LJM II L'i Living Standards Measurement Study Working Paper No. 116 Health Care in Jamaica Quality, Outcomes, and Labor Supply LSMS Working Papers No. 45 Gertler and van der Gaag, Measuring the Willingness to Pay for Social Services in Developing Countries No. 46 Vijverberg, Nonagricultural Family Enterprises in C6te d'Ivoire: A Descriptive Analysis No. 47 Glewwe and de Tray, The Poor during Adjustment: A Case Study of C6te d'Ivoire No. 48 Glewwe and van der Gaag, Confronting Poverty in Developing Countries: Definitions, Information, and Policies No. 49 Scott and Amenuvegbe, Sample Designs for the Living Standards Surveys in Ghana and Mauritania/Plans de sondage pour les enquetes sur le niveau de vie au Ghana et en Mauritanie No. 50 Laraki, Food Subsidies: A Case Study of Price Reform in Morocco (also in French, 50F) No. 51 Strauss and Mehra, Child Anthropometry in C6te d'Ivoire: Estimates from Two Surveys, 1985 and 1986 No. 52 van der Gaag, Stelcner, and Vijverberg, Public-Private Sector Wage Comparisons and Moonlighting in Developing Countries: Evidence from C6te d'Ivoire and Peru No. 53 Ainsworth, Socioeconomic Determinants of Fertility in C6te d'Ivoire No. 54 Gertler and Glewwe, The Willingness to Pay for Education in Developing Countries: Evidence from Rural Peru No. 55 Levy and Newman, Rigidit6 des salaires: Donnees micro6conomiques et macroeconomiques sur l'ajustement du march6 du travail dans le secteur moderne (in French only) No. 56 Glewwe and de Tray, The Poor in Latin America during Adjustment: A Case Study of Peru No. 57 Alderman and Gertler, The Substitutability of Public and Private Health Care for the Treatment of Children in Pakistan No. 58 Rosenhouse, Identifying the Poor: Is "Headship" a Useful Concept? No. 59 Vijverberg, Labor Market Performance as a Determinant of Migration No. 60 Jimenez and Cox, The Relative Effectiveness of Private and Public Schools: Evidence from Two Developing Countries No. 61 Kakwani, Large Sample Distribution of Several Inequality Measures: With Application to C6te d'Ivoire No. 62 Kakwani, Testing for Significance of Poverty Differences: With Application to C6te d'Ivoire No. 63 Kakwani, Poverty and Economic Growth: With Application to C6te d'Ivoire No. 64 Moock, Musgrove, and Stelcner, Education and Earnings in Peru's Informal Nonfarm Family Enterprises No. 65 Alderman and Kozel, Formal and Informal Sector Wage Determination in Urban Low-Income Neighborhoods in Pakistan No. 66 Vijverberg and van der Gaag, Testing for Labor Market Duality: The Private Wage Sector in C6te d'Ivoire No. 67 King, Does Education Pay in the Labor Market? The Labor Force Participation, Occupation, and Earnings of Peruvian Women No. 68 Kozel, The Composition and Distribution of Income in C6te d'lvoire No. 69 Deaton, Price Elasticities from Survey Data: Extensions and Indonesian Results No. 70 Glewwe, Efficient Allocation of Transfers to the Poor: The Problem of Unobserved Household Income No. 71 Glewwe, Investigating the Determinants of Household Welfare in C6te d'lvoire No. 72 Pitt and Rosenzweig, The Selectivity of Fertility and the Determinants of Human Capital Investments: Parametric and Semiparametric Estimates No. 73 Jacoby, Shadow Wages and Peasant Family Labor Supply: An Econometric Application to the Peruvian Sierra No. 74 Behrman, The Action of Human Resources and Poverty on One Another: What We Have Yet to Learn No. 75 Glewwe and Twum-Baah, The Distribution of Welfare in Ghana, 1987-88 No. 76 Glewwe, Schooling, Skills, and the Returns to Government Investment in Education: An Exploration Using Data from Ghana No. 77 Newman, Jorgensen, and Pradhan, Workers' Benefits from Bolivia's Emergency Social Fund No. 78 Vijverberg, Dual Selection Criteria with Multiple Alternatives: Migration, Work Status, and Wages (List continues on the inside back cover) Health Care in Jamaica Quality, Outcomes, and Labor Supply The Living Standards Measurement Study The Living Standards Measurement Study (LSMS) was established by the World Bank in 1980 to explore ways of improving the type and quality of household data collected by statistical offices in developing countries. Its goal is to foster increased use of household data as a basis for policy decision making. Specifically, the LSMS is working to develop new methods to monitor progress in raising levels of living, to identify the consequences for households of past and proposed government policies, and to improve communications between survey statisticians, analysts, and policymakers. The LSMS Working Paper series was started to disseminate intermediate prod- ucts from the LSMS. Publications in the series include critical surveys covering dif- ferent aspects of the LSMS data collection program and reports on improved methodologies for using Living Standards Survey (LSS) data. More recent publica- tions recommend specific survey, questionnaire, and data processing designs and demonstrate the breadth of policy analysis that can be carried out using LSS data. LSMS Working Paper Number 116 Health Care in Jamaica Quality, Outcomes, and Labor Supply Victor Lavy Michael Palumbo Steven N. Stern The World Bank Washington, D.C. Copyright @ 1995 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing June 1995 To present the results of the Living Standards Measurement Study with the least possible delay, the typescript of this paper has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. Some sources cited in this paper may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility whatsoever for any consequence of their use. The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this publication is copyrighted. Requests for permission to reproduce portions of it should be sent to the Office of the Publisher at the address shown in the copyright notice above. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the reproduc- tion is for noncommercial purposes, without asking a fee. Permission to copy portions for classroom use is granted through the Copyright Clearance Center, Inc., Suite 910, 222 Rosewood Drive, Danvers, Massachusetts 01923, U.S.A. The complete backlist of publications from the World Bank is shown in the annual Index of Publications, which contains an alphabetical title list (with full ordering information) and indexes of subjects, authors, and countries and regions. The latest edition is available free of charge from the Distribution Unit, Office of the Publisher, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66, avenue d'I6na, 75116 Paris, France. ISSN: 0253-4517 Victor Lavy is a professor of economics at the Hebrew University of Jerusalem; Michael Palumbo is an assis- tant professor of economics at the University of Houston; and Steven N. Stem is an associate professor of eco- nomics at the University of Virginia in Charlottesville. Library of Congress Cataloging-in-Publication Data Lavy, Victor. Health care in Jamaica : quality, outcomes, and labor supply / Victor Lavy, Michael Palumbo, Steven Stern. p. cm. - (LSMS working paper; no. 116) Includes bibliographical references. (p. ). ISBN 0-8213-3243-0 1. Medical economics-Jamaica. 2. Health planning-Jamaica. 1. Palumbo, Michael, 1964- . II. Stern, Steven N., 1958- III. Title. IV. Series. RA410.55.J25L38 1995 362.1'097292-dc2O 95-17080 CIP Contents Forew ord .. .. ... ..... .. ... ... ...... .. ..... ... .. ... . ... .. ..vii A bstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Acknowledgments ............................................ xi I. Intr odue on . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 II. Adult Health Status and Jamaican Health Care . . . . . . . . . . . . . . . . . . . . . . . . 3 III. D ata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 IV. Econometric M odel ........................................ 10 The Medical FacilityChoice ................................ 10 Health and Labor Force Participation . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 V . R esults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 M edical Facility Visit Results................................ 14 Health and Labor Force Participation Results . . . . . . . . . . . . . . . . . . . . . . 14 Comparative Statics ......................................21 VI. Conclusions .............................................27 References .... ... ........ ..... ... .. ... ..... ... .. .. ... .. ...29 Tables Table 1. Definitions and Summary Statistics for the Jamaican Survey of Living Conditions Variables ................................... 6 Table 2. Definitions and Summary Statistics for Health Care Characteristics . . . . . . . 9 Table 3. Medical Facility Choice Equation Using Principal Components of Facility Characteristics ....s........5...................... 15 Table 4. Medical Facility Choice Equation Using Staff Size of Facility . . . . . . . . . . 16 Table 5. Health Equations Using Principal Components of Facility Characteristics . . . 17 Table 6. Health Equations Using Staff Size of Facility . . . . . . . . . . . . . . . . . . . . 19 Table 7. Work Equations Using Principal Components of Facility Characteristics . . . 20 Table 8. ComparativeStatics ................................... 22 Table 9. Changes in the Distribution of Health . . . . . . . . . . . . . . . . . . . . . . . . . 24 Table 10. Availability: Comparative Statics . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 v Foreword This paper is part of a broader program of research in the Policy and Research Department on the effects of availability and quality of social services on outcomes of investment in human capital. This research program is located in the Poverty and Human Resources Division. The data used here are from the 1989 Jamaican Survey of Living Conditions, which is one of the many Living Standards Measurement Study (LSMS) household surveys that the World Bank has implemented in developing countries. This paper sets out to integrate the two sets of literature on (i) the effects of health on productivity and labor supply in developing countries and (ii) the impact of the quality of medical care on choice of health providers and on health status. Lyn Squire Director Policy Research Department vii I Abstract This paper investigates the relationships among labor force participation, health outcomes, and the availability and quality of health care in a developing country. We develop an econometric model that addresses the demand for health care in a choice model and then link it to health status outcomes and labor force participation decisions. The econometric model has two parts to it. First, we estimate a discrete choice model to determine how ill people choose among the various providers of health care. Using the parameter estimates from this model, we calculate the expected value of the best available medical facility which we then use as a measure of the quality of health care available to each individual. In the second stage, health is allowed to affect labor force participation in a simultaneous equations probit model where the potential endogeneity of reported health is controlled for with a set of instruments that includes the constructed health care quality measure. We report the results of simulating the impact on health outcomes and labor force participation of policy changes such as improving access to and quality of health care as well as reducing the price of health services. We use data from the 1989 Jamaican Survey of Living Conditions, the 1989 Jamaican Labour Force Survey, and a sample (also from 1989) of Jamaican health care facilities. We find that the quality of health facilities has a small improving effect on health but no significant effect on labor force participation. ix Acknowledgments This project was carried with financial support from the World Bank Research Committee (RPO 676-44). The Statistical Institute of Jamaica and the Ministry of Health carried out the extremely demanding field work for the survey, and made the data available to us. We have benefitted from the comments of John Newman, Andrew Foster, Mark Pitt, Mead Over, and other participants of a World Bank seminar. We are indebted to Margaret Grosh for her help and insights in understanding the sampling frame and the data from various surveys. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and should not be attributed in any manner to the World Bank. xi I. Introduction The potential market and productivity returns to improving adult health in developing countries is often noted as a major economic benefit of improving access to and quality of health care services. Several recent studies have demonstrated the link, mainly in the agricultural sector and in rural areas, between various measures of adult health and market productivity. Pitt and Rosenzweig (1986) have used a profit function approach to measure the effect of family morbidity on farm profits for a sample of farm households in Indonesia. They find no statistically significant effects of the number of sick days on farm profits but they find a strong effect on labor supply. In Strauss (1986), it appears that nutrient intake (a determinant of nutritional and health status) raises the current labor productivity of farm families in rural Sierra Leone. Deolalikar (1988) and Behrman and Deolalikar (1989) both have shown that health directly affects productivity and wage rates in low-income environments. Pitt and others (1990), using household data from Bangladesh, also have identified significant pecuniary returns to health as exhibited in linkages between health levels and productivity. All of the above studies have used nutritional and anthropometric measures as indicators of adult health, with the exception of Pitt and Rosenzweig who also examined the incidence of morbidity. More recently, Schultz and Tansel (1992) have examined the relationship between the incidence and length of morbidity spells on work experience in Ghana and Cote d'Ivoire. They report significant negative effects of morbidity on labor force participation and wages for men in both countries but not for women. All of these studies focus on the relationship between adult health and market returns. None of them however attempts to link the availability, quality, and utilization of health care to health and market outcomes.' Therefore, none of these results can provide a basis to evaluate the cost-effectiveness of certain health policies that impact on adult morbidity in terms of their payoff for labor productivity. The focus of this paper is to attempt to bridge this gap. This paper aims at investigating the relationships among labor force participation, health outcomes, and the availability of quality of health care in a developing country. We develop an econometric model that addresses the demand for health care in a choice model and then link it to health status outcomes and labor force participation decisions. The econometric model has two parts to it. First, we estimate a discrete choice model to determine how ill people choose among the various providers of health care. Using the parameter estimates from this model, we calculate the expected value of the best available medical facility which we then use as a measure of the quality of health care available to 1. A number of recent studies model the demand for health care in developing countries using discrete choice models (Akin and others 1985, Gertler and Van der Gaag 1990, and Lavy and Quigley 1992, are just few examples). But in very few, the impact of quality of care on utilization is estimated explicitly. Most importantly however, the related labor supply outcome, work days lost due to illness, is treated most often as an exogenous variable (measuring severity of illness) and not simultaneously determined. I each individual. In the second stage, health is allowed to affect labor force participation in a simultaneous equations probit model where the potential endogeneity of reported health is controlled for with a set of instruments that includes the constructed health care quality measure. In the comparative statics section of Chapter V (on p. 21), we report the results of simulating the impact on health outcomes and labor force participation of policy changes such as improving access to and quality of health care as well as reducing the price of health services. We use data from the 1989 Jamaican Survey of Living Conditions, the Jamaican 1989 Labour Force Survey, and a sample (also 1989) of Jamaican health care facilities. 2 H. Adult Health Status and Jamaican Health Care World Bank (1988) suggests that most Jamaicans enjoy better health than their counterparts in countries of similar income level. Life expectancy at birth is 74, and the crude death rate is 5.5 per thousand. The disease pattern underlying mortality and morbidity in Jamaica resembles that of the developed countries, being heavily biased toward chronic diseases. The leading causes of death for adults are cerebrovascular, heart, and hypertensive diseases and malignant neoplasms. AIDS is a growing health problem in Jamaica (Overholt 1992). The recent pattern of health problems suggests that Jamaica is moving out of an epidemiological environment dominated by infectious disease and malnutrition into one where the disease profile is dominated by the health conditions of adults and elderly. Three factors are central in explaining this transition: the change in the age structure due to declining fertility and mortality, changes in exposure and in magnitude of risks due to urbanization and industrialization, and changes in access to, use of, and effectiveness of curative health services that lead to changes in rates of chronic impairment and death. This epidemiological transition may have a direct effect on adult well being, but it may also lead to economic change through the impact of adult health on labor force participation and productivity. The analysis in this paper is designed to shed light on some of these economic effects. It specifically delves into the interrelationship between health care quality and adult health and their impact on labor force participation. Jamaica's health care services are provided free or at nominal charge to all citizens. The government owns and manages 95 percent of the hospital beds in the country. Public health services are delivered through 24 hospitals and 372 primary health care clinics (Lewis and Parker 1993). Each of the 14 parishes has at least one public hospital and over 20 health centers. Private health care is largely ambulatory. More than two thirds of the island's physicians are exclusively in private practice, while many of the doctors in the public sector have private afternoon practices. This system leads to wide and equitable access: over 90 percent of Jamaicans reside within ten miles of a health center, and 95 percent of all communities are within the same distance (World Bank 1988). However, recent evidence suggests that health status, at least for selected groups of the population, may be falling, perhaps as a result of the decline in public health spending between 1981/82 and 1985/86, from J$172 million to J$125 million in 1980 Jamaican dollars. Even though the government of Jamaica has placed recent priority on social sector activities, real per capita expenditures for health remained 10 percent lower in 1990/91 than their 1981/82 levels (Overholt 1992). The severe budget cuts have seriously diminished the capacity of the public sector to provide an acceptable level and quality of services in hospitals and health centers. The lack of maintenance during the years of budgetary stringency led to a severe deterioration in most infrastructure which is now in serious need of rehabilitation and modernization. Availability of medical supplies have been minimal, and essential drugs were not always available. The provision of health services has been impaired by erosion of staffing levels and quality owing to the deteriorating employment conditions relative to alternative potential employment and to severe limitations on other recurrent expenditures. 3 In pursuit of solutions to these problems, the government in Jamaica considered recently increasing user fees, particularly in hospitals, privatizing hospital and health care management, augmenting the role of private insurance in financial health costs, and restoring infrastructure and equipment in hospitals and health centers. 4 III. Data Our empirical work uses three sources of data to investigate the relationships among labor force participation, health outcomes, and the availability of quality health care. Labor force participation data are gathered in the Jamaican Labor Force Survey (LFS). The Survey of Living Conditions (SLC) reinterviews a subset of the same households and records health and demographic information. In this study, we merge data from the October, 1989, LFS and the November, 1989, SLC. Our third source of data is the Health Facility Survey, which is designed especially (by the Statistical Institute of Jamaica and the Ministry of Health) as part of the Expanded Health Module of the November, 1989, SLC. The facility survey, a one-time survey completed in August, 1990, is a census of all public health care facilities and a sample of private facilities. This section of the paper introduces the variables used in our empirical analysis which is described in Chapter IV. The Jamaican Survey of Living Conditions (SLC) is a detailed questionnaire designed to catalogue many aspects of households' economic conditions and decisions. The survey provides information regarding recent health experiences, many categories of consumption expenditures, housing characteristics, the availability of public services, and, among other things, demographic characteristics. In this paper, we use the demographic and health information and merge it with the same individuals' labor force participation data from the Labour Force Survey. Table 1 defines the SLC variables that we use and presents summary statistics for them. The first column of numbers are computed based on a sample of 9,840 individuals that report all of the necessary data for the labor force participation and health outcomes portion of our empirical work. Of the 16,642 individuals included in the Jamaican SLC roster, only 15,353 also are included in the LFS data file. We exclude 5,513 individuals that do not report their health status (the activities of daily living questions were asked only of individuals age 14 and above) which reduces our sample to 9,840 observations. More than ninety percent of the individuals who fail to report their health are children, so only about 500 adults are deleted because of missing health data. Except for being older (the average age in the original SLC sample is 26.32 years), our sample is very similar in other respects to the original sample. As is described in detail in the section on medical facility choice in Chapter IV (on p. 10), we require a subsample of the Jamaican SLC to estimate a model that predicts how persons with an illness or injury select a type of health care facility to visit. To provide data for the estimation of such a model, we select only those individuals in the SLC sample that report an illness or an injury during the last four weeks because those are the only individuals asked about health facility usage. Of the 15,353 individuals in the original SLC sample that report their parish of residence, only 2,249 report a recent illness or injury. Of these observations, 38 neglect to report whether they visited a health practitioner or in which type of health care facility the visit occurred. After deleting observations that fail to report other relevant information, 1,309 persons remain in the database that we use to estimate 5 Table 1. Definitions and Summary Statistics for the Jamaican Survey of Living Conditions Variables Health/Work Sample Facility Choice Subsample N=9,840 N=1,309 Mean Mean Variable Description (Standard Deviation) (Standard Deviation) AGE Age 36.44 47.03 (19.18) (21.17) FEMALE = 1, if person is female 0.500 0.60 EMPLOY = 1, if person is employ 0.570 0.53 EXHLTH =1, if person reports 0.31 0.15 excellent health VGHLTH = 1, if person reports very 0.26 0.19 good health GHLTH = 1, if person reports good 0.26 0.24 health FHLTH = 1, if person reports fair 0.11 0.23 health PHLTH = 1, if person reports fair 0.06 0.19 health INCOME family's annual income 3,373.47- 2,775 (9,472.18) (6,037) HHOUS =1, if person is head of 0.37 0.53 household MARRY = 1, if person is married 0.43 EDUC years of schooling (grade 8.35d completed or grade currently enrolled in) VIGACT = 1, if person reports limited 0.28 performance at vigorous activities (run, lift heavy objects) MODACT = 1, if person reports limited 0.15 performance at moderate activities (move table, do home repairs) BEND = 1, if person reports limited 0.12 ability to bend, kneel, or stoop WALKUP = 1, if person reports limited 0.14 ability to walk uphill or climb stairs 6 Health/Work Sample Facility Choice Subsample N=9,840 N=1,309 Mean Mean Variable Description (Standard Deviation) (Standard Deviation) WLKMIL = 1, if person reports limited 0.13 ability to walk more than 1 mile WLKYDS = 1, if person reports limited 0.09 ability to walk more than 100 yards ADL = 1, if person reports limited 0.05 ability to feed herself, bathe, or use toilet PRVPRIM = 1, if person visits private 0.32 primary health care facility PUBPRIM = 1, if person visits public 0.07 primary health care facility PRVSEC = 1, if person visits private 0.01 secondary health care facility PUBSEC = 1, if person visits public 0.10 secondary health care facility NOVISIT = 1, if person does not visit 0.51 any type of health care facility Notes: a. The standard deviation for the dummy variables simply equals Vf(1 - f) where f is the sample proportion. b. Among persons older than twenty years, sixty-seven percent are currently employed. c. This average excludes ten families that report annual income greater than one million dollars. d. These statistics are computed only for the persons that report years of schooling completed (or current enrollment). e. The remaining variables listed in the table are not included as covariates in the health care facility choice model. modal choices for health care facility visits.2 There is also a question in the survey identifying the specific health facility visited. Unfortunately, of the 647 people who report visiting a health care facility, the name of the facility visited is coded for only 258 of them. This may be due to survey design (visits to private facilities that are not included in the health facility sample would not be coded) or because the facility names may truly be "missing." Either way, the sample for whom we know the name of the facility visited is too small and potentially has large selection problems. Thus, we can identify only the type and the parish of the health care facility visited for our sample of 1,309 people. It is not 2. The other relevant information, listed in Table 1, includes demographic characteristics, employment, health status, and parish of residence. 7 surprising that members of the ill subsample are older and report worse health than persons in the main sample. The third source of data used in the empirical analysis consists of a survey of health care facilities in Jamaica. The facility survey records physical characteristics of the structures, personnel characteristics, a list of services offered by each facility, and an inventory of equipment and pharmaceuticals. Each facility is grouped into one of four types: private primary, public primary, private secondary, and public secondary. Primary facilities are doctors' offices, and secondary facilities are hospitals and large clinics. All secondary facilities, private and public, and all public primary care facilities are surveyed, but the sample includes only a random sample of privately owned primary care health facilities. The facility sample includes nearly twice as many public clinics and doctors' offices (362) as private ones (187) and three times as many public as private hospitals (23 versus 8). Looking back at the second column of numbers in Table 1, one can examine the distribution of health care facility visits among those persons that report a recent injury or illness. Private primary facilities are the most popular source of health care in Jamaica as nearly one-third of the SLC sample report such a visit. On the other hand, very few persons visit private hospitals (secondary types of health care facilities), and public primary facilities are visited slightly more frequently than private secondary facilities. However, more than half of the injured or ill persons in Jamaica do not visit a health practitioner in any care setting. Some characteristics of health facilities are presented in Table 2. The data reveal some interesting quality differences between public and private health care providers and between primary and secondary facilities. First, public primary and secondary health care facilities have more frequent incidences of leaky roofs, floors with holes, electrical problems, problems with plumbing, and security problems than their privately owned counterparts. Eighty-three percent of public hospitals, for instance, report security problems compared to none of the eight private hospitals. More than one quarter of public primary facilities report security problems compared to one tenth of the private doctors' offices. On the other hand, there is some evidence that public health care facilities offer more services than private facilities (for example, they more frequently offer immunizations, they have more ambulances, and they maintain a wider selection of pharmaceuticals). Additionally, public hospitals employ a larger number of physicians and registered nurses than private hospitals do. Hospitals, naturally, offer more services than primary facilities, and, while every hospital in Jamaica has electricity (private and public), only 84 percent of private doctors' offices and 74 percent of public ones can make the same claim. Finally, the ages of health care facilities in Jamaica differ substantially by type and by ownership (private versus public). On average, private primary care facilities have been operating for half as long as public ones (10. 10 versus 20.56 years). Private hospitals are about five years newer than public ones on average (23.00 versus 28.70 years). 8 Table 2. Definitions and Summary Statistics for Health Care Characteristics Variable Name Definition Mean Standard Deviation OFFSERV How long has this facility been offering services? 17.54 10.51 (in years) HAVEELEC Does the facility have electricity? 0.78 0.41 LEAKROOF Does the roof leak? 0.35 0.48 HOLEFLOR Are there holes in the floor? 0.10 0.30 ELECPROB Does the facility have electrical problems 0.18 0.38 PLUMBIN Are there plumbing problems inside the building? 0.26 0.44 TIMEHOLE How long have the floors had holes? (in months) 2.03 6.51 TIMEELEC How long have there been electrical problems? 3.28 7.96 (in months) TIMEPOUT How long has there been plumbing problems 3.02 7.68 outside the building? (in months) TIMESECU How long have there been security problems? (in 4.80 9.45 months) ACTIVEHC Is there an active community group working with 0.20 0.40 this facility? DAYSOPEN Number of weekdays open. 4.84 2.11 IMMUNIZ Does the facility offer immunizations? 0.73 0.44 NUMAMBUL Number of ambulances owned. 0.21 0.74 STAFF Total number of staff (personnel) on the 2.50 13.64 establishment/cadre. TOTLPROB Total number of problems (out of 7) the facility 1.61 1.67 is currently experiences. PCTEQUIP Percentage of equipment currently in working 59.46 16.71 condition. PCTDRUGS Percentage of 10 drugs kept in stock. 59.44 30.16 Note: The last four items of this table are constructed aggregates of the facility characteristics listed above them. 9 IV. Econometric Model The econometric model has two parts to it. First, we estimate the determinants of how people choose among health facilities. The parameter estimates from this procedure allow us to construct a measure of the quality of health care available to each person in our sample. We then use this measure, along with other variables, to estimate the determinants of health and work. The Medical Facility Choice Let y*jk be the latent value to person i of choosing the k'th medical facility of type j. We model yk as 4.1) yijk = Xpj + Zijk + e k =1,2,...,K j = 1,2,...,J where Xi are personal characteristics, Zij are characteristics of the k'th type j medical facility in the individual's parish, and e,k - iid GEV (Generalized Extreme Value).' If we observed the particular medical facility visited, then we could estimate Oj and -y using maximum likelihood or nested logit. However, we observe only the type of medical facility visited. As shown in Dubin and McFadden (1984), the expected value of visiting the best facility of type j is 4.2) y?. = E max y. 13 k ijk X. . + al.. + u.. where 4.3) I In exp{Z iy k and uij - iid EV (Extreme Value). The probability of choice j being chosen for a visit is 4.4) P . = exp{V }/ E exp{V. k} where 4.5) Vik = XiPk k C, 3. We assume people only use health facilities in their parish. 10 The log likelihood function is 4.6) L = iy In P iI j ii =E y,..V..- In Eexp{V} where y,, is the observed indicator of choice, yij =1(y > yk V k . j). The score statistics are 4.7) aL/a|. E (y. - P. )Xi and 4.8) aL/ay = a EE(y.. - P. .)(E P . Z .j) ij k ijk ijk where 4.9) P ijk exp{Zi jky}/E exp{Zim y} m and 4.10) aL/8a = EE (y - P. .I.. . ij 1)2. Estimation is by maximum likelihood estimation or method of moments. Once fl, y, and a are estimated, we can construct an estimate of the quality measure of medical care available to each person. Since y*4 is the E max yij, a good choice for the quality of medical care is k Ijk 4.11) I. = E max y#. I- . j = in exp{V } . This measure has the following attractive properties: a) it increases in each Vj, b) improvements in bad choices (i.e., where Vii is much smaller than the maximum Vjj) have very little effect on Ii, but c) improvements in good choices have large effects on t. This measure was used with mixed results in Stem (1992) and Stem (1993). Alternative simpler measures have some unattractive properties. For example, consider using an "average" facility quality index or any linear function of independent facility quality measures. This alternative has the unfortunate feature of decreasing when a very inferior (and therefore irrelevant) facility is added to the parish. In general, it does not give proper weight to the highest quality facilities. Another alternative is to use an estimate of the quality measure of 4. Proper choice of instruments for method of moments causes method of moments and maximum likelihood to be equivalent. 11 the best facility. This, though, does not allow for any unobserved variation among individuals in perceived or actual quality across facilities. For example, it assumes that distance from the facility is irrelevant. The most significant problem with using li to measure quality is that Vij cannot be measured when there are any Xi's. This is because we can estimate each Oj only relative to a base choice. Thus, what we are actually measuring in equation (4.11) is 4.12) I. = In E exp{V. . X = In exp{V } X b where b indexes the base choice. In Stern (1992) and Stern (1993), it was not clear how to choose a base choice. Here, the base choice is not to visit any type of health care facility. The other problem with using li as a measure of the quality of medical care is that one of the X's used in equation (4.11) may be the health or labor force participation of the individual. This causes an obvious problem when using li as an explanatory variable in predicting health. We solve this problem by excluding those variables in Xi that are endogenous when constructing I, in the section of Chapter V called, "Health and Labor Force Participation Results" (on p. 14). Health and Labor Force Participation This section essentially presents the model described in Stern (1989). Let p* be the latent value of working (participating) to person i and h be the latent measure of health for person i. We model p* and hi as 4. 13) p* = XPj0P + 6ph- + epi and 4.14) h = XhiOh + 6 hi A discrete indicator of p is observed, pi = 1(p* > 0). However, we observe a potentially biased indicator of health. Let h** be a latent measure of reported health, 4.15) h** = h* + 6p* + cr , and let hi = j iff cj _.. h** < cj+,, j = 0,1,...,M, co = -O, cl = 0, cm+l = o, and c2 through cm are parameters to estimate. The value of participating may affect reported health either because it affects true health (as in equation 4.14) or because people use health as an excuse not to work (as in equation 4.15). We cannot identify separately br and 6, 12 The method of estimation is a generalization of Amemiya (1978) and Heckman (1978) and is described in detail in Stern (1989). The estimation procedure does not require independence of the errors, ep, Eh, and e. First we estimate reduced form equations for pi and h*. We use other reported measures of physical well-being (reported activities of daily living, or ADLs) that are less subject to reporting bias (see Stern 1989).' Then we construct predicted values of p and h* using the reduced form parameter estimates and substitute the predicted values into equations (4.13) and (4.14). We can then estimate 61 and 0, using probit and Sh and 0h using ordered probit. Standard error construction requires a well-known correction described in Amemiya (1978), Heckman (1978), and Stern (1989). 5. One might argue that these instruments are not valid in that they are functions of some of the same unobservables that affect health. Excluding these instruments causes the coefficients in the employment equation to be identified only by the exclusion of the health facility quality variable. 13 V. Results Medical Facility Visit Results There are many choices of personal characteristics and facility characteristics to include in the specification of equations (4.2) and (4.3). Those characteristics that are discussed later in the paper are described in Tables 1 and 2. Based on preliminary probit and logit regressions, we determined that it was difficult to choose among the facility characteristics.' Thus we proceed in two different ways. First, we consider the first five principal components of the facility data. Only the second and third principal components have significant coefficients.' Second, we use four aggregated facility characteristics listed at the bottom of Table 2. Only the staff size of each facility is significant and has expected sign.8 Attempts to estimate a, the -y's, and the fl's simultaneously failed to converge. Therefore, we modified our procedure to estimate only the O's and the -y's simultaneously over a grid of values for o. To choose the best estimate of a, we evaluated several statistics: the sample log-likelihood values, two sums of squared residuals (one variance-weighted), and two hypothetical Newton-Raphson step sizes (assuming either that a was estimated alone or that it was estimated along with all of the other parameters). The final specification and coefficient estimates based on principal components are presented in Table 3. Our selection of a = 0.20 as the estimate arises because that value yields a large log-likelihood value, small sums of squared residuals, and because hypothetical step sizes for a > .2 are generally negative while hypothetical stepsizes for a < .2 are generally positive. The results indicate that a) older people are more likely to use private hospitals; b) women are more likely to use primary facilities and less likely to use secondary facilities relative to men, and c) the better one's health, the less likely one is to use any facility. Further analysis is delayed until the comparative statics section of this chapter (on p. 21). The final specification and coefficient estimates based on staff size are presented in Table 4. Results are almost exactly the same. The estimates in Tables 3 and 4 can be used to construct li in equation (4.11) for each person which is used to explain reported health outcomes in the next section. Health and Labor Force Participation Results Table 5 presents estimates of the structural coefficients for three different health equation specifications when the principal components of facility characteristics are used to 6. The facility characteristics were highly multicollinear, and estimates were very sensitive to the specification used. 7. The quality estimates do not change if we include all of the first five principle components. 8. This was based on much experimentation with different specifications of equations (4.2) and (4.3). Including the other three aggregate facility characteristics does not affect computed facility quality significantly. 14 Table 3. Medical Facility Choice Equation Using Principal Components of Facility Characteristics Personal Characteristics Variable Prv Prim Pub Prim Prv Sec Pub Sec CONSTANT -0.860* -2.293* -4.183* -1.601* (0.306) (0.527) (1.650) (0.479) AGE/10 -0.024 -0.024 0.460* -0.049 (0.037) (0.064) (0.204) (0.058) FEMALE 0.372* 0.691* -0.870 0.040 (0.135) (0.260) (0.583) (0.209) EMPLOY 0.194 -0.355 -0.192 -0.041 (0.136) (0.245) (0.624) (0.212) EXHLTH -0.530* -0.570 -0.425 -0.998* (0.259) (0.434) (1.277) (0.409) VGHLTH -0.690* -1.227* -1.331 -1.059* (0.241) (0.451) (1.213) (0.367) GHLTH -0.171 -0.394 -0.744 -0.765* (0.213) (0.353) (0.906) (0.333) FHLTH -0.006 -0.594 -0.002 -0.139 (0.201) (0.347) (0.666) (0.290) INCOME 0.005 0.002 -0.156 -0.033 (0.010) (0.019) (0.129) (0.023) Medical Facility Characteristics of 0.20 72 -0.1794 (0.092) 'Y3 0.670 (0.238) Notes: 1) Starred items are statistically significant. Numbers in parentheses are standard errors. 2) Variables are defined in Table 1. 3) -y, is the coefficient on the i'th principal component of the facility characteristics. a is defined in equation (4.2). 15 Table 4. Medical Facility Choice Equation Using Staff Size of Facility Personal Characteristics Variable Prv Prim Pub Prim Prv Sec Pub Sec CONSTANT -1.485* -2.955* -4.692* -1.641* (0.304) (0.526) (1.694) (0.485) AGE/10 -0.017 -0.024 -0.487* -0.037 (0.037) (0.066) (0.210) (0.058) FEMALE 0.377* 0.690* -0.867 0.032 (0.136) (0.261) (0.591) (0.212) EMPLOY 0.195 -0.362 -0.230 -0.014 (0.136) (0.245) (0.638) (0.214) EXHLTH -0.496* -0.562 -0.126 -0.914* (0.260) (0.435) (1.310) (0.413) VGHLTH -0.687* -1.211* -1.220 -1.015* (0.241) (0.452) (1.221) (0.371) GHLTH -0.183 -0.386 -0.711 -0.819* (0.214) (0.354) (0.911) (0.342) FHLTH -0.014 -0.582 0.143 -0.125 (0.202) (0.347) (0.680) (0.293) INCOME 0.006 -0.001 -0.157 -0.030 (0.010) (0.020) (0.130) (0.023) Medical Facility Characteristics a 0.40 'y 0.003111* (0.000553) Notes: 1. Starred items are statistically significant. Numbers in parentheses are standard errors. 2. Variables are defined in Table 1. 3. y is the coefficient on total number of staff in the health facility. a is defined in equation (4.2). 16 Table 5. Health Equations Using Principal Components of Facility Characteristics Variable Column 1 Column 2 Column 3 CONSTANT -0.314 0.041 1.820* (0.197) (0.200) (0.222) FEMALE 0. 108* -1.300* -0.837* (0.040) (0.069) (0.062) HHOUS -0.036 0.348* 0.171* (0.031) (0.034) (0.032) MARRY -0.051 -0.174* -0.151* (0.027) (0.027) (0.027) AGE/10 0.044 1.551* 0.997* (0.038) (0.071) (0.059) AGE*AGE/100 0.032* -0.157* -0.088* (0.004) (0.008) (0.007) EDUC -0.011 0.004 0.001 (0.007) (0.007) (0.007) MFCVAL 0.394* -0.806* -0.265 (0.179) (0.187) (0.183) EMPLOY -0.171* (0.030) EMPHAT -1.652* (0.064) EMPHATC -3.800* (0.181) C2 0.754* 0.765* 0.762* (0.015) (0.015) (0.015) C3 1.773* 1.828* 1.813* (0.024) (0.025) (0.025) C4 2.652* 2.828* 2.770* (0.034) (0.038) (0.037) Log Likhd -9670.4 -9406.2 -9492.5 Notes: 1. Starred items are statistically significant. Numbers in parentheses are standard errors. 2. Variables are defined in Table 1. 3. EMPLOY is the discrete indicator of employment. EMPHAT is the predicted value of the latent value of working. EMPHATC is the predicted expected value of EMPLOY. MFCVAL is the constructed quality of medical facilities based on the second and third principal components of the facility characteristics. 17 measure health facility quality. The first column presents estimates using the discrete indicator of employment as a regressor without controlling for endogeneity. Coefficients should be interpreted using equation (4.14) where increases in h* imply declining health. As expected, women and older people report being in worse health. Contrary to Stern (1989), employment has a significant effect on health. This is true even after controlling for endogeneity in the last two columns. This result has two possible interpretations. Either nonemployed people use poor health as an "excuse" for not working (see Parsons 1980 or Bazzoli 1985) or working actually improves one's health. Also, the first column implies that improving the quality of medical facilities (MFCVAL) diminishes health. The second column of Table 5 uses the predicted value of p (the value of working) based on reduced form coefficients instead of the discrete indicator of employment p, to control for the potential endogeneity of pi. When the potential endogeneity of labor force participation is controlled for in this way, improving the quality of medical facilities actually leads to improvements in health. The estimated coefficients in all other variables are essentially unchanged. The fit for this equation is better as well (as represented by the improvement in the log likelihood). The last column of Table 5 uses the reduced form predictor of pi, P(pi), to control for potential endogeneity of p, in a different way. This specification looks similar to the specification using Pi except that a) MFCVAL becomes insignificant (but remains negative) and b) the specification does not fit as well as column 2. For the remainder of the paper, we will use the specification in the second column. Table 6 presents health equation results when MFCVAL depends upon facility staff sizes. Results are almost exactly the same. However the estimates on MFCVAL become smaller and less significant. Still the preferred specification, the second column of Table 6, has a negative, significant coefficient estimate. Table 7 presents estimates of the structural coefficients for three different specifications of the employment equation when principal components are used. The first column has estimates using the discrete indicators of health as regressors without controlling for endogeneity. Coefficients should be interpreted using equation (4.13). Being female, married, or in bad health reduces the value of working. Aging increases the value of working up until age 41 and decreases it thereafter. The next two columns control for the possible endogeneity of health status. The second column uses the reduced form predicted value of hi as a regressor, and the third column uses the reduced form predicted values of dummies for the different values of h,. The results in the third equation are not reliable given the nonmonotone coefficient estimates on the health variables. But the first two equations are very similar implying that endogeneity of the health variables is not a qualitatively significant problem. Estimates based on facility staff size are almost identical and are not presented. 18 Table 6. Health Equations Using Staff Size of Facility Variable Column I Column 2 Column 3 CONSTANT -0.222 -0.485* 1.576* (0.121) (0.124) (0.150) FEMALE 0.117* -1.391* -0.873* (0.030) (0.067) (0.058) HHOUS -0.036 0.347* 0.169* (0.031) (0.034) (0.032) MARRY -0.054* -0.175* -0.152* (0.027) (0.027) (0.027) AGE/10 0.040 1.574* 0.998* (0.037) (0.072) (0.059) AGE*AGE/100 0.032* -0.160* -0.088* (0.004) (0.009) (0.007) EDUC -0.011 0.004 0.000 (0.007) (0.007) (0.007) MFCVAL 0.358* -0.310* -0.018 (0.087) (0.092) (0.089) EMPLOY -0.169* (0.030) EMPHAT -1.657* (0.064) EMPHATC -3.781* (0.182) C2 0.755* 0.765* 0.763* (0.015) (0.015) (0.015) C3 1.775* 1.828* 1.813* (0.024) (0.025) (0.025) C4 2.654* 2.828* 2.770* (0.034) (0.038) (0.037) Log Likhd -9664.2 -9405.1 -9490.5 Notes: 1. Starred items are statistically significant. Numbers in parentheses are standard errors. 2. Variables are defined in Table 1. 3. EMPLOY is the discrete indicator of employment. EMPHAT is the predicted value of the latent value of working. EMPHATC is the predicted expected value of EMPLOY. MFCVAL is the constructed quality of medical facilities based on facility staff size. 19 Table 7. Work Equations Using Principal Components of Facility Characteristics Variable Column 1 Column 2 Column 3 CONSTANT -2.432* -1.131* -3.231* (0.351) (0.365) (0.458) FEMALE -0.784* -0.839* -0.784* (0.062) (0.062) (0.065) HHOUS 0.245* 0.230* 0.234* (0.044) (0.044) (0.044) MARRY -0.107* -0.103* -0.115* (0.042) (0.042) (0.042) AGE/10 0.964* 0.969* 0.799* (0.073) (0.073) (0.092) AGE*AGE/100 -0.118* -0.107* -0.091* (0.007) (0.008) (0.009) EDUC 0.013 0.010 0.009 (0.010) (0.010) (0.011) EXHLTH 0.557* -0.257 (0.094) (0.574) VGHLTH 0.567* 7.196* (0.093) (1.748) GHLTH 0.498* -1.487 (0.089) (0.854) FRHLTH 0.411* 2.733* (0.092) (1.073) HLTHHAT -0.376* (0.047) EMPRATE 1.671* 0.860* 1.489* (0.321) (0.339) (0.403) ANNEARN/1000 -0.311* -0.318* -0.322* (0.065) (0.066) (0.066) Log Likhd -2755.7 -2743.3 -2733.2 Notes: 1. Starred items are statistically significant. Numbers in parentheses are standard errors. 2. Variables are defined in Table 1. 3. EXHLTH, VGHLTH, GHLTH, and FRHLTH are discrete indicators of excellent health, very good health, good health, and fair health respectively. The first column uses the reported values, and the third column uses the predicted values. HLTHHAT is the predicted value of the latent value of health. MFCVAL is the constructed quality of medical facilities based on the second and third principal components of the facility characteristics. 20 Comparative Statics In this section, we use the estimates from Column 2 of Tables 5 and 6 and Column 2 of Table 7 to perform some comparative static experiments. The experiments involve the effects of changing health care facility characteristics on health and labor force participation. Before proceeding we must decide whether the coefficient on employment, EMPHAT, in the health equation represents a real effect, a psychological effect, or both; this corresponds to assumptions about whether 6h in equation (4.14) or 6, in equation (4.15) are zero. If it is only a psychological effect, then changing health facility characteristics affect health only through the effect of MFCVAL on health, and they affect employment through the indirect effect of MFCVAL on health and health on employment. On the other hand, if it is also partially a real effect, then the health and employment equation must be solved simultaneously to measure the total effect of changes in MFCVAL on health and employment. Table 8 reports the effects on health and employment of increasing each listed characteristic by one at every health facility in Jamaica. For example, the health coefficient for LEAKROOF, 0.836, is the effect on health of starting with no leaky roof at any health facility in Jamaica and then causing every health facility in Jamaica to have a leaky roof. All rows other than STAFF/10 use the MFCVAL constructed from the principal components, and STAFF/10 uses the MFCVAL constructed from STAFF. The first two columns correspond to assuming that the coefficient on EMPHAT in the health equation is a psychological effect, and the last two correspond to assuming it is a real effect. Excluding the STAFF/10 row, the health facility characteristics included in the table include all of those listed in Table 2 that have the expected sign. "Good" health facility characteristics should decrease health and increase employment, and "bad" health facility characteristics should work the opposite way. These results should not be interpreted to mean that 14 variables independently have the correct sign because in fact there are only two degrees of freedom here (only two principal components of the health facility characteristics had significant coefficients). This lack of degrees of freedom is a serious problem with the data. It could be solved if there were information identifying which specific health facility each person used (rather than the available information about the type of health facility he used). Without the former information, there are only fourteen (the number of parishes) different health facility environments. With this information along with more specific geographic information, there would be a separate health facility environment for each person in the SLC. All of the effects reported in Table 8 are multiplied by 10. The results in the first two columns indicate that some health facility characteristics have significant effects on latent health but that none have a significant effect on the value of working (and therefore labor force participation). The results in the last two columns indicate that no effects are significant when there is a true employment effect on health. The effects in the last two columns are large but insignificant because they represent dividing the effects in the first two 21 Table 8. Comparative Statics With No Employment Effect With Employment Effect Variable Health Employment Health Employment OFFSERV -0.048 0.018 -0.127 0.048 (0.027) (0.017) (0.200) (0.171) HAVEELEC -0.511 0.192 -1.349 0.507 (0.340) (0.188) (2.235) (1.820) LEAKROOF 0.836 -0.314 2.206 -0.829 (0.542) (0.305) (3.626) (2.975) HOLEFLOR 1.720* -0.647 4.541 -1.707 (0.850) (0.579) (6.965) (6.108) ELECPROB 0.301 -0.113 0.794 -0.298 (0.224) (0.116) (1.364) (1.074) PLUMBIN 0.262 -0.099 0.692 -0.260 (0.556) (0.214) (2.148) (1.086) TIMEHOLE 0.925* -0.348 2.441 -0.918 (0.276) (0.299) (3.382) (3.318) TIMEELEC 1.108* -0.417 2.925 -1.100 (0.538) (0.372) (4.470) (3.935) TIMEPOUT 0.227* -0.086 0.600 -0.226 (0.102) (0.075) (0.902) (0.807) TIMESECU 0.271* -0.102 0.715 -0.269 (0.135) (0.091) (1.099) (0.962) ACTIVEHC -11.235* 4.224 -29.655 11.150 (5.677) (3.801) (45.705) (39.892) DAYSOPEN -0.168* 0.063 -0.444 0.167 (0.085) (0.057) (0.684) (0.598) IMMUNIZ -4.276* 1.608 -11.288 4.244 (1.695) (1.385) (16.599) (15.192) NUMAMBUL -5.463* 2.054 -14.420 5.422 (1.757) (1.738) (20.509) (19.459) STAFF/10 -0.039 0.015 -0.102 0.038 (0.084) (0.034) (0.709) (0.302) Notes: 1. All numbers are multiplied by 10. 2. Numbers in parentheses are standard errors. Starred items are significant at the 5 percent level. 22 columns by 1 - Ah p [see equations (4.13) and (4.14)]. This inflates the predicted effects but inflates standard errors even more. For the rest of this section, we focus on health effects assuming the effect of EMPHAT on health is psychological (Column 1 of Table 8). When the STAFF variable is used along with its corresponding construction of MFCVAL, the effect has the expected sign, but it is statistically insignificant. It is not yet clear how changes in the latent measure of health translate into changes in the distribution of reported health. Table 9 reports changes in the distribution of reported health (multiplied by 100) for each of the experiments reported in Table 8. For example, if each health facility in Jamaica were to acquire one more ambulance, the proportion of the population reporting excellent health would increase 4.95 percent and the proportion reporting poor health would decrease 2.81 percent. Again, these numbers are only suggestive because of the principal components-low degrees of freedom problem. Table 10 reports the results of improving the availability of health care facilities (multiplied by 100). Assume, for the sake of exposition, that the Health Facility Survey is actually a census of health facilities; i.e., all health facilities are observed. Consider adding one health care facility of a particular type in a particular parish of equal quality to the best facility of that type already existing in that parish. Table 10 reports the effects of performing such an experiment for each facility type and each parish. The effects are somewhat small (compared to those in Table 8). They are largest for public hospitals and smallest for public doctor's offices. To a large degree this is because there are very few public hospitals and many public doctor's offices. In particular, the results of regressing the log effects on log number of facilities, parish dummies, and facility type dummies are' 5.1) log Effect = 0.238 - 0.792 log NumFacs + 1.074 PrvPrm (0.099) (0.022) (0.063) + 0.009 PubPrm - 0.838 PubSec + e (0.061) (0.096) The regression indicates that adding another facility of given quality has a larger effect in parishes that currently have fewer facilities. This result is not surprising given the formula for the quality of medical care in equations (4.3) and (4.11). Whether it is better policy to increase the number of health facilities or improve already existing facilities depends on results in Tables 8 and 10 and on the relative costs of each type of improvement (we have no data on input prices). We also could consider the effect of making health care less expensive by transforming a private facility into a public facility (people do not pay for care provided in public facilities). For example, if the best private hospital in parish 8 became public (and 9. "log Effect" is the log of the reported effects in Table 10. "log NumFacs" is the log of the number of existing facilities of that type in that parish. "PrvPrm," "PubPrm," and "PubSec" are dummies for private primary, public primary, and public secondary facility types respectively. Parish dummies are not reported and are jointly not significant. The R2 is .985. 23 Table 9. Changes in the Distribution of Health Variable Excellent Very Good Good Fair Poor OFFSERV 0.044 0.005 -0.016 -0.008 -0.025 (0.024) (0.003) (0.009) (0.005) (0.014) HAVEELEC 0.463 0.057 -0.168 -0.089 -0.263 (0.309) (0.038) (0.112) (0.060) (0.175) LEAKROOF -0.758 -0.093 0.274 0.146 0.430 (0.492) (0.060) (0.178) (0.095) (0.279) HOLEFLOR -1.560* -0.191* 0.564* 0.301* 0.885* (0.771) (0.094) (0.279) (0.149) (0.437) ELECPROB -0.273 -0.033 0.099 0.053 0.155 (0.203) (0.025) (0.074) (0.039) (0.115) PLUMBIN -0.238 -0.029 0.086 0.046 0.135 (0.505) (0.062) (0.183) (0.097) (0.286) TIMEHOLE -0.839* -0.103* 0.303* 0. 162* 0.476* (0.250) (0.031) (0.090) (0.048) (0.142) TIMEELEC -1.005* -0.123* 0.364* 0. 194* 0.570* (0.488) (0.060) (0.176) (0.094) (0.277) TIMEPOUT -0.206* -0.025* 0.075* 0.040* 0. 117* (0.092) (0.011) (0.033) (0.018) (0.052) TIMESECU -0.246* -0.030* 0.089* 0.047* 0.139* (0.122) (0.015) (0.044) (0.024) (0.070) ACTIVEHC 10.189* 1.248* -3.685* -1.968* -5.782* (5.148) (0.630) (1.862) (0.994) (2.921) DAYSOPEN 0.153* 0.019* -0.055* -0.029* -0.087* (0.077) (0.009) (0.028) (0.015) (0.044) IMMUNIZ 3.878* 0.475* -1.403* -0.749* -2.201* (1.537) (0.188) (0.556) (0.297) (0.872) NUMAMBUL 4.954* 0.607* -1.792* -0.957* -2.811* (1.593) (0.195) (0.576) (0.308) (0.904) STAFF/10 0.038 -0.000 -0.012 -0.006 -0.019 (0.083) (0.000) (0.025) (0.014) (0.042) Notes: 1. All numbers are multiplied by 100. 2. Numbers in parentheses are standard errors. Starred items are significant at the 5 percent level. 24 Table 10. Availability. Comparative Statics Parish PrvPrm PubPrm PrvSec PubSec 1 -0.612 -0.135 -0.518 -0.810 (9) (13) (1) (3) 2 -0.240 -0.066 -0.165 -0.541 (35) (36) (4) (4) 3 -0.937 -0.183 na -1.340 (6) (15) (0) (1) 4 -0.594 -0.154 na -1.239 (11) (18) (0) (1) 5 -0.494 -0.075 na -0.832 (13) (30) (0) (2) 6 -0.504 -0.122 na -1.259 (12) (28) (0) (1) 7 -1.072 -0.149 na -1.366 (7) (21) (0) (1) 8 -0.499 -0.081 -0.323 -1.176 (15) (27) (2) (1) 9 -0.724 -0.131 na -1.377 (8) (19) (0) (1) 10 -0.459 -0.103 na -1.282 (13) (23) (0) (1) 11 -0.426 -0.100 na -1.258 (15) (29) (0) (1) 12 -0.589 -0.086 -0.756 -0.677 (14) (34) (1) (2) 13 - -0.454 -0.066 na -0.732 (14) (44) (0) (2) 14 -0.429 -0.090 na -0.827 (15) (25) (0) (2) Mean -0.574 -0.110 -0.448 -1.051 St.Dev. 0.215 0.036 0.259 0.296 Notes: 1. All effects are multiplied by 100. 2. Numbers in parentheses are the number of facilities of that type in that parish. 25 changed characteristics so that it appeared similar to the best existing public hospital in parish 8), health would improve by .01176 - .00323 = .00853. Such an improvement leads to slightly better outcomes than fixing leaky roofs in all health facilities. Results are similar for secondary facilities in other parishes, but they reverse for primary facilities. This implies that the private primary facilities are of sufficiently higher quality to more than make up for their higher prices. None of the results discussed in this paragraph would be relevant if the Jamaican government subsidized health care in such a way that the quality of private facilities did not diminish. 26 VI. Conclusions This paper has demonstrated how a measure of quality of health facilities can be constructed from available data so that it has reasonable properties. In particular, it is increasing in the quality of each relevant health facility, increasing with respect to the addition of new facilities, and weighted toward the best facilities. Furthermore, it has an appealing, intuitive interpretation. We show how to construct this measure and estimate its parameters using multinomial logit. Then we use it to help explain variation in health and labor force participation. The results of this paper show that improving the quality of health facilities has a significant but small effect on health and an indirect effect (through health) on labor force participation. Suggestive comparative statics simulations suggest an even tradeoff between improving existing facilities and building new, high quality facilities (ignoring cost). It is essentially impossible to say how to most efficiently improve health facilities unless a) data is gathered that can match individuals in the SLC with particular health facilities in the health facility census and b) data is gathered on prices of health facility inputs. Another limitation of the SLC data is that we do not know the nature of a person's illness or injury. Certainly, this information would significantly help to explain a person's choice of facility, especially the type of facility chosen. The collection of such data would increase the value of the current Jamaican data by an order of magnitude both to researchers and to Jamaican policymakers. 27 References Akim, John S., C. Griffin, D.K. Guilkey, and B.M. Popkin (1985). The Demand for Primary Health Care in the Third World. N.J.: Littlefield, Adams. Amemiya, T. (1978). "The Estimation of a Simultaneous Equations Generalized Probit Model." Econometrica. 46(5): 1193-1205. Bazzoli, Gloria (1985). "The Early Retirement Decision: New Empirical Evidence on the Influence of Health." Journal of Human Resources. 20(2): 214-234. Behrman, Jere and Anil Deolalikar (1988). "Agricultural Wages in India: The Role of Health, Nutrition and Seasonality," in David E. Shan, ed., Seasonal Variability in Third World Agriculture. The Consequences For Food Security. Baltimore: Johns Hopkins University Press. Deolalikar, Anil B. (1988). "Nutrition and Labor Productivity in Agriculture: Estimates for Rural South India." Review of Economics and Statistics. 70: 406-413. Dubin, Jeffrey and Daniel McFadden (1984). "An Econometric Analysis of Residential Electric Appliance Holdings and Consumption." Econometrica. 52(2): 345-362. Gertler, Paul and Jacques van der Gaag (1990). The Willingness to Pay for Medical Care. A World Bank Publication. Washington, D.C. Heckman, J. (1978). "Dummy Endogenous Variables in a Simultaneous Equations System." Econometrica. 46(4): 931-959. Lavy, Victor and John Quigley (1992). Willingness to Pay for the Quality and Intensity of Medical Care: Low Income Households in Ghana. Living Standards Measurement Study Working Paper No. 94. World Bank, Washington, D.C. Lewis, Maureen A. and Clover Parker (1993). "Policy and Implementation of User Fees in Jamaican Public Hospitals." World Bank, Washington, D.C. Overholt, Catherine (1992). "Literature Review of the Jamaican Public Health Sector." Collaborative For Development Action, Inc. Parsons, Donald (1980). "The Decline in Male Labor Force Participation." Journal of Political Economy. 88(1): 117-134. Pitt, Mark M. and Mark R. Rosenzweig (1986). "Agriculture Prices, Food Consumption, and the Health and Productivity of Farmers," in Indejit Singh, Lyn Squire, and John Strauss, eds., Agricultural Household Models: Extensions, Applications, and the Policy. World Bank, Washington, D.C. 29 Pitt, Mark M., M. Rosenzweig, and M.D. Nazmul Hassan (1990). "Productivity, Health, and Inequality in the Intrahousehold Distribution of Food in Low Income Countries." American Economic Review. 1139-1156. Schultz, Paul and A. Tanzel (1992). Measurement of Returns to Adult Health: Morbidity Effects on Wage Rates in Cote d'Ivoire and Ghana. Living Standards Measurement Study Working Paper No. 95. World Bank, Washington, D.C. Stem, S. (1989). "Measuring the Effect of Disability on Labor Force Participation." Journal of Human Resources. 24(3): 361-395. Stern, S. (1992). "Discrete Choice Models with Unobserved Heterogeneity: Demand for Transportation by Elderly and Disabled People." University of Virginia working paper. Stern, S. (1993). "A Disaggregate Discrete Choice Model of Transportation Demand by Elderly and Disabled People in Rural Virginia." Transportation Research, forthcoming. Strauss, John (1986). "Does Better Nutrition Raise Farm Productivity." Journal of Political Economy. 94(2): 297-320. World Bank (1988). Jamaica Summary Review of the Social Well-Being Program. Washington, D.C. 30 LSMS Working Papers (continued) No. 79 Thomas, Gender Differences in Household Resource Allocations No. 80 Grosh, The Household Survey as a Tool for Policy Change: Lessons from the Jamaican Survey of Living Conditions No. 81 Deaton and Paxson, Patterns of Aging in Thailand and C6te d'Ivoire No. 82 Ravallion, Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia No. 83 Ravallion and Datt, Growth and Redistribution Components of Changes in Poverty Measure: A Decomposition with Applications to Brazil and India in the 1980s No. 84 Vijverberg, Measuring Income from Family Enterprises with Household Surveys No. 85 Deaton and Grimard, Demand Analysis and Tax Reform in Pakistan No. 86 Glewwe and Hall, Poverty and Inequality during Unorthodox Adjustment: The Case of Peru, 1985-90 No. 87 Newman and Gertler, Family Productivity, Labor Supply, and Welfare in a Low-Income Country No. 88 Ravallion, Poverty Comparisons: A Guide to Concepts and Methods No. 89 Thomas, Lavy, and Strauss, Public Policy and Anthropometric Outcomes in C6te d'Ivoire No. 90 Ainsworth and others, Measuring the Impact of Fatal Adult Illness in Sub-Saharan Africa: An Annotated Household Questionnaire No. 91 Glewwe and Jacoby, Estimating the Determinants of Cognitive Achievement in Low-Income Countries: The Case of Ghana No. 92 Ainsworth, Economic Aspects of Child Fostering in C6te d'Ivoire No. 93 Lavy, Investment in Human Capital: Schooling Supply Constraints in Rural Ghana No. 94 Lavy and Quigley, Willingness to Pay for the Quality and Intensity of Medical Care: Low-Income Households in Ghana No. 95 Schultz and Tansel, Measurement of Returns to Adult Health: Morbidity Effects on Wage Rates in C6te d'Ivoire and Ghana No. 96 Louat, Gosh, and van der Gaag, Welfare Implications of Female Headship in Jamaican Households No. 97 Coulombe and Demery, Household Size in Cbte d'lvoire: Sampling Bias in the CILSS No. 98 Glewwe and Jacoby, Delayed Primary School Enrollment and Childhood Malnutrition in Ghana: An Economic Analysis No. 99 Baker and Grosh, Poverty Reduction through Geographic Targeting: How Well Does It Work? No. 100 Datt and Ravallion, Income Gains for the Poor from Public Works Employment: Evidence from Two Indian Villages No. 101 Kostermans, Assessing the Quality of Anthropometric Data: Background and Illustrated Guidelines for Survey Managers No. 102 van de Walle, Ravallion, and Gautam, How Well Does the Social Safety Net Work?: The Incidence of Cash Benefits in Hungary, 1987-89 No. 103 Benefo and Schultz, Determinants of Fertility and Child Mortality in C6te d'lvoire and Ghana No. 104 Behrman and Lavy, Children's Health and Achievement in School No. 105 Lavy and Germain, Quality and Cost in Health Care Choice in Developing Countries No. 106 Lavy, Strauss, Thomas, and De Vreyer, The Impact of the Quality of Health Care on Children's Nutrition and Survival in Ghana No. 107 Hanushek and Lavy, School Quality, Achievement Bias, and Dropout Behavior in Egypt No. 108 Feyisetan and Ainsworth, Contraceptive Use and the Quality, Price, and Availability of Family Planning in Nigeria No. 109 Thomas and Maluccio, Contraceptive Choice, Fertility, and Public Policy in Zimbabwe No. 110 Ainsworth, Beegle, and Nyamete, The Impact of Female Schooling on Fertility and Contraceptive Use: A Study of Fourteen Sub-Saharan Countries No. 111 Oliver, Contraceptive Use in Ghana: The Role of Service Availability, Quality, and Price No. 112 Montgomery, Kouam6, and Oliver, The Tradeoff between Number of Children and Child Schooling: Evidence from C6te d'Ivoire and Ghana No. 113 Menno Pradhan, Sector Participation Decisions in Labor Supply Models No. 114 Beegle, The Quality and Availability of Family Planning Services and Contraceptive Use in Tanzania No. 115 Lavy, Spratt, and Leboucher, Changing Patterns of Illiteracy in Morocco THE WORLD BANK A partner in strengthening economies and expanding markets to improve the quality of life for people everywhere, especially the poorest I -J15 Y~- 1 .5