Nutritional status in Ghana and its determinants

I-FIE 8706 SOCIAL DIMENSIONS OF ADJUSTMENT IN SUB-SAHARAN AFRICA W@lUM PAPE 1K. 9 Nutritional Status in Ghana and its Determinants Harold Alderman SDA Working Paper Series Editorial Board Chairman Ismail Serageldin Director, Occidental and Central Africa Department World Bank Members Ramesh Chander, Statistical Adviser, World Bank Dennis de Tray, Research Administrator, World Bank Yves Franchet, Director General, Statistical Office of the European Communities Ravi Kanbur, Editor of World Bank Economic Review and World Bank Research Observer and Senior Adviser, SDA Unit, World Bank Gabriel Kariisa, Chief Economist, African Development Bank F. J. C. Klinkenberg, Director, Directorate General for Development, Commission of European Communities Jacques Loup, Coordinator of Assistance to Developing Countries, United Nations Development Programme E. M. Morris-Hughes, Chief, Nutrition Planning, Assessment and Evaluation Service, Food and Agriculture Organization F. Stephen O'Brien, Chief Economist, Africa Region, World Bank Graham Pyatt, Professor of Economics, University of Warwick Paul P. Streeten, Director, World Development Institute, Boston University Victor E. Tokman, Director, Employment and Development Department, International Labour Office R. van der Hoeven, Senior Adviser, United Nations Children's Fund Editor Michel No6l Chief, SDA Unit World Bank Managing Editors Marco Ferroni Christiaan Grootaert Senior Economist, SDA Unit Senior Economist, SDA Unit World Bank World Bank SOCIAL DIMENSIONS OF ADJUSTMENT IN SUB-SAHARAN AFRICA WORKING PAPER NO. 3 Policy Analysis Nutritional Status in Ghana and its Determinants Harold Alderman The World Bank Washington, D.C. Copyright 0 1990 The World Bank Papers 1 and 2 in this 1818 H Street, N.W. series did not carry paper Washington, D.C. 20433, U.S.A. identification numbers. All rights reserved Manufactured in the United States of America First printing May 1990 The findings, interpretations, and conclusions expressed in this paper are entirely those of the author 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. In order to present the results of research with the least possible delay, the manuscript has not been edited in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. The material in this publication is copyrighted. Requests for permission to reproduce portions of it should be sent to Director, Publications Department, 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 reproduction is for noncommercial purposes, without asking a fee. Permission to photocopy portions for classroom use is not required, though notification of such use having been made will be appreciated. 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 Publications Sales Unit, Department F, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66 avenue d'Ina, 75116 Paris, France. ISSN 1014-739X Harold Alderman is a senior research associate, Cornell University Food and Nutrition Policy Program, and research fellow at the International Food Policy Research Institute Library of Congress Cataloging-in-Publication Data Alderman, Harold, 1948- Nutritional status in Ghana and its determinants. (Social dimensions of adjustment in Sub-Saharan Africa, ISSN 1014-739X; working paper no. 3. Policy analysis) Includes bibliographical references. 1. Poor-Ghana. 2. Malnutrition-Ghana. 3. Food supply--Ghana. 4. Ghana-Economic conditions- 1979- . 5. Ghana-Social conditions. I. Title. H. Series: Social dimensions of adjustment in Sub-Saharan Africa; working paper no. 3. III. Series. Social dimensions of adjustment in Sub-Saharan Africa; working paper no. 3. Policy analysis. HC1060219P623 1990 363.8'09667 90-12441 ISBN 0-8213-1555-2 SDA Working Paper Series Foreword Integration of social and poverty concerns in the struc- economic crisis in Africa on the one hand and the tural adjustment process in Sub-Saharan Africa is a adjustment response on the other hand affect the liv- major driving force behind the design of the World ing conditions of people. Empirically, major improve- Bank's adjustment lending program in the Region. To ments are needed in our knowledge of the social further the goal, the Social Dimensions of Adjustment dimensions of life in Africa, how they change, and (SDA) Project was launched in 1987, with the United whether all groups in society participate effectively in Nations Development Programme and the African the process of economic development. Gaining this Development Bank as partners. Since then many other knowledge will demand new efforts in data collection multilateral and bilateral agencies have supported the and policy oriented analysis of these data. Most im- project financially as well as with advice. The task portantly, policy actions are needed in the short term presents a formidable challenge because of the sever- to absorb undesirable side-shocks stemming from the ity of economic and social constraints in Africa and the adjustment process so that the poor and disadvan- intrinsic difficulty of tracing the links between eco- taged are not unduly hurt, and in the long term to nomic policies and social conditions and poverty. It is ensure that these groups fully participate in the newly essential to have a continuous professional dialogue generated growth. The SDA Project's mandate is to between all concerned parties, so that the best ideas operate, in a concerted way, in all three domains: get discussed by the best minds, and become, as concepts, data, actions. This working paper series will quickly as possible, available for implementation by report progress and experience in all three areas. I policymakers. This is the aim of the SDA working encourage every reader's active participation in the paper series. series and the work it reports on. It is meant to be a To fulfill its mission, the SDA Project operates on forum not only for exchange of ideas but even more different levels. Conceptually, contributions need to importantly to advance the cause of sustainable and be made which advance our understanding of how the equitable growth in Africa. Edward V.K. Jaycox Vice President, Africa Region iii 仳 r1r-71 i ne Social Dimensions of Adjustment (SDA) Project Working Paper Series The SDA Project has been launched by the UNDP The Surveys and Statistics subseries focuses on the Regional Programme for Africa, the African Develop- data collection efforts undertaken by the SDA Project. ment Bank, and the World Bank in collaboration with As such, it will report on experiences gained and other multilateral and bilateral agencies. The objective methodological advances made in the undertaking of is to strengthen the capacity of governments in the household and community surveys in the participat- Sub-Saharan African Region to integrate social dimen- ing countries to ensure an effective cross-fertilization sions in the design of their structural adjustment pro- in the participating countries. The subseries would grams. The World Bank is the executing agency for also include "model" working documents to aid in the the Project. Since the Project was launched in July implementation of surveys, such as manuals for inter- 1987, 30 countries have formally requested to partici- viewers, supervisors, data processors, and the like, as pate in the Project. well as guidelines for the production of statistical The Project aims to respond to the dual concern in abstracts and reports. countries for immediate action and for long-term in- The Policy Analysis subseries will report on the stitutional development. In particular, priority action analytical studies undertaken on the basis of both programs are being implemented in parallel with ef- existing and newly collected data, on topics such as forts to strengthen the capacity of participating gov- poverty, the labor market, health, education, nutrition ernments (a) to develop and maintain statistical data and food security, the position of women, and other bases on the social dimensions of adjustment, (b) to issues that are relevant for assessing the social dimen- carry out policy studies on the social dimensions of sions of adjustment. The subseries will also contain adjustment, and (c) to design and follow up social papers that develop analytical methodologies suitable policies and poverty alleviation programs and pro- for use in African countries. jects in conjunction with future structural adjustment Another subseries, Program Design and Implemen- operations. tation, will report on the development of the concep- The working paper series "Social Dimensions of tual framework and the policy agenda for the project. Adjustment in Sub-Saharan Africa" aims to dissen-d- It will contain papers on issues pertaining to policy nate in a quick and informal way the results and actions designed and undertaken in the context of the findings from the Project to policyrnakers in the coun- SDA Project in order to integrate the social dimensions tries and the international academic community of into structural adjustment programs. This includes economists, statisticians, and planners, as well as the the priority action programs implemented in partici- staff of the international agencies and donors associ- pating countries, as well as mediurn- and long-term ated with the Project. In the light of the three terrains poverty alleviation programs and efforts to integrate of action of the Project, the working paper series con- disadvantaged groups into the growth process. The sists of three subseries dealing with (a) surveys and focus will be on those design issues and experiences statistics, (b) policy analysis, and (c) program design which have a wide relevance for other countries as and implementation. well, such as issues of cost-effectiveness and ability to reach target groups. V Contents Executive Summary 1 1. Introduction 2 2. The Extent of Malnutrition in Ghana 3 Nutritional Status of Preschool Aged Children 3 Nutritional Status of Adults 8 3. Determinants of Malnutrition 10 Results of Regressions for Children's Nutritional Status 11 Results of Regressions Explaining Women's Nutritional Status 15 4. Food Expenditure Patterns 20 5. Conclusion 28 Appendix: Estimation of Instruments 31 References 34 vii Acknowledgement The author wishes to thank Paul Higgins for his contribution to this project and Paul Glewwe for helpful comments on an earlier draft. Executive Summary Social indicators for Ghana reflect the cumulative im- the influence of the womb environment more than pact of economic mismanagement since indepen- genetics and, as such, a lagged nutritional effect. No dence as much as current policies. Indeed, many of the gender bias was observed, consistent with other Afri- policies in place since 1983 represent a major break can evidence. Levels of chronic malnutrition (low with the previous decade. By a number of measures height for age) but not acute malnutrition (weight for this economic recovery program has been successful age) decrease as income increases, although the statis- in turning the economy around; GNP per capita has tical precision of this relationship is comparatively grown every year since 1983, a sequence which is low. There is, however, a strong effect of income on unprecedented in post independence history. The body mass indices (BMI) for adult women. The study magnitude of the decline prior to the recovery pro- also indicates that BMI decreases with higher parity, gram was sufficiently large, however, that this appre- which was introduced into the analysis as an in- ciable progress is still only a partial recovery. strumented variable to control for individual hetero- This report presents data based on the Ghana Living geneity and possible reverse causality. Standards Survey first year (1987-88) which indicate While the positive relationship between income and that levels of malnutrition in Ghana remain relatively nutritional levels observed in this study may be high compared to other African countries. The nutri- deemed obvious, it is surprisingly difficult to demon- tional status of children is nevertheless improved strate and often contradicted. The results, however, compared to a variety of data covering the period confirm a similar study based on C6te d'Ivoire data. 1982-86. These anthropometric measures of nutri- Many of the long term impacts of current economic tional status show a strong regional pattern, with mal- policies, although introduced for reasons other than nutrition increasing roughly from South to North. nutrition, will likely affect nutrition through this in- Econometric analysis also revealed strong effects of come relationship. A revival of Ghana's once notewor- household composition and intergenerational effects thy educational system may also affect nutrition via through the mother's height. Since this latter effect is income, rather than directly shifting the effectivendEs far larger than the similar positive correlation of height of input utilization. for age and the father's height, it is likely to indicate 1 i 1. Introduction While economists generally assume that changes in rate of 3 percent, a figure which is somewhat lower household incomes are unambiguous indicators of than the average for 1980-87, it would require an changes in the welfare of the household, many plan- average GNP growth rate of 5.8 percent in order to ners and government officials consider health and restore GNP per capita to its 1965 level by the end of nutritional status as additional indicators of social the century. There is nothing, of course, magic about welfare. This reflects both potential externalities from the year 2000, nor is this illustration necessarily robust health improvements as well as societies' specific to alternative starting points. It does, however, pro- aversion to malnutrition (Tobin). Consequently, infor- vide an indication of the scope of the real task of mation on the nutritional status of a population pro- recovery and puts short term measures of success in vides information on the health of the economy. perspective. Moreover, for many, full recovery only Moreover, insights into the etiology of undernutrition implies recovery into poverty. Such poverty is not a assist planners in effectively meeting their social ob- product of current policies, but is, nevertheless, its jectives. challenge. There is clearly a basis for concern for health and One means of addressing this challenge is the Pro- nutrition in Ghana. Life expectancy at birth in 1987 gram of Actions to Mitigate the Social Costs of Adjust- was only 54 years (World Bank). This is identical to the ment (PAMSCAD) designed by the Government with average for all low income countries, excluding India donor support in 1987. Mitigate is, perhaps, an inap- and China, according to World Bank estimates. Infant propriate word as many of the poverty issues the mortality rates are 90 deaths per 1,000 births. While program seeks to address predate the recovery pro- somewhat lower than the average for low income gram. Nevertheless, PAMSCAD addresses a variety of countries, this figure does not represent appreciable social concerns not always linked with economic re- progress relative to estimates from a decade before. In covery programs. a related vein, the World Development Report for 1989 As an input into these and other programs to allevi- indicates that food availability per capita in 1986 was ate poverty and its consequences, this study preserits not only one of the lowest in the world, it was appre- information on nutrition in Ghana in 1987/88 as indi- ciably lower than 10 or 20 years before (World Bank). cated in the first year data from the Ghana Living The comparatively low levels of social indicators Standards Study (GLSS) The indicators of nutrition reflect the cumulative impact of economic misman- derived from this survey are also compared with the agement since independence more than current poli- limited information available from earlier years in cies.1 Indeed, many of the policies in place since 1983 order to provide a perspective on changes in living represent a major break with previous economic poli- standards in recent years as well as to serve as a cies. By a number of measures this economic recovery baseline for further comparisons. The report also goes program has been successful in turning the economy beyond descriptive statistics and investigates the de- around; GNP per capita has grown every year since terminants of malnutrition. It is hoped that this anal- 1983, a sequence which is unprecedented in post inde- ysis can contribute to the design of nutrition programs pendence history. The magnitude of the decline prior in Ghana as well as to the general understanding of to the recovery program was sufficiently large, how- household behavior. ever, that this significant progress is still only a partial recovery. One illustration of the medium term task can be derived by noting that GNP per capita declined at an 1. See Rimmer for an economic history. average annual rate of 1.6 between 1965 and 1987 2. Methodology and descriptive statistics covering the entire (World Bank, 1989). Assuming a population growth survey are presented in Republic of Ghana 1989. 2 2. The Extent of Malnutrition in Ghana Nutritional Status of Preschool Aged Children d'Ivoire are strictly comparable not onlybecause of the similarity of survey methodology, but because the During the period October 1987 to September 1988, Ivoirian data are the only other data from Africa which 30.6 percent of preschool children aged under 5 years use a standard deviation criterion for determining were found to be chronically malnourished as indi- levels of malnutrition. To broaden the scope of com- cated by slow linear growth. Using a measure of ema- parison, Table 1 also includes a percent of median ciation or wasting, 7.8 percent of this preschool cutoff indicator of malnutrition in Ghana as well as an population could be classified as acutely malnour- indicator in terms of weight for age. Chronic malnu- ished. Before reporting how this national average dis- trition is much higher in Ghana than in its immediate tributes over age groups, gender or agroecological neighbor to the west, although levels of acute malnu- zones it is i.nteresting to compare the overall nutri- trition are similar. This pattern may reflect the fact that tional status measures with similar indicators from in 1986 the Ivoirian economy had declined sharply other countries and with other data from Ghana. This, from previous years while Ghana in 1987-88 had ex- in turn, necessitates a discussion of methodology, for perienced appreciable recent growth. Thus, they may different researchers chose different criteria for pre- have reached similar levels of short term malnutrition senting nutritional status. from different antecedents. This conjecture is sup- Most commonly, a low height for age relative to a ported by the fact that acute malnutrition is higher in reference population is used as a measure of past or C6te d'Ivoire in 1986 than the previous year, although chronic malnutrition while low weight for height in- the statistical significance of this difference is not dicates current or acute malnutrition.3 This, however, known. begs the question of the definition of "low". A cutoff The percentage of Ghanaians who are below 90 point of two standard deviations below the U.S. Na- percent of the reference height is roughly similar to the tional Center for Health Statistics (NCHS) reference percentages in other West African countries, except median is employed for this study. Alternative criteria Burkino Faso. The African levels are, however, far frequently employed are below 90 percent of reference below the levels of chronic malnutrition in South Asia. height for age and 80 percent of reference weight for Thus, it is somewhat surprising that the percentage of height. These percentage of median cutoff points, children in Ghana who are below 80 percent of the however, are not equivalent to the standard deviation standard for weight for height is closer to the levels in criteria. The differences between the two criteria are Nepal and Sri Lanka than to those in Sierra Leone, often not trivial. For example, using the percent of Togo, Liberia and Cameroons Current acute malnu- median cutoff criteria chronic and acute malnutrition trition in Ghana is also similar to rural Mali and Mau- levels were 19.6 percent and 5.1 percent in Ghana in ritania in post-drought recovery years but 1987-88. By this definition, then, levels of both acute substantially less than in those countries or neighbor- and chronic malnutrition are two-thirds of what they ing countries during the 1974 drought. A comparison are using a two standard deviation cutoff point. Since with nationwide surveys in Lesotho and Swaziland , an age invariant probability statement can be made for which an urban rural breakdown is not available, using the standard deviation presentation while the reveals a similar pattern; the Southern African coun- distribution of the percentages of the reference vary by tries had levels of chronic malnutrition roughly com- age and size, the former methodology is preferred. parable to Ghana but had levels of acute malnutrition Table 1 presents measures of malnutrition for se- which were far lower. Table 1 indicates, furthermore, lected African and Asian countries. Ghana and COte that the percentage of Ghanaian children who are 3 4 Table 1. Indicators of Malnutrition in Selected Developing Countries Survey Chronic Undernutrition Acute Undernutrition Underweightc Country Year Rural Urban Rural Urban Rural Urban Ghana 1988 34.8 22.0 8.6 6.1 22.9 14.3 1988 22.8a 12.3a 5.8 3.5b 34. 23.5d C6te d'Ivoire 1985 18.4 11.3 6.5 5.0 1986 19.4 11.2 6.8 8.4 ... ... Egypt 1978 23.8a 12.7a 0.7b 0.4 9.9 5.2 Cameroon 1977 22.4a 15.7a 1.1b 0.7b 23.0' 121d Liberia 1976 20.2a 13.8a 1.6b 1.7 25.5d 20.5d Togo 1977 20.5a 11.4a 2.3b 0.8b 16.5 8.9 Sierra Leone 1977 26.6a 13.8a 3.2 b 32.4 21.3 Nigere 1974 ... ... 11.4b Malie 1974 ... ... 10.7b 1975 ...... 5.3b Mauritaniae 1974 ... ... 9.9 1975 ... ... 6.1b Chade 1974 ... ... 22.5b 1975 ...... 12.1b Burkina Fasoe 1974 48.0a ... 9.1b 1975 43.8a ... 8.1b Kenya 1977 28.7a ... 4.4b Sri Lanka 1976 34.7a ... 6.6b ... 42.0 ... 1976 44.0 ... 8.4 ... 1980-82 36.3 ... 13.8 ... ... ... Nepal 1975 51.9a ... 6.6b ... 49.9 ... Note: Cut off at two standard deviations except where noted. a. Children below 90 percent of reference height for age. b. Children below 80 percent of reference weight for height. c. Children below 75 percent of reference weight for age, except where noted. d. Children below 80 percent of reference weight for age, e. Surveys covered only the rural sedentary population of that part of each country estimated to be most affected by the drought., The affected zone varied from a relatively small part of Burkina Faso to nearly all of Niger. Geographical coverage for the area was about one-third. Sources: USAID 1975, 1976a, 1976b, 1977a, 1977b, 1977c; USHEW 1974, 1975; Kenya 1979; Sahn, 1987, 1988. Table adapted from Kumar. underweight, either due to current or past malnutri- weight for age standard. This implies, then, that mal- tion, is also high relative to most of the African coun- nutrition has declined since 1986 but remains above tries for which comparable data are available. the levels reported for the early post-independence Despite the relatively high current levels of malnu- years. Note that the weight for age standard cannot trition in Ghana, the situation appears to have im- distinguish between chronic and acute malnutrition. proved compared with earlier years in the decade. For Many of the children who appeared malnourished in example, a nationwide survey of 14,000 children con- 1986 were likely stunted during the period of low food ducted by the Nutrition Dept, Ministry of Health and availability following the 1983 drought and bush fires. UNICEF in November and December 1986 found that By 1987 most, but not all, of these cohorts would have 58.4 percent of preschool children fell below 80 per- left the preschool bracket. cent of NCHS weight for age standards.7 This is Another indication of the changing pattern of mal- roughly twice the level of the first national survey nutrition in the decade comes from data collected by carried out in 1961-62. The current study found that the Catholic Relief Service. While these data suffer 31.4 percent of children fell below 80 percent of the from selection bias-the results are based on children 5 Figure la. Weight for Age of Children Ages 7-42 Months Attending Clinics by Month, 1980-83 Percent below 3rd percentile of Harvard W/Age standard 55 - 50 -*-1983 45 - - - -- - - -- - - -....1982 S- - - - - - - - - - - - - - - 1981 40 --. - 1980 .. ......... .................... ... 35-. -- -- Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Source: Catholic Relief Service Growth Surveillance System: Annual Report 1986. Accra, September 1987. Figure lb. Weight for Age of Children Ages 7-42 Months Attending Clinics by Month, 1984-86 Percent below 3rd percentile of Harvard W/Age standard 46 - 44 - 42 - - 40 ---.- 1986 38 - *-1985 36' - ........ ....1984 34....... 32 . 301 11111 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Source: Catholic Relief Service Growth Surveillance System: Annual Report 1986. Accra, September 1987. who attended clinic based feeding centers-the infor- high relative to CRS data for that year, while the GLSS mation in Figures la and lb indicate the severity of the data are similar to the most recent year of the CRS nutritional problem between 1982 and 1984 relative to series. the beginning of the decade or to 1986. The CRS data Table 2 breaks down the GLSS data by age and by use the 3rd percentile from Harvard weight for age gender. While it is often observed that girls in South standards. These are roughly comparable to 80 per- Asia are more likely to be malnourished than are boys, cent of NCHS weight for age standards. Conse- females do not appear to be at a relative disadvantage quently, the 1986 National Nutrition Survey appears in nutrition in Africa in general8, nor in Ghana in 6 Table 2. Nutritional Indicators by Age and Gender, Ghana 1987-88. (Percent of cell population in each category) Age Group (in months) Category 0-6 6-12 12-24 24-60 All Males Height for Agea Z-score:5 -2 4.35 11.02 33.33 39.50 31.19 -2 < Z-score 5 -1 13.77 29.13 25.67 26.82 25.33 -1 < Z-score ! 0 39.13 30.71 19.16 18.66 22.36 0 < Z-score 5 1 22.46 20.47 11.11 9.04 12.21 1 < Z-score:5 2 13.04 5.51 7.66 3.50 5.69 Z-score > 2 7.25 3.15 3.07 2.48 3.22 Weight for Height Z-score!5 -2 4.35 14.17 18.77 4.66 8.66 -2 < Z-score 5 -1 15.22 47.24 37.16 24.05 28.30 -1 < Z-score 5 0 27.54 27.56 28.74 45.04 37.71 0 < Z-score 1 39.13 7.87 12.64 22.89 20.96 1 < Z-score 2 11.59 3.15 2.68 2.33 3.55 Z-score > 2 2.17 0.00 0.00 1.02 0.83 Females Height for Agea Z-score -2 4.70 10.64 31.69 39.11 30.22 -2 < Z-score 5 -1 11.41 18.44 33.74 25.79 24.78 -1 < Z-score 0 24.16 38.30 22.63 21.06 23.72 0 < Z-score ! 1 33.56 19.86 9.88 9.31 13.57 1 < Z-score:5 2 17.45 8.51 0.82 3.15 5.04 Z-score > 2 8.72 4.26 1.23 1.58 2.68 Weight for Heighth Z-score:5 -2 4.03 9.22 13.99 5.44 7.39 -2 < Z-score 5 -1 12.08 41.13 33.33 26.07 27.54 -1 < Z-score 5 0 32.89 40.43 32.92 40.83 38.26 0 < Z-score : 1 31.54 7.09 14.81 23.35 20.80 1 < Z-score 2 16.78 1.42 4.12 4.15 5.36 Z-score > 2 2.68 0.71 0.82 0.14 0.65 a. Low height for age Z-score indicates chronic undernutrition. b. Low weight for height Z-score indicates acute undernutrition. Some percentages may not add to exactly 100.00 due to rounding. Source: GLSS data. specific. Levels of acute malnutrition for boys are ap- decade (Levinson, CRS). This pattern was also ob- preciably higher than for girls in the 6-24 month age served in the early 1960s (Rimmer). Greater Accra has brackets. This is the case for the most severe cases (Z- the lowest levels of both acute and chronic malnutri- scores less than -2) as well as more moderate cases (Z- tion in the current study. The remainder of the coastal scores between -2 and -1). The gap closes and actually agroecological zone has appreciably less chronic mal- reverses in the older age bracket. This age bracket, nutrition than the forest or savannah regions but has however, has relatively little acute malnutrition. On relatively high levels of chronic malnutrition. Note, the other hand, the older children have the highest however, that the cell size for each ecological zone is levels of chronic malnutrition, consistent with the fairly small. Thus, confidence intervals are wide. view that low stature represents the cumulative effects Levels of malnutrition by per capita expenditure of nutritional shocks. In a similar vein, levels of deciles are indicated in Table 4. The deciles were chronic malnutrition for youngest children do not ranked using a pooled sample of all children below 5 differ from levels of current malnutrition, as the only years. Since rural areas are relatively poorer, children past these children have is a recent past. in the lower deciles are disproportionately from rural Table 3 indicates that malnutrition is particularly regions (including those classified as semi-urban). high in the savannah agroecological zone.9 This is Using this manner of ranking, the urban and nonur- consistent with studies which report higher levels of ban populations in a given decile have the same real malnutrition in the northern regions earlier in the expenditures. This would not be the case if rankings 7 Table 3. Nutritional Indicators by Agroecological Zone and Gender, Ghana 1987-88. Males Females All Percent Percent Percent Percent Percent Percent Agroecologi- with chronic with acute with chronic with acute with chronic with acute cal zone Number malnutrition malnutrition Number malnutrition malnutrition Number malnutrition malnutrition Coast (ex- cluding Accra) 256 27.6 7.4 298 21.8 9.1 554 24.0 8.3 Greater 150 20.7 6.7 113 23.9 6.2 263 22.0 6.5 Accra Forest 531 34.1 7.7 559 32.7 6.4 1,090 33.4 7.1 Savannah 275 35.3 12.0 261 36.8 6.9 536 36.0 9.5 Notes: Chronic malnutrition refers to height-for-age Z-score less than -2.0; acute malnutrition refers to weight-for-height Z- score less than -2.0. Source: GLSS data. Table 4. Indicators of Child Malnutrition by Per Capita Expenditure Decile Rural Urban Per Capita Average per Percent with Percent with Percent with Percent with expenditure capita chronic acute chronic acute decile expenditure malnutrition malnutrition N malnutrition malnutrition N 1 12,226 33.5 11.1 215 25.9 7.4 27 2 19,545 33.8 7.3 204 33.3 2.3 42 3 24,628 37.9 10.1 208 22.9 2.9 35 4 29,342 41.2 7.2 194 24,.0 10.0 50 5 34,930 30.9 2.7 188 20.0 3.6 55 6 40,497 34.3 9.5 169 21.1 2.6 76 7 47,170 36.9 5.5 146 26.8 5.1 97 8 55,969 28.8 9.9 142 22.5 7.8 102 9 67,789 37.6 9.4 117 17.9 7.8 128 10 100,428 22.6 16.0 106 18.3 7.3 137 Notes: Individuals are ranked by household per capita expenditure adjusted for monthly inflation. N refers to the number of individuals within each cell. Source: GLSS data. were based only on relative expenditures within a was also conducted by the Ghana Statistic Survey in sector. The data indicate that at a given expenditure 1988 (Ghana Statistic Survey). The DHS reports levels level urban children are less likely to be either stunted of malnutrition among 1,841 children between the or wasted. This may reflect differences in sanitation ages of 3 and 36 months. The 30 percent of the sample and access to health care, There may be an element of reported with chronic malnutrition is comparable to self-selection of parents into urban areas as well. the data reported here for children in the same age It is somewhat surprising that Table 4 reveals no bracket. While the survey did not have data on income clear pattern of nutritional status by expenditure lev- or expenditure, the regional breakdown indicated els. There is only a slight tendency towards lower higher levels of malnutrition in the North, and two levels of chronic malnutrition as expenditure per ca- Upper Regions, comparable to the savannah pita increases. Moreover, the pattern is clearly not agroecological zone in this paper. The study also indi- monotonic. There is no pattern at all regarding acute cated relatively high levels of malnutrition in the Cen- malnutrition. While small cell sizes again reduce the tral Region, regional breakdown which is not accuracy of any decile average, aggregation into quin- indicated in the current study. Also, the overlapping tiles does not lead to a different picture. Similarly, the study provides verification of the absence of any gen- picture does not change if households are ranked by der specific pattern of nutrition. expenditures rather than the ranking being con- Other tabulations indicated in the DHS anticipate structed on an individual basis according to the expen- results investigated using multivariate techniques ditures of the household in which the individuals live. which are reported below. For example, the report It is useful to discuss these results in the context of indicates that children which are first born or who the Demographic and Health Survey (DHS) which were born more than 2 years after their older sibling 8 are less likely to have either chronic or acute malnutri- uire and Austin present evidence of economic and tion. Children who recently had diarrhea have higher health consequences of being small, as opposed to rates of malnutrition, although the statistical signifi- consequences of growth faltering or becoming small.10 cance of this and similar differences is not indicated. A particular concern is the nutritional status of women Finally, the DHS reveals a pronounced difference in as an indicator of adult health as well as the possible the rates of malnutrition of children whose mothers consequences for birth outcome and subsequent in- have more than a middle level of education. Differ- fant mortality (Martorell and Gonzalez-Cossio, ences at lower levels of education, however are less Thomas et al., 1988). apparent. Nor is it possible to say whether the effect One difficulty in studying the nutrition of adults is of education works through child care or through the that, to a large degree, adult nutrition, particularly presumably higher incomes of the more educated height, is determined in childhood, possibly as early households. as the first two years of life. As such, there is relatively little of current policy relevance that is indicated by Nutritional Status of Adults studying adult stature. While indices such as the body mass index (BMI-weight in kilograms over height in Relatively little attention is given to the nutritional meters squared) are relatively independent of height, status of adults, in part because they are less vulnera- there is less information on which to base expectations ble to severe consequences of undernutrition. Indeed, for a healthy population. Payne, however, reports a Seckler argues that smallness in adults is actually few studies from actuary tables in developed coun- adaptive (see also Beaton). On the other hand, McG- tries which indicate that adult mortality risk rises with Table 5. Ranges of Body Mass Indices for Adults Age 20 and Over, 1987-88 (Percent of cell population) BMIa Range Classificationb Urban Rural Pooled Males (n=1,029) (n=1,710) (n=2,739) BMI ! 18.4 Health Risk 12.9 19.0 16.7 18.4 < BMI 19.9 Underweight 20.4 25.3 23.4 19.9 < BMI:5 25.0 Acceptable 56.9 53.5 54.8 25.0 < BMI 29.9 Overweight 8.6 1.9 4.4 BMI > 29.9 Obese 1.2 0.3 0.6 Females (n=1,127) (n=1,962) (n=3,089) BMI 5 17.5 Health Risk 6.5 11.1 9.4 17.5 < BMI 5 18.6 Underweight 5.8 11.4 9.4 18.6 < BMI!5 23.8 Acceptable 54.1 64.3 60.6 13.8 < BMI 28.5 Overweight 20.7 10.4 14.1 BMI > 28.5 Obese 13.0 2.8 6.5 a. BMI is calculated as the ratio of weight in kgs to height squared (in square cms). b. While BMI cutoff points for a healthy population are still provisional, the above classification was proposed in 1983 by the Royal College of Physicians in London to characterize the limits of the acceptable weight range for height by gender. Source: GLSS data. Table 6. Body Mass Indices and Adult Heights by Expenditure Deciles Males Females Urban Rural Urban Rural Decile BMI Height Number BMI Height Number BMI Height Number BMI Height Number 1 19.8 164.1 40 19.7 164.2 252 20.8 155.7 56 20.4 155.9 214 2 20.8 167.9 51 19.5 165.1 260 22.7 156.9 58 20.1 156.6 206 3 19.6 167.0 58 19.9 164.9 239 21.3 156.3 51 20.7 156.4 190 4 20.5 166.5 73 19.8 166.3 210 21.1 158.2 86 20.4 156.0 191 5 20.4 166.4 108 20.0 165.7 188 21.6 156.3 106 20.5 156.4 161 6 20.2 166.5 135 20.1 165.1 181 22.7 158.0 114 20.8 156.7 124 7 20.7 168.3 117 19.6 166.0 199 22.5 158.0 115 21.7 156.7 149 8 20.7 169.0 130 20.1 166.8 205 24.1 157.5 147 21.2 156.9 138 9 21.0 168.9 181 20.2 167.1 174 23.0 158.2 144 20.5 157.3 125 10 21.3 170.1 259 20.5 169.0 162 24.3 159.2 159 22.2 158.3 62 Source: GLSS data. 9 BMI below 19 or 20 or above 22. Dugdale offers 19 as although this could also reflect the normal process of a lower limit for adequate health, but in view of data aging. When heights are regressed in a very simple from India, Indonesia and Thailand, Payne suggests a model using age, age squared and location dummy lower cutoff at 18. James et al., while stating that any variables, a quadratic pattern emerges for adult males. standards are to a fair degree arbitrary, suggest that This has a peak age between 28-36 years depending levels above 18.5 should be considered normal. They on the age cutoff at the lower tail, that is, the age at suggest individuals with levels below 18.5 should be which an individual is considered adult. The lower subdivided into three subgroups, with those having age for the turning point was associated with the BMIs below 16 being classified as being in a third, or higher age for the cutoff (25 years). This pattern could, most severe, grade of chronic energy deficiency. The however, be generated by the inclusion of males who Royal College of Physicians (1983) presume a risk for have not yet finished their growth or by a fairly uni- health with BMI levels below 18.4 for males and 17.5 form pattern in a portion of the data and a slight for females. This group also has a higher cutoff point decline amongst the oldest. For females the time trend for either overweight or obesity. is linear with older women significantly shorter. There Table 5 indicates the distribution of BMI for adult does, then, appear to be some evidence of a secular males and females (excluding those reported as preg- trend in height, although the data and technique are nant or lactating") for urban and rural populations not sensitive enough to determine whether the chil- following the classification of the Royal College of dren who came of age in the 1970s or later had their Physicians. In general, rural residents are thinner than growth affected by the economic decline. their urban counterparts. Moreover, males are leaner than females overall, although more women are ob- served in the lowest categories. There is, in addition, a Notes surprisingly large number of females with BMIs in the higher brackets indicating overweight or obesity. 3. Waterlow, J. C. 'Note on the Assessment and Classification Table 6 breaks the population into expenditure groups of Protein Energy Malnutrition in Children." Lancet 1: pp. 87-9 and indicates that increased BMI among adults is more (1973). closely associated with increased total household ex- J.Waterlow, R.Buzina, W.Keller, J. Lane, M. Nichaman, and closelyJ.Tanner, "The Presentation and Use of Height and Weight Data for penditures than are similar measures of nutrition Comparing the Nutritional Status of Groups of Children Under the among preschoolers. The pattern is, however, not Age of 10 Years," Bulletin of the World Health Organization, 55(4) pp. monotonic. Note that adult heights also increase with 489-98,1977. current family expenditures. Under the assumption 5. Intercountry comparisons are not likely to reflect ethnic dif- that adult heights are influenced by the level of family ferences. It is widely observed that growth of middle class children in various countries conforms to the international standards at least income in their childhood, this likely indicates a strong until the prepuberty growth spurt. correlation of expenditures between generations. Re- 6. USAID (1986, 1977) cited in Svedberg. P, table 9.1. verse causality-from stature to expenditures-is log- 7. Levinson, F.J. "Combating Malnutrition in Ghana." mimeo, ically possible and in keeping with research on March 1988. Details of this study have not yet been published. productivity referred to above, but is probably not the 8. Svedber& P. "Undenutritionin Sub-SaharanoAfica: Is There a Sex Bias?" Stockholm, Institute for International Economic Studies, only pathway. When individuals are ranked by pre- Seminar Paper #421, University of Stockholm, October 1988. dicted income a similar pattern is observed. 9. Although the sample was not designed for regional compar- In many countries there has been a secular trend isons, malnutrition appears higher in the three northern regions towards taller generations. This may be due, in part, compared to savannah agroecological zones within other regions. income growth as well as improved health environ- 10. See Strauss (forthcoming) for a review of nutrition and pro- tov ductivity from an economist's perspective. ments. Such a trend, however, should be modest in 11. Lactating women were found to have significantly lower Ghana given the uneven, and often negative, income BMI than non-lactating women although the difference is smaU. The growth. Older adults in the sample are clearly shorter, difference in the means of BMI is 0.37 (t=2.9). 3. Determinants of Malnutrition The standard model of individual utility maximiza- instruments but sacrifices some of the potential to tion using home produced goods (including nutrition) elucidate pathways to nutritional status that would be as arguments provides the underlying basis for most revealed by a health production function. empirical studies of nutrient consumption or nutri- Both approaches are presented in the analysis that tional outcome.12 These models are now well known follows. The distinction is more in concept than in and only a few features need be discussed here. Of practice as there are few variables available in the data particular importance is the fact that food enters the which can be considered as inputs into the production utility function in two manners-directly as a con- of nutrition. The GLSS data is the most complete data sumer good and indirectly asan input into the produc- set available for Ghana and is, in many respects, supe- tion of health. Other items, such as health care, enter rior to most household surveys from developing coun- into health production as inputs but do not directly tries. Nevertheless, the structure of the analysis is contribute to household utility. severely constrained by limitations in data that are Ideally, the role of income or other variables which currently available. indirectly influence nutrition could be derived simul- The household survey per se does not contain con- taneously from production functions and input de- sumer prices. While these were collected in a comple- mands. The nutritional health derivative is the sum of mentary cluster or market survey, such prices were coefficients for the inputs in the production function not available at the time of the analysis. Moreover, times the respective derivatives in the input demand with a maximum of three observations per commod- functions. Improvements in nutritional status with ity, in order to utilize this supplementary data one increases in income or other changes in the constraints would have to account for missing prices as well as whichinfluencea household'sbudget allocation, then, within cluster price variation. Similarly, the survey depend not only on the demand for nutrients or other design included the collection of information on com- inputs but also on the magnitude of the response in munity infrastructure. Such data were also unavail- the production function. able for this study. One limitation is the difficulty of identifying instru- While one generally cannot determine the coeffi- ments for inputs into the nutritional status production cients of variables for which observations are not function. Since inputs are endogenously determined, available, it is possible to include such information there is a potential for bias in estimates which use implicitly in the instrumenting equations.13 Each vil- observed levels of inputs. On the other hand, pre- lage or cluster is assumed to represent a single market dicted inputs are often predicted with low precision. and is surveyed over an interval of only a few days. Moreover, identification may depend on arbitrary re- Consequently, there is virtually no within-cluster strictions or on variables such as infrastructure and price variation, nor, for the large part, any variation in distances to various facilities which may have little infrastructure or other community variables. Cluster explanatory power. One alternative is a reduced form fixed effects estimates, then, allow one to remove the approach or variations of that approach which in- potential missing variable bias that might result from cludes predicted income (Thomas et al., forthcoming; the exclusion of price and infrastructure variables. Sahn, 1989). This differs from a conventional reduced Moreover, clusteraverages on observed variables con- form model in that it employs a linear combination of tain information on the impact of these excluded vari- the independent variables which explain income ables which help provide variation useful in studying rather than those variables themselves. Such an ap- the impact of endogenous variables. Such an approach proach is useful for explaining the impact of policy is enhanced when the number of clusters is relatively 10 11 large and when the clusters cover an extensive range logarithm of predicted per capita household food ex- of economic and physical environments as is the case penditures is used as a proxy for food inputs.'s Health here. inputs are generally difficult to measure-one needs Such fixed effects models are formulated in terms of to distinguish between curative and preventative in- deviations from village means. The general form of puts-and are indicated here by predicted days of this model is: illness per child. Both food and illness are in- strumented using cluster fixed effects models which Y(vo - Yv + (X(vi - Xv)B (1) include, among other variables, household composi- tion and predicted per capita expenditures. Separate where the i subscripts denote observations on individ- urban and non-urban prediction equations for this uals and the v's denote villages or clusters. This model latter variable are reported in the appendix. can, however, be equivalently expressed in terms of While the discussion above indicates that the fixed dummy variables: effects approach allows the analysis to proceed with- out a set of community variables, it is nevertheless fair Yi= XiB + K,Cv (2) to at least indicate what factors may influence the illness instrument that are not expected to enter di- where K, denotes a vector of dummy variables which rectly into the nutrition production function. This set are defined as one if the individual is from the vth may include information on the distance to clinics and village or cluster and zero otherwise. When all cluster the quality of health care personnel. These factors observations are collected during the same time pe- could, in principal, influence nutrition directly riod then equation (2) is virtually equivalent to an through public health education and maternal nutri- equation with a seemingly more complete specifica- tion. Other variables such as malarial prevalence, co- tion of exogenous price (P,) and infrastructure vari- liform bacterial counts in water, and other disease ables (Z,): vectors, however, would affect nutrition only through disease patterns. Yi= XiB + PvD + ZvO (3) In addition to these inputs, nutritional status may be influenced by a number of characteristics of the Equations (2) and (3) differ mainly in terms of their child including age, gender, and number of siblings. error structure which has some potential conse- While community factors are implicitly included in quences for instrumental variables derived from the input functions, additional regional differences are them. possible. Thus the regressions include dummy vari- A variation of this approach which is used here is to ables for agroecological zones as well as for Accra and include the cluster mean value of the left hand side Tema and a variable for semi-urban communities. At variable on the right hand side of the estimating equa- a given level of inputs, the level of education of the tion. This average implicitly contains information on parents may influence the efficiency with which these prices and infrastructure which is useful for identify- (and other) inputs are used and are, therefore, in- ing the impact of the variable which is being in- cluded in the regressions. strumented. This average could also be added to the Behrman and Wolfe have argued that maternal ed- predicted value from equation (1) before conducting ucation may proxy for other environmental and the second stage estimation. In this case, however, it human capital factors. Thomas et al. (1988) find partial would be difficult, if not impossible, to correct the support in that they observe a decline in the coeffi- standard errors using two stage least squares estima- cients of education when parental heights are in- tions or the methods discussed byMurphy and Toppel cluded. Education, however, remained highly as well as Duncan.14 significant in their study of Brazilian children. The current study, therefore, includes variables for both Results of Regressions for Children's the education and size of the parents. A strong prior Nutritional Status assumption is that the mother's height will have a stronger impact than the father's. This is based not on The nutritional status of a child, as measured by factors related to non-formal training or experience, either height for age or weight for height is produced but rather on biology, although not necessarily genet- by two main inputs, food and health care. Since the ics. While the genetic contribution of both parents is GLSS data does not measure food in quantity terms, equal, the mother also has an environmental effect household consumption of nutrients cannot be di- through the womb environment. As one does not rectly calculated. Nor, of course, is the individual con- observe the genotype but only the phenotype,6 mea- sumption by the child observed. Accordingly, the surement of the genetic pathway is biased downward 12 Table 7. Children's Height for Age Equations-Pooled Rural and Urban Sample (Standard errors in parenthesis) Variable (1) (2) (3) (4) (5) Intercept -15.60 -16.37 -17.48 -17.03 -14.19 (2.84) (2.91) (2.97) (2.92) (1.72) Age in Months -0.101 -0.099 -0.098 -0.097 -0.097 (0.009) (0.009) (0.009) (0.009) (0.009) Age Squared 0.00115 0.00111 0.0011 0.0010 0.00111 (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Female 0.034 0.026 0.027 0.025 0.025 (0.077) (0.077) (0.077) (0.077) (0.077) Household Size 0.061 0.061 0.065 0.066 0.041 (0.027) (0.027) (0.027) (0.028) (0.019) Older Sibling -0.228 -0.233 -0.231 -0.230 -0.244 (0.062) (0.062) (0.062) (0.062) (0.061) Older Half Sibling 0.003 0.005 0.030 0.017 0.024 (0.068) (0.069) (0.069) (0.069) (0.069) Younger Sibling 0.386 0.371 0.373 0.375 0.352 (0.095) (0.095) (0.095) (0.095) (0.093) Younger Half 0.159 0.191 0.189 0.186 0.177 Sibling (0.109) (0.109) (0.108) (0.109) (0.108) Urban 0.109 0.080 0.098 -0.053 0.109 (0.113) (0.116) (0.116) (0.117) (0.107) Semiurban -0.202 -0.193 -0.200 -0.210 -0.193 (0.111) (0.111) (0.111) (0.112) (0.111) Accra -0.291 -0.330 -0.371 -0.414 -0.350 (0.193) (0.195) (0.197) (0.201) (0.193) Tema -0.102 -0.101 -0.093 -0.130 -0.097 (0.330) (0.333) (0.332) (0.334) (0.333) Forest Zone 0.033 0.023 0.019 0.040 0.017 (0.187) (0.187) (0.187) (0.190) (0.189) Forest*Age -0.011 -0.011 -0.011 -0.012 -0.012 (0.005) (0.005) (0.005) (0.005) (0.005) Savannah Zone -0.514 -0.470 -0.475 -0.410 -0.471 (0.219) (0.219) (0.219) (0.233) (0.227) Savannah*Age 0.0002 0.002 0.002 0.001 0.001 (0.006) (0.006) (0.006) (0.006) (0.006) Mother Primary 0.113 0.130 0.122 0.115 0.107 Education (0.126) (0.126) (0.126) (0.126) (0.127) Mother Middle 0.158 0.134 0.123 0.113 0.122 Education (0.110) (0.110) (0.110) (0.111) (0.111) Mother Secondary 0.356 0.369 0.355 0.299 0.335 or Higher (0.347) (0.350) (0.350) (0.352) (0.351) Father Primary -0.098 -0.087 -0.083 -0.085 -0.073 Education (0.145) (0.147) (0.147) (0.148) (0.147) Father Middle -0.160 -0.187 -0.185 -0.221 -0.191 Education (0.109) (0.110) (0.110) (0.112) (0.109) Father Secondary -0.215 -0.213 -0.194 -0.226 -0.177 or Higher (0.181) (0.181) (0.182) (0.184) (0.179) Mothers Height 0.0047 0.0045 0.0045 0.0045 0.0045 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) 13 Table 7. Continued Variable (1) (2) (3) (4) (5) Fathers Height 0.0021 0.0023 0.0023 0.0022 0.0022 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) Mothers Age 0.017 0.018 0.018 0.018 0.017 (0.006) (0.006) (0.006) (0.0061) (0.006) Log* Expenditures 0.399 0.454 0.312 0.293 Per Capita (0.223) (0.231) (0.242) (0.245) Days 111* - -0.031 -0.037 -0.042 -0.039 (0.021) (0.021) (0.021) (0.021) Expenditures* Per - 0.040 0.032 0.035 0.041 1lness (0.043) (0.043) (0.043) (0.043) Log Food* Expendi- - - 0.264 - - tures Per Capita (0.139) Log Cereal* Expen- - - - -0.056 0.0673 ditures Per Capita (0.080) (0.080) Log Root Crop* - - - -0.023 -0.013 Expenditures (0.052) (0.051) Log Meat* Expendi- - - - 0.147 0.147 tures Per Capita (0.089) (0.089) Log other food* - - - 0.113 0.139 (0.131) (0.129) R2 0.217 0119 0.221 0.223 0.206 Number 1,560 1,526 1,526 1,526 1,526 by errors in variables while the environmental path- rately for rural and urban populations. Differences in way is not. incomes or disease prevalence in the communities are Inclusion of parental characteristics presents an ad- incorporated in these instrumental variables. ditional problem; not every parent was present in the Children from the savannah zone are significantly household. Slightly more than 10 percent of all moth- smaller relative to age and gender specific standards ers were not measured for height. Conditional on the than other children. This difference is large in magni- mother being present, 27 percent of the fathers in rural tude (approximately half a standard deviation). More- areas were not resident, either due to travel or estab- over, regional differences in illness prevalence and lishment of separate households, while the corre- average income also work to the detriment of these sponding number in the urban sample was 34 percent. children. There is not an apparent difference between Preliminary tests with models which exclude father children from the forest zone and those from the specific variables indicate that parameters do not dif- coastal zone (the excluded regional variable) on the fer between sample subsets with and without fathers average. The age interaction term, however, indicates present. Although there is not an obvious difference that older children from the forest are smaller than in the subsamples, the truncation does have a cost in their cohorts. A similar interaction is not significant for terms of reduced sample size and correspondingly the savannah region. One interpretation that can be larger standard errors. tested on subsequent surveys is that the stature of The results of the regression explaining heights for older children reflects former conditions which have age are presented in Table 7. Chronic malnutrition is subsequently improved relative to the coastal zone.17 more or less cumulative and, therefore, the heights for As expected, there is a strong relationship between age decline relative to standards until the child is a mother's height and her child's. Since another corre- above age 4. No gender bias is observed, nor is an late of the mother's human capital-her education-is urban bias indicated. This latter result is surprising also in the model, this influence is likely through the given the descriptive statistics. Note that predicted biological pathways referred to above. Also, as ex- variables entered as instruments are estimated sepa- pected, the significant influence of the father's height 14 is less than the mother's. The BMI of both parents are nents of food expenditures are included is compara- simultaneously determined with the child's nutrition tively small. and, therefore, the variables are not included in the The total food expenditure variable predicted in model. Alternative specifications (not presented here) cluster fixed effect models is significant only when the including BMI, however, show that the combination standard errors are unadjusted. The coefficients of the of genetic, behavioral, and economic factors which individual components of food expenditures are im- influence a parent's BMI also correlate with the long precisely measured with or without the inclusion of term nutrition of a child. The relationship is, however, the total expenditure variable. While there is a concep- only significant in the case of the mother. Parental tual difficulty in using food expenditures as a proxy education is a puzzle. The education of the father has for nutrients, 21 the low coefficients are still difficult to a negative influence and the positive influence of the explain. Moreover, the difficulty is highlighted by the mother is not significant. This departure from prior fact that expenditures have an apparent influence that expectations may reflect low quality education for is not mediated through the purchase of food or health women currently in childbearing years. Education, inputs. A number of recent studies by both economists however, has an indirect effect through the demand and nutritionists have indicated that a marginal in- for other inputs and through income as well as the crease in household food consumption has only a direct effect. There is, in addition, one other pathway small impact on nutritional status although a few of from the mother's background to the child's stature; these studies indicate a stronger association with in- older women have taller children. It is not possible to come (Alderman 1989). Not only does the absence of distinguish whether this has to do with the biological a pronounced and robust impact of food consumption hazards of adolescent pregnancies or the experience challenge concepts about nutrition or at least the mea- and maturity of the mother. Note, furthermore, that surement of it, the income effect poses a challenge as holding the size of the household constant, children income should influence nutrition through inputs. with a full sibling less than two years older (including While these inputs may include housing (or the choice twins) are significantly shorter than cohorts without of residence) or other variables which are not included such a sibling. This again, can be a long term influence in this model, the pathway is not well understood. of prenatal conditions or current competition for re- To put these results in perspective, Model 2 implies sources. The presence of half-siblings, however, has no that a 10 percent increase in income will reduce the effect on heights for age.18 number of children in the Savannah who indicate Children with younger siblings are significantly chronic malnutrition by 4 percent. This reduction in- taller than their cohorts. The magnitude of the effect is creases to 9 percent with a 25 percent increase in larger if the sibling is a full sibling. This may be an income, but the reduction is over 11 percent with the artifact of nonlinearities in the age relationship; the smaller increase in income and some primary school- youngest children in the sample have no full siblings ing for each mother. The percentage decline in malnu- and fewer half siblings within two years of their age.19 trition in other agroecological zones following An additional pathway that would influence full sib- increase in income is slightly higher, although starting lings is through the age of weaning. While early wean- from a smaller base. Note, however, that with a higher ing is potentially detrimental, very late weaning is as percentage of women already having an education, a well. Further analysis of such relationships may be 10 percent income increase plus a minimum of pri- illuminated using household fixed effects models sim- mary schooling for each woman does not have as large ilar to Strauss and to Horton. an impact in these areas as in the savannah. Predicted days of illness has the expected negative Unlike height for age, which measures the cumula- impact, implying that investments in health will indi- tive effect of nutritional shocks, weight for height is a rectly increase stature relative to standards. The aver- short term indicator. As most of the variables in the age amount a household spends per disease incident regression are state variables rather than time varying does not appear to have an impact in any of the spec- measures it is more difficult to explain weight for ifications including that which excludes the weakly height than height for age.2 Such is the case here. Few correlated income instrument (5). This variable, total variables in the regressions reported in Table 8 are household resources as measured by the logarithm of significant. The significant age pattern confirms that total expenditures per capita, has an appreciable im- weight for height declines until about three years of pact on nutritional status.20 The significance, however, age. As in the regressions explaining stature, children is marginal after adjusting for the fact that the variable with either full or half siblings less than two years is estimated in a previous step. Note that the change older are lighter relative to height than their cohorts. in the coefficient when food expenditures or compo- Also, children with younger siblings appear better off, 15 Table 8. Children's Weight for Height Table 8. Continued Equations-Pooled Urban and Rural Sample (Standard errors in parenthesis) Variable (1) (2) (3) Variable (1) (2) (3) Intercept -0.919 -0.835 -2.01 Mothers Height 0.0001 0.00002 0.00004 (1.05) (1.05) (1.30) (0.0004) (0.0004) (0.0004) Age in Months -0.037 -0.038 -0.037 Fathers Height 0.0005 0.0005 0.0005 (0.0001) (0.007) (0.007) (0.0004) (0.0004) (0.0004) Age Squared 0.00074 0.0007 0.0007 Mothers Age 0.006 0.005 0.005 (0.00010) (0.00010) (0.00010) (0.004) (0.004) (0.004) Female 0.043 0.067 0.067 Log expenditures * -0.077 -0.068 0.109 (0.052) (0.052) (0.052) per capita (0.052) (0.053) (0.059) Household Size 0.003 0.002 0.011 Days 111* - -0.116 -0.014 (0.011) (0.011) (0.013) (0.014) (0.014) Older Sibling -0.081 -0.088 -0.085 Expenditures Per - -0.003 -0.001 (0.041) (0.041) (0.041) Illness* (0.029) (0.029) Older Half Sibling -0.065 -0.065 -0.073 Log food - - 0.152 (0.046) (0.046) (0.046) expenditures* (0.099) Younger Sibling 0.081 0.091 0.097 R2 0.05 0.06 0.06 (0.041) (0.062) (0.062) Number 1,566 1,532 1,532 Younger Half -0.046 -0.051 -0.049 * predicted variables Sibling (0.073) (0.072) (0.073) Urban -0.161 0.152 0.155 although the half-sibling variable in this case is anom- (0.071) (0.071) (0.071) alous. Semi Urban 0.065 -0.053 0.045 The influence of either predicted illness or food (0.07) (0.07) (0.074) expenditures is inconsequential, as is any of the com- Accra 0.266 0.255 0.222 ponents of food expenditures (not reported in Table (0.126) (0.127) (0.128) 8). Total expenditures also appears to be a poor pre- -0.149 -0.164 -0.163 dictor. The coefficient of the logarithm of expenditure (0.223) (0.223) (0.233) is, however, between 0.8 and 1.1, depending on the specification, in an unpooled urban equation. Despite Forest Zone 0.238 0.245 0.249 this difference, pooling could not be rejected on the (0.125) (0.125) (0.125) basis of the appropriate statistical test. Forest* Age -0.007 -0.007 -0.007 (0.003) (0.003) (0.003) Results of Regressions Explaining Women's Savannah Zone 0.252 0.249 0.259 Nutritional Status (0.142) (0.142) (0.142) Savannah* Age 0.0072 0.0070 -0.007 There is a special concern for the nutrition status of (0.004) (0.004) (0.004) women inasmuch as the maternal malnutrition may translate into low birth weight children and, conse- Education (0.085) (0.085) (0.084) quently, high infant mortality rates (Martorell and Gonzales-Cussio).23 As mentioned, a mother's height, Mothers Middle -0.007 -0.023 0.016 that is, nutrition lagged one generation, may influence Education (0.074) (0.073) (0.074) child survival (Thomas et al. forthcoming). Adult Mothers Second- 0.183 0.216 0.020 height is, however, to a fair degree, a product of early ary or Above (0.23) (0.23) (0.23) childhood nutrition as is discussed above. Height may Fathers Primary 0.127 0.119 0.120 also be influenced by nutrition during adolescence, Education (0.098) (0.099) (0.099) especially if pregnancy occurs before growth is com- Fathers Middle 0.095 0.088 0.085 pleted. Given the variability of timing of both the start Education (0.071) (0.072) (0.072) and completion of puberty, however, it is difficult to use age specific growth standards to model nutrition Fathers Secondary 0.118 0.107 0.111 for adolescents. The study of the nutrition of women or Above (0.119) (0.119) (0.119) attempted here, then, concentrates on body mass index (BMI) of adult, non-pregnant women. To the 16 degree that this short term measure correlates with expectation of a maternal depletion syndrome and birth outcome or other outcome variables, determi- consequently, offers an additional explanation for the nants of BMI are also likely correlates for these mea- negative relationship between a child's birth order sures. Note, however, that BMI is uncorrelated with and long-run nutritional status modelled by Horton,s adult heights. Note, however, that a maternal depletion hypothesis The model employed is similar to the model used does not necessarily imply a direct effect on subse- for children, in that cluster fixed effects are used to quent children. estimate instruments for inputs into the production of Birth spacing as well as parity likely contributes to nutritional status. There is one principal difference: maternal depletion. A dummy variable for women the influence of parity (number of pregnancies) on who are currently lactating, which correlates to some short term nutritional status is tested. While modem degree with recent delivery, is insignificant and posi- forms of family planning are not widely practiced24, tive. The variable, however, is not precise. Although traditional methods including temporary abstinence virtually every mother breastfeeds her child, and delayed marriage also influence the number of breastfeeding continues, on average, until one and a births. Consequently, the variable is treated as endog- half years. The lactation variable, then, includes the enously determined. Parity was determined using the post-partum period, as well as the cumulative effects fertility module in the questionnaire which was ad- of nursing a child. ministered to one woman of child bearing age (includ- BMI increases significantly with education, particu- ing post-menopause) per household. A regression larly post-secondary education. This is in addition to based on this subsample was used to estimate predic- the effect of education that works through income. tors for all adult women in the sample. Education not only influences the wage rate of an While a maternal depletion syndrome of worsening individual but also the type of work. If physical inten- nutrition with multiple pregnancies has been postu- sity decreases with education, as is most plausible, the lated (Jelliffe and Jelliffe) and tested (Huffman et al., education variable would capture the effect of reduced Merchant and Martorell), multivariate approaches to energy expenditure. This hypothesis is supported by adult women's nutrition are not common in the liter- the coefficient of land. Although land is an indicator ature. It is, therefore, difficult to disassociate the effects of wealth, the coefficient is negative and significant, of age (or lagged economic conditions for which age albeit small in magnitude. Given the relatively limited may proxy) and parity. Moreover, as parity is likely use of hired labor in Ghana, the size of a household's affected by education and assets, as well as a variety land holding likely correlates with the labor effort of of factors known to the woman but not the researchers women within the household. Furthermore, women which may influence both the number of children and who live in urban areas, with a higher density of the woman's nutrition, the instrumented multivariate markets and services as well as greater access to public approach used here represents a potential method- transportation, have significantly higher BMI at the ological improvement over earlier approaches. same income levels or levels of food expenditures. The results indicate a negative coefficient for parity. The probability of illness is not significantly corre- Parity is predicted from a subsample who reported lated with BMI. While a negative coefficient is ex- their reproductive history largely on the basis of age, pected, a predictor based on self reported illness may income, and education, all of which are also included not be a good proxy for an objective measure of illness. in all variations of the model presented in Table 9. As is often also observed in both developed and de- Unlike the food and commodity fixed effects equa- veloping countries, in the GLSS data self-reported tions, the price and infrastructure variables which are illness increases with income or education or both. implicit in the parity model provide little additional This reflects both differences in the concept of illness information, and it is likely that correction for bias in and the possibility that the poor cannot afford to be the standard error will have an appreciable impact on ill-in the sense of labor foregone. Since the instru- the currently significant t-statistics. Subsequent re- ment used here correlates self-reported illness with search which pools the second year sample may add education and income, it does not purge the potential to the variation of the instrument as well as reduce the bias. standard error of the coefficient through the increase On the other hand, the significant and positive co- in sample size. efficient of expenditure per disease incident is ex- The parity parameter is reasonably robust across pected. Note that the fixed effect instrumenting variations of the estimates in Table 9 and most plausi- equations explained little of the intracluster variance. ble missing variable biases are controlled. Conse- Consequently, the variance of this regressor comes quently, it offers some support for the prior almostentirelyfromthebetweenclustereffects.These 17 Table 9. Body Mass Indices Equations-Pooled Urban and Rural Sample (Standard errors in parentheses) Probit Variable (1) (2) (3) (4) (5) (6) Intercept 2.530 1.733 2.804 -0.182 2.149 3.399 (1.%2) (1.996) (2.627) (2.252) (2.201) (0.788) Height in centimeters 0.019 0.019 0.019 0.018 0.021 -0.008 (0.009) (0.009) (0.009) (0.009) (0.010) (0.004) Age in years 0.364 0.357 0.355 0.362 0.370 -0.070 (0.042) (0.043) (0.043) (0.043) (0.043) (0.017) Age squared -0.003 -0.003 -0.003 -0.003 -0.004 0.001 (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0001) Parity* -0.279 -0.281 -0.280 -0.295 -0.310 0.040 (0.093) (0.094) (0.094) (0.095) (0.095) (0.038) Lactating dummy 0.049 0.034 0.023 0.076 0.066 -0.299 (0.167) (0.167) (0.169) (0.169) (0.169) (0.073) Primary education 0.415 0.384 0.378 0.416 0.433 0.033 (0.215) (0.220) (0.221) (0.221) (0.221) (0.087) Secondary education 0.307 0.338 0.342 0.320 0.367 -0.139 (0.171) (0.172) (0.172) (0.173) (0.173) (0.073) Post-secondary 0.844 0.807 0.814 0.750 0.891 -0.068 education (0.432) (0.432) (0.432) (0.433) (0.433) (0.199) Log per capita 0.789 0.738 0.785 0.607 - -0.167 expenditure (0.113) (0.115) (0.137) (0.131) (0.046) Predicted illness - 0.001 0.005 -0.015 -0.014 probability* (0.040) (0.040) (0.041) (0.041) Log expenditure - 0.293 0.296 0.299 0.324 per illness (0.087) (0.087) (0.089) (0.089) Log food expenditure - - -0.145 - - - per capita* (0.232) Log cereal expenditure - - - 0.275 0.353 - per capita* (0.124) (0.123) Log root crop expendi- - - - 0.013 0.056 - ture per capita* (0.086) (0.086) Log meat, fish, and - - - 0.247 0.309 - poultry expenditure (0.133) (0.133) per capita* Log other food expendi- - - - -0.157 0.070 - ture per capita* (0.228) (0.223) Urban dummy 1.139 1.012 1.019 0.915 0.944 -0.135 (0.157) (0.161) (0.162) (0.166) (0.166) (0.065) Land owned -0.002 -0.003 -0.003 -0.003 -0.003 0.001 (0.0009) (0.0009) (0.0009) (0.0009) (0.0009) (0.0003) Forest dummy -0.732 -0.756 -0.764 -0.629 -0.676 0.129 (0.145) (0.145) (0.146) (0.154) (0.154) (0.060) Savannah dummy -1.261 -1.155 -1.162 -1.051 -1.072 0.333 (0.187) (0.194) (0.194) (0.231) (0.232) (0.075) April-June -0.264 -0.289 -0.281 -0.335 -0.337 0.095 (0.170) (0.170) (0.171) (0.175) (0.175) (0.070) July-September 0.464 0.464 0.454 0.457 0.471 -0.061 (0.167) (0.168) (0.169) (0.172) (0.172) (0.070) October-December 0.169 0.153 0.140 0.120 0.089 0.002 (0.182) (0.183) (0.184) (0.190) (0.190) (0.074) Household Size 0.143 0.136 0.127 0.155 0.152 -0.039 (0.020) (0.020) (0.024) (0.023) (0.023) (0.008) R 0.153 0.156 0.156 0.158 0.153 - Number 3,692 3,673 3,673 3,673 3,673 3,692 * predicted variable 18 stem from differences in the availability of medical and high levels decrease productivity and quality of care as well as the expected quality per unit price. life. Accordingly, equation 6 presents a probit analysis Predicted expenditures per illness, then, likely proxies which indicates the probability of a women having a for the quality of local health infrastructure. low BMI as an alternative to the functional forms As discussed above, only specification (5) in Table 9 which imply "more is better". The, admittedly some- is a nutrition production function and, hence, pre- what arbitrary, cutoff used for low levels of BMI is a ferred on conceptual grounds. Nevertheless, it is in- level below 18.5 (James et al.). As can be seen in the structive to compare the coefficient of total per capita table, the signs of most of the coefficients are the expenditures in the alternatives presented in the opposite of the corresponding coefficient in the other table.6 The coefficient of expenditures declines as regressions as is consistent with the dependent vari- more information is included, although the standard able being a measure of nutritional risk. Higher parity error increases only slightly. The coefficient is, there- increases the probability of a low BMI while higher fore, significant in all specifications. This raises a ques- income reduces the probability. The former coefficient tion which remains unanswered: by what pathway is, however, no longer statistical significant. The urban does income influence BMI that is distinct from dis- dummy variable and other regional patterns remain ease incidence, health expenditures and food expen- important. ditures? It is not through investment in sanitation; that should work through the illness variable which, in any Notes case, is not apparently a function of the presence of piped water or flush toilets. A plausible explanation is 12. See, for example, Pitt, Mark and Mark Rosenzweig "Health that energy expenditure per unit of time worked de- and Nutrient Consumption Across and Within Farm Households" which correlate with expenditures Review of Economics and Statistics 67, May 1985, pp. 212-23. creases with assets wRosenzweig Mark and T. Paul Schultz "Estimating a Household per capita. Moreover, if leisure is a normal good and, Production Function: Heterogeneity, the Demand for Health Inputs furthermore, less energy intensive than work, the ef- and Their Effects on Birth Weight" Journal ofPolitical Economy 91 (no. fect would be enhanced through the demand for lei- 5) pp. 723-46, 1983, describe a similar model for the household sure.27 The unexplained total expenditure effect is production of health. 13. Zvi Griliches correctly observes that although econometri- particularly large relative to other variables. cians often lament the absence of adequate data, it is precisely such The food purchases variable has no apparent effect issues that make their work interesting. See "Economic Data Issues" in specification (3) which includes the income effect. in Griliches, Z. and Michael Intrilligator, eds. Handbook of Economet- The four subgroups of food expenditure instrumental rics vol. 3, North Holland, 1986. variables are, however, jointly significant when they 14. While, in large samples, the coefficients of such instruments d win subsequent OLS regressions are consistent, the standard errors are included without the total expenditure variable. are biased downward and, thus, any t-statistics are biased upwards. Although there is some independent variation in the The standard adjustments require information about the covariance four variables due to the implicit price effects modeled matrix of the parameters in the first step (instrumenting) equations. in the fixed effects approach, it is, nevertheless, diffi- This would be readily available from estimates derived from eq. (2) cult to estimate the effect of the different components oreq. (3),butisnotobtainedwheninstrumentsarethesumofvillage means plus the predicted deviation derived from eq. (1). Note, also, of food expenditures with precision. They are likely that if one assumes that the cluster mean value of the instrumenting fairly correlated with each other. Nevertheless, it is equation is less likely to be correlated with the error in the subse- plausible that expenditures on root crops have a quent OLS regression than a village dummy variable would, the smaller impact than expenditures on cereals or meats. approach used here is less likely to give a biased result. Even with a fair amount of regional variation con- 15. Within a commodity group quality differences are generally small. Food expenditures, however, also vary due to shifts between tained within the instruments for health and food commodity groups which are often appreciable as incomes rise. expenditure, there is a highly significant difference Accordingly, food expenditures are only weakly correlated with the between BMI status of residents of the coastal amountofanygivennutrientspurchased.Foodexpenditures,there- agroecological zone and the other two zones. This fore, should be disaggregated to the greatest degree possible. parallels the effect observed for children. While there 16. That is, the physical manifestation of the genotype in inter- is no direct policy conclusion that can be derived from action with the environment. 17. A decline in conditions in the coast is unlikely. Such a trend this regional pattern (which is also observed in the should also be indicated in the interaction term for the savannah. equations explaining total expenditures) it does reflect 18. Half siblings are defined as living in the same household but a degree of heterogeneity in the country which may havingdifferentmothersandthesamerelationshiptothehousehold exceed the preferences of planners. head. If they are in a three generational household with two married sons, they may, in addition, have different fathers, and hence, not Unlike the population of children, there are many be siblings at all. There are, however, very few cases in the sample women in the sample who have BMI which are high where such an extended household was observed. enough to warrant some concern. Although difficult 19. In a personal communication, Paul Glewwe has pointed out to quantify, there is likely a loss function where low that the absence of a younger sibling may reflect childhood mortal- 19 ity. The factors which contribute to this may be correlated among of the 4,488 women surveyed indicated that they were currently family members. using birth control. 20. Although the Hausman test for exogeneity gives only weak 25. Horton interpreted her results in terms of competition for justification for the use of such an instrument, conceptually such a resources within the family. This interpretation is plausible as birth long run measure is more appropriate for regressions explaining weights in developed countries generally increase with birth order. height. There is less evidence on this relationship from low income coun- 21. This is discussed below. tries. 22. Even illness, which is a short term event, is replaced with 26. While there is a strong a priori case for using instrumental predicted illness which is a long term probability. Total expendi- variables for inputs as well as for total expenditures when a subset tures is, however, the observed current value. of expenditures is the dependent variable, there is not a strong case 23. Joanne Leslie, "Stronger, Better Nourished Women: The Key for using an instrumented variable when BMI is the dependent to Stronger, Better Nourished Children," paper prepared for Rocke- variable. Unlike height, there are few studies which indicate reverse feller Workshop on Beyond Child Survival, October 1989, argues causalityfromBMItoincome,andthelaborleisurechoiceisunlikely that there is a need for a concern with women's nutrition not just to strongly affect the model. A Hausman test rejected theendogene- maternal nutrition. This is obviously true in that they are half the ity of expenditures at the 5 percent level. In the interest of efficiency population, but the main measurable consequences are generally in of estimates, therefore, observed rather thanpredicted expenditures terms of fetal outcome or maternal depletion. are used in these regressions. 24. Ghana Statistical Service Ghana Demographic and Health Sur- 27. The total expenditure variable includes the effect of work vey 1988. Preliminary Report, Accra 1988, indicates that 20 percent of efforton expenditures andis, therefore,not ideally suited to test this women aged 15--49have ever used a modem method. Only 5 percent hypothesis. 4. Food Expenditure Patterns The regressions presented above consider food expen- maize than do the coastal and forest zones where ditures only in so far as they explain nutritional status. maize is mainly produced. The savannah region also Food policy, however, has a wider scope than nutri- spends more in proportion to its total expenditures on tion policy (Timmer, et al.). Food is a substantial com- rice, although in this case this is in keeping with the ponent of real income and its price and availability is distribution of cultivation. Neither rice nor wheat, a major indicator of the health of the economy. Thus, however, have a major share in the food budget, even policy makers have an additional concern for levels of in urban areas where they are more heavily consumed. consumption of food in general and for specific food While this study cannot devote sufficient attention commodities. to the complex issue of price policy and market inter- Two types of descriptive statistics are particularly ventions for food commodities, it should be noted that useful for food policy analysis: budget shares to vari- price stabilization is easier for commodities which are ous commodities, and the percentage of major food traded. Root crops, due to their bulk, are rarely traded items which are purchased in the market as opposed in international markets in a form suitable for human to produced at home. While more complete food pol- consumption. Both maize and sorghum are traded icy analysis requires a knowledge of price and cross commodities, but generally of grades which are used price elasticities, as well as information on marketable for animal feed. This should indicate the relative diffi- surplus, the type of information in Tables 10 and 11 cultly Ghana will have in using international markets assist in approximating real income changes attendant to assist in keeping food prices low and in stabilizing to food price movements and, hence, indicate relative the real incomes of the majority of its population. vulnerability. Such vulnerability is of particular con- On a more positive note, the diversity of the diet is cern in Ghana, as well as most other African countries, indicative of substitution possibilities. As indicated by due to relatively high food budget shares and the Pitt, increases in the price of a commodity may reduce comparatively high intra- and inter-year variability of consumption of that food, even though total nutrient prices. consumption may not change, indeed, may increase The average Ghanaian household spends between due to substitutions into other foods, Alderman (1986) 61 and 76 percent of its budget on food commodities argues that this inherent stability of nutrient con- depending on the area of residence (Tables 10a and sumption with respect to price changes is less likely in 1Ob). The differences among agroecological zones and cultures where a single staple dominates the diet. Such between urban and rural areas reflects, in part, differ- concentration is, however, not the case in Ghana. If ences in average incomes. It is, however, often ob- prices for foods are not highly correlated, substitution served that an urban resident will spend less on food between food commodities can contribute to main- than a rural counterpart with the same income, as the taining food security. In this respect, the savannah price of food in terms of nonfood is relatively higher regions with the relatively higher concentration in in urban areas. This includes the cost of housing, coarse grains (less diversified diet) are most likely transport and fuel, which generally claim more of the vulnerable to fluctuations in prices, as well as to fluc- budget of urban dwellers than of rural residents. tuations in real incomes around their low averages. Roots, tubers and plantains dominate the budget in In theory, the income elasticities for food expendi- the forest zone as well as the rural coastal zone, while tures in Table 12 should convey roughly the same in the savannah families spend more on grains than information as is reported in the breakdown of bud- on roots. Even within the broader categories strong gets by expenditure quintile in Tables 11a and b. Arc regional patterns are obvious. Millet and sorghum are elasticities may be derived from changes in budget virtually confined to the savannah zone, although that shares. For example, a constant budget share across region also devotes a larger share of its income to expenditure quintiles implies an elasticity of one while 20 21 Table 10a. Rural Food Budget Share Means by Agroecological Zone Agroecological Zone Pooled Rural Commodity Coastal Forest Savannah Sample Cereals 0.152 0.118 0.351 0.180 Maize/Kenkey 0.114 0.083 0.117 0.098 Rice 0.015 0.019 0.024 0.019 Millet/Sorghum 0.000 0.000 0.202 0.047 Wheat bread/Pasta 0.023 0.015 0.009 0.016 Roots/Tubers 0.195 0.262 0.141 0.217 Cassava/Gari/Fufu 0.148 0.117 0.061 0.113 Yams 0.010 0.025 0.066 0.031 Sweet potato/Potato 0.002 0.000 0.001 0.001 Cocoyam 0.013 0.047 0.007 0.028 Plantain 0.022 0.073 0.006 0.044 Meats/Fish/Dairy 0.165 0.164 0.082 0.145 Beef 0.005 0.006 0.009 0.006 Poultry 0.011 0.013 0.008 0.011 Other meats 0.009 0.020 0.014 0.016 Fish/Shellfish 0.135 0.121 0.049 0.108 Milk/Cheese 0.005 0.004 0.002 0.004 Other Foods 0.189 0.187 0.191 0.188 Oilpalm oil/Nuts 0.031 0.026 0.012 0.024 Other oils/Fats 0.006 0.003 0.002 0.004 Groundnuts 0.008 0.011 0.020 0.012 Fruits 0.016 0.019 0.007 0.016 Vegetables 0.076 0.088 0.095 0.086 Alcoholic beverages 0.015 0.016 0.026 0.018 All Foods 0.700 0.730 0.765 0.730 Mean Hhold Per Capita 85,313 72,289 56,315 72,500 Expenditures # of observations 524 949 444 1,917 Note: Cassava budget shares include purchases of prepared fufu. As this product may be made from other rootcrops,as well, the cassava shares may overestimate actual cassava consumption slightly. Source: GLSS an increasing budget share implies an elasticity are presented to illustrate the potential shortcoming of greater than one. Note, however, that the elasticities the widely used double log functional form; elastici- are estimated using an instrumental variable for ex- ties derived from budget shares are preferred for the- penditures in order to remove some of the correlation oretical reasons.28 Moreover, whenever the two of errors that may influence patternsinTable 11. More- estimates differ appreciably, the elasticities derived over, the elasticities are estimated using cluster fixed from budget share equations are closer to arc elastici- effects to remove any missing variable bias stemming ties calculated from cell means. from the absence of explicit price information. Conse- The elasticities for expenditures on food items are quently, the elasticities reported in Table 12 differ quite large. While elasticities for expenditures on a somewhat from a straightforward derivation of arc commodity are greater than for the physical quantity elasticities using the means reported in Tables 11a and consumed of that commodity due to quality effects, it 11b. All equations include variables for family size, is unlikely that the difference is large.29 This implies percentage of children under five years, number of that the demand for food commodities in Ghana is members with primary and with secondary schooling likely to grow in the near future as incomes increase. or above by gender, land holding, and a dummy vari- This is consistent with the comparatively low food able for female headed households. consumption for the country as well as data from other Note that a few of the elasticities are not robust with parts of West Africa (see, for example, Strauss, 1982). respect to functional form. The alternative estimates This, however, differs from the comparatively low 22 Table 10b. Urban Food Budget Share Means by Agroecological Zone Agroecological Zone Pooled Rural Commodity Coastal Forest Savannah Sample 0.162 0.135 0.258 0.162 Maize/Kenkey 0.096 0.089 0.116 0.094 Rice 0.029 0.024 0.032 0.028 Millet/Sorghum 0.001 0.001 0.091 0.010 Wheat bread/Pasta 0.037 0.022 0.019 0.030 Roots/Tubers 0.115 0.177 0.102 0.135 Cassava/Gari/Fufu 0.075 0.080 0.056 0.075 Yams 0.014 0.021 0.036 0.019 Sweet potato/Potato 0.001 0.000 0.000 0.001 Cocoyam 0.006 0.031 0.003 0.014 Plantain 0.019 0.045 0.007 0.026 Meats/Fish/Dairy 0.163 0.150 0.100 0.152 Beef 0.009 0.014 0.020 0.012 Poultry 0.018 0.014 0.009 0.015 Other meats 0.012 0.021 0.017 0.015 Fish/Shellfish 0.111 0.095 0.045 0.099 Milk/Cheese 0.013 0.007 0.007 0.010 Other Foods 0.173 0.191 0.193 0.181 Oilpalm oil/Nuts 0.025 0.022 0.014 0.023 Other oils/Fats 0.009 0.007 0.006 0.008 Groundnuts 0.008 0.010 0.017 0.009 Fruits 0.019 0.018 0.010 0.018 Vegetables 0.064 0.086 0.090 0.074 Alcoholic beverages 0.014 0.014 0.013 0.014 All Foods 0.613 0.653 0.651 0.630 Mean Hhold Per 122,619 93,342 79,663 108,446 Capita Expenditures # of observations 681 397 127 1,205 Note: Cassava budget shares include purchases of prepared fufu. As this product may be made from other rootcrops as well, the cassava shares may overestimate actual cassava consumption slightly. Source: GLSS estimates derived from time series data by Asante et ciency or marketed surplus ratio in that they only al. While there are no obvious errors in the Asante et report the share of consumption that comes from own al. study, the difference from the results here is larger production; data on the portion of production that was than generally observed when time series and cross marketed was reported in a different section of the sections are compared. The Asante results imply a survey and is not analyzed here. While there are sub- slow growth of demand for agricultural commodities stantial differences across agroecological zones, differ- and, hence, a potential for surplus production in the ences across expenditure groups are relatively small short run. The present study implies higher levels of in the rural area. Exceptions to this generalization are food consumption (and possible nutrition) with in- for relatively minor commodities. There is a pro- come growth and a need for agriculture to strive to nounced difference between expenditure groups, keep pace with demand. The difference has major however, in the urban area. Consistent with the wider policy implications. It would, therefore, be useful to range of income earning opportunities available to use combined time series and cross sectional estimates them, upper income groups are far less reliant on their from a combination of GLSS rounds when they be- own agriculture for their food consumption. One ex- come available to verify the present results. ception is the relatively high reliance on own produc- Tables 13a and 13b indicate the reliance of house- tion for root crops amongst the few well off forest and holds on markets for food consumption. The ratios savannah urban dwellers. reported differ from the more common self-suffi- 23 Table 11a. Rural Food Budget Share Means by Quintile Per Capita Expenditure Quintile Pooled Rural Commodity 1 2 3 4 5 Sample Cereals 0.206 0.193 0.187 0.165 0.150 0.180 Maize/Kenkey 0.098 0.100 0.107 0.101 0.087 0.098 Rice 0.018 0.018 0.017 0.020 0.021 0.019 Millet/Sorghum 0.077 0.063 0.047 0.028 0.020 0.047 Wheat bread/Pasta 0.014 0.012 0.015 0.017 0.023 0.016 Roots/Tubers 0.187 0.226 0.228 0.221 0.220 0.217 Cassava/Gari/Fufu 0.099 0.123 0.112 0.109 0.120 0.113 Yams 0.032 0.028 0.034 0.029 0.031 0.031 Sweet potato/Potato 0.001 0.001 0.001 0.000 0.001 0.001 Cocoyam 0.021 0.029 0.033 0.033 0.026 0.028 Plantain 0.034 0.045 0.048 0.050 0.042 0.044 Meats/Fish/Dairy 0.138 0.135 0.148 0.145 0.160 0.145 Beef 0.004 0.005 0.006 0.006 0.008 0.006 Poultry 0.008 0.009 0.012 0.012 0.014 0.011 Other meats 0.015 0.011 0.016 0.015 0.022 0.016 Fish/Shellfish 0.108 0.106 0.110 0.107 0.110 0.108 Milk/Cheese 0.002 0.003 0.003 0.005 0.007 0.004 Other Foods 0.191 0.179 0.182 0.182 0.207 0.188 Oilpalmoil/Nuts 0.023 0.021 0.026 0.025 0.025 0.024 Other oils/Fats 0.003 0.003 0.004 0.004 0.005 0.004 Groundnuts 0.013 0.012 0.012 0.012 0.014 0.012 Fruits 0.011 0.014 0.014 0.017 0.021 0.016 Vegetables 0.097 0.089 0.082 0.078 0.086 0.086 Alcoholic beverages 0.016 0.015 0.016 0.018 0.024 0.018 All Foods 0.722 0.733 0.744 0.713 0.737 0.730 Mean Hhold Per Capita 23.1 39.5 55.3 780.9 163.9 72.1 Expenditures ('000) # of observations 383 384 383 384 383 1,917 Note: Cassava budget shares include purchases of prepared fufu. As this product may contain other rootcrops as well, the cassava shares may overestimate actual cassava consumption slightly. Source: GLSS Even in rural areas, households in the coastal and harvest season and purchases some of the same com- forest zones purchase more cereals than they consume modity later in the year using wage or other earnings from their own production. A similar pattern is ob- would indicate a comparatively low share of own served for meats and other foods. The share of root production in total consumption. Nevertheless, the crops obtained from the market in the rural coastal data indicate a larger reliance on traded foods than zone, while less than half of total consumption, is commonly assumed for small farmers in Africa. These nevertheless larger than the comparable share in the farmers, then, would be affected by changes in mar- other two zones. To a degree, the high reliance on keting margins both as producers and consumers. marketed foods for rural areas reflects the allocation of land to cocoa production and other export crops in Notes the forest region. This, however, is only part of the story. In addition, many maize and rice producers sell 28. The advantages include the fact that budget share equations the bulk of their production and purchase cassava and aggregate across commodities and are less prone to heteroscadastic- ity. Moreover, they allow elasticities to vary over expenditure levels. yams. Moreover, as the ratio is calculated here, a 29. Deaton found only a small difference for C6te d'Ivoire. household which sells some of their production at 24 Table 11b. Urban Food Budget Share Means by Quintile Per Capita Expenditure Quintile Pooled Urban Commodity 1 2 3 4 5 Sample Cereals 0.187 0.162 0.161 0.161 0.147 0.162 Maize/Kenkey 0.119 0.100 0.093 0.088 0.080 0.094 Rice 0.027 0.027 0.026 0.032 0.026 0.028 Millet/Sorghum 0.020 0.012 0.015 0.003 0.001 0.010 Bread/Pasta 0.021 0.023 0.027 0.038 0.039 0.030 Roots/Tubers 0.156 0.152 0.130 0.118 0.113 0.135 Cassava/Gari 0.089 0.081 0.071 0.064 0.068 0.075 Yams 0.017 0.019 0.020 0.020 0.017 0.019 Sweet potato/Potato 0.001 0.001 0.001 0.001 0.001 0.001 Cocoyam 0.019 0.021 0.014 0.009 0.007 0.014 Plantain 0.030 0.030 0.024 0.024 0.020 0.026 Meats/Fish/Dairy 0.129 0.142 0.153 0.160 0.175 0.152 Beef 0.010 0.010 0.013 0.013 0.013 0.012 Poultry 0.009 0.012 0.016 0.019 0.021 0.015 Other meats 0.011 0.014 0.014 0.016 0.021 0.015 Fish/Shellfish 0.096 0.098 0.102 0.097 0.101 0.099 Milk/Cheese 0.004 0.007 0.008 0.014 0.018 0.010 Other Foods 0.187 0.181 0.178 0.170 0.190 0.181 Oilpalm oil/Nuts 0.025 0.025 0.024 0.019 0.021 0.023 Other oils/Fats 0.006 0.008 0.008 0.009 0.010 0.008 Groundnuts 0.011 0.010 0.009 0.008 0.008 0.009 Fruits 0.014 0.016 0.015 0.022 0.023 0.018 Vegetables 0.090 0.080 0.075 0.062 0.064 0.074 Alcoholic beverages 0.008 0.012 0.014 0.013 0.022 0.014 All Foods 0.660 0.637 0.621 0.608 0.625 0.630 Mean Hhold Per Capita 36.2 60.0 84.2 124.7 236.8 108.4 Expenditures ('000) # of observations 241 241 241 241 241 1,205 Note: Cassava budget shares include purchases of prepared fufu. As this product may contain other rootcrops as well, the cassava shares may overestimate actual cassava consumption slightly. Source: GLSS 25 Table 12. Expenditure Elasticities of Various Foods: Estimates from Log-Log and Budget Share Modelsa Log-Log Model Budget Share Model Commodity Rural Urban Rural Urban Cereals 0.9180 1.0637 0.6904 0.9139 Maize/Kenkey 0.8390 0.8285 0.6049 0.7875 Rice 0.9984 1.5026 1.2669 1.2017 Millet/Sorghumb 1.2234 -1.1010 0.7041 0.5789 Wheat bread/Pasta 1.8053 1.9076 1.3488 1.4772 Rootcrops/Plantains 0.9346 1.2027 0.8974 1.0929 Cassava/Gari/Fufu 0.7255 1.4567 0.8795 1.0284 Plantain 0.1265 1.0979 0.4854 1.0983 Meats/Fish/Poultry/Dairy 0.8995 0.9493 0.9665 0.9047 Beef 1.5232 0.2296 2.6154 1.3069 Poultry 2.0698 2.5224 2.6154 1.3069 Other meats 2.0475 0.9459 1.7592 1.2827 Fish 0.3838 0.5103 0.6580 0.6041 Dairy Products 1.7009 2.7676 1.8563 1.7433 Other Foods 1.1503 0.7814 1.0861 0.8569 Oilpalm oil/Nuts 0.9577 -0.8326 0.9094 0.3105 Other oils/Fats 0.6946 1.3475 0.9885 1.1754 Fruits 1.7006 2.0705 1.3975 1.9454 Vegetables 1.0612 -0.4389 1.0946 0.4557 Groundnuts 1.1992 1.4404 0.9869 1.0289 Alcoholic beverages 0.1122 1.2328 1.4986 1.9379 Total Food 0.9097 0.9274 0.9115 0.9390 # of Observations 1,914 1,188 1,914 1,188 Notes: The dependent variable in the log-log model is the natural logarithm of per capita expenditures on the food commo- dity; in the budget share model it is the proportion of total household expenditures going toward the food commodity. In both models the income instrument is the predicted natural log of total per capita household expenditures. a. Insofar as the standard errors of the estimates for the two models are in different units and are thus not strictly compara- ble, they are not reported here. They are available from the author upon request. b. Millet/sorghum elasticities were estimated using the subset of households in the Savannah zone (n=441 for rural, n=111 for urban). Source: GLSS 26 Table 13a. Share of Home Production in Rural Food Expenditure-Top and Bottom Quintiles by Agroecological Zone Coastal Forest Savannah Commodity Lower Upper Lower Upper Lower Upper Total Food 0.266 0.281 0.486 0.415 0.663 0.650 Cereals 0.279 0.199 0.479 0.452 0.876 0.849 Maize/Kenkey 0.356 0.303 0.615 0.658 0.906 0.887 Rice 0.000 0.048 0.126 0.129 0.521 0.570 Millet/Sorghum .. .. .. .. 0.932 0.973 Roots/Tubers 0.592 0.589 0.824 0.738 0.868 0.831 Cassava/Gari/Fufu 0.650 0.606 0.817 0.726 0.813 0.861 Yams 0.572 0.597 0.746 0.563 0.935 0.819 Sweet potato/Potato 0.683 0.122 0.667 0.112 0.754 0.481 Cocoyam 0.808 0.775 0.944 0.927 0.977 0.884 Plantain 0.404 0.576 0.887 0.836 0.936 0.848 Meats/Fish/Dairy 0.059 0.129 0.066 0.119 0.218 0.345 Beef 0.000 0.055 0.000 0.000 0.000 0.000 Poultry 0.429 0.335 0.638 0.554 0.828 0.730 Other meats 0.474 0.296 0.284 0.334 0.498 0.262 Fish/Shellfish 0.027 0.082 0.001 0.011 0.059 0.416 Milk/Cheese 0.000 0.000 0.000 0.008 0.089 0.000 Other Foods 0.227 0.241 0.412 0.414 0.430 0.579 Oilpalm oil/Nuts 0.250 0.274 0.537 0.501 0.302 0.297 Other oils/Fats 0.091 0.152 0.025 0.179 0.009 0.221 Groundnuts 0.000 0.110 0.196 0.161 0.639 0.927 Fruits 0.337 0.442 0.848 0.820 0.444 0.230 Vegetables 0.297 0.313 0.482 0.422 0.633 0.790 Alcoholic beverages 0.000 0.000 0.106 0.247 0.019 0.009 # of observations 67 144 179 182 137 57 Notes: Calculations are ratios of consumption from home production to total consumption. These differ from self-sufficien- cy ratios, which are ratios of total production to total consumption. Ratios for Millet/Sorghum consumption are calculated only for the Savannah zone, since only small amounts are consumed in the other regions of the country. Source: GLSS 27 Table 13b. Share of Home Production in Urban Food Expenditure-Upper and Lower Quintiles by Zone Coastal Forest Savannah Commodity Lower Upper Lower Upper Lower Upper Total Food 0.067 0.039 0.252 0.097 0.454 0.105 Cereals 0.066 0.067 0.174 0.065 0.679 0.092 Maize/Kenkey 0.103 0.142 0.269 0.142 0.707 0.131 Rice 0.001 0.000 0.021 0.000 0.204 0.090 Millet/Sorghum 0.000 0.000 0.000 0.000 0.872 0.030 Roots/Tubers 0.113 0.081 0.520 0.214 0.546 0.233 Cassava/Gari/Fufu 0.135 0.117 0.483 0.163 0.717 0.251 Yams 0.000 0.018 0.432 0.326 0.352 0.069 Sweet potato/Potato 0.341 0.000 0.000 0.025 0.000 0.000 Cocoyam 0.407 0.196 0.827 0.575 0.951 0.195 Plantain 0.078 0.078 0.560 0.267 0.369 0.538 Meats/Fish/Dairy 0.042 0.028 0.047 0.062 0.188 0.006 Beef 0.000 0.000 0.000 0.000 0.321 0.000 Poultry 0.149 0.067 0.404 0.253 0.613 0.046 Other meats 0.000 0.043 0.134 0.177 0.313 0.014 Fish/Shellfish 0.041 0.024 0.000 0.000 0.032 0.000 Milk/Cheese 0.000 0.000 0.000 0.000 0.000 0.000 Other Foods 0.020 0.029 0.144 0.088 0.307 0.101 Oilpalm oil/Nuts 0.026 0.008 0.171 0.061 0.075 0.038 Other oils/Fats 0.002 0.000 0.041 0.000 0.029 0.000 Groundnuts 0.014 0.000 0.051 0.007 0.479 0.021 Fruits 0.049 0.133 0.287 0.243 0.135 0.009 Vegetables 0.024 0.030 0.186 0.147 0.483 0.214 Alcoholic beverages 0.000 0.000 0.026 0.014 0.000 0.000 # of observations 92 175 101 54 48 12 Note: Calculations are ratios of consumption from home production to total consumption. These differ from self-sufficiency ratios, which are ratios of total production to total consumption. Ratios for Millet/Sorghum consumption are calculated on- ly for the Savannah zone, since only small amounts are consumed in the other regions of the country. Source: GLSS 5. Conclusion This study indicates a positive relationship between less, targeted programs may counteract some of these income and nutritional levels. While this may be handicaps. deemed obvious, it is surprisingly difficult to demon- For example, the nutrition of children can be af- strate and often contradicted (Behrman, Deolalikar, fected in the short run through improved prenatal and Wolfe). The results here confirm a similar study nutrition. There is ample evidence that medical care based on C6te d'Ivoire data (Sahn). Many of the long and intervention programs, especially those targeted term impacts of current economic policies, although to mothers exhibiting less than normal growth during introduced for reasons other than nutrition, will likely pregnancy, can reduce the probability of a low birth affect nutrition through this income relationship. A weight child. While food supplementation programs revival of Ghana's once noteworthy educational sys- designed for children often fail to indicate measurable tem may also affect nutrition via income, rather than impacts on nutrition, partially due to imprecise target- directly shifting the effectiveness of input utilization. ing, programs aimed at women in the later stages of We are not, however, able to investigate the impact of pregnancy are more likely to have significant results any particular existing government program. This is, (Kennedy and Alderman, Anderson). For example, in part, due to the absence of infrastructure variables, one of the most noted studies of maternal supplemen- although there is some indirect evidence of the role of tation indicated a major reduction of low birth weight health infrastructure through the probability of dis- children in the Gambia, although little effect on lacta- ease. Moreover, the Government of Ghana is still in tion was observed (Prentice et al.). the process of restoring the overall economy as well as This discussion has touched upon two forms of social infrastructure and does not currently have na- targeting. A priority for women in the later stages of tionwide programs aimed at improving nutrition, the pregnancy is a form of life cycle targeting, while the impacts of which might be investigated in a similar regional pattern implies geographic targeting.30 An- analysis of nutritional status. Similarly, the GLSS was other type of targeting that is often considered, target- not designed to investigate the impact of those types ing by income or other measures of economic status, of supplementary feeding programs that are being often requires more data and trained administrators undertaken. One can, however, extrapolate a bit from than is currently available in Ghana. the evidence in this data as well as from other pro- The suggestion of a focus on pregnant women is a grams to consider potential interventions. divergence from the main form of nutritional inter- For example, the regional patterns verified here con- vention that was included in the PAMSCAD package. firm the persistence of a nutritional disadvantage for To be sure, no one wants to argue, in an either/or children born in the savannah and also provide a basis manner, against providing extra food to under weight for determining priorities. The higher levels of malnu- children. Nevertheless, it is often difficult to design trition in the North is not a new disparity-it was supplementary feeding programs which are sharply observed in nationwide surveys nearly 30 years ago. focused on children in the vulnerable 6 to 24 month The legacy represents a vicious circle; the effect which bracket. Such programs work best when coupled with is referred to above as the womb environment or health care centers. They have their main impact phenotype pathway carries the effects of malnutrition through the incentives for monitoring and preventa- to subsequent generations. Regional differences may tive care. The potential for an impact of the food also reflect ecological constraints as well as the rela- distributed is enhanced if the quantity increases when tively high costs of health infrastructure and utiliza- a child indicates failure to thrive. This has been at- tion in a region with difficult transportation and, in tempted by voluntary agencies in Ghana with mixed places, relatively low population densities. Neverthe- results. The success of such programs is dependent on 28 29 increased utilization of health care outlets which, in the age at which a mother begins her family, an effect turn, is dependent on availability and quality. on children's nutrition should be observed as indi- Like child oriented programs, interventions tar- cated by the maternal age variable in the regressions. geted to pregnant women can be subtargeted by other Similarly, any program which reduces the probability criteria, including low weight gain. Moreover, they of childhood disease will have an additional affect on share a feature in that a portion of the benefits comes health. Preventative health programs, then, may be a from the increased incentives to obtain prenatal care.31 more direct pathway to improving nutrition than sup- Although the number of women in the later stages of plementary feeding programs. pregnancy is lower than the number of preschool It is not possible, however, to say much about the children or the number of children in the more narrow potential for food pricing policy to improve nutrition. band of maximum vulnerability, universal coverage is Not only are there few cost effective instruments likely infeasible for the short or medium term. which can influence marginal prices for vulnerable This, then, returns the discussion to geographic tar- groups in Ghana, it is hard to interpret the available geting. Such targeting is administratively simple and food expenditure variables in a policy context. They virtually free of incentives to alter behavior or conceal are not strong predictors for children's nutritional information that influences the efficiency of some tar- status, although they are for adult women. Even in this geting schemes. It, however, has the potential to add case, however, one would be foolish to venture that an to inter-regional political tensions. Moreover, geo- increase in the price of foods which are inelastic-im- graphic targeting often assumes away major differ- plying increased expenditures following such a price ences in the costs of delivering a service. When the cost increase-is a sound nutritional policy. More conclu- of delivering services is greater in remote regions, the sive results, and more useful policy levers, might be most effective means of achieving national objectives uncovered if the nutrients for which food expendi- are not necessarily indicated by poverty lines or sim- tures are believed to proxy can be calculated. This may ilar targeting aids. With semi-urban areas exhibiting be possible when the full set of prices which were comparatively high levels of chronic malnutrition, collected are linked with the household data. these areas may be a convenient entry point for pro- While the strength and types of conclusions that can grams, particularly in the savannah regions, to be be drawn from the GLSS data is limited by data prob- augmented if, or when, experience and infrastructure lems referred to above, the surveys are designed for permit, in high priority rural regions. monitoring as well as for research which elucidates While the GLSS data collection is not designed for pathways. Repeated surveys as envisioned by the program evaluation, the data may assist in targeting GLSS project can provide such a monitor for overall criteria. In particular, the increased sample size that economic progress as well as specific progress in im- will be possible when the second (and subsequent) proving social welfare. Repeated surveys may, fur- years of the survey are available can conveniently thermore, provide increased information to isolate provide statistics on nutrition at greater levels of geo- effects only weakly illuminated with the current data. graphic disaggregation than currently advisable. The analysis above points to a few additional gen- Notes eralities which can augment the planning of health and nutrition strategies. For example, given the effect 30. For a review of targeting methods and results see Alderman, of parity on the mother's nutrition and the sibling forthcoming, as well as Pinstrup-Andersen. effect observed in the children's regression, programs 31. The GLSS data indicates that 88 percent of all mothers (84 percent in rural areas) reported prenatal care. This is surprisingly, which influence both the number and spacing of births even suspiciously, high and does not indicate the type or the quality will likely improve nutrition.32 If, furthermore, educa- of care obtained. tion or specific family planning programs influence 32. For a recent review, see Haaga. Appendix: Estimation of Instruments While the estimating of instrumental variables is un- ture equation this should measure the long run net dertaken primarily to obtain regressors for the main returns). Land, which is allowed to vary by ecological estimates, there are a number of points of note in these zone due to quality factors, also has a small but posi- equations. These are discussed briefly below. tive impact. The coefficient of land in cocoa, while Total Expenditures (Appendix Table 1) are used as a positive, is also surprisingly small, and not significant. proxy for unobserved permanent income. In prelimi- Dummy variables for agroecological zones and type nary estimates (not shown) it was observed that ex- of urban setting indicate that locale has a significant penditure equations were more plausible than income effect, even controlling for different levels of assets. predictors. While most coefficients in the expenditure Finally, female headed households have significantly functions are quite consistent with a permanent in- smaller expenditures in both rural and urban areas. come interpretation, the same can not be said with the Parity Regressions were estimated using OLS (Ap- income equations. For example, urban household ex- pendix Table 2). While the dependent variable is a penditures increase 100,000 cedi for every university count variable (integer) rather than a continuous vari- degree in the household (male) and 30,000 for primary able and is, furthermore, truncated at zero, Ainsworth completers, with rural results being surprisingly sim- has indicated that techniques to correct for the depar- ilar. In the urban income equations the corresponding ture from OLS assumptions that these conditions en- numbers are 9,000 and -2,000 and in the rural they are tail do not have a noteworthy influence on the results. not significant at -17,000 and 8,000 respectively. The The dependent variable is the number of children ever problem is also indicated looking at the average level born, regardless of survival. of income for the entire sample which is roughly only While the age variables are both highly significant half of average expenditures. It is more disturbing that in the two equations, this merely confirms the cumu- the underreporting of income (for a number of fairly lative pattern of the dependent variable. obvious reasons it is unlikely that expenditures are The main concern, however, is with the income overreported) is clearly not uniform. If one looks at variables which indicate increasing parity at a given regional means (urban/non urban) or at means by age and education as incomes increase. A similar pat- agroecological zones one sees that Accra has the high- tern was observed by Ainsworth for C6te d'Ivoire, est expenditures and the lowest incomes while the although it is not commonly observed in developed situation is roughly reverse for the rural savannah. countries. Unlike Ainsworth, however, the sp6cifica- Perhaps, government employees are more likely to tion here includes a quadratic term. The turning point underreport incomes. at which the marginal impact of income becomes neg- While the comparison between expenditure and ative occurs when the logarithm of total expenditures income regressions was done in total cedis terms, it is roughly one and a half standard deviations above was noted that these regressions were heteroscadastic. the mean which is well within the sample. This problem was reduced when dependent variable Post-primary education has a strong effect on parity was transformed into natural logarithms. The results at a given age.' The coefficients for post-secondary reported in Appendix Table 1 are, furthermore, trans- education are higher than for secondary education. In formed into per capita terms. contrast to C8te d'Ivoire, education has a significant The equations indicate that, in general, education influence on parity even in rural areas. had the expected positive impact on total family ex- penditures, with the coefficient in the rural equations being often higher than in the urban. Note, also, that Notes an additional woman with postprimary education has anadiina omn th o primaf eedire se 1. The study did not determine the impact on completed births, a larger impact on the logarithm of expenditures per that is the totalnumber of children after a woman has completedher capita (or total family expenditures) than does an childbearing. More educated women generally start their childbear- additional male with a similar education. Nonland ing later, but may not necessarily have less children over their assets have small but positive impacts (in an expendi- lifetime. 31 32 Appendix Table 1. Expenditures Equations (Standard errors in parentheses) Log Total Expenditures Per Capita Variable Rural Urban Pooled Intercept 11.501 11.859 11.177 (0.063) (0.074) (0.045) Males 0-5 yrs. -0.157 -0.188 -0.086 (0.018) (0.022) (0.011) Males 5-10 yrs -0.112 -0.182 -0.086 (0.020) (0.023) (0.012) Males 10-20 yrs -0.091 -0.155 -0.078 (0.016) (0.020) (0.009) Males 20-65 yrs -0.041 -0.086 -0.023 (0.028) (0.037) (0.017) Males 65+ yrs -0.039 -0.212 -0.074 (0.058) (0.085) (0.037) Females 0-5 yrs -0.142 -0.207 -0.089 (0.018) (0.023) (0.011) Females 5-10 yrs -0.117 -0.154 -0.092 (0.021) (0.024) (0.013) Females 10-20 yrs -0.077 -0.093 -0.054 (0.017) (0.021) (0.009) Females 20-65 yrs -0.111 -0.093 -0.035 (0.023) (0.028) (0.013) Females 65+ yrs -0.220 -0.102 -0.116 (0.049) (0.063) (0.026) Males with primary 0.075 0.068 0.051 education (only) (0.025) (0.027) (0.015) Males with second- 0.144 0.197 0.133 ary education (0.083) (0.062) (0.042) Males with univer- 0.697 -0.013 -0.017 sityeducation (0.288) (0.156) (0.131) Females with pri- 0.013 0.103 0.055 mary education (0.025) (0.026) (0.013) (only) Females with sec- 0.220 0.291 0.265 ondary education (0.167) (0.092) (0.065) Land in Coastal 1.187a 0.784a 0.445a Zone (0.446) (1.097) (0.313) Land in Forest Zone 0.767a 0.712a 0.912a (0.255) (1.313) (0.216) Land in Savannah 2.919a 7.000a 3.035a Zone (0.460) (3.905) (0.395) Cocoa Acreage 0.413a 3.213a 0.479a (0.479 (3.8681 (0.413k Value of Livestock 5.73 3.107 3.701 VaueofVeices(1.654Z (5.8131 (1.029 Value of Vehicles 2.983 3.396 3.643 (1.548 (1.702 (0.727) Value of Other 6.623 1.176 1.610b Capital (1.786) (0.577) (0.439) Semi Urban dummy 0.071 - 0.062 (0.029) (0.024) Urban dummy - - 0.213 (0.023) Forest Zone dummy -0.119 -0.089 -0.112 (0.033) (0.041) (0.023) Savannah Zone -0.352 -0.178 -0.350 dummy (0,043) (0.062) (0,030) 33 Appendix Table 1. Continued Log Total Expenditures Per Capita Variable Rural UJrban Pooled Female head of -0.119 -0.203 -0.071 household (0.040) (0.046) (0.024) dummy Age of head of -0.003 -0.003 -0.003 household (0.001) (0.001) (0.0008) Interview month 0.025 0.018 0.025 trend (0.004) (0.005) (0.003) Accra dummy - 0.200 0.247 (0.042) (0.037) Tema dummy - 0.020 0.068 (0.064) (0.055) R 0.335 0.466 0.346 Number 1,914 1,188 3,673 a. Times 10- b. Times 107 Appendix Table 2. Parity Estimates (Cluster Fixed Effects) Rural Urban Intercept -144.912 -155.427 (31.315) (36.909) Age 0.393 0.385 (0.034) (0.044) Age Squared -0.002 -0.003 (0.0005) (0.0007) Highest Schooling = primary 0.042 -0.069 (0.158) (0.197) Highest Schooling = secondary -0.275 -0.873 (0.118) (0.151) Highest Schooling = post secondary -1.629 -1.604 (0.474) (0.271) Log Total Expenditure 20.980 22.356 (4.947) (5.718) (Log Total Expenditure) -0.795 -0.835 (0.197) (0.221) Land Ownership -0.0005 -0.001 (0.0006) (0.004) Cluster Mean Parity 0.460 0.560 (0.054) (0.064) R of Equation 0.663 0.584 Number 1,248 782 References Alderman, Harold: "Food Subsidies and the Poor", in Duncan, Gregory (1987): "A Simplified Approach to Psacharopoulos, George. ed. Essays on Poverty, Eq- M-Estimation With Application to Two-Stage Esti- uity, and Growth, World Bank, Washington, D.C. mators", Journal of Econometrics 34, pp. 373-89. Forthcoming. Dugdale, A. E. (1985): "Family Anthropometry: a New S (1989): "New Research on Poverty and Malnu- Strategy for Determining Community Nutrition", trition. What are the Implications for Research and The Lancet, September 21, pp. 672-73. Policy?", paper presented at World Bank-IFPRI con- Ghana Statistical Service (1989): "Demographic and ference on Poverty, Arlie House Oct 25-28. Forth- Health Survey 1988", Accra. coming in Lipton, Michael and Van der Gaag, Ghana, Statistical Service, and World Bank Social Di- Jacques (eds.), untitled volume of poverty research. mensions of Adjustment Unit (1989): "Ghana Living Alderman, Harold (1986): "The Effects of Income and Standards Survey, First Year Report", Accra. Food Price Changes on the Acquisition of Food by Government of Ghana (1987): "Programme of Actions Low-Income Households", International Food Pol- to Mitagate the Social Costs of Adjustment", Accra. icy Research Institute, Washington, D.C. Griliches, Zvi (1986): "Economic Data Issues", in Anderson, Mary Ann (1989): "The Relationship Be- Griliches, Z. and Michael Intrilligator (eds.), Hand- tween Maternal Nutrition and Child Growth in book of Econometrics, Vol. 3, North Holland. Rural India", unpublished Ph.D Thesis, Tufts Uni- Haaga, John (June 1989): "Family Planning as a Child versity, School of Nutrition. Survival Program: Evidence, Implications, and the Ainsworth, Martha (1989): "Socioeconomic Determi- Role of Information", paper prepared for Rockefel- nants of Fertility in C8te d'Ivoire", Living Standards ler Foundation conference on "Beyond Child Sur- Measurement Study, Working Paper No. 53, World vival", Cornell University. Bank, Washington, D.C. Horton, Susan (1988): "Birth Order and Child Nutri- Asante E., S. Asuming-Brempong, and P. Bruce (1989): tional Status: Evidence From the Phillipines", Eco- "Ghana; Grain Marketing Study", World Bank, nomic Development and Cultural Change 36, pp. Washington, D.C. Mimeo. 341-54. Beaton, George (1989): "Small but Healthy? Are We Huffman, Sandra L., Mark Wolff, and Sarah Lowell Asking the Right Question?", Human Organization (October 1985): "Nutrition and Fertility in 48, pp. 30-9. Bangladesh: Nutritional Status of Nonpregnant Behrman, Jere R. and Barbara L. Wolfe (1987): "How Women", The American Journal of Clinical Nutrition Does Mother's Schooling Affect the Family's 42, pp. 725-38. Health, Nutrition, Medical Care Usage and House- James, W., A. Ferro-Luzzi, and J. Waterlow (1988): hold Sanitation?", Journal of Econometrics 36, pp. "Definition of Chronic Energy Deficiency in 185-204. Adults", European Journal of Clinical Nutrition 43, pp. Behrman Jere R., Anil Deolalikar, and Barbara Wolfe 969-81. (September 1988): "Nutrients: Impacts and Deter- Jelliffe, D. and E. Jelliffe (1978): "Human Milk in the minants", The World Bank Economic Review 2 (no. 3) Modern World", Oxford University Press, Oxford, pp. 299-320. UK. CRS (1987): "Growth Surveillance System. Annual Re- Journal of the Royal College of Physicians 17 (Nov. 1, port 1986", Accra, Mimeo. 1983), pp.1-53. Special Issue on Obesity. Deaton, Angus (1987): "Estimation of Own- and Kennedy, Eileen and Harold Alderman (1987): "Com- Cross-price Elasticities from Household Survey parative Analysis of Nutritional Effectiveness of Data", Journal of Econometrics 36, pp. 7-30. Food Subsidies and Other Food-Related Interven- 34 35 tions", IFPRI and WHO-UNICEF Nutrition Support Macroeconomic Restructuring", World Development Program, Washington and New York. 15, pp. 809-30. Kenya, Republic of (1979): Child Nutrition in Kenya, - (1989): "The Prevalence and Determinants of Ministry of Economic Planning, Government Malnutrition in C6te d'Ivoire", World Bank, Wash- Printer, Nairobi. ington, D.C., Mimeo. Kumar, Shubh (1985): "The Nutrition Situation and its Seckler, David (1982): "Small but Healthy: A Basic Food Policy Links", in John Mellor, Christopher Hypothesis in the Theory, Measurement and Policy Delgado, and Malcolm Blackie, (eds.), Accelerating of Malnutrition", in New Concepts in Nutrition and Food Production in Sub Saharan Africa, Johns Hopkins their Implication for Policy, Maharastra Association University Press, Baltimore, MD. for the Cultivation of Science, Pune, India. Levinson, F. J (1988): "Combating Malnutrition in Strauss, John. "Households, Communities and Pre- Ghana", Mimeo. school Children's Nutriton Outcomes: Evidence Martorell, Reynaldo and Teresa Gonzalez-Cossio from Rural C6te d'Ivoire", Economic Development (1987): "Maternal Nutrition and Birth Weight", and Cultural Change, Forthcoming. Yearbook of Physical Anthropology 30, pp. 195-220. - Forthcoming: "The Impact of Improved Nutri- McGuire, Judith and James Austin (1987): "Beyond tion in Labor Productivity and Human Resource Survival: Child Growth for Nutrition, Health and Development: an Economic Perspective", in Per Development", UNICEF, New York. Pinstrap-Andersen, Political Economy of Nutrition Merchant, Kathleen and Reynaldo Martorell (1983): Policy, Johns Hopkins. "Frequent Reproductive Cycling: Does it Lead to - (1982): "Determinants of Food Consumption Nutritional Depletion of Mothers?", Progress in Food and Production in Rural Sierra Leone", Journal of and Nutrition Science, Vol. 12, pp. 339-69. Development Economics 14 pp. 77-103. Murphy, Kevin M. and Robert H. Topel (1985): "Esti- Svedberg, Peter (1988): "Undernutrition in Sub- mation and Inference in Two-Step Econometric Saharan Africa: Is There a Sex Bias?", Institute for Models", Journal of Business & Economic Statistics, International Economic Studies, Seminar Paper Vol. 3, No. 4. #421, University of Stockholm. Payne, Phillip (1986): "Appropriate Indicators for Thomas, Duncan, John Strauss, and Maria Helena Project Design and Evalution", in Food Aid and the Henriques "Child Survival, Height for Age and Well-Being of Children in the Developing World, UNI- Household Characteristics in Brazil", Journal of De- CEF, New York. velopment Economics, Forthcoming. Pinstrup-Andersen, Per (1988): "Consumer-Oriented - (1988): "How Does Mother's Education Affect Food Subsidies: Cost, Benefits and Policy Options Child Height?", University of Warwick, Coventry, for Developing Countries", Johns Hopkins Univer- Mimeo. sity Press, Baltimore, MD. Timmer, Peter, Walter Falcon and Scott Pearson Pitt, M.M. (1983): "Food Preferences and Nutrition in (1983): "Food Policy Analysis", Johns Hopkins Uni- Rural Bangladesh", Review ofEconomics and Statistics versity Press, Baltimore, MD. 65:1 (February), pp. 105-14. Tobin, J (1979): "On Limiting the Domain of Inequal- Pitt, Mark and Mark Rosenzweig (1985): "Health and ity", Journal of Law and Economics 13, pp. 263-77. Nutrient Consumption Across and Within Farm UNICEF (1988): "Adjustment Policies and Programs Households", Review of Economics and Statistics 67, to Protect Children and Other Vulnerable Groups in pp. 212-23. Ghana", in Cornia, Giovanni, Richard Jolly and Prentice, Andrew, etal. (1983): "Pre-natal DietarySup- Frances Stewart (eds.), Adjustment With a Human plementation of African Women and Birthweight", Face, Oxford University Press. Lancet (i) pp. 489-92. USAID (United States Agency for International Devel- Rimmer, Douglas. "Poverty, Equity and Growth in opment) (1975): Nepal National Nutrition Survey, Of- Ghana: 1950-1988", World Bank draft manuscript fice of Nutrition, USAID, Washington, D.C. of a forthcoming book. (1976a): Arab Republic of Egypt National Nutri- Rosenzweig, Mark and T. Paul Schultz (1983): "Esti- tion Survey, Office of Nutrition, USAID, Washing- mating a Household Production Function: Hetero- ton, D.C. geneity, the Demand for Health Inputs and Their - (1976b):Liberia National NutritionSurvey,Office Effects on Birth Weight", Journal of Political Economy of Nutrition, USAID, Washington, D.C. 91 (no. 5), pp. 723-46. (1976c): Sri Lanka National Nutrition Survey, Sahn, David E (1987): "Changes in the Living Stan- Office of Nutrition, USAID, Washington, D.C. dards of the Poor in Sri Lanka During a Period of 36 (1977a): United Republic of Cameroon National Waterlow, J. (1973): "Note on the Assessment and Nutrition Survey, Office of Nutrition, USAID, Wash- Classification of Protein Energy Malnutrition in ington, D.C. Children", Lancet 1, pp. 87-9. (1977b): Sierra Leone National Nutrition Survey, Waterlow, J., R. Buzina, W. Keller, J. Lane, M. Nicha- Office of Nutrition, USAID, Washington, D.C. man, and J. Tanner (1977): "The Presentation and (1977c): Togo National Nutrition Survey, Office Use of Height and Weight Data for Comparing the of Nutrition, USAID, Washington, D.C. Nutritional Status of Groups of Children Under the (1977d): Lesotho National Nutrition Survey, Age of 10 Years", Bulletin of the World Health Organi- Washington, D.C., Mimeo. zation 55 (4), pp. 489-98. (1986): Swaziland National Nutrition Survey World Bank World Development Report 1989, Oxford 1983, Full Report, Washington, D.C., Mimeo. University Press. USHEW (United States Department of Health, Educa- tion, and Welfare) (1974, 1975): Sahel Nutrition Sur- veys, Centers for Disease Control, USHEW, Atlanta. s Distributors of World Bank Publications ARGENTINA FINLAND MALAYSIA Forsulscriptionrdrs, Carlos Hirsch, SRL Akateemnen Kirjakauppa University of Malaya Cooperative International Subscription Service Galeria Guemes P.O. Box 126 Bokshop, Limited P.O. Box 41095 Florida 165,4th Roor-ac 453/465 SF-0101 P.O. Bax 1127, Jala Pantai Baru Craighall 1333 Buenos Aies Helsinki 10 Kuala Laumpur Johannesburg 2024 AUSTRALIA, PAPUA NEW GUINBA, FRANCE MEXICO SPAIN FIJI; SOLOMON ISLANDS, Worid Bank Publications INFOTEC Mundi-Prena Libras, SA. VANUATU, AND WESTERN SAMOA 66, avened'lda Apartado Postal 22-960 Castello 37 D.A. Books &Journals 75116 Paris 1406oT1alpan, Mei D.F. 2601 Madrid 648 Whitehorse Road Mitcham 3132 GERMANY, FEDERAL REPUBLIC OF MOROCCO Libreda Internadonal AEDOS Victoria UNO-Verlag Societe d'Etudes Marketing Marocaine Cnsell de Cent, 391 PoppeldorferAlleeSS 12 reMaar,B d. d'Afa 009Barcelons AUSTRIA D-530 Bonn 1 Casablanca Gerald and Co. SRI LANKA AND THE MALDIVES Gram-31 GREECE NETHERLANDS LakeHouseBookshop A-1011 Wien KEME InOr-Publikatiesbv. PO. Box 244 24, Ippodamon Street Platia Plastiras P.O. Box 14 100, Sir Chittampalom A. Gardiner BAHRAIN Athena-11635 7240 BA Lochem Mawatha Bahrain Research and Consultancy Colombo 2 Assodates Lid. GUATEMALA NEW ZEALAND P.O. Box 22103 ibreinas Piedra Santa Hills Ubrary and Information Service SWEDEN Manama Town 317 Centro Cultural Piedra Santa PrivateRag For as ing ilWas 11 mile 650 an 1 New Market Fritzes Fackbokforetagt BANGLADESH GuatealaCity Auckland Reginggatan 12, Bx 16356 Micro Industries Development S,103 27 Stockholm Assistan- Society (MIDAS) HONG KONG, MACAO NIGERIA House 5, Road 16 Asia 200D ld. Univerty Pees lumited Forouc ion o,ders Dhanmondi R/Area Mangkok Post Office Three CrownsBuildingJericho Wernergren-WillisaeAB Dhaka 1209 Bute Street No.37 PrivateMail Bag5095 Box 3DD4 Mo-gkok. Kowloon Ibada S-104 25 Stockholm Banc, ofk: Hong Kong 15% Nor Ahmed Sarak NORWAY SWITZERLAND Chittagong 4000 HUNGARY Narvesen Information Center Forisle timl. Kaltura BertrandNarvesensvei2 Librairie Payat 76, KDA. Avenue P.O. Box 139 P.O. Box 6125 Etterstad 6, =e Grenus Kulna 1389 Budapest 62 N4)602 Oslo 6 Case postal 381 CH 1211 Ceneva 11 BEIGIUM INDIA OMAN Publications des Nations Unies Allied Publishers Private Ltd. MEMRB Information Services Forsamction ordens Av. du Ro* 202 751 Mount Road P.O. Box 1613, Seeb Airport Librairie Payot 1060 Brussels Madras - 600 002 Muscat Service desAbormements Casepostal3312 BRAZIL Bnh officer PAKISTAN CH 1002 Lausarme Publicaces Te-nicas Internadonai. ISJ.N. Henedi Marg Mirza Book Agency Ltda. Ballard Estate 65, 5hhrah--quaid-eAa TANZANIA Roa Peixoto Gnid 209 Bonbay-401 038 P.O. Box No. 729 Oxford Universty Press 01409 Sao Pal, SP Lahore3 P.0 Box S29 13/14 Asaf Ali Road Dar . Salasm CANADA New Delhi-11000D2 PERU Le Difuseur Editoial Desarrollo SA THAILAND C.. 85,.1501B rue Amphre 17 Chittaranjan Avenue Apartado 3824 Central Department Store Boucherville, Quebec Calcutta - 700 072 Lima 3D6 Silom Road J4B 1E6 Bangkok Jyadev HostelBuilding PHILIPPINES CHINA 5th Main Road Gandhinagar National BookStore TRINIDAD & TOBAGO, ANTIGUA China Fnancial & Economic Publishing Bangalore - 560 009 701 Rizal Avenue BARBUDA, BARBADOS, House P.O. Box 1934 DOMINICA, GRENADA, GUYANA, 8, Da Fo SiDongJie 35-1129 Kacidguda Cross Rood Metro Manilo JAMAICA, MONTSERRAT, ST. Beijing Hyderabad -50027 KITTS & NEVIS, ST. LUCIA, POLAND ST. VINCENT & GRENADINES COLOMBIA Prarthana Flats, 2nd Floor ORPAN Systenatios Studies Unit Enlace Ltda. NearThakoreBaug, Navrangpura PatacKulturyiNauki 49WattsStreet Apart.do Acreo 34270 Ahmedabad-380009 00-601 Waresawa Curepe Bogota D.E. Trinidad, West Indis Patiala House PORTUGAL COTE IYIVOIRE 16-A Ashok Marg Livraria Portugal TURKEY Centre dEdition et de Diffuelon Lucknow -226 031 Run O Ca0o 7074 Haset Kitapevl, AS. Africaines (CEDA) 1200 Usban ltilal Cadded No. 469 04 B.P. 541 INDONESIA Beyoglu AbidjanO4Plateau Pt Indira Umited SAUDI ARABIA, QATAR Istanbul J. Sam Ratalangi 37 Jari Book Store CYPRUS P.OBox101 PO. Box 3196 UGANDA MEMRB Information Services Jakarta Pusat Riyadh11471 Uganda Bookshop P.O Box 2098 P.0 Box 7145 Nicosia IRELAND MEMRB Information Services Kampala TDC Publishers Bramch ofes DENMARK 12 North Frederick Street Al Also Street UNITED ARAB EMIRATES SamfundsLitteratur Dublin I Al Dahna Center MEMRB Gulf Co. Rosenoerns All 11 First Floor P.O Box 6297 DK-1970 Frederikebrg C ITALY P.O. Box 7188 Sharjah Ucosa Commissionana Sansoni SPA Riyadh DOMINICAN REPUBLIC Via BenedettoF.rtini 120/0 UNITED KINGDOM Editora Taller, C. porA. Casella Postale 552 Hai Abdullah Alimesa Building Microinfo Ltd. Restauracion e Isabel la Catolica 309 50125 Florence King Khaled Street P.O. Box 3 Apartado Postal 2190 P.O. Box 3969 Alton, Hampshire GU34 2PG Santo Domingo JAPAN Damman England Eastern Book Service EL SALVADOR 37-3, Hango 5-Chome, Bunkyo-ku l113 33, Mohammed Hassan Awad Street URUGUAY Fusades Tokyo P.O. Box 3978 Instituto Nacional del Libra Avenida Manuel Enrique Araujo 03530 Jeddah San Jose1116 Edifido SISA, ler. Piso KENYA Montevideo San Salvador AfricaBook Service(E.A.) Ltd. SINGAPORE, TAIWAN, MYANMAR, P.O. Box 45245 BRUNEI VENEZUELA EGYPT, ARAB REPUBLIC OF Nairobi Information Publications Ubreria del Este Al Ahram Private Ltd. Aptdo. 60.337 Al Galax Street KOREA, REPUBLIC OF 02-06 1st Fl, Pi-Pu Industrial Caracas 1060-A Cains Pan Korea Book Corporaion Bldg. P.O. Box 101, Kwangwhamun 24 New Industrial Road YUGOSLAVIA The Middle East Oteserver Seoul Singapoe1953 Jtgoaloveneka Kniga 8 COwarbi Street P.O. Box 36 Cairo KUWAIT SOUTH AFRICA, BOTSWANA Trg Republike MEMRB Information Services Forsingle il: YU-1000Belgrade P.O. Box 5465 Oxford Univsity Prese Southern Africa P.O. Box 1141 Cape Town 80 г The World Bank Headquarters European Office Tokyo Office 1818 H. Street, N.W. 66, avenue d'16na Kokusai Building Washington, D.C. 20433, U.S.A. 75116 Paris, France 1-1 Marunouchi 3-chome Chiyoda-ku, Tokyo 100, Japan Telephone: (202) 477-1234 Telephori: (1) 40.69.30.00 Facsimile: (202) 477-6391 Facsimile: (1) 47.20.19.66 Telephone: (3) 214-5001 Telex: wui 64145 WORLDBANK Telex: 842-620628 Facsimile: (3) 214-3657 RCA 248423 WORLDBK Telex: 781-26838 Cable Address: INTBAFRAD WASHINGTONDC Cover design by Bill Fraser 0-8213-1555-2