ROOK NUMRFR HC.1iS.S3> ACCESSION NO The Balme L ibrary T H E EC ONOM IC COST OF M A LA R IA AND TH E BEHAV IOUR O F FA RM ER S TOW ARDS M ALAR IA CA RE IN AM AN S IE EA ST AND KWAEBIBBREM D ISTR IC T S O F GHANA BY JA N E T AMA DZATOR (M rs.) A THESIS SUBMITTED TO THE DEPARTMENT OP AGRICULTURAL ECONOMY AND FARM MANAGEMENT, UNIVERSITY OF GHANA, IN PARTIAL FULFILLMENT OF THE REQUIREMENT FOR THE AWARE OF THE DEGREE MASTER OF PHILOSOPHY (M. PHIL) IN AGRICULTURAL ECONOMICS, SEPTEMBER, 19 9 4 To my Dad, brothers and sister, husband and to the memory of my late mother. Above all I dedicate this work to God Almighty for seeing me through. DEDICATION ii ACKNOWLEDGEMENT I extend my sincere thanks to the Almighty God for helping me to complete this work successfully. I am greatly indebted to my Supervisors, Dr. W. K. Asenso- Okyere, Dr. (Mrs.) R. Al-Hassan and Dr. G. T-M Kwadzo, especially Dr. W. K. Asenso-Okyere for his guidance, constructive criticisms and valuable suggestions throughout the course of my work. I am grateful to my supervisors for the encouragement given to me. My sincere gratitude also goes to Mr Michael Dzator (my Husband), for his valuable contributions and comments, support in prayer and for the inconveniences suffered throughout the course of my work. My profound appreciation goes to all my friends and colleagues especially James Benhin, Isaac Osei Akoto, Anarfi, Augustina Adukonu, Felix Asante and Joseph Offei for their support throughout the study. Finally, I thank the UNDP/World Bank/ WHO Special Programme for Research and Training in Tropical Diseases for supporting this research and the Health Social Sciences Research Unit based at the Institute of Statistical, Social and Economic Research (ISSER), University of Ghana, Legon. Janet A. Dzator September, 1994 iii ABSTRACT The study focused on malaria care seeking behaviour of farm households in general. Time and money costs of care, the impact of socio-economic factors affecting health care seeking behaviour and household choice of malaria care providers were investigated. The findings show that households appear to be more sensitive to time costs in choosing malaria care providers. The report presents direct costs and indirect costs of treating an episode of malaria and also traced malaria treatment behaviour. TABLE OF CONTENTS Chapter one: INTRODUCTION......................................1 1.0 The setting .....................................1 1.1 Problem statement .............................1 1.2 Objectives of the study ...................... 3 1.3 Data sources .................................. 3 1.4 The study areas ...............................6 1.5 Organisation of thesis ....................... 8 Chapter two: THE SOCIO-ECONOMIC IMPORTANCE OF MALARIA ...... 10 2.0 Introduction ................................. 10 2.1 Socio-economic importance of malaria ...... 12 2.2 The shadow effect of malaria on agriculture. .21 Chapter three: THE ECONOMIC COST OF SEEKING MALARIA CARE ...22 3.0 Introduction................................22 3.1 Cost sharing in the health sector ......... 22 3.2 The economic costs of malaria ............. 24 3.2.1 Direct cost ................................ 25 3.2.2 Indirect cost ..............................25 3.3 Methodology ................................27 3.4 Estimation of direct cost ................. 28 3.4.1 Results on direct cost of malaria ......... 29 3.5 Estimation of indirect cost of malaria ....3 2 3.5.1 Results on indirect cost of malaria .......35 3.6 General components of costs of malaria ....37 Chapter four: THE CHOICE OF MALARIA CARE PROVIDER......... 3 9 4.0 Introduction .............................. 3 9 4.1 Health care providers in Ghana ........... 3 9 4.1.2 Underlying reasons for choosing providers..42 4.2 Choice of provider for treating malaria ...43 4.2.1 Choice of first provider................... 4 6 4.2.1 Choice of second provider .................4 6 4.2.3 Choice of third provider .................. 47 Chapter five: THE DEMAND FOR MALARIA CARE - A MULTINOMIAL LOGIT APPROACH ............................... 4 8 5.0 Introduction ......................... ....48 5.1 The theory of demand ..................... 48 5.1.1 The consumer .............................4 9 5.1.2 The concept of separability...............52 5.1.2.1 Weak separability ........................ 53 5.1.2.2 Strong separability..................... .,53 5.1.2. 3 Pearce-separability ................... ...53 5.2 Binary models ........................... .55 iv Page V5.2.1 The linear probability model ................ 56 5.2.2 The probit model...............................60 5.2.3 The logit model................................ 62 5.2.4 The multi-nomial logit model................. 66 5.3 Specification and estimation of the malaria care demand function................. 7 0 5.4 Determinants of choice between health care providers: ............................. 7 4 5.4.1 Self medication and government/ community providers ...................... 74 5.4.2 Self medication and private/misswion providers ................................. 74 5.4.3 Self medication and drug stores ...........77 5.4.4 Private/mission and government/ community providers ..... -................77 5.4.5 Drug stores and government/ community providers ...................... 79 5.4.6 Drug stores and private/mission providers 80 Chapter six: SUMMARY AND CONCLUSIONS ..........................82 6.0 Introduction ................................. 82 6.1 Summary of major findings ................... 83 6.1.1 Costs .........................................83 6.1.2 Health care seeking behaviour ...............83 6.1.3 Impact of socio-economic factors on choice of malaria care providers ................... 84 6.2 Implications for policy ....... 84 6.3 Conclusion ........................•............. 87 Page R E F E R E N C E S 88 LIST OF TABLES Table 2 .1 Direct and indirect economic cost of malaria and the equivalent days of per capita output lost in Africa......................... 13 Table 2.2 Reported cases of malaria in Ghana 1985-1991 ....... 17 Table 2.3 Types of diseases and their impact on health status..18 Table 3.1 Components of economic cost of malaria at facility ...26 Table 3.2 Average direct costs of services and drugs per case for treating malaria at health care facility......... 30 Table 3.3 Average travel and waiting time to seek malaria treatment at orthodox health care facilities......... 36 Table 3.4 Summary of costs of malaria in both Amansie East and Kwaebibirem districts..............................37 Table 4.1 Choice of for treating mild malaria in Amansie East district......................................... 44 Table 4.2 Choice of for treating severe malaria in Amansie East district......................................... 44 Table 4.3 Choice of for treating mild malaria in Kwaebibirem district..................................45 Table 4.4 Choice of for treating severe malaria in Kwaebibirem district .................................45 Table 5.1 Variable description.................................73 Table 5.2 Multi-nomial logit results for the combined districts........................................... 7 6 vi Page vii LIST OF FIGURES Page Fig.2.1 The vicious cycle of malaria and household poverty 2 0 Fig.3.1 Economic impact of malaria............................. 24 Fig.5.1 Predictions with the linear porbability model......... 59 Fig.5.2 Comparison of logit and porbit cumulative distributions.65 CHAPTER ONE IN T R O D U C T IO N 1.0 The setting The amount of money a household is willing to devote to malaria care is related to the behaviour of the household towards general malaria care. Knowledge of such behaviour is useful in designing malaria control programmes and cost sharing schemes. Although malaria is the major health problem in Africa, there is very little research on its economic impact. Factors promoting certain patterns of health care seeking behaviour such as the user charges borne by' individuals and families for seeking health care were introduced recently. Comprehensive estimates of the economic costs of malaria to individuals and communities are rare in many African countries although they are important for decision making. 1.1 The Problem Agriculture is the main source of income in most rural areas in Ghana and labour is the most important single input in rural agricultural production apart from land. Malaria has important economic consequences for rural agricultural communities. The relatively high prevalence of the disease during and just after the rainy season which unfortunately coincides with peak agricultural activities have serious implications for agricultural labour supply 1 which has already been diminishing with increasing migration of young men and women to the cities. Often demand for agricultural labour supply overshoots its supply during the wet season and thus pull up wages for farm labour. With the increase in farm wages, households are expected to substitute hired labour with its own labour. However, this source of labour has been dwindling with children going to school and polygamy reducing in many communities. The households labour supply situation can be exacerbated when diseases such as malaria inflict on some of the members of the household. It is therefore important that immediate care is sought so that household productivity is not unduly compromised. The dramatic change in peoples behaviour towards health care observed after the introduction of the Hospital Fees Regulation (1985) in Ghana as documented by Waddington and Enyimayew (1988) suggests that there exist some form of relationship among cost of health care services, income and health care seeking behaviour of people. However,quantitative evidence is limited in the literature in many African countries. This makes it difficult to devise equitable health care financing schemes. The availability of the cost of care and knowledge about the malaria care seeking behaviour of individuals and or households will therefore be useful in guiding the enactment of pragmatic policies that will ensure the utilisation of health services for human resource development. A study which is therefore aimed at analysing the health care seeking behaviour of farmers who form the majority of the people in the rural areas is considered timely in the current era when major 2 health financing reforms are being introduced in most developing countries. 1.2 Objectives of the study The general objective of the study is to estimate the economic cost of malaria and to determine the impact of time, monetary costs of malaria and other socio-economic factors on health care seeking behaviour of agricultural households in two districts of Ghana. The specific objectives are; 1. to estimate the direct and indirect costs of seeking malaria care; 2. to ascertain the malaria care seeking behaviour of households; and 3. to estimate the determinants of the demand for health care. 1■3 Data Sources This study formed part of a large research project which was undertaken by the Health Social Sciences Unit of the Institute of Statistical, Social and Economic Research. The data for the study was obtained from primary sources through a survey conducted in the Amansie East and Kwaebibirem districts of Ghana. After selecting the study areas in collaboration with personnel from the Ministry of Health, 3 reconnaissance visits were made to these areas. Twenty-one communities were selected for the survey. Amansie East and Kwaebibirem have some similarities. They are both agricultural districts with tree crops as the major products. Whereas Kwaebibirem district has a commercial diamond mining activity, Amansie East district is adjacent to Adansie West district where one of the richest gold mines in the world exist. Information on household malaria care seeking behaviour, costs of seeking care, income and other socio-economic variables were obtained through personal interview of a sample of farmers using a pre-coded structured questionnaire. The sample was the same one being used by the Policy, Planning, Monitoring and Evaluation Department (PPMED) of the Ministry of Food and Agriculture (MOFA) which has been collecting monthly agricultural price statistics from farmers in all districts. According to the PPMED the sample is a random sample of all the farmers in a district. This study was household based although certain aspects were devoted to collecting data based upon individuals. A household in this study is defined as all people who eat from the same pot and sleep under the same roof in three months of the 12 months preceeding the interview. In a few of the communities where the PPMED sample list did not exist, a random-convenience sampling technique was used in the selection of households. In this technique communities were mapped out by land marks, for example, roads and every second or third house depending on the size of the community was selected. Interviews were granted to the first 4 household head encountered and who was willing to participate after explaining the selection procedure. Questionnaires were administered on a total of 1300 households heads but 1289 questionnaires were used for the analyses after cleaning the data. The distribution of usable questionnaires was: 747 from Amansie East district and 542 from Kwaebibirem district. A time and motion chart was designed and used to collect data on waiting time from health care facilities within the two districts. A random accidental sampling technique was employed to select patients at the out-patient-departments (OPD). Selected patients were followed and timed at the various points of activity such as arrival, registration, consultation, laboratory, dispensary and departure. Timing was clandestinely done without the knowledge of patients. This procedure was adopted to allow normal behaviour and or to prevent the patients from hurrying through the activities or otherwise. Information on agricultural wage rates and public goods like schools, sources of drinking water and sanitation in general was gathered using a semi-structured community questionnaire. All the questionnaires were pre-tested in Asuaba and Adeiso in the Eastern region before they were used for the actual field survey. The communities chosen for the pretesting have similar climatic conditions and economic activities like the selected communities for the study. 5 1 . 4 The study areas The Amansie East district is in the Ashanti region and is bounded on the north by the Bosomtwi-Kwawoma district assembly, south by the Adansi West district assembly, West by the Amansie West district assembly, North-West by Upper Denkyira district assembly, and on the East by Asante-Akyem district assembly. The district has three major rivers which are the Offin, Oda and Domkra and an important lake, Lake Bosomtwi. These rivers and the Lake offer fishing opportunities for many farmers in the area. The weather has a double maxima rainfall pattern. The main wet season is from March to mid July followed by a short dry season and then a minor wet season from mid September to the end of November. The vegetation cover is mainly that of the tropical rain forest type. Farming is the main occupation and cocoa, maize, cassava and plantain form the bulk of agricultural produce in the district. Other economic activities include surface gold mining at Afoako and private transport operation. The district has a total population of about 166,658 [census, 1984] with an average growth rate of about 2.6 per cent per annum. In terms of health care facilities the district is endowed with 2 government, 3 mission and 1 private hospitals; 2 health centres; 23 Public Health Care [PHC] centres and 4 maternity homes of which one each is under private and mission management and the remaining two are under the management of the district assembly. From the available statistics disease morbidity figures follow the national trend with malaria ranking highest 6 among the reported cases at the Outpatient Department (O.P.D.). It is the highest cause for admission and second highest cause of death after anaemia in the district. Eleven communities were selected for the field survey in this district. They are: Kokotrour, Afoako, Abodom, Dedease, Asamang and Kokofu, Tweapease, Jacobu, Adamso, Samanhyia, Fiankoma and Dominase. All the communities have year round motorable roads and they are served by public transport services except for Tweapease. Apart from Samanhyia where there is no junior secondary school, the rest of the communities have primary and junior secondary schools. Apart from Dadease, Tweapease, Adamso, and Samanhyia (some of which are cluster villages) all the communities have either treated pipe borne water or wells with pump. Only three of the communities namely Tweapease, Adamso and Samanhyia do not have public latrines. Hospitals are situated at Kokofu, Dominase and Jacobu. Like other rural areas in Ghana, there are a number of traditional and spiritual healing centres in the Amansie East district. Kwaebibirem district is one of the 15 districts in the Eastern region with a population of 126,649 [Census, 1984]. There are two well marked wet seasons in a year with a typical rain forest vegetation cover. The heaviest rainfalls are recorded between June and August and the minor between September and November. Agriculture is the main occupation of the inhabitants and the major crops cultivated include cocoa, palm fruit, plantain, cassava, maize and citrus. Diamond is also mined in the district. The health care facilities in the district are; l quasi­ 7 government hospital [the Mines Hospital], 1 mission hospital, 4 government health centres/posts and 1 mission health post, 3 private clinics, 1 mission clinic and 5 Maternal and Child Health [MCH] clinics. Between the years 1990 and 1991, malaria featured prominently as the number one ailment for the cause of admission and for OPD attendance. Communities selected for the survey in this district are Takyimang, Abenaso, Asuom, Otumi, Kade, Atobriso I and II, Nkwantanang, Apinamang, Akwatia and Kusi. The roads linking all the communities are motorable throughout the year. Primary and junior secondary schools are also available in all the communities. Senior secondary and or technical schools are available in Kade, Asuom and Akwatia. The two hospitals in the District are located in Kade and Akwatia. Few of the communities also have clinics and maternity homes but traditional and spiritual healing facilities exist in all communities except Kusi where there is none. 1. 5 Organisation of the Thesis The study is organised into 5 chapters. Chapter 2 summarizes the socio-economic importance of malaria. It focuses on the perceived importance of the disease especially to agriculture and other economic development activities both in Ghana and elsewhere. In chapter three the costs of seeking malaria care is presented. Chapter 4 focuses on choice of malaria care seeking behaviour and the demand for malaria care is discussed in chapter 5. A summary 8 of major findings, policy implications of the findings for health care financing and conclusions are drawn in chapter 6. 9 CHAPTER TWO THE SOCIO-ECONOMIC IMPORTANCE OF MALARIA 2.0 Introduction A labour intensive agricultural economy like Ghana's may suffer crop output losses as a result of malaria morbidity and mortality. Knowledge of the rainfed agricultural calendar reveals that there are certain periods of the year in which some particular agricultural work must be done. For instance, harvesting, sowing and planting cannot be done throughout the year. Illness during such specific periods may endanger crop yield. Unfortunately, the peak of malaria transmission and morbidity coincides with the onset of the rains and some few months after the rains when most of these important activities are performed. Across the eco-epidemiological areas identified by the epidemiology division of MOH there is in general a high endemicity of malaria with intense perennial transmission. Even in the Northern savannah where transmission is noted to be seasonal it was noted in the MOH Malaria Action Plan (1992) that transmission is changing from seasonal to perennial. Increasing construction of large numbers of small to medium scale dams for agricultural purposes have contributed to the intensive perennial transmission of the malaria parasite because of the presence of water and the possible year round cultivation of crops and fishing. Apart from dams for agricultural purposes some economic 10 activities in many parts of the country render the areas malarious. The numerous small-scale diamond mining activities in the Eastern Region and small-scale gold mining activities in the Eastern, Ashanti and Western Regions are usually accompanied by the creation of pools of water which serve as effective breeding sites for the vector Anopheles gambiae . In Akwatia in the Eastern region where large scale diamond mining occurs, the open cast mining operations are accompanied by areas which collect water during the rainy season and serve as effective breeding sites for the mosquito vector. The debilitating effects of the disease on victims and the shadow effect of the disease on other household members through the provision of care during illness for example, may induce labour shortage when it is most needed. Households are thereby faced with low levels of production, hunger and poverty in the long run. National statistics from Rwanda, Chad and Ghana, showed that malaria accounted for more cases , more deaths and more potential days of healthy life lost than any other one cause (Shepard et al, 1991) . Long term data from Zambia, Togo and Rwanda similarly showed that the incidence of malaria increased by 7, 10, and 21 percent, respectively, every year over the 1980s (Ibid) . It was suggested that the increasing incidence of the disease was associated with population migration, growing resistance of the parasite to antimalarial drugs such as chloroquine, limited resources for malaria treatment and control, irrigation and other development activities that create mosquito breeding sites. 11 Economic stagnation and decline were also identified as causes for the general health problem (Shepard et. al, 1991). 2.1 Socio-economic importance of malaria Estimates of the cost of particular diseases are intended to provide information on the value to be gained from reductions in their prevalence or incidence. Such information may be useful, for example, in determining priorities for planning and financing health care services, and in evaluating whether the benefit from a particular disease prevention programme justifies its cost. In an effort to focus on a long neglected aspect of malaria in Africa, four case studies across the continent had been performed by Shepard et al. (1991), to quantify the impact of malaria on Africa, a summary of which is shown in table 2.1. The average cost per case of malaria for all sub-Saharan Africa was estimated to be $9.84 (in constant US dollars): $1.83 for direct costs and $8.01 for indirect costs. Given the average value of goods and services produced per day in Africa to be $0.82 during the time of the study, the estimated cost per malaria episode was found to be equivalent to 12 days of output. By 1995 the average cost of a malaria case was projected to rise to $16.40 due to increasing case severity and chloroquine resistance. At the same time per capita output was predicted to fall to $0.77 thereby increasing the burden of a case of malaria. In terms of time the per capita cost of the disease was estimated to be equivalent to 12 21 days of output. The burden is forecasted to rise from $1.34 in 1987 to $4.02 by 1995. iao le i.v. D irect and indirect econoi output lost in Africa. rue cost oi mai;aria anu me iequivalent uav s ui utsi D irect cost per capita Indirect cost per capita Total cost per capita Output per capita Equivalent No. o f days o f output lost Area (US $ r (US $V (US (U S $V to malaria Sub-Saharan A frica 1.83 8.0 9.84 0 .82 12.0 Rwanda 0.63 2.25 2.88 0 .83 2.5 Burkina Faso (Solenzo Dist..) 0 .26 0.89 1.15“ 3 .0 Chad (Mayo-Kebbi D ist.) 0.01 0.59 0.60 5 .0 Congo (Brazzaville') 0 .27 0.47 0.74 < 1 Source: Compiled from Trop. Med. Parasitol. 42 (1991) * Values are in 1987 U S dollars ‘ Health care providers pay about 14% o f the direct cost. 13 In Africa as a whole, the annual economic burden of malaria was estimated to be $0.8 billion as at 1987 and it is estimated to rise to $1.7 billion in 1995. The economic burden represented a 0.6 per cent of the share of GDP in 1987 and a 1 per cent share of GDP for 1995 (Shepard et.al. 1991). In Rwanda, Shepard et. al, (1991), estimated the total cost of malaria for 1989 to be $2.88 per capita (in 1987 U.S. dollars). Of this cost $0.63 per capita represents the direct cost of treatment including care for out-patients and hospitalised cases in both government and private facilities as well as self treatment. The other $2.25 per capita represent the indirect costs of productive time lost to malaria morbidity in adults and in providing care for sick children. The indirect cost also include the value of life time earnings lost through premature malaria mortality. Average output per day per capita of the Rwandan economy at the time was $0.83 resulting in the loss of 3.5 days of production or an equivalent of 1 percent GDP. Across the country (Rwanda) the average cost of each of the 1,772,271 reported malaria cases in 1989 was $11.82. The break down gave $2.58 as direct cost and 9.24 as indirect cost. The incidence of the disease was noted to have risen eight-fold from 1979 to 1990 (Ettling and Shepard, 1991) . In rural Rwanda where it was reported that daily output per capita was less than half the national average, user fees to patients for treatment in public health facilities in the rural areas were found to be only 5 per cent of operating expenses. It was reported that if the entire direct cost per case were passed on 14 to patients, it would represent 29 days of individual rural cash income (Ibid). In Burkina Faso, Sauerborn et.al, (1991) used household data on health service utilisation and agricultural production to estimate the direct and indirect costs of malaria. Cost per case averaged $5.96 and cost per capita was $1.15. Indirect cost due to mortality was the largest component and was noted to be $0.89 per capita. The direct cost for providing malaria care per capita was $0.26. Travel and drug costs to households accounted for 86 per cent of the direct cost and the remaining 14 percent represented Ministry of Health expenditure on other services provided for treating a case of malaria. The daily marginal productivity of labour was used to estimate the indirect cost of malaria. Out of the total work days lost to malaria, 75 per cent of the days were lost during the agricultural food crop season, 11 per cent during the cash crop season and 14 per cent during the maintenance season. The value of the production lost was estimated to be $0.10 per capita (Ibid). Nur and Mahran (1988) estimated a total of 9,741 work hours to be lost to malaria from a sample of 250 tenant farmers in Sudan over a period of one year. Although 62 percent of this loss were noted to be compensated by healthy family members, a negative shadow effect of malaria is reported to emerge in the form of abstracting women from household activities and children from school. It was also found that malaria reduced tenants efficiency by 50 per cent through debilitating effects and the occurrence of 15 the disease was highest within seasons where demand for agricultural labour was at its peak. Malaria is often noted to be one of the most serious obstacles to human efforts to develop agriculture, establish permanent settlements or in any way modify the environment. The disease is not only one of the most widespread diseases in the world but also the number one killer of people, especially in the tropics. Figures provided by the World Health Organization (WHO) indicate that some 2.2 billion people or 40 per cent of the worlds' population are at risk from malaria. Of this population 90 per cent of them carrying the parasite are in tropical Africa. The disease is also said to claim over one million lives per year of which 80 per cent are under five. In Ghana malaria accounts for about 9 per cent or 16,000 of all deaths per year. Over 30 per cent of all hospital admissions in the country is attributed to malaria. In table 2.2, malaria is shown to account for about 40 per cent of all diseases reported in the country's health institutions. The proportion of malaria to other disease did not vary much even with different reporting rates. This supports the findings of the Ministry of Health (MOH) that malaria is hyperendemic in Ghana (MOH, 1991). Table 2.2: Reported Cases of Malaria in Ghana. 1985-1991_______ Year No. of Cases Percent of all Reporting _______ diseases_____________rate 1985 593,368 40 41 1986 807,019 39 61 1987 1,141,893 43 58 1988 1,232,882 43 65 1989 1,556,169 44 77 1990 1,438,713 38 83 1991 1. 372 . 771 40 82 Source: Centre for Health Statistics/Epidemiology Division, Ministry of Health, Ghana, Various Years. The importance of the disease is further supported by general epidemiological data. With respect to prevalence, malaria is the most important disease in Ghana, followed distantly by diarrhoea and urinary tract infection (Asenso-Okyere, 1994). Being the largest disease problem malaria is also the number one disease to claim the highest number of days of healthy life among other diseases. Malaria accounts for 15.4 percent or 58,427 days of healthy life lost per year per 1000 population in Ghana (table 2.3). With these evidences the importance of malaria to productivity among the economically active population especially in agriculture cannot be overemphasised. 17 Table 2.3: Types of Disease and their Impact on Health— Status--- Days of Healthy Rank Disease Life Lost Per Year Percent of Order Classification_______per 1000 Population______Total----- 1 Malaria 58,427 15.4 2 Prematurity 34,432 9 . 1 3 Measles 23,033 6 . 1 4 Birth injury 22,612 6.0 5 Sickle Cell Disease 21,797 5 . 9 6 Pneumonia,Child 20,857 5.5 7 Kwashiorkor, Marasmus 19,312 5.1 8 Dysentry and Gastroenteritis 17,322 4 . 5 9 Neonatal Tetanus 14,847 3.7 10 Accidents (All Kinds) 11,137 2 . 9 11 Tuberculosis 10,097 2.7 12 Stroke 8 , 915 2 . 4 13 Pneumonia, Adults 8 , 743 2 . 3 14 Psychiatric Disorder 8 , 542 2 . 3 15 Cancer 7,315 1.9 16 Preanancv, Complications of C . 005 1.6 Source: Ministry of Health, 1978. A Primary Health Care Strategy for Ghana, Accra. 0 Currently the authorities of the Ministry of Health in Ghana have designed an elaborate strategy to mount a malaria control programme. At the same time, the Ministry of Food and Agriculture is also about to vigorously pursue an intensive programme on small scale irrigation projects which paradoxically may increase the breeding sites of the mosquito vector. Although the link between increased malaria transmission and increased irrigation activities is not documented in Ghana the disease has now become a perennial health problem in Northern Ghana instead of its seasonal nature in the recent past (MOH, 1991). In line with the recommendations of the World Bank [1991] concerning differential application of user fees for rural dwellers and urban dwellers it is expected that knowledge of health care seeking behaviour of farmers especially in 18 the current era when irrigation development and malaria prevention are being highly.encouraged in the country can be a useful guide to decision making. The relationship between agricultural development programmes such as irrigation and the incidence of malaria noted above explicitly illustrates the argument that health affects development and development also affects health. Figure 2.1 illustrates how malaria infection affects productivity which in turn may lead to low incomes (or increased poverty) which may pre-dispose people to malaria infection or increased severity. Thus, malaria and poverty are engaged in a vicious cycle. Malaria infection results in poor health which can lead to loss of life (or loss of healthy productive days) or reduced labour efficiency thereby causing reduction in household output and income which negatively affects ability to pay for malaria care services. People become more sick because they cannot afford to pay for efficient treatment services and therefore they become poorer because they cannot work efficiently (Asenso-Okyere, 1992). The question one would like to ask therefore is how can this vicious cycle be broken so that access to health care, especially the poor, can be increased so as to improve upon the welfare of people. 19 Fig 2-1: THE VICIOUS CYCLE OF MALARIA AMD HOUSEHOLD POVERTY, 'Source: As'enso-*.Q} 0 (5.7) U, u 2 0 The diagonal elements (i.e. Uu) are negative and the cross price elasticities of substitutes (Uy) are positive. Also, complementary goods have negative cross price elasticities and independent goods have zero cross price elasticity. The matrix provides a technique for studying the interrelationships among commodities. However, if a large number of commodities are involved, the interrelationships among commodities become complex because there would be too many parameters in the matrix. To overcome this problem, a number of constraints can be imposed on the demand system (see George and King, 1971). The major constraints include homogeneity constraint, 51 Engel aggregation, Cournot aggregation, Slutsky or Symmetry condition, and Separability. As a matter of relevance the study will concentrate only on separability concept of demand. 5.1.2 The concept of separability The concept of separability was introduced independently by Leontief (1967) and Sono (-1961) . The concept assumes that the set of (N ={1, n}) commodities available to the consumer can be partitioned into s mutually exclusive and collectively exhaustive subsets, {N,, ..., N J , so that the commodities within each subset possess some common characteristics. According to Stroz (19 57) the importance of separability assumptions for the consumer allocation problem stems from the fact that, consumption decision can be viewed as occurring in two stages. At the first stage, the consumer allocates income between subsets of commodities. The second stage involves allocation within subsets or commodity groups. This type of hierarchical classification as indicated by Asenso-Okyere (1982) has been contended by many authors (Hassan and Green, 1982; Hassan and Johnson, 1976; George and King, 1971) as an alternative rationalization for aggregation of commodities when data is limited. Depending on assumptions, several types of separability is included in the literature. Three of these; (i) weak (ii) strong and (iii) Pearce separability as noted by Goldman and Uzawa (1964) are discussed. 52 5.1.2.1 Weak separability This concept implies that a utility function U(X) can be divided into subsets {N,, NJ such that the marginal rate of substitution between two commodities i and j from the same subset I is independent of the quantities of commodities not belonging to subset H (George and King, 1971). Thus U-, 3( ----) U; dXt = 0 for all i,jd and k^I (5.8) 5.1.2.2 Strong separability The utility function U(X) is termed strongly separable with respect to a partition {N,, ..., NJ if the marginal rate of substitution between two commodities i and j from subsets I and J respectively does not depend upon the quantities of commodities outside of I and J. Thus U, d( ----) U; dXt = 0 for all id, jeJ and k£l and J (5.9) 5.1.2.3 Pearce-separabilitv A utility function U(X) is termed Pearce separable with respect to a partition {Nw NJ if the marginal rate of 53 substitution between two commodities i and j from the same subset I does not depend upon the quantities of all other commodities including other commodities in the same subset. That is Ui d( --- ) uj . ------ = 0 for all i,jel and k^i and j (5.10) d x k With reference to the theorems of Goldman and Uzawa (19 64) it can be deduced that strong separability implies an additive utility function, whereas weak separability, does not. Furthermore, in the case of Pearce-separability, the utility function includes both weak and strong separability concepts which implies that the utility function is separable between groups and additively separable within groups. One advantage of the concept of separability is that it greatly reduces the number of parameters to be estimated. Johnson, Hassan and Green (1982) for example, have shown that using strong separability there are n+1 parameters to be estimated (where n is number of commodities) as a basis for fully characterizing consumer responses to price and income in such demand systems. For weak separability the required number of parameters is n+l/2(s2-s) (s is number of commodity group. The various types of separable utility functions imply special conditions which may be used as a non­ linear restrictions in the estimation of a complete system of demand equations. The separability restrictions therefore provide a means of greatly reducing the econometric problems associated with demand systems estimation. However, separability restrictions 54 imply strong consumption pattern, similarities within and across groups which may not exist. By the separability concept for instance, Johnson, Hassan and Green (1982) used the concept to study four groups of commodities defined as: durable goods, semi-durable goods, non-durable goods and services. In studying the consumer demand for food commodities George and King (1971) identified 15 separable groups within a list of 49 commodities. Similarly, the separability concept was assumed in this study whereby health care was considered as an identifiable subset of commodity in the consumer basket and demand for the various actors or providers within the health sector was therefore studied. 5.2 Binary Models Binary choice models are modifications of the linear regression model classified under qualitative response models. Qualitative response models are synonymous with quantal, categorical, qualitative choice or discrete models. They are built to alleviate the difficulties associated with the interpretation of the probabilistic models in which the explanatory variables are categorical or non-continuous in nature. In most empirical works done, suitable modifications of these theories have been used extensively in analyzing and forecasting the behaviour of economic agents in different economic contexts (Schmidt and Strauss,1975; Amemiya, 1985; Li,1975). The existence of different health 55 facilities both within the orthodox and the traditional systems with different characteristics provide households with alternatives from which a choice can be made. Binary choice models usually deal with a choice between two alternatives i.e. a dichotomous framework for example of a 'yes' or 'no' situation and that the choice they make depends on the characteristics of the- individual (Judges et.al; 1980). More generally, well specified binary choice models allow predictions to be made from the relationship between a set of attributes describing individuals and the type of choice they make. Similarly, it is also possible to determine the probability that an individual will make a certain choice depending on a given set of attributes about the person. The options available are whether a particular provider is chosen or not and this choice is dichotomous (or binary) in nature. The purpose of qualitative response models however is to determine the probability that a decision maker, with a given set of attributes makes one choice rather than the alternative. To effectively link thoughts it is necessary to give an overview of the underlying principles of the qualitative (binary) choice models. 5.2.1 The Linear Probability Model The linear probability model in a regression form is specified as; Yi = X'i 5 + e; 10) 56 where Y; = 1 if the ith decision maker selects the first alternative say a type of health care facility and 0 if the ith decision maker selects the second alternative. X; is a vector of regressors; typically, attributes and characteristics of the ith decision maker. 13 is a vector of parameter coefficients, and e; is the error term which is an independently and identically distributed random variable with mean zero. Since Y; can take on only two values 1 and 0 the probability of the distribution of Y is described by letting P< = Prob (Y;=l) and 1-Pj = Prob (Y;=0) (Pindyck and Rubinfeld, 1991) . Considering the zero-one nature of the regressand, the linear probability model is interpreted as conditional probability; that is the ith decision maker selects an alternative e.g. a type of health care facility given his or her X's. The predicted value of yj is interpreted as the estimate of this conditional probability. The parameter vector S> measures the effect of unit changes in the X's on the conditional probability. Estimating by ordinary least squares [OLS] method yields consistent and unbiased estimates of 5. However, the model has some shortcomings. The variance of the disturbance term is heteroscedastic so that its value depends on 6 thus resulting in a 57 loss of efficiency of the parameter estimates(Goldberger, 1964; Greene, 1990). Because of the heteroscedasticity problem the standard deviation of the OLS parameter estimates are biased and the distribution of the disturbance term is also not normal (Pindyck and Rubinfeld, 1981) . The classical statistical tests of significance are therefore not applicable to linear probability models since these tests depend on the normality of the disturbance terms. Also an extensive analysis of the properties of the linear probability models by Aldrich and Nelson (1984) noted that even if X'6 is constrained between 0 and 1 the formulation allows the estimates to fall outside the (0,1) interval as shown in figure 5.1. 58 Fig 5-K Ph ed ic ti on wi th th e li ne ar p ro b ab il it y m o d e l 00 O o LL u Eo c o uLU ■D* c a Oj u o 2 Qj Eoc­ o CD cd 5 a; c JDD cr T7 C O j*: o>i*U c a_ With regard to the difficulties surrounding the otherwise appealing linear probability model, alternate approaches to analyzing qualitative response models have been specified. The alternative specifications make use of monotonic transformations of the linear probability model by the cumulative distribution function to ensure that the predictions lie in the unit interval for all X's (Goldberger, 1964). Among the infinitely many transformations that the cumulative distribution function can provide, the probit and the logit transformations will be highlighted here as a matter of interest and relevance to the current study. 5.2.2 The Probit Model The probit model, is associated with the standard normal cumulative distribution function. Consider the (linear function of regressors) model; Z; = X' R (5.11) where Z; is a random variable with probability density function f(Z) and assume Z,* is a certain threshold level of Z, and translates the index Z into empirical decision about alternatives given the individual attributes. Also let the value of y; be such that ; |l if Z, > Z,* y< = I and (5 .1 2 ) 1 0 if Z; < Zj* 60 Each yj is thus a function of X's by way of Z. Let f(Z) be the value of the cumulative distribution function at Z such that P ( y, = 1 /Z,) = P(z; < Z,) = F(Z) (5.13) and P ( yj = O/Zj) = P (Zj* > Zj) = 1 - F(Z;) (5.14) be a transformation of a linear probability model by use of cumulative distribution'function f(Z). The probit model assumes that Zj* is a standard normal variate such that the probability that Z;’ is less than or equal to Zj can be computed from the standard normal cumulative distribution function (Greene 1990 ) . Based on the normal distribution Pi =Pi(yi = 1) = G(X 'B ) = U x'“ (Z) 3Z (5.15) = u (X'6 ) By the monotonic transformation, Pj will lie within the unit interval. Pj represents the probability of an event occurring; in this case the probability of an individual choosing the first alternative. The change in probability P,- with respect to a change in Xj is given by dPj/dXj = (dF/dZj) (dZj/aXj) = f(Zj)6 (5.16) where f(Zj) represents the value of the standard normal density function associated with each possible value of the underlying index Z. 61 5.2.3 The Logit Model The term, logit analysis derives from the idea of logistic function which came from the early work of Berkson (1944). The method has been refined by Theil (197 0) and has thence been referred to in some literature as the Berkson-Theil method(Li, 1975). Logit analysis employs the cell frequency distribution to derive the logit estimate of qualitative model parameters and it is proved to be asymptotically equivalent to the maximum likelihood (ML) estimation. On computational grounds the logit estimation is also found to be more preferable than the Maximum Likelihood (ML) procedure (McFadden, 1974; Amemiya, 1985). The logit probability model which accounts for the heterogeneity associated with linear probability models, is associated with logistic cumulative distribution function of the form; eZl Prob (Yi-1) = F (Zj) = ---------- (5.17) 1 + ezi The logit model specifically is defined as the natural logarithmic value of odds in favour of a positive response. In other words the odds that an event will occur are given by the ratio of the probability that it will not occur. The conditional probability P; is the probability that an individual will make a certain choice given X; and is specified as; 1 62 Letting X'B = Z, equation [9] can be estimated by multiplying both sides of the equation by 1 + e'2 to obtain the relation below. That is; Pi ( 1 + e'z) = 1. (5.19) Rearranging equation [10] gives Pi ( e"z) = 1 - P; . (5.20) Dividing through equation [5.20] by P, and taking logs of both sides gives the relation specified below which carries the label logit. Thus; Pi Z| = log --------- = X ' R (5.21) 1 " Pi where P; is the conditional probability of a positive response for a household with characteristic X;, and B ' s are the parameters. The dependent variable in this relation is defined as the logarithm of the odds of choosing an alternative. According to Li (1975) since the true logit Z; is not observed the logit estimate Zt based on cell relative frequency is used instead, that is ; - fi Pi Z; = log -------- = log + Uj (5.22) 1 - f, l - p. 63 where f; is the observed relative frequency in the cell i and u, is the error term. The estimated effects in the logit model are linear in the parameters but non-linear in probability due to the logistic transformation (Goldberger, 1964), In addition to the fact that the probability implied by the logit model can never exceed unity, interaction effects are much less inherent in the logit model than in the ordinary least squares model (Li,1975). As indicated by Pindyck and Rubinfeld (1991) both the logistic and probit formulations are quite similar. The only difference between them is shown in figure 5.2 where the logistic graph has slightly flatter tails. The logistic distribution is shown to resemble the t-distribution with seven degrees of freedom. Although the logistic function is similar to the cumulative normal function it is easier to use computationally. The logit model is therefore often used as a substitute for the probit model. Another important appeal of the logit model highlighted by Pindyck and Rubinfeld (1991) is that it transforms the problem of predicting probabilities within a (0 ,1 ) interval to the problem of predicting the odds of an event's occurring within the range of the real line. 64 f jg. 5-J.: C om pa rison of loqft and p rob it c um u la t iv e - d is t r ib u t io n s- Value of z Source: pindyck and Rubinfeld ( 1991 )i Econom etric Models and Economic Forecasts. 5.2.4 The multi-nomial logit model. With some modification the results of the univariate dichotomous model typically apply to the multi-response models. Instead of the 'yes' or 'no' alternatives facing the decision maker, for example, whether he consumes or does not consume health care services, the decision maker can also be faced with the problem of choosing from "among k alternatives where k represents different types of health care service providers. In each instance all choices are mutually exclusive. That is the "either or" choice model has been transformed into a range of options among which the decision maker can choose. The dependent variable y; for such multi-nomial or multi-response models takes k values where j = 1, 2, . . .k. The dependent variable then equals k if the ith decision maker selects the kth alternative. Similar to the binary choice model the multi-response model is specified as P( Yi = ) ) = Fs (X'B) (5.23) where i = 1,2,...n; j = 2,3,....k and X and 3 represent vectors of independent variables (or regressors) and unknown parameters, respectively. The explanatory factors include the attributes of the alternatives as well as characteristics of the decision makers. The subscripts i and j denote the ith decision maker and the jth alternative, respectively. Consequently, the selection of the jth alternative represents the action of drawing from a multi-nomial distribution with probabilities P,...Pn. 66 The multinomial logit model via Nerlove and Press (1973) , and Schmidt and Strauss (1975) mostly applies to data that are individual-specific (Greene, 1990). For example if there is a choice variable (say health care provider) of n options (e.g orthodox, traditional, etc.) open to an individual and the data for each individual in a sample contains information on all the options then the model for the choice of health care provider is specified e“J"xi P(Yi = j) = -------------------------- (5.24) Z"j_0 eBjxi Greene (1990), pointed out that the model has an indeterminacy such that for any Si’ = + q for any non-zero vector q, an identical set of probabilities result because all the terms involving q drop out. A suggested normalization is to assume that 60 = 0 such that the probabilities are therefore e Bj'Xi P(y; = j) = ------------------------- (5.25) 1 + 2" =0 e"*'*' and 1 P (Yj = 0) = ----------------------------- (5.27) 1 + Skk=0 enit’Xl By differentiating equation (5.23), the marginal effects of the regressors on the probabilities are obtained from an expression 67 such as; dPj/dX; = Pj [Bj - 2k Pk 6k] (5 -28) which can also be computed from the parameter estimates( Greene, 1990; Pindyck and Rubinfeld, 1981 ) . For example, given the J log-odds ratio to be p ij In -------- = B/X (5-29) PiO and assuming there is a normalization on any probability k, the odds ratio will be In pij = X/[B: - Bk] (5.30) P* where the parameters [Bj - Bk] denote the marginal effect. The relation also allows possible pairing and prediction of relative probability that an individual with given characteristics is more likely to choose alternative j other than k. For notational simplicity let the subscript i designating individual observations be dropped. In this case, Pj (where j = 1...J) indicates the probability that the jth choice will be made. The equation (5.28) presumes that the logarithm of the odds of one choice relative to a second choice is a linear function of the attribute X. For n such equations the odds of one is associated with the remaining equations in the sense that the system must be constrained so that the sum of the individual probabilities equal one (Pindyck and Rubinfeld, 1981). 68 As in the linear probability model, it is necessary to estimate each equation separately for the fact that the choice of logit form forces constraints on the model which reduces the number of parameters. For example, in the system of equations in the model represented as P3 P3 a vlog ----- = 621X ; ’ log = 631X and log = 332X P, P, Pi it can be shown that P3 P3 P, P3 P2 log ---- = log + log = log - log---- P 2 Pi P2 Pi pi — (®31 ®2l)^’ (5.31) From the point of view of estimation, Greene( 1990) noted that it is useful that the odds ratio, Pj / Pk, does not depend on the other choices because of the independence of the disturbances in the original model. Although this is not very attractive from a behavioural point of view, the estimation of the multinomial logit model is straight forward. The multinomial logit model can be estimated by the maximum likelihood method which is a generalization of that for the binomial probit and logit models (Schmidt and Strauss, 1975; Pindyck and Rubinfeld, 1981; Greene, 1990). It should also be noted that the logistic transformation allows linearity and much less interaction effect than the ordinary least squares models. 69 5.3 Specification and estimation of the malaria— care— demand function In order to make a meaningful progress in financing modern health care in low income countries, a good understanding of the health care seeking behaviour of people in terms of demand for malaria care becomes very crucial. The health care seeking behaviour of people can be translated into demand model because the choice and subsequent utilization of a health care facility by consumers could be likened to their revealed demand for the service. Akin et al (198 6) proposed that consumers are assumed to maximize a utility function over a vector of purchased medical services and a composite of all other goods subject to their incomes. The outcomes predicted as a result of this maximization process are that consumers' choice of health care provider is determined by a number of exogenous socio-economic variables and the economic costs associated with the consumption of health care services. It can be hypothesized that consumers make choices depending on the impact or the influence of the socio­ economic factors associated with the options available -to them. Essentially, the multinomial logit function used is given by: In (Pj/PJ = /3jk Xjk + 7j z (5.32) where Pj is the probability of choosing a provider, k=0 ...j Xjk is composite of independent variables that vary by facility eg facility price, proximity, travel and 70 waiting time Z is a vector of socio-economic variables with regard to malaria patients and their households /3jlc, 7 are coefficients. P0 is the default probability for normalization and it becomes a default option for not seeking care at a facility with coefficients normalized to zero. In such a case own elasticity of the probability of selecting j with respect to any of the exogenous variables could be calculated (ie k=j ). It should be noted that normalization could be done on any other probability k (where j?^). The odds ratio, lnP^/P^ = X; ' (fy - j8k) could then be used to calculate the marginal effects or cross elasticities with respect to the exogenous variables. From previous studies it was noted that the natural logarithm of the odds in favour of a household (as a decision making unit) in choosing a particular health care provider (public, mission, private or traditional) is a function of income, transport cost to facility, cost of drugs and other user charges, opportunity cost of travel time and the opportunity cost of waiting time at the facility (Akin et al, 1986; Van der Gaag et al, 1988) . In addition, personal or household characteristics such as occupation, gender, education, family size and household assets were also hypothesed to influence household choice of health care providers. In line with the above assumptions a system of demand equations in which the choice of health care service providers is a function of exogenous time and cash prices and other socio- 71 economic variables is summarized in a reduced form as shown below Y; = f (Pi Tr D Wt Q Z) (5.33) Yi = type of facility, e.g. government, mission etc Pi = facility price( i.e.lumpsum charge per episode of malaria) Tr = travel time to facility in hours D = distance to facility in km Wt = total waiting time at facility Q = dummy for provision of drugs at facility (proxy for quality of facility; yes=l, no=0 ) Z = composite of household and individual socio-economic variables( such as age, household income, respondent's level of education, and severity of the disease among others). The model is a modification of that used by Akin et. al. (1986) and Van der Gaag et. al. (1988) . It is assumed that health care service providers or health care facilities form a- set of mutually exclusive variables which represent the dependent variable. Separability is invoked whereby the demand for health service (malaria care in this case) is assumed not to be affected by the demand for other commodities or services. Such an assumption is possible if it is considered that it is the desire of people to live and so they will allocate some money to take care of their health. However, within the health care delivery system they will make the necessary adjustments to choose the provider that satisfy their expectations given the constraints facing them. 72 Table 5.1: Variable description Facility-specific variables . Price:- Actual fees paid per malaria case (in cedis) at facility Travel time:- Travel time in hours from the home of patient to a health care facility Waiting time:- Total time spent at a health care facility to seek malaria care Distance:- Distance in kilometres from the resident community of patient to a provider Drugs provision:- Whether patients were given drugs at facility (yes=l, no=0) Individual and household characteristics Age:- Age of patient in years Ethnicity:- Ethnic origin (Akan=l, non-Akan=0) Religion:- Religion of patient (christian=l, non-christian=0) Respondents' education:-Years of formal schooling of the respondent Household expenditure:-The annual household per capita expenditure as a proxy for income Severity:- Perceived severity of malaria (mild=l, severe=0) Gender:- Gender of patient (male=l, female=0) Survey district:-Whether the patient is in Amansie East district or in the Kwaebibirem ____________________________ district (Amansie East=l. Kwaebibirem=01 Unlike the Akin et. al. model, specific variables representing monetary and time prices for seeking malaria care at alternative facilities were not included in this model. This limitation makes it impossible to calculate cross-price elasticities for the various providers. 73 5.4 Determipants of choice between health care providers,^ The relationship between the choice of these facilities and their respective characteristics in addition to the attributes of people who make the choice were determined by the multi-nomial logit technique. The model allowed the measure of the effect of time, money and other exogenous variables on the choice of alternate health care providers. The coefficients of variables indicate how changes in each of the variables affect households utility and represent the effect of these factors on choice of provider. The signs of the coefficients show the direction of the propensity of households in choosing a provider relative to an alternate provider. A negative coefficient on a variable for example indicates that an increase in that variable will decrease the probability of selecting a provider relative to the other alternative (Akins et al, undated). 5.4.1. Self medication and government/community providers As shown in table 5.2 in most cases an increase in facility price appears to make households more inclined to choose orthodox health care facilities. Facility price and district or location of respondents are negative and statistically very significant in the relation that compares the choice of self medication with government/community providers of malaria care. The direction of the signs on the coefficients with regard to the providers indicate that households are more likely to seek care at government or 74 community health care facilities rather than self medicate if there is an increment in cost of self medication. Government health care facilities are more likely to be used instead of self medication if they are located near the consumers. There is also significant difference between the two districts in terms of choice of the two providers. For example, households in Kwaebibirem district are more likely to self medicate than those in Amansie East district which tends to support the finding of the choice of a provider summarized in chapter 4. Although not significant, an increase in travel time to seek health care will lead to the adoption of more self-medication. As per capita income increases, people tend to self medicate more when they get malaria. One startling result is that as the distance to seek health care increases people prefer to government clinics to self-medication. It is plausible that clinics that are far away are perceived to offer high quality service and so are well patronised. 75 Table:5.2 Multi-nomial logit results for the combined districts. ( figures in parentheses are t -stats) logS/G logS/P logS/D logP/G logD/Gi logD/p FaciIity Price (7.8093)*** -1.3744 (26.0380)*** -3.3943 -0.0535 (6.1367)*** (0.3802) 0.0874 (2. -0.8572 0386)* -1.7999 (4.8360)** Travel time to facility 2.6803 (1.3355) 0.1249 (0.0000) 0.1871 (0.0000) 0.4050 -1.1828 (13.6350)*** (0.9156) -2.2282 (1.6290) Waiting time at facility (22.5073)** -139.7574 (0.0526) ■254.1402 -25.7555 -0.0476 -2.0679 (0.0028) (0.0000) (0.6427) -3.0253 (29.1971)*** Given drugs at facility (Yes=1, otherwise=0) 1.3434 (0.0000) 1.7649 (0.0000) 0.0991 (2.2069)* 0.3296 0.1279 (7.4095)*** (0.6525) Distance to facility -2.2599 (2.0025)* 1.1682 (0.0028) 8.0784 (0.00000 -0.3129 0.7983 (9.0676)*** (0.9989 0.5018 (0.5234) Ethnicity; (Akan=1 otherwise=0) -3.6273 (0.0000) -2.1001 (0.0004) -3.9176 (0.0000) 0.0037 (0.0041) -0.2030 (2.4184)* 0.0103 (0.0069) Religion; (christian=1 otherwise=0) 0.0105 (0.0077) -4.3544 (0.0020) -0.2276 (2.6793)* -0.0528 (0.9005) -0.0242 (0.0586) 0.1203 (0.8648) Age 0.0851 (0.6763) 0.3311 (0.71-60) 0.0148 (0.0483) -0.0998 (2.8437)* 0.0686 (0.6300) 0.2864 (3.4385)** Household per capita income (4.9306)** 0.0593 (1.1366) 0.5393 0.0803 0.0229 (0.6453) (2.7784)* (0. 0.0642 1399) (0, 0.3190­ .5258) Respondent's years of education 0.0161 (0.0214) 0.0769 (0.0309) 0.0109 (0.0229) 0.0229 0.0642 (6.4000)*** (2.3647)* -0.1815 (1.6746) Severity (severe=0 mild=1) 0.0301 (0.0542) -0.0178 (0.0019) -0.1353 (1.1537) 0.0106 (0.0292) 0.1745 (1.9264) 0.2297 (2.0526)* Gender (female=0 male=1) -0.0738 (0.5248) 0.6826 (1.4095) 0.0065 (0.0092) 0.1098 (3.8394)** -0.0251 (0.0957) -0.1908 (2.2195)* District (Amansie East=1, Kwaebibirem=0) -0.7855 (18.9881)*** -0.0931 (0.0495) -6.9461 (0.0000) -0.1866 (7.4405)*** 0.2606 (4.3745)** i 0.6869 (20.4208)*** Source: Survey data. *** Significant at a = 0.01 ** Significant at a 0.05 * Significant et a =0.10 where S = probability of self-medication (consulting no provider for prescription) D = probability of visiting drug store for malaria prescription G = probability of seeking malaria care at government/community health care facility P = probability of seeking malaria care at private or mission health care facility. 76 5.4.2. Self medication and private/mission providers Decision concerning choice of treating malaria at private or mission facilities as against self medication depends mostly on the treatment cost at these facilities. Results indicate that, majority of households would switch from self-medication to private/mission providers for malaria treatment if there should be an increase in the cost of treatment at these facilities. Private and mission clinics have been charging full cost fees and many people associate high quality with the services they provide. A higher price may be presumed to be higher quality. 5.4.3. Self medication and drug store. Religion appears to influence household decision with regard to choice between these two providers. Christians according to the result prefer to go to private/mission facilities as compared with drug stores more than non-christians. In addition to religion, household per capita income is also significant at the 10 per cent level. As income increases people tend to do self-medication than to consult a drug store operator. 5.4.4. Private/mission and government/community providers These categories of providers are both orthodox. From table 5.2, the most important factors for choice of either of these facilities include; travel time to facility, proximity and 77 respondent or the head of the household's years of education. These factors are all statistically significant at one per cent level. Government/community facilities are chosen mostly because of shorter travel time otherwise households would prefer to use private or mission facilities. Similarly, government or community facilities would be used if they are cited very close to the people. The number of years of education that a household head had also affects health care treatment choice decision. Results from table 5.2 indicated that longer years of education of a household head is a positive factor for selecting a private or mission health care facility for treatment. As revealed by the district factor, there is locational difference in the choice of either private/mission and government/community facilities. For example more educated heads of households from Amansie East district would prefer government/community facilities while the same category of households in Kwaebibirem district would prefer private/mission facilities. The finding actually, portrays the perceived strength of mission and government facilities in both districts. By the status of amenities, the district government hospital in Amansie East is perceived to be the best provider of health care services among others in the district and the same view is held about the St. Dominic mission hospital in Kwaebibirem district. Facility price and availability of drugs at facility are significant at 10 per cent level. When treatment charges and 78 availability of drugs are considered in selecting malaria care provider, the propensity to choose private/mission facility is higher. The observation on price factor may be because people are used to paying fees at private/mission facilities. Gender and age are also important in the private versus government malaria care provider choice decision equation. Males according to the results have more propensity to choose private/mission facilities for treatment. Children are also more likely to be sent to government facilities than adults. As people grow older they tend to attend government facilities instead of private/mission facilities. However, when compared with drug stores, older people tend to choose them over government clinics for malaria consultation. 5.4.5. Drug store and government/community providers. Waiting time and availability of drugs have much influence on choice of the above providers. People are sure of drugs at drug stores and therefore more inclined to seek treatment there instead of government facilities. The sign on the coefficient however, indicate that increase in price of drugs and longer waiting time at drug stores would lead to a shift in demand of treatment services from drug stores to government/community health care providers. Although the sample comprise Akan majority, it was observed that people of Akan origin are more likely to visit drug stores to seek malaria treatment than other ethnic groups. Heads of households who have more years of education were also found to 79 prefer consultations at drug stores than government/community facilities in treating malaria. The two districts also vary in view of the choice between these providers. 5.4.6. Drug store and private/mission providers The major factors ' determining household choice in this relation are facility price, waiting time for treatment, age of the sick, income and district affiliation. With regard to facility price and waiting time it can be deduced that many people seek health care currently from drug stores or chemical shops because of the relatively cheaper cost of seeking care and shorter waiting time at this facility. Both waiting time and facility price are significant at 1 % level of significance. From the results, children suffering from malaria are more likely to be sent for consultation at private and or mission facilities than being sent to a drug store. Another important factor in the decision of choice between drug store operators and private/mission health care providers is household per capita income. As shown in table 5.2 households with higher per capita income have higher propensity to seek care at drug stores. Significant differences also exist in behaviour of people in the two districts with regard to choice between drug stores and private/mission facilities. More people will seek malaria care at private/mission facility as opposed to drug store consultations in Kwaebibirem district than Amansie East district. This is not 80 surprising since St. Dominic Catholic Hospital in Kwaebibirem district is attended by patients from all over the district. Severe malaria tends to be sent to private/mission facilities as compared with drug store consultation. Significant gender differences were also observed. Males are more inclined to seek malaria treatment at private/mission facilities than drug stores. 81 CHAPTER SIX SUMMARY, POLICY IMPLICATIONS AND CONCLUSION fe. u introduction This study was set up to calculate the cost of malaria care in Ghana to investigate the'estimate the behaviour of people towards malaria care in terms of a choice model in a multinomial logit framework. Data for the study was obtained from administering a structured questionnaire to 1300 households in two districts in two regions of Ghana. This chapter is devoted to a summary of the major findings of the study and a discussion of the conclusions that can be drawn from these findings. The presentation is done in three sections. The first section is devoted to a summary of major findings while the second covers the policy implications for financing health care, and the last part presents conclusions and recommendations of the study. The results must be interpreted with caution since the cases of malaria considered were self reported. It may include other fevers. The one year recall period used in the study may also affect the values reported. 82 6•1 Summary of results 6.1.1 Costs • The general average cost per case of malaria for the combined sample was estimated to be 05987.41 or US $8.67. Of this cost 01247.88 or US $1.81 represent the direct cost of treatment and 04739.53 or US $6.87 represent indirect costs. On the average 5 days are lost to the di'sease in terms of patient and caretaker time. Drugs were one of the major specific items on which households spend money to treat malaria. The proportion of drug costs in total malaria treatment costs in general range between 37.2 per cent and 67.9 per cent in the two districts. Consultation and laboratory services were among the activities on which patients spent most time in seeking care at orthodox facilities. It was found that most households obtained drugs around their communities without travelling. However, travel cost for consultation is the second highest expenditure made by households in treating malaria. 6.1.2 Health care seeking behaviour Analysis of the actual behaviour in terms of choice of provider to treat malaria however, showed that the first treatment attempt by most people as revealed in the survey was self medication. This behaviour does not depend on age or severity of the disease. Upon the failure of the first treatment most people 83 switched from self medication to an orthodox provider. Interestingly all those who had to take a’ third action to treat an episode resorted to self medication again. 6.1.3 Impact of socio-economic factors on choice of malaria care provider * The major factors determining household choice of malaria care providers are facility price, waiting time for treatment, age of the sick, income and district affiliation. With regard to facility price and waiting time it can be deduced that many people seek health care currently from consulting at drug stores or chemical shops because of the relatively cheaper cost of doing so and the shorter waiting time at these stores. Travel ' time to facility, proximity and head of the household's years of education are very important to decision making with regard to choice between orthodox facilities. 6.2 Implications for policy Facility price stands out as a very important factor in the choice of health care service provider. Because of the price self medication has been on the ascendancy. This has implications for the correct diagnosis of disease and therefore the administering of the right treatment. The increase in self medication is also due to the perception that quality at orthodox facilities is low as many 84 of them do not have drugs and prescribers spend very little time with them in the consulting room. It was obvious that patients would travel long distances and pay higher fees to seek malaria care if they were sure the quality was good. At least this was the case for St. Domonic's Catholic Hospital in Akwatia in Kwaebibirem district. This implies that the cost sharing scheme should go along with improvements in the-services provided. One aspect of the high cost of seeking treatment is that many people would delay seeking care until the disease becomes severe. That is, they would consult orthodox health care providers after home remedies have failed to cure the disease. The knowledge that agricultural households have seasonal flow of income suggests that many of them have difficulty in raising sufficient funds to support a member to promptly seek malaria care at orthodox health care facilities at the onset of the disease. Support from extended families which used to reduce the shock of emergency health needs have now dwindled. The consequence of this is that the period of delay in seeking effective malaria/fever treatment has been extending as people spend some time to raise the needed funds. By such behaviour the World Health Organisation's (WHO) recommendation of controlling malaria through prompt and effective treatment will be in jeopardy. Apart from the prolonged suffering of the sick when households initially resort to treatment which are comparatively ineffective, more days are also lost to production by both patients and their caretakers. Although the direct impact of malaria morbidity on 85 agricultural output is not clearly documented in the literature at the moment the disease has far reaching consequences for society as a whole. As noted by Asenso-Okyere (1994) the chances of schooling of most children is bleak if a breadwinner dies. The frequent withdrawal of children from school either because they are sick of malaria or for them to substitute for a task of a sick household member could cause' them to be disinterested in schooling or lead to low academic standards. As already indicated, malaria morbidity and mortality are associated with low productivity and a fall in output. Malaria is therefore linked to poverty in a vicious cycle and thereby making it more difficult for households to afford treatment charges at reputable health care facilities. The observations on household malaria treatment seeking behaviour and determinants of choice of provider suggest that malaria control programmes could be successful if malaria or health education programmes are vigorously pursued jointly with poverty reduction programmes. It was also found that the nearer health care facilities are to people the better the chances are that they would use them. It is therefore suggested that future plans for health care development should consider citing health care facilities close to the people. Long waiting time at most government orthodox health care facilities was also found to be have a negative influence on facility utilization. Health facilities should therefore expedite the way they perform their duties, especially in government 8 6 facilities so that patients do not spend long times waiting for treatment. 6.3 Conclusion The high cost of malaria treatment at Cl,247.88 or US $1.81 appears to deter many rural households from promptly seeking care at orthodox health care facilities. This is evident in the importance attached to facility treatment prices before households select their malaria care providers. Although other socio-economic factors such as long waiting times at facility, proximity and household head's level of education were statistically significant for the various demand models, it can be inferred that income promotional activities could improve health care seeking behaviour and health in general in rural communities. 87 REFERENCES 1. Agyepong, I. A. (1991). Medical Treatment Choice in a Hill District of Eastern Nepal: A dissertation to the University of Liverpool School of Tropical Hygiene, Dept, of International Community Health. 2. Agyepong,I .A ., Wellington, E., and M. Ablordey. (1990). Socio­ cultural and clinical Features in Malalria Transmission. A study among adolescent girls in rural Ghana- Dangbe West District. Ministry of Health, Ghana/World Health Organisation, Geneva. 3. Aldrich, J. and Nelson, F. (1984). "Linear Probability, Logit and Probit Models." Beverly Hills Calif., Sage. 4. Akin, J.S.; Griffin, C.; Guilkey, D.K. and Popkin , B.M. (1986). "Demand for Primary Health Care in Bicol Region of the Philippines". Economic Development and Cultural Change, 34(4): 755-328. 5. Amemiya, T. (1985). Advanced Econometrics. Cambridge, Mass. Havard Universty Press. 88 6 . Amemiya, T. (1985). "Qualitative Response Models: A survey. Journal of Economic Literature. 7. Asenso-Okyere, W. K. (1982). "A Quarterly Econometric Model of the U.S. Beef Sector." Unpublished Ph.D dissertation, University of Missouri-Colombia. 8 . Asenso-Okyere, K.W. (1992). Disease and Health in four Selected Areas in Ghana with Special Reference to Malaria and Schistosomiasis. Health Social Sciences Research Unit, ISSER, University of Ghana. Monograph Series No. 1. 9.Asenso-Okyere, K.W. (1993) . Health Financing and Social Security in Ghana. Health Social Sciences Research Unit, ISSER, University of Ghana. Monograph Series No. 2. 10. Asenso-Okyere, K.W. (1994) . Socio - Economic Factors in Malaria Control. World Health Forum, World Health Organisation, Geneva: 15(3) 11. Audibert M. (1986). Agricultural non-wage production and health status: a case study in a tropical milieu. Journal of Development Economics 24 (1986) 275 - 291. 89 12. Avi Dor and van der Gaag; (1986). " Quantity rationing and demand of adults for medical care in rural Cote d'Ivoire": - in Health economic research in developing countries. 13. Avi Dor, Gertler, and van der Gaag, J. (1987). " Non-Price Rationing and Medical Care provider choice in rural Cote d'lvore". Journal of Health Economics 6 : 291­ 304 . 14. Behrman, J. R. and Deolalikar A. B. (1988). "Health and Nutrition". In Handbook of Development Economics, edited by H. Chenery and T.N. Srinivasan. vol.l. Elsevier Science Publishers, B. V. 15. Brad Schwartz J., Akin J. S., and Barry M. Popkin (undated). " Economic determinants of demand for modern infant delivery in low countries: the case of the Philippines. In Health economic research in developing countries. 16. Brandt, H. (1980). Work capacity constraints in tropical agricultural development. Frankfurt. Verlag Peter Lang(1980). 90 17- Brinkmann, u., and Brinkmann, A,(1991). Malaria and health in Africa: the present situation and epidemiological trends. Trop. Med. Parasitol. vol.42 204-213. 18. Castro E.B. and Mokate K.M. (1988). Malaria and its Socio­ economic Meanings: The study of Cunday in Colombia (ed) Herrin and Rosenfield, In: Economics, Health and Tropical Diseases. University of Philippines School of Economics. 19. Conly G.N. (1975). The impact of Malaria on Economic Development: A case study. Scientific Publication No. 297. Washington, D.C. Pan American Health Organisation. 20. de Ferranti, D. (1989). "Paying for Health Services in Developing Countries: An Overview ", World Bank Publication. 21. Diesfeld, H. (1970). The evaluation of hospital returns in developing countries. Methods of Information in Medicine vol.9 27-34. 91 22. Ellis., Randall, P. and Germano Mwabu., (1991). " Demand for out-patient medical care in Kenya". Mimeo, Boston University. 23. Ettling, M. B.,Shepard, D. S. (1991). Economic cost of malaria in Rwanda. Trop. Med. Parasitol. vol.42, 214-218. 24.,Femi Olokesusi, (1991). The Impact of Man and His Environment \ on Malaria Incidence in Ondo State. (A case study) . NISER. Monograph Serise No. 7. 25. Gertler, P., and van der Gaag. (1988). The willingness to pay for medical care; Evidence from two developing countries; The John Hopkins University Press, Baltimore. 26. Goldberger, A. (1964). Econometric Theory. New York: Wiley 27. Greene A.W. (1993). " Econometric analysis." Second edition, Published by Macmillan, N.Y. 28. George, P. S. and King, G .A . "Consumer Demand for Food Commodities in the United States with Projections for 1980." Giannini Foundation Monograph No. 26, University of California, 1971. 92 29. Hassan, Z.A. and Johnson, S.R. (1976). "Consumer Demand for Major Foods in Canada." Agriculture Canada, Economics Branch Publication No. 7 6/2, April. 30. Joseph F. Hair (Jr). Rolph E. Anderson, Ronald L. Tatham and William C. Black. (1992). " Multivariate data analysis with readings". Third edition. Published by Macmillan N.Y. 31. Johnson, S.R., Hassan,Z.A. and Green, R.D. (1982). "Demand System Estimation— Methods and Application." Ames: Iowa State University Press. 32. Judge G.G. et. al. (1988). Introduction to the Theory and Practice of Econometrics, 2nd ed. New York: Wiley. 33. Lennox, R.W. (1991). Malaria in Africa: the need for economic analysis. Trop. Med. Parasitol. vol.42, 198. 34. Leontief, W.W. (1967). "Introduction to the Theory of the Internal Structure of Functional Relationships." Econometrica 15(361-373). 35. Li Mingche M. (1977). " A logit model of home ownership" Econometrica, vol 45, No. 5 [July 1977]. 93 36. Macdonald, G. (1950). The economic importance of malaria in Africa. [Unpublished WHO document]. Geneva, World Health Organization, WHO/MAL/60 Afr/Mal/Conf/16. 37. Maddala G. S. (1992). Introduction to Econometrics. Second edition. Published by Macmillan N.Y. 38. McFadden, D. (1973). "Conditional Logit Analysis of Qualitative Choice Behaviour." In P. Zarembka, ed. Frontiers in Econometrics, New York Academic Press. 39. McFadden, D. (1974). "The Measurement of Urban Travel Demand." Journal of Public Economics 3(303-328). 40. McFadden, D. (1984). "Econometric Analysis of Qualitative Response Models." In Z. Griliches and M. Intriligator, eds., Handbook of Econometrics, vol. II Amsterdam, North Holland. 41. McPake, B., Hanson, K., and Mills, A., (1992). "Implementing the Bamako Initiative in Africa". A review and five country case studies. London School of Hygiene and Tropical Medicine, Department of Health and Policy. 94 42. MOH, Ghana. (1991). Malaria Action Plan 1993 - 1997- Prepared by the Epidemiology Division , Ministry of Health, Accra. 43. Mwabu, G., Ainsworth, M., and Nyamete, A., (Forthcoming). "Quality of medical care and choice of medical treatment 'in Kenya: An empirical analysis; Journal of Human Resources. 44. Mwanzia, J., and Mwabu, G., (1993). "User charges in public health facilities in Kenya11- Effect on revenue and attendance without improvement in medical care quality". Mimeo. Ministry of Health, Kenya. 45. Nerlove, M. and Press, S. (1973). "Univariate and Multivariate Log-linear and Logistic Models. " RAND-R1306-EDA/NIH, Santa Monica, California. 16. Nur E.T.M.and Mahran H.A. (1988). The effect of Heath on Agricultural Labour Supply: A Theoretical and Empirical Investigation, (ed) Herrin and Rosenfield In : Economics, Health and Tropical Diseases. University of Philippines School of Economics. 95 47. Pampana. (1963). "International implications of malaria eradication". In a textbook of Malaria Eradication. Oxford University Press, London. 48. Peter Kennedy (1979). "A guide to Econometrics", third edition. Printed by T. J. Press Ltd. Padstow; Cornwall. 49. Pindyck, R. and Rubinfeld, D. (1981); " Econometric Models and Economic Forecasts." New York McGRAW-HILL Book Company. 50. Pindyck, R. and Rubinfeld, D. (1991); " Econometric Models and Economic Forecasts." 3rd ed. New York McGRAW-HILL Book Company. 51. Pitt. M. , Rosenzweig and M. Hassan. 91990). " Productivity, Health and Inequality in the Intrahousehold Distribution of food in Low-income Countries". American Economic Review, vol. 80. 52. Salako, L.A. (1987). Malaria in Nigeria. In:Buck, A.A. (ed) . Proceedings of the conference on malaria in Africa. Practical considerations on malaria vaccines and clinical trails, Washington, D.C. December 1-4 1986. Washington, D.C., American Institute of Biological Sciences. 131-141. 96 53. Sauerborn, R., Shepard, D. S., Ettling, M. B., Brinkmann, U. , NoUgtara, A., Diesfeld, H. J. (1991). Estimating the direct and indirect economic costs of malaria in rural district of Burkina Faso. Trop. Med. Parasitol. vol 42, 219-223. 54. Sauerborn, R.(1990).A n analysis of the variable cost of providing in-patient and out-patient care in the medical district of Solenzo.Mimeo Boston. Harvard School of Public Health. 55. Schmidt, P. and Strauss, R. (1975a). "The Prediction of Occupation using Multiple Logit Models." International Economic Review 16(471-486) 56. Schmidt, P. and Strauss, R. (1975b). "Estimation Models with Jointly Dependent Qualitative Variables: A Simultaneous Logit Approach." Econometrica 43 (745-755). 57. Schwartz, J.B., Akin, Popkin, B.M., (1988). "Price and Income Elasticities of Demand for Modern Health Care: The case of Infant Delivery in the Philippines". The World Bank Economic Review 2(2): 4 9 - 76. 97 58. Shepard, D.S., Ettling, M. B., Brinkmann, U., Sauerborn, N. (1989). The economic cost of malaria in Africa. Trop. Med. Parasitol. Vol.42 199-203. 59. Sono, M. (1961). "The Effects of Price Changes on the Demand and Supply of Separable Goods." International Economic Review 2(239-271). 60. Strotz, R.H. (1957). "The Empirical Implications of a Utility Tree." Econometrica 25(269-280). 61. Subedi J. (1989). "Modern Heath Services and Health Care Behaviour ; A Survey in Kathmandu, Nepal". Journal of Health and Social Behaviour vol. 30 (December): 412 - 420. 62. Theil, H. (1971). Principles of Econometrics. New York. John Wiley and Sons, Inc. 63. Waddington,C.J.K. and Enyimayew (1989) . A price to pay: the impact of user charges in Ashanti-Akim district of Ghana. International Journal of Health Planning and Management, vol.4 17-47. 98 64. Waddington, c. , and Enyimayew, K. A., (1990). "A price to pay, Part 2. "The impact of user charges in the Volta Region of Ghana," International Journal of Health Planning and Management 5(4): 287 - 312. 65. Wang'ombe, J. (.). "Economic Evaluation in primary Health Care: The case of Western Kenya Community Based Health Care Project" Social Science and Medicine 18(.) 375 - 385. 66. World Bank (1987). Financing Health Services in Developing Countries: An agenda for reform, World Bank Washington, D. C. 67. World Bank (1986). Population, Health and Nutrition. Sector Review. Report No. 7597-Gh. 58. Yoder, R.A. (1989). "Are people willing and able to pay for Health Services ?. Social Science and Medicine 29(.) 35 - 42. 99