it exists in other cities as well. Clearly, all cities have their poor and rich neighborhloods; the question is whether they are clustered in the same wav as they are in Bogota. We also need to ask how much spatial mobility there is across differenet parts of the cities. To what extent does current location affect the current earnings and streams of futtire earnings? Aill large cities in the world have rich and poor neighborhoods. Peo- ple choose their place of residence according to their income, the loca- tion of their work, and their preferences regarding the amenities and other characteristics of the neighborhood. Normally, income is the main factor in the choice of residential location. People also sort themselves out by ethnic origin and, to some extent, by occupation and class. The issue here is a feedback mechanism that reverses the causa- tion somewhat so that a person's location affects his or her earning potential. We have documented that both Bogota and Cali are characterized by a high degree of spatial inequality in incomes. Recalling the pattern in Bogota, the north (sector 8) is particularly rich, and the south (sector 2) is particularly poor. The other sectors are more heterogeneous, but the ranking of the radial sectors in ascending order of household income per capita is 2, 3, 6, 4, 5, 7, and 8. Although there are virtually no rich people in the south, there are poor people in almost every part of the city. Historically, it seems that the rich have continuously moved north while the poor have located in the south and later in the west. This  WORKERS AND THEIR EARNINGS I 13 makes it even more likely that the better schools would largely be in the nor-th. When large areas of a city are known as rich and poor, as in Bogota, people's addresses can become screening devices. Employers, for example, may use an applicant's address to gauge his or her likely characteristics-reliability, home background, and so forth-in much the same way that names of schools are used as screening devices. In summary, if the residential location variables are found to hiave a significant effect on earnings, there are at least three explanationis. First, in the human capital tradition, it may be argued that these location vari- ables act as proxies for ability, schooling quality, and the like, which are unmeasured otherwise. Second, also in the humani capital tradition, it may be argued that they act as proxies for other productivity characteris- tics of workers (class, statLis, aspiration, attitudes, and contacts) that are correlated with their residence location. The third possibility is that location is being tised as a partial screening device in a city character- ized by notable spatial differences in income. There are essentially two location variables that have been used. First are the dummies for each residential sector, usinig sector 2, the poorest, as the comparator base. Second is the distance of residence from the center of the city. Dummy coefficients for sectors 3 through 7 are all between about 0.15 to 0.25 and are not significantly different from 0.2. Sector 8 is obviously different, with coefficients of about 0.6. These coef- ficienits measure the systematic deviation in log income from sector 2 means, keeping everything else constanit, including distance from the city center. Thus, workers in sectors 3 through 7 receive about 20 per- cent more in earnings than otherwise equivalent workers in sector 2, whereas workers in sector 8 receive about 60 percent more. The addi- tion of these variables does not appreciably increase the level of expla- nationi (R2) that is the proportion of log variance of earnings that is accounited for. But all these coefficients are lhighly significant. We have also discussed at some length the location of many of the poor at the periphery. The estimated coefficient of the distance variable implies that workers' earnings decline by about 2.5 percent per kilometer fromli the citv center, on average. Furthermore, the addition of the distance variable makes the differences in earnings between radial sectors more pronounced. As before, we also estimated the returns to education by stratifying the sample by residential sector. Although this procedure stiffers from truncationi bias, particularly for the largely poor sector 2. the estimated returns are fouLid to vary systematically by sector; the coefficients for schooling are 0.08, 0.09, 0.10, 0.1 1, 0.11 . 0.14. and 0.16 for sectors 2, 3, 4, 5, 6, 7, and 8, r espectively. These differences are not statistically signif- icant between sectors 2 and 3 and between sectors 4, 5, and 6, but the overall pattern is consisterit with dummy variable coefficients.  1 14 UNDERSTANDING THE l)EVELOPING MIETROPOLIS These results were further investigated in different ways. First, do the low-income workers who live in richer areas earn more than similar workers in the poorer areas? The indication is that they do not. It seems that the differences are mainly among the better-educated workers. A worker from the poorer areas who manages to get secondary and higher education does earn less thani someone from the richer areas with an equivalent education. Schooling quality differences presulmably explail this finding. For the relatively unskilled, less-educated workers, school- ing quality makes little differences. Second, how mobile are people? I)o people who move "down" from richer to poorer locations carry their work characteristics with them, and vice versa? The data contained information about the last move of the household. We classified as upwardly mobile movers those who had moved from a poorer sector to a richer sector in the last ten years; those who had moved from a richer to a poorer sector were classified as downwardly mobile. For the 1978 data, the former comprised 40 percent of the workers and the latter about 15 percent. Most of the moves were between adjacent sectors.8 Bogota households were therefore quite mobile, but the intersectoral mobility is between relatively similar sectors. The addition of these diummies did not alter other results but did increase the level of explanation of earn- ings slightly. The downwardly mobile movers appear to earn about 10 percent more than similar people elsewhere. This suggests that school- ing quality does matter and that people fi-om higher-income locations who move to lower-income ones retain some of the higher-income char- acteristics. The upwardiv mobile people seem to be no different than other similar workers. These results argue against the labeling or screen- ing hypothesis and support the hypothesis that location acts as a proxy for schooling and other aspects of background. The last supply-side segmenting variable is that of union membership. Only about one-quarter of the male workers in both Bogota and Cali belonged to unions. Wheni union membership was introduced as another- dummy variable in the earnings functionis, the average earnings premium for union members was about 6 percent in Bogota and 12 per- cent in Cali. It is striking that this appears to be more important for migr-anits from riural areas and small towns than for others. Given the earlier finding that it was only these migrants who were disadvantaged, and not other migrants or natives, this finding suggests that union mem- bershiip is more important and serves as a screening device when there are few other distinctions of labor skill or quality. Results from both the place-of-origin variable and the location-of- residence investigations support the idea that these variables are essen- tially proxies for backgrotind, particularly schooling quality and ability as iifliteniced by background. Although migrants are no worse off than natives in general, rtiral migranits do appear to be somewhat worse oft:  WORKERS ANI) THEIR EARNINGS 115 The results imply that there might be income feedback effects from pro- longed residence in disadvantaged neighborhoods. One way of compen- sating for some of these effects is to belong to a union. The importance of the background effects should not, however, be exaggerated; they add only about 3 to 6 percent to the level of' explanation in the log vari- ance of earnings, even though each of the coefficients estimated is highly significant statistically. There seems to be little evidence of the spatial disadvantage hypothe- sis usually expounded for cities in the United States. High inner-city unemployment rates are often explained by the flight ofjobs to the sub- urbs and the resulting deterioration in the access of poor workers, Utis- ally blacks, to these jobs. In Bogota, even though the poorer parts of the city have far fewer jobs than richer areas do in relation to their popula- tion, there is no measurable difference between participation rates and unemployment rates. The differences are essentially those in e arnings. Access to jobs does not seem to be a problem for the urban poor in Colombia. This may be partly because of Bogota's relatively cheap pub- lic transportation system and flat bus fares (see chapter 8). The Protected Sector On the demand side, we have attempted to identify the protected sector by a number of different variables. First, it was hypothesized that firms with formal social security schemes and/or formal employment con- tracts must belong to the formal sector. Second, consistent with much other work on the subject, it is expected that large firms pay their work- ers more than other smaller firms do. Third, different types of work sta- tus (for example, employees versus self-employed workers) may' give different returns. Fourth, we test to determine if different kinds of industry give varying returns. Last, we see if there is any measurable dis- tinction between the government and the private sector, government being an easily identifiable protected sector and often alleged to pay much more for comparable jobs than the private sector. It is difficult to find labor market segmentation in Bogota according to any of these cri- teria. There are some measurable differences, but protected-sector pre- miums or returns to education, measured as earnings received, are seldom higher than 10 to 15 percent. We first introduce dummy variables for the existence of social security schemes or employment contracts in the workers' place of'work. About half the workers, both male and female, work for employers who have these benefits. The male workers who have contracts earn about 5 per- cent more than other equivalent workers. Similarly, workers in firms with social security schemes earn about 7 to 9 percent more than other equivaleint workers. The correlation coefficient between these two vari-  1 16 UNDERSTANDING T'HE DEVELOPING METROPOLIS ables is 0.5. Next, a variable measuring the logarithm of firm size is introduced.9 Its coefficient implies that a doubling of firm size increases earnings by abouLt 2.5 percent. This means that an employee of a firm with 100 employees would earn about 10 percent more than a similar employee of a firm with fewer than 5 employees. These results mean that workers in large firms (with, say, more than 100 employees) that also have other formal sector characteristics would earn about 10 to 20 percent more than similar employees in very small firms (with I to 5 employees). Although these differences are significant, they are not large and can scarcely be said to imply the existence of a strong pro- tected sector in Bogota. The existence of relatively high labor mobility is consistenit with this finding, as are the relatively low returns to firm spe- cific experience. Similar workers in different kinds of firms, formal and informal, receive roughly similar earnings. We next examine other characteristics of work that could possibly seg- ment the labor market in an identifiable manner. First, earnings func- tions were estimated for different types of workers: blue-collar and white-collar employees, and employers and the self-employed. Because the earnings and education of blue-collar workers are uniformly low, the estimationi for these workers suffers from truncation bias, and returns to education for them are found to be low. But there is no statis- tically significant difference between the returns to education for white- collar employees and those for employers and the self-employed. Sec- ond, earning functions were again estimated, stratified by industry of activity classified by the standard industrial classification at the one-digit level. Again, no statistically significant differences were found in the returnis to education between manufacturing, construction, trade and commerce, public administration and other services, and financial establishments. Workers in the tranispor-tation and communiication sec- tors, however, seem to get somewhat lower returns; in this case it does not appear to be a truncationi problem. Finally, different estimates were also made for the government and private sectors. The mean earnings for workers in government were significantlv higher than for workers in the private sector. The rate of return to edu- cation, however, was not significantly different. The main difference is that union members in government get about 20 percent higher earn- ings (compared to about 6 percent higher in the private sector). A nota- ble feature of the earnings function estimation for government workers was that R2, at 0.68, was the highest earnings function estimated in this study. The human capital variables-in particular, years of education- explained more than two-thirds of the variance in earnings in govern- ment jobs. This reflects the higher degree of codification in earnings structure in the government: earning levels are more clearly related to  WORKERS AND THEIR EARNINGS 117 formal qutalifications. The returns to work experience were lower in the government than in the private sector. This probably reflects a tendency in the government to do more formal screening at the time of entry and then to provide slow growth in earnings with experience. Government employees accounted for about 15 percent of the total in Bogota. Given these results, it is difficult to argue that the Bogota labor mar- ket is highly segmented. Most of the variation in labor earnings is explained by the traditional htlinan capital variables. Although differ- ences in background add little to the overall level of explanation, they make a substantial difference in the prediction of earninigs for workers with specific backgrounds. Although these differences have been mea- sured as the place of origin or current residence location, they are attributed to schooling quality and other background influences. On the demand side, a strong protected sector is simply difficult to identify. althouglh workers in formal sector firms do earn about 10 to 20 percent rnore than other equivalent workers. The only sector that seems to be particularly protected is the government sector, witli preiniums of about 15 to 20 percent being paid to workers thanks to the efforts of strong government unions. Outside the government, union membership makes little difference except among relatively unskilled r ural migrants. Operation of the Urban Labor Market: What Have We Learned? The results reported in this chapter are in marked contrast to most other studies of urban labor markets and to popular impressions of the employment situation in cities in developing countries. The improve- ment in the urbani employment situation in Colombia in the 1970s was partly related to the existence of boom conditions there in the latter part of the decade. The rapid growth in GDP dturing this period resulted from policies that favored agriculttiral growth and expansion of export- related manufacturing activities. Governmental programs for infrastruc- ture investment also helped provide high growth in tertiary-sector employment in urban areas. The boom in coffee prices during the late 1970s contributed to prosperity in the agricultural sector and to unprec- edented tightening of the labor market in rural areas, which was reflected in higher wages for tunskilled workers in urban areas. The late 1960s and the 1970s were also periods of significant demo- graphic changes: fertility declined remarkably, resulting in slowdowns in urban population growth, overall populationi growth, and rural-urban migration. In Colombia this also coincided with the turning point in the pattern of urbanizationi; the effervescent urbanization between 1940 and 1980, when the urban population rose from 30 to 65 percent of the  118 UNDERSTANDING THE DEVELOPING METROPOLIS total population, gave wav to a rather slower rural-urban transforma- tion. This is a familiar S-shaped urbanization pattern. The other reason our findings in this chapter differ from those in other studies is that most other studies, particularly those of the infor- mal sector, suffer from the use of biased samples. The informal sector is often identified by analyzing the characteristics of workers from slums, from specific occupations or types of firms, and so on. Appropriate coin- parisons are seldom made. Doing so means keeping other things con- stant, and this is difficult without using a careftilly drawn, citywide or all- urbani sample. The labor market was found to operate quite efficiently. There is scant evidence of segmentation; similarly qualified people earn similar incomes in different activities; there is high labor mobility'; and return1s to humani capital are similar to those in other countries. Part of the ten- dencv toward better income distribution is attributed to the "deepen- ing" of education, with consistent and rapid improvements in average educationi attainment (see Mohan and Sabot 1988). There is evidence of declines in the returns to secondary and higlher education as the stock of people in these categories has increased. Concurrently, the real wages of the unskilled have improved-a consequence of tightening labor market conditions in both rural and urbani areas and improved overall labor quality with the universalizationi of primary education. These forces have resulted in a decline in the ratio of earnings between the mor-e-educated and the less-educated. These results suggest very strongly that the continued expansion of education, along with a reduc- tion in the dispersion of completed education, will be quite effective in reducing the levels of inequality, at least in personal labor earnings over the long run. The evidence from Bogota and Cali indicates that the labor market is not characterized by a strong protected sector. Individuals working in es- tablishments with formalized employment arrangements (with unions, social security schemes, and written contracts) and those in large enter- prises earn only slightly' more than others. Moreover, the stability of the estimated rates of returni to schooling and experience across the differ- ent occupational and industrial categories argue against the existence of a highly segmented labor market in Bogota and C.ali. Much of the writing on1 the existence of informal and protected sectors in cities in developing countries is connected with high rural-urban migrationi. Contrary to popular nmisconceptionis, migrants are not especially poor; thev do not concentrate in specific areas of the city-the center or the periphery; they are not concenitrated in particular occupations or activi- ties; and thev are not less educated or less skilled than natives, on aver- age. In short, they are not found to be disadvantaged in any respect.  WORKERS AND) THEIR EARNINGS 119 Women earn less than men in every occupation and industry. The ratio of average female earnings to male earnings is between one-half and twvo-thirds in most dfeveloping countries. In Bogota and Cali it is nearer one-half. WAhat evidence thele is suggests that women get rates of r eturni for cducation similar to those of mcn but do not get comparable retur ns for experience. Thus, women's earnings differentials increase with age. Womeni with higher edtucation are far more likelv to work than others. The fact that womens access to education is now similar to that of men suggests that women in developing coun1tries are likelv to enter the labor force in unprecedented numbers, mtch as they did in devel- oped countries buit perhaps even faster. Notes 1. We can express the basic humani capital model as follows: (5-1 ) Y = Ph- H.e" (5-2) 1I= e-e" (5-3) y = lnY = lnPh + S+ u +-1 where Y is labor earninigs, 11 is the unobserved quanititv of humall capital, Ph is the market rental price of a unit of humani capital (which ma, vary over time and space), and u represenlts other influenices on wages. Equation 5.2 may be interprete(l as a production functioni for htuman capital, uSilig schooling (S) as onle input and usinig abilitv, efficiency, and so forth (denoted bv v) as other inputs. Ph5f then gives the flow of returns from the stock of humani capital /J, with u subsuminig other influences on earnings. 2. The equation estimated was ln(Y) = P,, + XI PI + X2 ,f2 + X< 033 + X4P4 + X5 + E, where XI = education variables, X, = experience variables, .Y3 = place-of- origin shift variables, X4 = location-of-residenct shift variables, and X5 = charac- teristics of employment. See Mohiani (1986, chap. 8) for a more detailed specifi- cation, the actual variables used, and other details. 3. Smith and iWelch (1977) also usecl a splined educationi variable to obtain estimates of the differential returins to different levels of education. In our spec- ification, the splined variable essentially decomposes the educationi Variable into stepwise dummyv (the certification premium) andl slope variables (vwar-s of cdui- cation at each level). 4. EXPER = (age - years of education - 6) years. 5. That is, IlnY= ,5,, + -1 RSEDtI + 021 EXPER + 022 (FXPFR)2 + X 03 + X4 4 + X5 05 + E (earnings) (schooling) where X3 = place-of-origini variables; X4 = location-of-residenice variables, X5 = employmenit chiaracter-istic variables, and E is the error term.  120 UNDERSTANDING THE DEVELOPING METROPOLIS 6. This section summarizes work found in Fields (1980) and in Mohan (1980, 1981, 1984, and 1986, chap. 9). 7. See McGregor (1977) for similar arguments in relation to Glasgow. 8. See flamer (1981) for more details on residential mobility. 9. The firm size itself was also tried, but the coefficient was negligible and i nsignifican t.  Chapter 6 Firms and Their Location Behavior The preceding chapter examined the supply side of the labor market. This chapter looks at the demand side of urban employment-more specifically, at the location patterns of firms and the behavior underly- ing these patterns.' Household behavior is, in general, better understood than the behavior of firms. This is partly because the characteristics of households are more homogenous than the characteristics of firms. Firms in retail or wholesale trade are quite different, for example, from manufacturinig firms, which themselves can be different in each indus- try. The location imperatives for banks are quite distinct from those of government departnments. Moreover, bank branches are quite different in behavior from their head offices. The study of employment location therefore has to account for a number of differences between firms and in different dimensions-in terms of their function, size, age, and so forth. Understanding workplace location is crucial to understanding residence location and emerging urban travel patterns. Although we established the general location patterns of the different economic activities that existed in Bogota and Cali, we paid special attention to manufacturing activities because they are usually regarded as leading other activities in a city. Their extensive backward and for- ward linkages often determine the location of other activities. In order to focus on behavior, we intensively studied firms in a couple of selected industries. Before we discuss our findings, however, let us consider the composition of employment as a whole, the spatial distribution of differ- ent types of employment, and the changes that have recently taken place in Bogota and Cali. Decentralization in the two cities has been so thorough that in the late 1970s, the central business districts of Bogota and Cali provided only about 15 percent and 18 percent of employment, respectively, with 121  122 UNDERSTANDING THE DEVELOPING METROPOLIS net new employment almost entirely outside the center. Employment location trends are vital components of' the urban dynamic. For one thinlg, place of residence is closely related to place of work: families look for housinig in areas not too far from their jobs. For another, the different economic activities liniked in a production process often need to be located near one another to reduce costs and facilitate tasks to be perf'ormed. If one of the segments (that is, firms) forming part of such a technological chain moves to another location, the other segments are likelv to move to the same area, thus instituting a substantial change in the spatial distributioni of employment and probably also in the populationi, housing, physical infrastructure, trip patterns, and demand for transportation-in other words, altering the spatial struc- ttie of the city. In charting the changes that have taken place in Bogota and Cali in employment location, we resorted to a large array of data sets (see the appendix). The broad patterns of employment location were estab- lished from the World Bank-DANE Household Survey of 1978 in which each worker was asked to provide the characteristics of the firm where he or she worked, including its location within the city. Trend changes were established by comparing the 1978 pattern with that in 1972, when another lhousehold survey-the Phase 11 Survey-was conducted in BogotA. We also had access to the 1976 Social Security establishment files for l3ogota and the 1978 files for Cali. This source naturally pro- vides poor coverage of small firms but is useftul for checking the quality of iifor-mation in the 1978 Household Survey. For a detailed analysis of the manuftacttuinig sector we resorted to two additional sets of data. I)ANE publishes an annual directory of industrial firms that employ ten or more employees. The directorsv incltzdes informationi about individ- ual establishments, including locationi, production, sales, and inputs ised in manufacturing. Based on these files for each year from 1970 to 1975, we were able to trace the movement of each firm over this period. A special survey of a sample of nmanufacturing firms was conducted for the City Study in 1978 to obtain further detail. It is unusual to have such rich data in any city in developed or developing countries. We hope, therefore, that this analysis of employment location behavior has appli- cability beyond Bogota and Cali. Among the questions this chapter raises and attempts to answer are these: Whlat kinds of firms move? Which remain stationary? Where do new firms locate? Why do the firms that move do so? How do new firms decide where to locate? Are there differences in the behavior of small and large firms? What can we expect in the future? What is the magni- tiude of capital land substitution? Ilow does this affect location in the presence of'lan(d price gradients?  FIRMS AND THEIR l.OCATION BEHAVIOR 123 Trends in the Location of Employment In both Bogota and Cali, employment in all sectors (manulactuijig, commerce, and services) has historically been concentrated in and aroutid the CBD, but this situationi has beeni chanlginig rapidly since the 1960s as the two cities grew. The most significant anid best-documented locational shift may be seen in the case of manufactur-ing firms that have been movinig out of the CB[) to peripheral locations, thus reducinlg the absolute number of manutactur-inig jobs in the (BD. This is consistent with the worldwi(de trend observed in all large cities in both developed and developing counitries for which data about employment location are available. In Bogota and Cali this decentralization has been occur- ring in conjunictioni with rapid growth in mantu,lfctUIring employment: annual average growth rates were almost 8 percent between 1973 and 1978. Employmenit in the commerce and service sectors also decentral- ized but more slowly. The larger retail and wholesale trade firms contin- ued to locate in or near the CBI) while ther-e was somiec decentralization of the smaller firmns. This was also true of the service sector, although the pattern was not as clear as for trade and commercial firms. Conse- quently, the CBD became more specialized in its commerce and service f'lnctionis, wlhile the periphery is specializinhg in manufacturinig. Overall CBI) emplovnielt has remained constant in absolute termns, but because of total employmnent growth, its share has been falling. The employmenit structure of Bogota and C.ali is not very different from that of other large citics in the world (see Mohan 1986, chapter 3). The share of manufacturinig employment in Bogota is 24 percent, that of commerce 20 percent, and that of services (includinig finance) 41 percent. The last is a litde higher than in many cities because of the presence of the nationial government in Bogota, but it is similar to the share in othe r large capital cities in the world. Cali is a little different in that the share of manufacturing is about 30 percent and that of services about 33 percent. Almost all large cities in the developing world have more than 20 percent employmenit in manufacturing. Noncapital cities suLch as Calcutta, Bombay, and Sao Paulo have a higher shaire of manu- facturinig employment, as might be expected. Similarly, in a typical large North American city the structure is about 30 percent manufactrz-itig, 30 percent commerce, and 25 percent services; utilities, co1istruction, transport, and communication account for the rest. The structule of employrnent in Bogota has remained relatively stable since at least 1950, and it may be regar-ded as a well-balanced metropolitan city. No particu- lar sector is overrepresented in Cali either. About half of all em)ploy- ment in both the cities was in small firms, that is, firms employing fewer than ten employees (see table 6-1). The commerce and service sectors  124 UNDERSTANDING THE: DFVL.OPING CMFTROPOLIS Table 6-1. Employment by Firm Size and Major Industry Group, 1978 (percent) A.1 Mlanular- industrips' luring Commerre J,nanice Servires Bogota Smallb 52.0 42.0 71.6 43.2 59.8 Iarge' 48.0 58.0 28.4 56.8 40.2 Total 100.0 100.0 1)0.0 100.0 10(.0 Thousands of people employed 1,212 286 246 98 394 Percentage share of total emplovmenit 100.0 23.6 20.3 8.1 35.0 Cali Smallb 55.2 36.3 77.0 55.6 72.9 Largec 44.8 63.7 23.0 44.4 27.1 Total 100.0 10(.( 100.0 100.0 100.0 Thousands of people employed 368 115 80 13 107 Percentage share of total employment 100.( 31.2 21.5 3.7 29.0 a. InCludes other sectors in adlition to manufacturing, colmmerce, finiance, and scr- 'ices. h. Firms with fewer than ten employees. c. Firms withi ten or more emplovees. Source: K. S. Lee (1989. table 2-3), based On World Bank-i)ANF Househioll Survey 1978. were characterized by the presence of many small firms, whereas 60 to 65 percent of total manufacturinig employmiient was in large firms. How was this employment distributed across the two cities? WAe employ the same spatial scheme used earlier-that is, rings and sectors (see table 6-2). As might be expected, Cali, the smaller city, is more cen- tralized overall than Bogota. The large firms in the cominerce and finanec sectors locate in the center, with rings I and 2 accounting ior half or more of total employment in these groups. Manufacturing is the most decentralized, and the small and large firms seem similar in their location behavior. Small retail (commerce) firms are almost as decen- tralized as firms in the maniufactiuring sector. The distribtition of employment by radial sectors (see table 63) reflects the pattern of land use specialization referred to earlier. In Bogota, manufacturing employmtnt is concentrated in the "induistrial corridor"-sectors 4 and 5. Sectors 3 (the Bosa area in the southwest) and 6 (near the airport) have recently begtun to attract manufactul-inlg activities at their peripheries. jobs in the comrnerce, finance, and ser- vice sectors are located primarily in the r ich sectors 7 and 38 in the north, as well as in the CBD.2 T his reflects the northwar( inovement of high-  FIRMS AND THEIR LOC(ATION BEHAVIOR 125 Table 6-2. Employment Distribution by Ring, Firm Size, and Major Industry Group, 1978 (percent) Alan tfactunng Commerce Ftnance Servires Small Large Small Large Small Large Small Large Ring firmn.s firms firms firtns firmis firms firms firms Bogota 1 4.66 7.00 13.13 22.57 52.14 33.99 8.77 18.60 2 13.42 13.17 17.69 24.84 14.52 39.77 16.02 22.52 3 16.16 24.60 13.18 17.99 11.23 11.29 16.00 18.19 4 26.70 23.58 19.50 18.68 11.58 11.03 24.75 20.11 5 34.37 24.88 33.74 14.00 8.52 3.17 27.93 13.71 6 4.21 1.09 1.95 0.21 1.51 0.00 5.50 3.18 N.ie. 0.49 5.68 0.82 1.70 0.49 0.76 1.02 3.69 Total 10(.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 tali 1 7.02 12.61 17.02 52.76 57.20 76.04 9.53 23.51 2 28.74 22.81 23.29 23.42 28.51 13.51 35.38 29.87 3 41.14 39.06 36.64 15.03 6.25 0.00 37.76 15.26 4 18.24 9.31 14.42 4.77 3.12 0.00 12.89 22.61 5 2.58 0.63 6.06 0.00 4.91 10.45 2.59 1.53 N.i.e. 2.27 15.57 2.57 4.03 0.00 0.00 1.85 7.22 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.(00 N.i.e. Not included elsewhere. Note: Small firms ale those *ith fewer- than ten emplovees; large fimns. thiose with ten or more emplovees. ,Source: K. S. Lee (1989, tables 2-5 and 2-6). based oni World Bank-n.FNE Household Sur- vey 1978. income residents over the past several decades and the subsequent growth of financial and services activities in that area. This pattern of employment location corresponds remarkablv closely to the residence patterns of workers shown in the last chapter. Production workers were found to be concentrate(d in sectors 2 and 6, and professionals were found in sectors 7 and 8. Firms have followed the rich to some extent, and the poor have adjusted their residence locations to the resulting pattern of employment location. An examination of Cali shows the existence of an "industrial corridor" in sector 3 and the tendency of ser- vice employment to follow the rich households south. In both Bogota andi Cali there was little manufacturing employment in the rich residen- tial sectors. The rich succeeded in keeping their environimenit relatively pollution-free. Again, it has been useful to study the spatial distribution of employ- ment bv both concentric rings and radial sectors. An examination by rinlgs reveals the decentralized nature of employment and the concen- tration of certain activities in the C:BD. An examination of radial sectors  126 UNDERSTAN DING THE DEVELOPING METROPOLIS Table 6-3. Employment Distribution in Bogota by Radial Sector, Firm Size, and Major Industry Group, 1978 (pelrcent) AUa ufactaring (CSmmere,' tinan ce Ser vire. Sm(ll Large .Small Large, Small L.argr Smaill Large J?adial secto fi rmns firms ft rrns fiZoZn firrns firms fitms firno I 4.66 7.00 13.13 22.57 52.14 33.99 8.77 18.6() 2 14.36 4.65 14.91 1.16 2.10 1.57 10.85 4.84 3 24.97 14.89 18. 16 8.26 6.52 0.82 12.90 7.98 4 8.75 20.33 12.54 7.46 2.81 2.98 7.60 4.21 5 8.10 27.28 7.36 17.09 4.45 13.59 7.52 12.12 6 18.75 7.77 11.74 4.75 5.07 2.27 11.77 16.87 7 11.50 5.81 8.09 13.53 8.44 11.73 13.92 9.23 8 8.4'3 6.60 13.26 23.48 17.98 32.29 25.65 22.44 N.i.e. 0.49 5.68 0.82 1.70 0.49 0.76 1.02 3.69 Total I00.O 1(0.0( 100.(0 10(.0( 1((1.0( 1 (.0() 10(0.0( 10.0( N.i.e. Not iocluded elsewfiere. NAle: Si nall fi1 Ins are those mwtlh fevwer than ten emiaplor>ees; large fir ins, those w%ih te,n o- mol-e em ployees. Soar': K. S. Lee ( 1989, table 2-8), based or)n World Btnk-n)..NFr Hotusehol]d Surrey 1978. shows the previously noted patterin of land use specialization and the relationishiip of employment andc residential location a little more cl early. WVe now use the 1972 Phase 1I Survey and the 1978 W'orld Bank-DANE Hiouselholcd Sur vey to observe Bogota's clhatiging structure of employ- melnt locationi (see table 6-4). The two sur-vevs were not exacllv compa- rable, but thev give an idea of the broad decentralization trend in employment. Specifically, because of the nature of the questioni asked in that survey, employmenit in the CBD is exaggerated for 1972. The main difference from 1972 to 1978 is the increased importance of ring 5 in its share of employment. Althotigh detinitiornal patterns alluded to previ- oisly do not allow strict comparisonis, it is likely that r i1g 2 also increased its share in employment over this period. We miglht say that, in effect, the (.Br) has expanded into ring 2 and that we should really think of both rings I and 2 as the CBD now. The constrtmctioni of the new city center, the Centro Internacionial in the northern part of ring 2 has contributed to this expansion. In fact, a detailed analysis of the C(BD and the CentrLo Internacionial reveals that the sumn of employment in the two locations remained rotgilly constanit over the period. We conclutde that overall the center of Bogota can be regarded as stable in absolute mag- nittide of employmenit but that its share in the total has decreasedl. It is clear, however, that manutfacturing employment has decreased in abso- lute terms as well. The 1.BD is now more and more a specialized trade and finanicial center. It is difficult to track trends in Cali because earlier  Table 6-4. Employment Location in Bogoti, 1972 and 1978 (peircellt) .All dudtstn7es Aa Manuactl 1ig Corn ,Merce Ft nan cv .Servzcf Ririg 1972 1978 1972 1978 1972 1978 1972 1978 1972 1978 1 23.03 13.95 18.20 6.01 19.43 1 5.75 42.11 41.43 22.62 12.91 2 I3.61 17.74 16.07 13.47 12.18 19.77 13.69 29.38 12.74 18.68 3 14.62 16.40 18.94 21.54 13.35 14.83 6.89 11.13 15.88 16.87 4 18.80 20.60 20.27 24.89 21.83 19.37 10.00 10(.88 20.74 23.18 5 18.61 24.94 21.76 28.25 21.52 27.72 14.64 5.65 17.89 21.79 6 1.67 3.43 1.04 2.19 3.42 1.51 (1.55 0.60 1.88 4.41 N.i.e. 9.67 2.96 3.72 3.66 8.27 1.05 12.12 0.63 8.25 2.15 Total 100.00 100.00 100.0 100.00 100. 1.0 1(0.00 100.00 10(.(0 100.00 Ni.e. Not in-cluided elsewhere. ai. Iicilu(les other sectors in addition to inaiinfi actirinig. conienic ce. finance, and ser-xices. Soure: KI S. Lee (1989. table 2-10). based oni World Banik-m NE I lonsehold Stit Nec 1978 and Phase 11 Survev 1972.  I28 t'NDERSTANDING THE DEVEL.OPING METROPOLIS data are simply not available. A comparison of social security files between 1976 and 1978 suggests that the (CBD there has also lost manu- facturing jobs but has attracted more trade and finance jobs. The 1978 survey allowed a rough tracking of firms in both Bogota and Cali over the preceding five years. This was very uisefil, because informationi on movement wvas otherwsise available only for mantifacttir- ing firms. The information is, however, inexact because it was derived fromii the workers' own knowledge about the current and previouis loca- tion of their employers. The decentralizationi pattern was clear in both cities. Existing ftiriis were founld generally to move outward, whereas new firms also generally located in outer rings. In Bogota there was a net movemenit of firms out of the CBD. Net losses also occurred in ring 3. The outer rings 4 and 5 clearly gained in all activities. It was interesting to notc that ring 2 gaiiecI in coinmerce, finance, and service jobs becatise of the expansion of the Centro Interniacional adjacent to the old (:Bt). The city center had effectively expanded and moved slightly northward. Outer ring 5 received about one-third of all new jobs. In Cali, too, there was a net loss of jobs from the city center, but all the other rings experienced net increases in employment. Because it is a smaller city, each ring outside the (CBD was experiencing net employ- melit growth. The decentralization trend was strongest in manufactur- ing jobs in both cities. It should be noted that the decentralization trend is composed of both existing firms moving out toward the periphery aid new firms locating towarcl the periphery. Nonmanufacturinig firms replaced some of the maniufactuiriing employmenit that Inoved out. Patterns of Employment Location in Manufacturing Given the availability of the time series data in the DANE industrial direc- tory files from 1970 to 1975, the changing location of manufacttiring jobs could be charted over this period in detail. The methods employed were similar to the empirical studies of employment location in the United Kingdoin and the Unitedl States, where a typical approach was both to investigate not only the growth aind decline of stationary firms but also the location choices of newly established firms and relocating firms and to analvze location patterns.3 It is perhaps tiseftil to compare the manufacturing firmns in Bogota and Cali with those in large l.S. cities. The employment dvnamics in the Colombian cities are quite different. Reflecting the vibrant growth of manufacturing employment documented earlier, the birth rate of firms in Bogota and Cali was between 7 andl 9 percent anntially from 1970 to 1975. The rate for typical American cities was more like 3 to 4 percent. The death rates of firmns, however, were similar: betveen 3 ancl  FIRMS ANI) THEIR LOCATION BEHAVIOR 12 9 5 percent. Thus, both Bogota and Cali have been going through a phase of continual increase in manufactturing establishmenits, whereas the number of establishrnents in American cities is now essentially stable. It is also notable that the majority of firms that die and the ones that are born are small. In the mid-1970s the average firm size in Rogota was about forty, approximately half the average for U.S. cities. The average firm size in Cali was a little higher, about fifty, reflecting the prescnce of a few very large firms employinig more than 500 people. Such firmlis (with at least 500 employees) typically accouniit for 35 to 50 percent of all manutfacttur-inig employnment in U.S. cities, while in BogotA they account for only about 15 percenit, and in Cali 25 percent. These figujres ignore the employmllent in the smallest firms, those employing fewer than ten people. In order to trace the spatial changes in manufacturinig employmenit, the industrial directory data were again classified by location according to the ring and sector framework in bothi Bogotd andl Cali. As might be expected, the decentralization trend is confirmiied (see table 6-5). Mov- ing outward, the rate of growthl increases with each ring (except f-or a slight decline in growth rate from ring 5 to ring 6 in BogotA). Note that in Bogota the absolute increases in manufacturing employment were similar in rinigs 3 and 5. It would seem that the center of the city, rings I and 2, had become saturated, anid firms were moving otit to get largel space. NMany moved small distances to ring 3, and others moved still far- ther otUt. Cali was more centralized, with the bulk of manuLfacturing employmenit still concentrated in r ings 1, 2, and 3. Similar tabulation of sectors confirmedl earlier findinigs. The industrial corridor (in sectors 4 and 5) continued to provide abotut three-quarters of all manufacturing employment in BogotA; in Cali the inclustrial corri- dor (located in sector 3) provided about 60 percenit. There was clear- evi- dence of the increasing importance of sector 3 in sotutlhwestern Bogota, the Bosa area, where manuifacturing employment expandecl from about 8 percent in 1970 to II percenit in 1975. Most of this increase is in the periphery of the sector. Both Bogota and Cali reveal clear land use specialization, with little mixing of residential areas with manufacturing employment. In RogotA the residenitial areas are predominiantly in the north (rich) and in the south and southwest (poor), and these areas are separated by the indtus- trial corridor in the west. Much of this patter n originally resulte(d from lanid-tuse zoninig but has been emphasized by people's residential choices in recent years. We now analyze changes by the underlying components: changes in employment in stationary firms; birth of new firms; disappearance firojm the directory and presumned death of firmns; and relocation of firms within the city between 1970 and 1975 (see table 6-fi). About 25.000 jobs  1301 U'ND)ERSTANDING THF: DEVELOPING METROPOlIS Table 6-5. Distribution of Manufacturing Employment by Ring: Bogotii and Cali, 1970-75 1970 1975 Annual average growtlh rate Rilag e'rtyOns /'erceattge 1'erMon.1 Percentage (percetlt) B(gotei 1 4,538 5.60 4.102 3.47 -2.00 2 11,767 14.53 14,898 12.59 4.83 3 34,351 42.42 47,858 40.44 6.86 4 18,112 22.37 25,958 21.94 7.46 5 11,548 14.26 24,047 20.32 15.80 6 391 0.48 729 0.62 13.27 N.ie. 266 0.33 741 0.63 - Total 80,973 10.00 118,333 10(.00 7.88 (Cali 1 2,600 7.65 3,064 6.95 3.34 2 13,836 40.74 14,381 32.60 0.78 3 15,192 44.73 21,704 49.21 7.40 4 1,761 5.18 3,361 7.62 13.80 5 367 1.05 10( 0.23 -22.90 N.it. 219 0.64 1,499 3.40 - Total 33,965 10(0.(0 44,109 10(.00 5.37 - Not ;vailable. N.i.e. Not included elsewhere. Vo/e: Data are fiomn establishments with ten or more emplo)ees. Swi,rm: K. S. lee (1989. table 3-3). based onl DANE industrial directors file. were added by the birth ofl new firms and about 15,000 new jobs resulted from the growth of stationary firms during the period. The rate of growthi in jobs in each ring is similar tor stationary and new firms except in ring 2, where there is high growth in new firms and low growtil in stationary firms. The total employment of firms that moved wits higher in their new locations than in their old ones. This patterni supports the incubator hypothesis, which suggests that small new firms locate in centr-alized areas where thiere are already many similar firm1s and where it is easy to obtain specialized essential services such as rea(dy production spaces and finiancial serNices (see Hoover and Vernon 1959; Struyk and James 1975). NeIw firms need the benefit of agglomeration economileis to start. A more disaggregated spatial analysis showedl that the incubation area was concentrated in the inner area of the indtustrial corridor, that is, the part of the corridor that fell in ring 2. This area was characterized by a high concentration of firm births, by a higher- increase in employment due to births rather than to the expart- sion of existing firms, and by the small average size of new firms. An area with similar characteristics was also found in the Chapinero district in noi-thwest Bogot5, in ring 2, sector 7. Cali had a similar area just  FIRMS AND THEIR LOCATION BEHAVIOR 131 south of the CBD. In both cities the average size of new firms in the outer areas was large-about 100 employees. An examination of the firms that relocated strengthened the incuba- tor hypothesis. A large number of small firms moved out of the incubator areas, but most moved short distances-40 percent stayed within the rinig, and most of the others moved from one ring to the next iing otit. Most of the firms that moved greater distances, to rings 4 and 5, were large. To understand the location patterns better, we further disaggregated the data by industries. Decentralization occurred in most industries, except plastics, which was already predominantly found in ring 5. As in other cities, only the printing and apparel industries were concentrated in the centet; with more than 30 percent of employment in rings I and 2.4 These industries did decentralize between 1970 and 1975 but to a lesser extent than other industries. An attempt was also made to mea- suL-e the cluster-ing of industries: how close to each other do firms in the same industry locate? It was surprising to find no particular pattern among industries. Firms do not seem to gain bv locating near similar firms in any industry.? The evidence of agglomeration economies is then only present in the incubator area, where small new firms gain by locating where there are all necessary facilities. In summary, then, we found overwhelming evidence of decentraliza- tion in manufacturing employment in Bogota and Cali in even the short period of 1970-75. Firms that relocate generally move outward and expand their size, but they usually move only short distances. New, small Table 6-6. Changes in Manufacturing Employment by Ring: Bogota, 1970-75 Annual death rate Annualgrrowth rate Annuailiirthrate (percenitage offirnis Orignri/destinatzon ofstationarfirnuL (percentage of 7newi going omt of ratio of relocating Ring (percent) firnmsv) huiness)l firms" 1 3.6 4.1 2.4 ?.8 2 2.9 6.4 2.7 1.5 3 4.5 4.5 2.1 1.0 4 3.9 4.6 1.7 0.5 5 7.3 7.9 2.8 0.1 6 n.a. 17.6 2.2 1.2 Total 4.7 5.4 2.2 0.8 .a. Not appli(able. Note: Data are for firms with more than ten employees. a. Based on total 1970 manufacturing employment. b. Calculated as the ratio of the number of people employed in firms that moved oUt of the ring to the number of people emploved in firms that moved into the ring. Source: K. S. Lee (1989, table 3-6). based on DANE industrial directory file.  132 UNDERSTANDING TIHE DEVELOPING METROPOLIS firms locate in the incubator area-in the industrial corridor but near the C(BD. They replace other new firms that either succeed and move out or fail and die. Firms locating at the far periphery are usually large enough (wvith an average of 100 employees) not to need the agglomera- tion economies of the city center. The decentralization trend is com- mon to all industries. Factors Influencing the Location of Manufacturing Firms To explain these changinig location patterns, a special sample survey of Bogota's manufacturing establishments was conducted in 1978, using the industrial directory as the sample base. Data from this survey were used to model the location behavior of manufactur-ing firms. The underlying theoretical framework is that a particular type of firm will choose a site with particular attributes that are optimal to the firm in terms of such criteria as profits and costs. The determining characteris- tics of the firms include type of product, production process, building characteristics, lot size, floor space, and level of skill in the work force. Important site attributes incltide proximity to product markets and sup- pliers, comnllutinIg clistanice for employees, traiisportationt modes, and the qLuality and availability of public utilities anic municipal services. A sample ot 126 firms was chosen firom the roughly 2,600 firm records in the clirectory. Thie sample was stratified by location history (new fir ms, movers, stationar-y firms), current zone of location, type of indus- trv, and establishment size as defined by the numLLber of employees. To obtain reasonable homogenieity among firms, the sample was focused on textile- and metal-fabricating firms, which covered about 50 percent of total manufacturing employment in Bogota. Both industries can be regarded as relatively footloose in that they require no special site char- acteristics for their location. Being footloose, they are also presumnably more susceptible to governmenit policy regarding location. A third groLlp. "other indtistries," was aclded for descriptive variet.v. In order to maximize coverage of employees in this limiteld sample, larger firms were oversamnpled in the survey. The geographical coverage was good in that twentyv-seven of the thirty-eight conmiuas, located mostly in rings 3, 4, and 5 but with some in ring 2, were covered in the survey. Of the 126 firms, 50 were movers (that is, firms that were relocating), 18 were new, and 58 were stationary. The homogeneity of firms from similar indus- tries allowedl the testing of behavioral hypotheses with reasonable degrees of freedlom. The incubator hypothesis is supported by firms in this sample. The new firms were smaller than others; they worked in single-shift batch processes: thev were housed in relatively older buildings, often with  FIRMS AND THEIR LOCATION BEHAVIOR 133 more than one floor; and they had little land for expansion. Thus small, new firms begin operations in the more centralized locations and move to the periphery into larger sites only after they have achieved some success. The use of trucks has been well documented as a major explanatory factor in the decentralization of employment in the United States (Hoover and Vernon 1959; Moses and Williamson 1967). The same phe- nomenon was of great importance in Bogota: 95 percent of all goods going to and from firms were carried in trucks in Bogota at the time of this survey. This is remarkable considering that the train station in the center of the city was an important reason for the central location of firms until just two decades ago. The change is all the more impor-tarit given that 40 to 50 percent of the manufacturing output of these firrns is exported outside Bogota. To gain insight into why firms located where they did, managers were asked to evaluate their firms' locations. On the whole, the respondenits were very satisfied with road access, proximity to clients and suppliers, and the availability of unskilled workers, but they were dissatisfied with the quality of municipal services, zonal amenities, the availability of skilled workers, and the cost of land for expansion. Managers of firms that had moved were usually satisfied with their current plant capacity, in contrast to managers of many of the stationary firms. Reflecting their more centralized locations, managers of the new firms were largely happy with the quality of mtunicipal services. 1ndeed, that quality is yet another reason for the centralized location of new firms. Dissatisfaction with muriicipal and public services increased with distance from the center. This suggests that when firms move to the periphery they lead infrastructure investments, which follow them with some lag. This nega- tive aspect of the new locations is traded off against advantages associ- ated with greater space. Moves are usually associated with changes in technology and expan- sion of production; the new technology often requires more space to spread out. Thus the larger firrns move longer distances to the periph- ery, whereas smaller firms move just one or two kilometers. This is pre- sumably important in order to keep the same labor force: about 80 percent of employees remain with the same firnm even after the move. The average commuting distance does not change after the move, which suggests either that firms move short distances or that employees change tlheir places of residence to suit their new job location. In the firms' own evaluation of factors affecting their move, the expansion of plant capacity was the most significant. Rent paymnents, proximity to suppliers, and zone amenities were also important. The proximity of clients, quality of public services, and cost of land carried less weight but did enter into their (lecisionmaking. The firms'  134 lNDERSlAND)ING TtlE DEVEILOPIN( METROPOIIS responses suggest that they pay little attentioni to public sector decisions concerning infrastructure availability; they assume that infrastructture will follow with a lag. Modeling the Intra-Urban Location Behavior of Manufacturing Firms How can these firms' location decisions be modeled and estimated econometrically? Unfortunately, theoretical and empirical literature on firm location behavior is quite rare, which makes it difficult to draw gen- eralizations from our work.' The general idea, however, is that because Firms can be assumed to be profit-maximiizing, they would be expected to respond to the existing lanid price gradlients and any other inptzt price gradients in deciding their location. The results obtained were certainly consistent with such a hypothesis. Although firms did not explicitly give much importance to land prices, the clesire for plant expansion was often given as the main reason for relocation. I.ower land prices at the periphery attract such relocating firms: hence land prices are indirectly seen as an important variable affectilg location. Careful considerationi was given in the City Study to developing a sys- tematic modeling framework in order to fulther the unlderstanidinig of manufacturing firms' location behavior. Because the fill derivation of the model and its detailed estimates are available elsewhere,7 here we merely sketch the principles behind the modeling effort and provide its main results. A firm's production funlction may be written as 0 =J( l.,X;Z), where 0 = OUtplut, L= lot size, X = vector of variable inputs such as labor and capi- tal, and Z = vector of site characteristics. These site characteristics are independenit of lot size and represenlt local public goods at a particular location . Thle relevant cost components are wages, capital costs, input materi- als' costs, delivery costs of inputs and outputs. and land rent. These would be the main variables of interest for calculating the optimum combination of inputs and for optimum site selection. Following stan- dard urbani economic theor-y based on bid rents, a particular plant site can be seen as being occupie(d by the firm willinig to pay the highest price. The bid price will depend on the attractiveness of the site to a par- ticular firm. In a locational eqtiilibrium, all firms of a particular type in an urban area would make the same rate of profit, and no firm would have an incentive to relocate. In this case the cost tradeoff calculations by indi- vidual businesses determine the spatial distribution of firms. For exam- ple, a large manufacturinig plant may choose a site in a low-ren-t area  FIRMS AND THEIR LOCATION BEHAVIOR 135 near the urban periphery to meet its need for more space, in spite of a greater delivery distance. Small firms, in contrast, may prefer a central location where the availability of various externalities more than com- pensates for high rents. The preceding theoretical firamework leads to an empirical fi-ame- work for predicting the probability that a firm of a particular type will occupy a site withi particular attributes (Z). Because thie firm with the highest bid will occupy a given site, the relevant random variable for determininig this probability is the maximum amount a group of firms with similar attributes would pay (that is, the maximum bid). The prob- ability (listribtifion of the random variable associated with the maximum bid leads to a multiiioinial logit specification for the firm location model.8 Such specifications are useful in modeling any behavior dealing withi choice. As chiapter 8 (isCLIsses, the issue in travel demandl modeling is how to model an individtual's choice between different modes. Similarly, the issue in residential location is how to moclel a consumer's choice of par- ticular types of hotises and locations. Here we have the employment location problem of firms having to choose among locations. An appli- cation of suchi models to residential location choice by Ellickson (1977, 1981) has suggested the approach adopted in this study. He formulated his problem: given a site of certain characteristics, what is the probabil- ity that a h)ousehcold of a certain tvpe wvill occupy it? Our problem here is similar: given sites of certain characteristics (Z), what is the probability that certain kinds of firms will occupy it? Another approach would be to define certain zontes in the city and to find the probability associated with each type of firm locating in each zone. This would necessitate dividing the city into arbitrar-y zones such as rings and sectors and find- ing the type of firm most likely to locate in each zone. The approach that has been adopted hel-e is more general; sites are specified by more general characteristics. The most important advantage of this approach is that thle model allows the observation of a wide range of spatial varia- tion in site character-istics and the sample does not hiave to be broken up by zones. This type of specification also allows policy analysis. It predicts the probability that a particular tvpe of- firm will occupy a site with par- ticular attributes. The marginal effect of changes in these attributes on the probability of these firms' location can also then be estimated and anialvzed. Hence, coniclusions can be drawn on the efficacy of different kinds of policy instrumnienits that affect these attributes. In this study, the data collected in the survey of 126 manufacturing firms were utilizedI for estimatinig the model. The firms were grouped into four types: small and large; textile-mantifacturing and metal- fabricating. Of the 126 firms surveyed. 87 fell into these categories. Large metal-fabricating firms were usecl as the reference group. The  136 UNDERSTANDING THE DFVELOPING METROPOLIS effect of different location attributes on the probability of location of the remaining three types of firms was seen in relation to the location attributes' effect on the reference group. The independent variables included: * Access to local markets for output (PRODSOLD) = percentage of product sold in Bogota * Access to local markets for inputs ([NPUTBT) = percentage of input bought in Bogota * Proximity to residential location of production workers (WKSOUTH) = percentage of workers living in southern part of Bogota * Proximity to residential location of administrative workers (ADM- NORTI-I) = percentage of workers living in northern part of Bogota * Index oF quality of local public services (ELECINT) = frequency of electricitv interruption * Presence of agglomeration economics (LOCQT) = location quotient defined as comuna's () share of industry (i) in relation to its share in total manufacturing employment * Intensity of economic activity (POPDENS) = population density of the comuna in which the firm was located * Measure of accessibility to the city center (DISTCBD) = airline dis- tance from the CBI) of the center of the comuna in which the firm was located The estimation procedure provides measures of the influence of each independent variable on the probability of a firm being in the group specified in the dependent variable. For example, the coefficient of the proximity of firms to residential areas provides a measure of the impor- tance of this variable to small textile firms as compared to the base. The estimated coefficients are translated into elasticities such that policy conclusionis can then be drawn. The overall goodness of fit was found to be high, providing some confidence in the results. The multinomial logit estimation method was used to perform these investigations. It became clear that small firms are quite distinct from large firms in their location behavior and that textile firms display behavior different from that of metal-fabricating firms. These results indicate the interaction between generic differences found in different industries according to their size, the nature of their relationship with their suppliers, types of product markets, and their location. Small textile firms were the most likely to locate near the city center and in close proximity to their input suppliers and the homes of their production workers. In Bogota this meant locations in rings 2 and 3 and nearer the southern and southwestern parts of the city. Large textile firms, in contrast, choose sites farther from the city center-about three  FIRMS AND THEIR l.OCATION BEHAVIOR 137 times farthier away, in fact-and proximity to thie homes of their profes- sional and administrative workers seems to be more important. They are also more sensitive to locating in areas with better public services (for example, a more reliable electricity supply). They are not sensitive to proximity to local markets and are farther away fiom dense residential areas. Small metal-fabricating firms are similar, in many wavs, to small textile firms: they like to be near their inputs and product markets but are not likelv to be near the citv center. All these findings are consistent with the idea that smaller firms need to take advantage of agglomera- tion economies that are realized by locating in areas wher-e both input and produce markets are close by and where productioni workers are easily available. I.arge firms, on the other hand, are oriented more toward markets outside the city and are less (lependent on the city in which they are located. They tend to interinalize manv of the agglomera- tion economies that smaller firms take advantage of by being near other firmins. To strenigtheni confidence in these findings, the dependenit variable could be specified in different ways (that is, firms could be grotiped in differenit ways). Location choice could be modeled more dlirectly by grouping the firms by ring location. In this case, firms were dixidedl into four grotips: rings I and 2 combined; ring 3; ring 4; and rings 5 and 6 combined. As noted earlier, this was an arbitrary grouping proceduire, bLut it was tised here to bolster the results found earlier. The estimations were found to be very robtist. Firms that sold a large proportion of' their output within Bogoti tend to choose sites near the center, and export- oriented firms locate nearer the periphery. Similarly, firms that rely on local suppliers tend to be near the center. Moreover, locally oriented firms tend to be small. Firms that recenitlv moved are extreme;ly unlikely to be in or near the CBD: they almost always move to outer areas where more space is available, thus contributing to the decentralization of manufacturing employment. All of these results imply that mantifacturing firms in Bogota behave in an optimizing fashion in choosing their location to minimize overall costs. But we still have little sense of how responsive they are to changes in factor prices. When firms move to more peripheral locations in order to expand, how responsive are they to the land price gradient that exists in the city (see chapter 3)? Is there also a wage gradient? Additional esti- inations were made to address these questionis. It is clear that a land price gradient exists. The 1976-78 land price gradient was about 0.10 to 0.12; that is, land price can be expected to decline by about 10 to 12 percent with each kilometer from the city cen- ter. The sample of 126 maniufactuL-inig firms also had direct infolrmation about the value of the land where these firms were located. Remarkably, the gradient calculated from these observationis-about 0.11-was  138 UNDERSTANDING THE D)EVELOPING METROPOLIS exactly consistent with the other independent data. The other major factor of production is labor. There are few good reasons to expect the existence of a wage gradient in a city. Firms located in the center of a city need to pay slightly higher wages to attract workers who might have a longer commute firom outer areas. But the same justification could be found for higher wages in outlying areas. Hence there are few estimates that suggest the existence of clear wage gradients in cities. Nevertheless, in Bogota, even after adjusting for the usual human capital variables, wages are found to have a gradient of about 0.05 as one goes out from the center of the city. Ilow do firms respond to these locational changes in factor prices? The elasticities of substitution between capital and land and between land and labor are good measures for this response. A low elasticitv of capital/land substitution would suggest a greater degree of decentrali- zation in the presence of a land price gradient. The elasticity of substi- tution measures the percentage of change in capital/land ratio in response to a change in factor prices-that is, the ratio of land price to capital price. As land prices vary, then, how does the capital/land ratio vary? Again, there were few estimates of these elasticities for manufacturing firms in other cities. The capital/land elasticities that have been esti- mated are mostly for urban housing. They range between about 0.4 and 0.8 in most cases (see McDonald 1981 for a review). In this studv we uti- lized the 84 firms (out of the 126) who were owner-occupiers, and we used the market value of plant and equipment as the capital value. The resulting elasticity of substitution was about 0.31-much less than esti- mates made for urban housing in the United States. This result supports the a priori assumption that the elasticity of capital land substitution for manufacturing is likely to be smaller than for housing because the possi- bilities of locating manufacturing activity in multistoried structures are limited. Similarly, the land/labor elasticity was calculated by using the wage bill for each firm. The value of this elasticity was also about 0.30; that is, as land price increases, firms increase the intensity of labor used per unit of land. As expected, we see greater crowding in factories in central areas of the city. In summary, the r esults from all these econonmetric investigations sug- gest that firm location behavior is by no means random. The estimated model was derived for profit-maximizing firms, which were found to behave much as would be predicted. The goodness of fit was satisfac- tory, and the estimated model was capable of predicting, in probability terms, which types of firms are likely to occupy a site with particular characteristics. The location patterns that the model predicted were consistent with those expected. Atmong simnall firms, accessibility to local inlput and output markets was the most important factor in the location  FIRMS ANI) THEIR l.OCATION BEHAVIOR 139 decision; the benefits of accessibility to the central area compensated for the high rents and congestion. Large establishments, which were more export oriented and required more plant space with modern pro- duction technology, located in outer areas, where more space was avail- able at lower cost. The estimation results also indicated that the quality of public utility services is very important for large firms and that prox- imity to the residences of administrative workers is more important to such firms than is proximitv to the homes of production workers. The precise modeling framework used here was specific to Bogota. The independent Variables chosen were those available fi-om the survey and considered important in Bogota. Similar investigations in other cit- ies could involve different specifications and different variables suitable for those cities. Our study has, however, indicated a inodeling frame- work that is svstematically derived from urban economic theory and contributes to an understanding of the behavior of manufacturinig firms. Trade and Service Employment Patterns It is much easier to model the location of establishments that are in some sense people-serving-for example, schools, local government agencies, retail trade establishments, and banks-than it is to model the location of manufacturing companies. Access to local markets is the main determinant of the location of employment in people-serving activities. Considerable work has been done in this area,9 and the mnodel used is a simple analog of Newton's Law of Gravity: the level of interac- tion is directly proportional to the mass of interacting bodies (size of the groups of sectors) and inversely proportional to the distance between them. This approach is therefore referred to as the social physics approach. Although the model is then not couched in terms of eco- nomic behavior, it may be interpreted as reflecting firm-optimizing behavior. A firm chooses its location in order to optimnize its access to the customers who are likely to demand its services. Implicit in this for- mulation is that proximiiity to inputs is less important to such firms than proximity to customers or that prices of inputs are invariant wvith respect to location. The model framework assumes that each stubarea in the metropolis attracts the various activities in relation to its relevant loca- tion attributes. These attributes will vary with the activity group consid- ered but will include both "policy" attributes such as available land for the activity aind "given" attributes such as current populationl or employ- ment levels and land value (Lakshmanan 1964). Only the broad outline of the model and the main results are provided here; details are avail- able in K. S. Lee (1989).  140 UND)ERSTANDIN(; 'tHE DEVELOPING METROPOLIS Because we did not conduct a special firm survey of firms in these activities at the establishment level. information was not available for this exercise. Instead, the activity location information in the 1978 Househiold Survev had to be used. From this survey we knew the loca- tion of employment of all the workers in the sample. 'I'he behavior of firms was therefore deduced from the location of workers in those trade and service sector activities. The specific activities considered in these exercises were wholesale trades; retail trades; restaurants and hotels; financial establishments; real estate and business services; government services; sanitary services; social and commtinitv services; recreation and entertainment; and personal and household services. The standard model repr-esents the hypothesis that the trade anld ser- vice firms in the ith zone serve people in all zones in the city but the likelihood of service declines with distance.10 This implies that a firm of a particular type will locate in a zone with particular attributes. Three meastur-es of market potential wer-e used in the estimates. The first two were the numnber of hotLisehlolds and population in each zone. As might be expected, thev gave similar results. The third was puL-chlas- ing power of a zone. The dependent variable used was employment den- sitv of each clifferenit activity in each of the thirty-eight comunas in Bogoti. The model was quite successftul in explaining the location of these tracle and service activities. The more directly people-serving an activitv was, the better the model performned. Thus, the best fits were obLained for reuail trade, social and commnunity services (which included educationi), and personal and household services (which included domes- tic workers). The least satisfactory explanation was found for the loca- tion of government workers and sanitar-y service workers. This is as expected, since these latter groups serve the public less directly. The purchasinig-power model performed better than the population andl household measures of market potential for all activities except retail trade, where it was about the same. This is consistent with the over- all employment location pattern mentioned earlier-that is, the loca- tion of maniy more employment activities in the richer parts of the city. Specificallv. the purchasing-power measure improves the degree of explanation significantly for- the location of employment in restaurants and hotels, finance, recreationi and entertainment, and personal and household service. Clearly, all of these activities are highly dependelnt on customers with high incomes. Banks, restaurants, and domestic ser- vants are all Inore likely to be found in the richer areas of the city. One other experiment was performed to test the degree of associa- tion between the market potential proxv and employment density of each zone. Sensitivity tests were performed by varving the value of g- that is. changing the decay factor of market potential by distance. If a highel g resullts in better goodness of fit, the more localized the activity  FIRMS AND TIIEIR LOCATION BEHAVIOR 141 is. Recreation and entertainment, social and community services, per- sonal and household services, and restaurants and hotels turn out to have significantly improved goodness of fit with an increase in g: these activities are more dependenii on local clientele. In summary, it was again found that the location of employment in different activities was as woulld be expected from rational profit- maximizinig kinds of behaviolr Employment in retail trades, social and community services, and personal and household services follow the patterns of residential location. Other activities, such as recreation and entertainmentt, restaurants, hotels, and finanicial services, nee( to be near high-income districts of the city. Government services and wholesalers are relatively independent of the residenitial location of hiouseholds. These investigations illuminiate the interactioni between residential and employment location and highlight the imipor-tance of lhigh-income res- idential locations. Implications for Location Policy This chapter emphasizes the pheniomenoni of decentralization that accompanies the growth of cities. It focuses oni the role of the manufac- turing sector as the leading factor in the decentralization of growing cit- ies. In tryinig to uLnderstand the economic forces at work in this process, we have also begunl to better understand that certain activities arc likely to continue to locate near the center of the city as it grows and that therefore the (CBD is likely to remain an important concentration of employmnent in large cities. What do we learn from all these findings that could help shape policy in regard to the structure of cities? TIhe growth and diversification of industr-y in a large city would almost inevitably lead to decentralization of the more successful segments of industry. The rnost important reason for firms to relocate toward the periphery of Bogota was their desire for expansion, usually accompa- nied by techniiological change in the productioni process. Mass produc- tion requires assembly lines or continuotus flow process technology, which implies the need for single-floor plants and therefore more space. This is another way of saying that for such processes the elasticity of cap- ital/land substitution is low, and( firms move down the rent gradient to the periphery to acquir e the larger plots they need for productioni, gain- ing space for future expansion in the bargain. Moreover, large plots of land that are not btiili up are usually available only at the periphery. In contrast to larger firms, smaller manufacturing firms, especially newer ones, still prefer locations close to the city center. They locate in premises vacated by other firms and near to the many input services- financial institutions, raw materials suppliers, equipment repair shops,  142 UNDERSTANDING THE DEVELOPINC. MET ROPO LIS and residences of prospective workers-that they typically require for their dav-to-day funictioniing. A clue to this behavior is found in the high birth and death rates of firmns in Bogota during the period under study. New, small entr-ants into the manufacturing business are inherenitly unstable and therefore naturally likelv to economize on overhead and fixed costs of all kinds. Many services that a larger firm would inter-nal- ize are r ented temporar ily by such a firm until it reaches greater stabil- itv. Many new entrepreneulls, for example, are technicians with little capital. starting out on their owin after breaking away from their original employers (see Cortes, Berry, and Ishaq 1987). They can afford to take few risks. A thriving, concentrated area of manufacturing firnms that gives agglomeration economies to budding entrants is therefore likely to impart health to a city seeking industrial expansion. Many urban policies are clesigned to thwart the location of new manu- facttur-inig firms in large cities. In couLntries as dlifferent from one anotlher as Indlia, the Republic of Korea, and Venezuela, similar policies prohibiting manufacturing firms from locating in the large metropoli- tan cities have been followed. These policies have ignored the impor- tant role of large cities as incubators of entreprenieuirslhip. The' have resulte(d from a lack of understaniding of the locational dynamics of industrial growth. Prohibiting new manufacturing activity in large cities also has the effect of reducing effective death of sick firms. The death of firms is easier for a city's economy to sustaini if the displaced labor has a good opportuLnity to find work in new firms. If births are prohibited, resistance to death increases and the city acquires an environment of stagnationi. Apart from enforcing pollution regulationis and( preventing hazardotus industries fromn locating in densely populated areas, it should be understood that the locational dynamics of maniufactturinig firms are such that large firms or expanding and successful small firms will them- selves decentralize away from central cities. It is also commonplace to find in(lustrial estates located on the periphery of cities; these are ofteni designed to attract small firms. They have seldom been successful. Our results provide persuasive explana- tions for suchi failure. Industrial estates are unllikely to become incuba- tors of entreprenLeurship until a large nulnber of successftil firms are located there already. Moreover, new firms cann1ol. tolerate interrup- tions in basic services like electricity, water, andl transportation. New developments in cities typically suffer from teething u'otubles in early stages; if they do not, it is at very high investment costs. Enliglhtenied industrial location policv in cities would therefore not aim for premna- ture decentralizationi of industrial activities; it would occur anyway with successf'tl growth. The role of policy is to facilitate suCh) growthi and plan for the provision of public services in response to suchi growth. It is the larger firms that are more likely to move to periplheral locations andf  FIRMS AND THEIR LOCATION BEHAVIOR 143 they are also more likely to be able to pay for the various necessary pub- lic services. Attempts are also often made, at very high initial infrastructure costs, to persuade small and medium-size manufacturing firms to locate in new "satellite" cities, which are typically 30 kilometers to 100 kilometers from the large city. Examples of such attempts include Banweol, near Seoul in Korea, and Kalvan, near Calcutta in India. Attempts at such development are usually miserable failures. The actual cost of locating in such places tisually far exceeds the benefits from any tax or other incentives offered by the government. The pressures for decentralization are relentless with growth, how- ever, so public policy must also be careful not to thwart it. Public trans- portation routes, for example, are typically radially oriented toward the center of the city, in response to the traditional location of jobs. An expanding city with a growing and decentralizing manufacturinig sector also suggests the need to instittite circumferenitial routes so that poor workers can reachi their destination without wasteful trips through the center (see chapter 8). Many of the relocating large firms in Bogota planned their moves well before the availability of public services in the new location. There seemed to be a general belief that once they moved, public services like roads, water supply, and power would follow. This may have been pecu- liar to Bogota, where the decentralized agencies (see chapter 9) are par- ticularly responsive to demand and have the autonomy to act. It was also the case that much industrial location in and around Bogota conformed to the industrial zones provided accordinig to land use regulationis. This points to the value of realistic land use zoning that is itself responsive to finms' preference and able to give public authorities the flexibility to provide services in response to demand. W'hat abouLt other activities? ManLifacturing, after all, seldom employs more than one-third of the workers in a city. For most commerce and service employmilent, we have found a pattern opposite to that of manul- factturinig. Larger firms locate much more in the center, whereas smaller- firms are more decentralized. Overall, as we also saw in chapter 3, employmenit in manufacturing is more decentralized than in services and commerce. Indeed, it can be said that the CRD is specializing more and more in nonmanufacturing employment, and the periphery, in manufactur-inig. The reason, although this has not been determiiecl empirically, seems to be that office activities, trade activities, and other similar activities are likely to have a higher capital/land substitution elasticity. It is easv to pile up such activities in multistory buildinigs. At the same time, many activities that are directly people-serving, sucIh as small retail stores, other personal services, local goverriment services, and schools, hiave to be near their- customiiers and therefore decenitralize  1 44 UNDERSTAND)ING THF. DEVELOPING METROPOLIS as their customers do. Residential location and manufactturing location therefore lead these other activities in their decentralization pattern as a city grows. But most of these activities are sensitive more to purchasing power rather than to mere mass of people, so they are much more likely to be found in the richer areas than in the poorer ones. In fact, the many retail activities found in the C:BD are located there because of the lack of effective demand in the many poor areas of cities in developing countries. A poor neighborhood cannot support many activities. This explains why some activities are likely to continiue to be more central- ized in cities in developing countnes until incomes increase in those countr-ies and cities. It has also been observed that competing peripheral "city centers" emerge as a city expands. The indication is that the relative poverty of cities in developing countries makes this process slow, however. City planners often prematturely plan competing centers for retail, whole- sale, office, and other activities, whereas they attempt to restrict the accretion of new activitv. in the traditional city center. Our study suggests that this should be done onlv after careful consideration of the popula- tion's demand for such activities. Otherwise, much inefficiency can result from the loss of agglonmeration economies inherent in the city center. These observations illustrate the strong linkage between the pattern of residence location, the spatial distribution of income, and the loca- tioin of employvment, and show how their mutual interaction produces the city structure. The spatial inequality of incomes influences the distri- bution of' emplovyment location and explains the lack of employment in the poor areas of Bogota. Once again, the effect on the design of public transportation is clear. We can better predict the course of a city's growth once we understand these linkages. Notes 1. This chapter summarizes the work reported in K. S. Lee (1989). 2. The traditional commercial district (Chapinero area) in sector 7 used to be the "Fifth Avenue" of Bogota; in recent years, however, retail stores and restau- rants have sprung tip along 15th Avenue northward, and a large shopping cen- ter (called Unicentro) was developed in the northern part of sector 8. The determinanits of commercial and service employment location were also tested using a gravity model. For details, see K. S. Lee (1989. chapter 6). 3. As reviewed in Kemper (1973). See also Struvk and James (1975), Hanushek and Quigley (1978), Leone (1971), Schmenner (1973, 1982), and Cameron (1973). 4. See Henderson (1988) for a broader discussioni of the kinds of firms that are found to continue in large cities.  FIRMS ANI) THP IR I OCAFI ON R.HAVIOR 145 5. Henderson (1988) r eached a similar conclusion. 6. Works on this subject as it applies in the United States include Hlanushek and Song (1978), Erikson and Waysleniko (1980), Schneniner (1973, 1982), and Carlton (1979, 1983). 7. The derivation was first provided in K. S. L.ee (1982) and is also available in K. S. Lee (1989, chapter 5). 8. The application of the multinomial logit method to urbani economic rcsearch became popular withi McFadden's work on travel demand studies (1973, 1974. 1976) and wtith work by Friedman (1975), Lerman (1977). and Quigley (1976) on residential location studies. 9. There was muci initerest in the location of'service-oriented establishments in the United States in the earlv 1960s (for example, Lowrv 1964; Lakshmanan 1965; W. (;. Hansenl 1959; W. B. Hansen 1961: Pendletoni 1963; Huff 1961). 10. Details of the model may be found in K. S. Lee (1989. chapter 6). Briefly, the model f'ramework is as follows: k E f(G,, ) where Ek = number ot'jobs in activity k in zone i; G1 = a gravity measul-e of the market potential of zone i; and ET = the total nuImber ofjobs in zonie i. The rneasure of m'iarket potential can be expressed as z (; = , i, j = .I dg where Z j = proxy fotr market poten tial of zonie i with respect to zone; dg = dis- tance between zonie i and zonC]: and g= distance decav weight. If, for example, we measure market proxy by the population in each zoniC, and( if there are, say, three zones in the city and g= 2, thenl: p, PP2PI P.2 I= _ + + 22 dil d12 " 143 where d, I is a measure of the size of zonie I-for example, the radius of zonie 1. GI