Health Policy and Planning Advance Access originally published online on December 16, 2005
Health Policy and Planning 2006 21(2):110-122; doi:10.1093/heapol/czj008
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Original article |
Geographic aspects of poverty and health in Tanzania: does living in a poor area matter?
School of Public Health and Tropical Medicine, Tulane University, New Orleans, LA 70112, USA
Correspondence: David R Hotchkiss, Department of International Health and Development, School of Public Health and Tropical Medicine, Tulane University, 1440 Canal Street, Suite 2200, New Orleans, Louisiana 70112, USA. Tel. +1–504–988–3289. E-mail: hotchkis{at}tulane.edu
 |
Abstract |
|---|
Previous studies have consistently found an inverse relationship between household-level poverty and health status. However, what is not well understood is whether and how the average economic status at the community level plays a role in the poverty–health relationship. The purpose of this study is to investigate the concentration of poverty at the community level in Tanzania and its association with the availability and quality of primary health care services, the utilization of services, and health outcomes among household categories defined by wealth scores. A principal component method has been applied to rank households separately by urban/rural location using reported levels of asset ownership and living conditions. The household wealth scores were also used to classify communities into three cluster-types based on the proportion of households belonging to the poorest wealth tercile. On average, all the wealth terciles living in low poverty concentration areas were found to have better health outcomes and service utilization rates than their counterparts living in high poverty concentration clusters. Consistent with the finding is that high poverty concentration areas were further away from facilities offering primary health care than low poverty concentration areas. Moreover, the facilities closest to the high poverty concentration areas had fewer doctors, medical equipment and drugs. Among the high poverty concentration clusters, the 10 communities with the best women's body mass index (BMI) measures were found to have access to facilities with a greater availability of equipment and drugs than the 10 communities with the worst BMI measures. Although this study does not directly measure quality, the characteristics that differentiate high poverty concentration clusters from low poverty concentration clusters point to quality as more important than physical access among the study population.
Key Words: poverty, geographic location, service utilization, health status
|
Introduction |
|---|
In both developed and developing countries, poverty or low income is an important determinant of the health status of the population (Backlund et al. 1996
; Polednak 1997; Marmot 1999; Shaw et al. 1999; Schellenberg et al. 2002; Gwatkin et al. 2003). In fact, poverty and poor health status appear to reinforce each other, making it difficult for individuals with poor health to break out of a ‘poverty–health’ trap (Whitehead et al. 2001). Thus, a lower socioeconomic situation leads to poorer health, which in turn, exacerbates the incidence and intensity of poverty.
A number of studies have also found quite a strong geographic concentration of poverty, i.e. poor households tend to live in areas where most of the people are poor (Polednak 1997). The geographic concentration of poverty is not surprising. Fertility of agricultural land and other agro-climatic variables affect the economic status of the population in a geographic area (Minot and Baulch 2002), and the economic status of the population may affect the infrastructural development of the locality. The presence of improved facilities, infrastructure and quality services increases the cost of living, thereby pushing poor households out of the area. Despite these underlying dynamics, geographic units are rarely exclusively poor or non-poor, especially in developing countries of the world.
In a private market-oriented system, purchasing power affects service availability and its quality. However, it is not clear if the average economic status of geographic regions also affects the spatial distribution of primary health care services, especially when health services are organized and delivered predominantly by the public sector or not-for-profit agencies. The purpose of this study is to examine the relationships between the concentration of poverty and the availability of primary health care services, quality of care, service utilization and health outcomes in small geographic areas. The main questions we want to address are: Is increased geographic concentration of poverty associated with lower health status after controlling for household socioeconomic status? Is there any systematic bias against relatively poor areas in terms of physical availability of primary health care facilities? Does the quality of services delivered vary significantly with the average economic status of geographic areas? Are increases in the supply of health care services, especially of maternal and child health services, associated with improved health status of the population in relatively poor areas?
Since the research questions of the study follow a sequential pattern, the paper has been organized to address the issues in a recursive step-by-step manner. After defining relative poverty in terms of wealth scores, we examine the relationship between relative poverty and health status. The effect of geographic concentration of poverty on health can be identified once we control for the household-level poverty and health relationship. Following an analysis of the geographic concentration of poverty and health, potential factors influencing the poverty–health relationship among small geographic units are presented by looking at the availability and quality of health care services. The analysis also explores whether the geographic concentration of poverty has an independent effect on population health. Finally, the last section presents a summary of the results as well as a discussion of the policy implications and the limitations of the study.
|
Background |
|---|
We address the questions posed above using data from Tanzania, an East African country with high levels of poverty but with a history of experimentation with socialism and egalitarian efforts at wealth redistribution. The Tanzanian Government's 1967 Arusha Declaration proposed a decentralized system of government and a rural development plan based on cooperative farm villages. Heavy emphasis was also placed on the development of primary education and primary health care (Ofcansky 1997
).
The effectiveness of these reforms was limited, and Tanzania remains one of the world's poorest countries (World Bank 2001), with 41% of the population below the poverty line of about US$1 a day. According to the United Nations Development Program, Tanzania's Human Development Index ranking was 151 out of 173 countries (UNDP 2002). The health indicators of the population, especially the ones related to child and maternal health, have remained extremely poor. Life expectancy of population at birth in 2000 was only 51 years, mainly due to the extremely high rate of infant mortality (at 104 per 1000 live births). Nutritional status of the surviving children is quite poor as well; 29% of under-five children are underweight and 44% have low height for age. The maternal mortality ratio per 100 000 live births remained at around 530 in recent years, about 100 times the rate for Western European countries. More than 45% of the population of Tanzania is below 15 years of age and the total fertility rate per woman is 5.5 (UNDP 2002). Clearly, given the high fertility rate in the country, the overall improvements in the health status of the population will depend crucially on the ability of the country to improve the health and nutritional status of women and children.
Access to quality health services varies geographically, but on average, rural populations in Tanzania tend to be closer to health facilities than rural populations in neighbouring countries (Beegle 1995). Even so, the availability of key supplies, equipment and services at those clinics varies considerably (Turner 1994; World Bank 1999; Chen and Guilkey 2002). According to the UNDP, immunization against measles among infants was about 72% in 1999 and less than 79% of the population had access to essential drugs. The contraceptive prevalence rate was only about 24% and only about a third of all deliveries were attended by skilled health staff. Efforts over the past decade have sought to improve public sector administration and health system responsiveness by devolving key responsibilities for health system management to local governments (Gilson et al. 1994; Hutchinson 2002; De Savigny et al. 2002). Unfortunately, these efforts have not produced the intended results in terms of health service utilization and health status of the population. A recent study indicates that although the likelihood of being ill does not vary significantly among socioeconomic quintiles in a poor region in Tanzania, the children from relatively poor quintiles are less likely to receive appropriate treatment for illness (Schellenberg et al. 2003).
|
Methods |
|---|
Survey design and sample size
The data used in the study have been obtained from two sources: household information was obtained from the 1996 Tanzania Demographic and Health Survey (TDHS), while information on the health care supply environment comes from the 1996 Tanzania Service Availability Survey (TSAS) (Bureau of Statistics [Tanzania] and Macro International 1997). The TDHS, a nationally representative survey, provides detailed information on household demographics, asset ownership, dwelling conditions, health status of women and children, utilization of maternal, child health and other selected health care services, and knowledge and practices related to health. The TDHS was conducted from July to November 1996, but about 90% of households were surveyed during August to October. Therefore, the cross-sectional analysis results should not be biased due to seasonal variability of economic and health status.
The TDHS is based on a three-stage sampling design and consists of 357 sample enumeration areas, the same clusters used by the Demographic and Health Survey carried out in 1991/92. Ninety-five of the clusters were located in urban areas, and the remaining 262 were rural. The selection of these enumeration areas or sample clusters was made in two stages. At the first stage, small administrative units called ‘wards’ were selected randomly, and at the second stage, clusters were randomly selected within the selected wards. A list of households residing in those clusters was prepared and, at the third stage, households were randomly selected from each of the clusters in proportion to the number of households in the clusters. In total, the TDHS selected 8900 households for the survey, and out of this sample a total of 8120 women aged 15 to 49 years (2088 urban and 6032 rural) were successfully interviewed. The average number of households surveyed in a cluster was about 23.
In addition, information on the availability and quality of health care facilities was obtained from the 1996 TSAS, which was implemented in conjunction with the TDHS. For each sample cluster, the closest of each type of facility (hospital, health centre and dispensary) was visited, and information on the facilities and their service delivery operations was obtained. The TSAS includes detailed information on staff training and availability of services, infrastructure, equipment, supplies and medical personnel. The TSAS did not collect information on the distances between the surveyed clusters and the health care facilities, but the geographical coordinates of communities and facilities were collected in a previous survey, the 1991/92 Tanzania Service Availability Survey. Geo-coordinates from the 1991/92 survey have been used to calculate the straight-line distances between the centre of each of the clusters and the health facilities that serve the cluster. Unfortunately, neither the 1991/92 TSAS nor the 1996 TSAS were carried out on the island of Zanzibar. As a result, we exclude households in Zanzibar from the parts of the analysis that are based on linked household and facility information.
Research approach/variable definitions
The research approach used in this study consists of simple cross-tabulations to assess the associations between household poverty, health status and health service utilization patterns after controlling for community-level poverty.
Household asset scores and relative poverty
Household socioeconomic status has been defined by using information on asset ownership and dwelling conditions. Because the TDHS did not collect information on household income or consumption, we used alternative measures of economic status based on asset or wealth indicators. A number of studies have demonstrated the validity of using wealth-based indicators for categorizing households (Filmer and Pritchett 1999; Montgomery et al. 2000; Filmer and Pritchett 2001). Since the TDHS questionnaire collected information on both the dwelling conditions (i.e. type of toilet, floor construction, availability of electricity, source of water) and asset ownership (i.e. radio, refrigerator, bicycle, motorcycle, car), all these variables were used to construct a composite measure of economic well-being or wealth by applying principal component analysis. These indices are often referred to as household wealth scores. Details on the construction of the wealth scores are summarized in the Appendix. Note that we are using household asset position as a proxy for the economic status of all members of the household.
Since the wealth scores are based on patterns of asset ownership among households rather than the monetary value of assets owned, the scores can be used to define relative socioeconomic position and relative poverty. In order to define the socioeconomic groups, we have ranked households in terms of their wealth scores (described in the Appendix). Most previous studies that used this type of approach divide the sample into wealth quintiles (Filmer and Pritchett 1999; Montgomery et al. 2000; Filmer and Pritchett 2001). In this analysis, dividing the sample into quintiles (or quartiles) would require arbitrary allocation of a significant number of households to adjacent quintiles due to the heaping of wealth scores at the lower tail of the distribution. In order to avoid this problem, we opted to divide the sample into wealth terciles. The lowest tercile defines the ‘poorest group’ for this analysis. Appendix Table A1 lists all the variables on assets and living conditions used for constructing the wealth score, and the percentage of urban and rural households that report owning these items for the three wealth terciles defined.
|
Cluster-level poverty
Our principal interest in this study is to examine the degree of poverty concentration and its association with access to and quality of health care services, health care utilization and health outcomes. We have classified each cluster into one of three groups based on the proportion of cluster households belonging to the poorest tercile. Clusters were classified as being of ‘high poverty concentration’ if more than 60% of households in the cluster were in the poorest tercile, as ‘medium poverty concentration’ if 33 to 60% of households were in the poorest tercile, and ‘low poverty concentration’ if fewer than 33% of households belonged to the poorest tercile.
Although the cut-off points chosen are quite arbitrary, two practical concerns were balanced to identify the cut-off levels. To define the high poverty concentration areas, we started with 75% as the cut-off. The number of clusters above this level was so low that we decided to use a lower cut-off level to increase the sample size for empirical analysis. The choice of 60% as the cut-off yielded 48 clusters in rural areas and 20 clusters in urban areas in the high poverty concentration category. The lower cut-off point used to distinguish between low and medium poverty concentration areas was arbitrarily set at 33%. In poverty mapping, researchers have used different head-count ratio cut-off levels for different countries of the world. For example, Woldemariam and Mohammed (2003) used a head-count ratio of 33% to define the least poor areas for Ethiopia, but for Malawi a higher proportion was used (49%). Bigman et al. (2000) did not define any explicit cut-off level but grouped 3871 communities of Burkina Faso into four poverty concentration categories by assuming that each category should have 25% of the population. Minot and Baulch (2002) mapped the poverty areas of Vietnam by using a head-count poverty ratio of less than 40% to define lowest poverty concentration areas (the two least-poor geographic groups).
In the cluster level analysis, we have dropped those clusters with fewer than 10 sample households. These clusters are likely to be located in low population density areas. The exclusion of the small clusters reduced the number of clusters in the sample from 327 to 298. Exclusion of the clusters, however, did not affect the results of the analysis.
Availability and quality of health care services
The health facility surveys associated with the TDHS collected information on facilities located in or near each of the survey clusters. Access to health facilities was defined by the facility's distance from the centre of the cluster. Distance has been used as the proxy for access in many other studies as well (Stock 1983; Booth and Verma 1992; Muller et al. 1998; Noorali et al. 1999; Nemet and Bailey 2000; Brabyn and Skelly 2002; Buor 2003). The presence of a facility, by itself, does not indicate adequate access to or delivery of quality services. A number of additional variables were used to indicate the degree of access to and the quality of services delivered through the facilities. Following the literature on quality indicators (Mwabu et al. 1994; Alderman and Lavy 1996; Akin et al. 1998; Sahn et al. 2003), we measure facility quality by the number of health care providers in the facility, the availability of essential medical equipment and supplies, and the availability of essential drugs.
In order to incorporate information on the availability of essential medical equipment and drugs into a measure of facility quality, we again used the principal component approach. Alternative measures of equipment availability were also defined by adding the type of different drugs and equipment available in the health facility. For drugs, we considered not only the availability of drugs but also information on their regular availability. For example, if the facility did not have a specific drug from the list during the survey, the value assigned to that facility for that drug was zero. The drug availability index was assumed to be 0.5 if the facility had the drug at the time of the survey but reported a stock-out during the past 6 months, and 1.0 if the facility had the drug without any stock-out. After assigning the values to each of the drugs in the list, an overall facility-specific drug availability index was constructed by adding the values for all drugs. The first principal component scores were also used to construct an alternative index of drug availability for a health centre.
Health status and service utilization measures
In this study, a number of anthropometric measures have been used as proxies for the overall health status of the population. In developing countries, anthropometric measures of women and children are found to be much more sensitive to changes in economic situation than other health indicators (Gwatkin et al. 2003). Due to the higher variability of women's and children's nutritional status among socioeconomic groups, these were chosen as the preferred measures to use in the statistical analyses. Moreover, the health statuses of women and children correlate quite well with the overall health status of the population (Gray 1993). It should be noted here that the relationship between nutritional status and body mass index (BMI) is not a linear one; just like low BMI, high BMI also reflects poorer nutritional status. Despite this potential problem, Gwatkin et al. (2003) have demonstrated the usefulness of the indicator for the developing country context.
In the analysis, we have also used several health status indicators based on disease incidence for women and children. The survey collected information on incidence rates for fever and diarrhoea among under-five children within the 2 weeks prior to the survey. The DHS defined diarrhoea as three or more watery stools in a day.
|
Wealth scores and poverty in Tanzania |
|---|
Household-level poverty and health status
Table 1 shows the mean values of BMI by household wealth tercile and by rural and urban residence. In developing countries, lower socioeconomic status should be associated with lower BMI values. The table indicates that for both rural and urban areas, the average BMI value increases with improved wealth scores. Table 1 also shows the distribution of women by three BMI categories: low BMI (less than 21), medium BMI (21 to less than 27) and high BMI (27 or greater). In rural areas, only about 3% of women from poor families had high BMI compared with 7% among women in the top socioeconomic category. At the other extreme, over 44% of women from the poorest tercile have low BMI levels compared with 39% among women in least poor households. In urban areas, 38% and 27% of women from the poorest and least poor terciles, respectively, had low BMI. Therefore, in terms of the BMI of women, relative poverty appears to be related to health status. The results of one-tail significance tests imply that the average BMI of the poorest tercile was significantly lower than the average BMI of the least poor tercile.
|
Two indicators of children's nutritional status were calculated from height, weight and age information. The average weight-for-height (WFH) and height-for-age (HFA) z-scores are reported in Table 2. Notice that, in both rural and urban areas, the average scores tend to improve with better household wealth scores. In addition, in both rural and urban areas, the average HFA z-score among children in the least poor tercile was significantly higher than for children in the poorest tercile. The difference of WFH scores between the poorest tercile and least poor tercile was statistically significant for children in rural areas but not for children in urban areas.
|
Table 2 also reports the prevalence of common childhood illnesses by household wealth terciles in both urban and rural areas. Among children in urban areas, the reported prevalence of diarrhoea or fever declines with higher household wealth, and this result is statistically significant. No such pattern was found among children in rural areas.
Reported prevalence of illnesses is often quite unreliable. Perceptions of illness vary significantly among socioeconomic groups due to the differences in educational status of household members and the extent of knowledge about health and disease (Sen 2000). Poor households are less likely to seek medical care for childhood illnesses, and when medical care has not been sought, households may not consider the episode as an illness. Since BMI and child anthropometry, health indicators that are not subject to reporting bias, are highly associated with socioeconomic status, we can conclude that relative household poverty affects the health status of household members adversely.
Cluster-level poverty and health status
The negative relationship between economic status and health is well known and widely reported. However, in this paper, we want to investigate the effect of concentration of poverty within a geographic area on the health status of the population living in that area. As mentioned earlier, we have defined the concentration of poverty by using household wealth scores. The three geographic clusters we have defined are high concentration, medium concentration and low concentration of poverty. In the clusters categorized as high concentration of poverty, 60% or more of households are from the poorest wealth tercile group.
Table 3 reports the health status of the population by cluster categories. Again, we have used anthropometric measures of women and children, and prevalence of childhood illnesses, to define the average health status of the population in a geographic area. Note that, with the exception of weight-for-height in urban areas, the mean anthropometric z-scores of children were found to be significantly lower in the high and medium poverty concentration areas than in the lowest poverty concentration areas. The analyses of the incidence of childhood illnesses and women's BMI reveal a similar pattern. For example, in urban areas, a significantly lower reported prevalence of diarrhoea and fever is observed on average in the low poverty concentration areas than in high poverty concentration areas.
|
However, it is not surprising to see a correlation between the degree of concentration of poverty and the health status of the population. We have seen before that health status measures are related to relative poverty levels. A higher proportion of households from the poorest tercile in an area should imply lower health status in that area. Therefore, lower average health status of high poverty concentration areas may not necessarily imply an independent effect of geographic poverty concentration on health.
To test the hypothesis that poverty concentration at the community level influences health status of the population after controlling for household wealth, we compared the health status measures by household-level wealth terciles and by cluster-level poverty concentration categories (Table 4). While these results are somewhat mixed, they generally support the premise that differences in health status among the population are not due to the wealth scores of households alone, and that the concentration of poverty at the community level appears to play a role in determining the health status of the population. For example, consider the results on height-for-age. Among the children in the poorest and least poor household wealth terciles, those living in the low poverty concentration areas were found to be significantly better off than those living in the high poverty concentration areas. However, no such relationship was detected among children living in the middle tercile. In addition, for the indicator weight-for-height, children of households in the middle household wealth tercile living in the low poverty concentration areas were better off than comparable children in the high poverty concentration areas. However, this finding does not hold for children in the poorest and least poor wealth terciles.
|
Among women in each of the three household wealth terciles, the percentage of women with low BMI levels (below 21) was found to be lower in the low poverty concentration areas than in the high poverty concentration areas. The results were found to be statistically significant for women in the poorest and least poor terciles, but not in the middle tercile.
|
Poverty concentration, access to care and quality of care |
|---|
 TopAbstractIntroductionBackgroundMethodsWealth scores and poverty... Poverty concentration, access to... Does geographic access affect...Variability within cluster...ConclusionBiographiesAppendixReferences |
|---|
One of the related questions is whether access to and quality of health care services is different between clusters with low and high poverty concentrations. If the health care supply environment is worse in high poverty concentration clusters, then this might be an important factor in explaining why the concentration of poverty is associated with poorer health outcomes. Using a number of measures of both physical access to services and the readiness of health care facilities to provide services, we find evidence that high poverty concentration areas are indeed at a disadvantage.
Consider Table 5, which compares the physical availability of services for households living in the high poverty concentration clusters with that for households in the low poverty concentration areas. Distances were calculated between the clusters and the closest facility of any type, as well as to the closest hospital, health centre, dispensary and UMATI clinic. Dispensaries are administered by the Ministry of Health, while UMATI clinics are administered by the Family Planning Association of Tanzania.
|
In rural areas, the high poverty concentration areas were found to be more than 50% further away from the closest health care facility of any type than the lowest poverty concentration areas (5.6 km vs. 3.5 km). High poverty concentration clusters in rural areas were also found to be further away from the closest hospital, health centre and UMATI clinic. For dispensaries, no systematic relationship was found between the degree of poverty concentration of an area and the distance.
Within urban areas, the results are somewhat mixed. Households in high poverty concentration clusters were found to have significantly less favourable physical access to dispensaries, health centres and hospitals than households in lower poverty concentration areas. However, unlike rural areas, there does not appear be a difference between high and low poverty concentration areas in the distance to the closest facility of any type. In addition, physical access to UMATI clinics was found to be more favourable for the high poverty concentration clusters than for better-off clusters.
If we consider the number of health care personnel available in all facilities nearest to the sample clusters (excluding hospitals) as another proxy for access to care, the high poverty concentration clusters were found to be worse off than the medium and low poverty concentration clusters. We compared the clusters with respect to the mean number of doctors, medical assistants, medical aides and trained midwives working in the nearest primary health care facilities. The results indicate that as the concentration of poverty declines, the average number of each type of personnel increases. The differences in the number of personnel across the poverty concentration areas are statistically significant.
In addition to having less physical access to services, households in high poverty concentration clusters may also face lower quality of care at health care facilities than households in low poverty concentration clusters. We explored this issue by comparing high and low poverty concentration areas by the availability of equipment and drugs in health care facilities. The results indicate that, in both rural and urban areas, facilities nearest to high poverty concentration clusters had significantly lower levels of supplies of equipment and drugs than facilities nearest to the other, lower, poverty concentration clusters (see Appendix Table A2).
|
|
Does geographic access affect utilization of services? |
|---|
|
TopAbstractIntroductionBackgroundMethodsWealth scores and poverty...Poverty concentration, access to...Does geographic access affect...Variability within cluster...ConclusionBiographiesAppendixReferences |
|---|
Lower access and inferior quality of service may adversely affect health outcomes among households living in highest poverty concentration clusters compared with other clusters via the utilization of health care services. To examine utilization patterns by the poverty concentration of clusters, we calculated the proportion of respondents using three types of primary health care services: prenatal care, deliveries and immunizations (Table 6).
|
For prenatal care and deliveries, utilization rates in both rural and urban areas were found to be significantly lower in the high poverty concentration clusters than in low poverty concentration clusters. For example, the percentage of rural women delivering in a health care facility was reported to be 33.9% in the high poverty concentration clusters, compared with 47.0% in the low poverty concentration areas. In addition, women in high poverty concentration clusters were less likely to have their births attended by a trained provider than women in low poverty concentration clusters (43.7% vs. 55.8%), and in rural areas, utilization rates for tetanus toxoid injections and child vaccinations were found to be significantly lower in the high poverty concentration areas than in low poverty concentration clusters. (These findings do not hold among individuals residing in urban areas.) Unfortunately, we were not able to investigate the utilization rates of curative health care services, as the TDHS did not collect this type of information.
|
Variability within cluster categories |
|---|
It is interesting to note that some of the high poverty concentration clusters fared quite well in terms of BMI of mothers and anthropometry of children. Are these clusters different from others in the same category in terms of access to medical care, educational status or knowledge of household members on health and diseases? To examine the differences between the two sub-groups of high poverty concentration clusters, we identified the 10 worst and the 10 best clusters using BMI of women as the basis for selection. Since the access indicators for urban clusters were skewed due to the presence of a large tertiary hospital, we limited the analysis to high poverty concentration clusters in rural areas.
Table 7 shows the characteristics of these two sub-groups within the high poverty concentration areas. Note that the women and their husbands in the best clusters are better educated than their counterparts in the worst clusters. Moreover, the percentages of women having knowledge on diarrhoea and family planning methods were also significantly higher in the best 10 clusters.
|
In terms of the structural factors thought to influence the quality of health care services, the 10 worst clusters were found to have lower levels of equipment and drugs in the closest health care facilities compared with the 10 best clusters. For example, the principal component scores for medical equipment were –4.804 and –0.267 for best and worst poor clusters, respectively, while the principal components for drugs were –1.018 and –0.504. However, these differences were not statistically significant. Not reported in Table 7 is the finding that the differences between the two groups of clusters on physical distance to the closest health care facilities were also statistically insignificant.
|
Conclusion |
|---|
Summary
Using an index of asset ownership for categorizing households into three socioeconomic groups, this paper shows that the poorest tercile of households in Tanzania had worse health status indicators compared with those in the least poor tercile. This is consistent with the findings in the literature that indicate an inverse relationship between the degree of poverty and health status.
We also investigated the relationship between geographic poverty and the health status of population. The extent of geographic poverty was defined by the concentration of households from the poorest wealth tercile in a locality. A comparison of the health status measures of the population, especially the BMI of women and height-for-age of children, showed that degree of concentration of poverty and the health status measures are related. The fact that the differences in health status measures between high and low poverty concentration clusters became greater when we controlled for the wealth terciles of households implies that geographic poverty concentration has an independent effect on health. In other words, relatively poor households living in low poverty concentration areas were, on average, better off in terms of a number of health status measures than poor households living in high poverty concentration clusters. Similarly, least poor households living in low poverty concentration clusters had better health status than their counterparts living in high poverty concentration clusters.
We also investigated a number of factors that might help to explain why poverty concentration is associated with health status, even after controlling for household-level socioeconomic status. For example, the analysis of the cluster-level data revealed that the health facilities were located further away from the high poverty concentration clusters than from the low poverty concentration clusters. For Tanzania, we expected a relatively egalitarian geographic distribution of primary health care facilities because of its socialistic past. The presence of unequal geographic distribution in Tanzania probably implies that the situation is even worse in other developing countries.
Distance is only one aspect of unequal access to care for the high poverty concentration clusters; another important aspect is the availability of health care providers and drugs in the facilities. Households from the high poverty concentration clusters not only had to travel longer distances to reach primary health centres or other health facilities, but these facilities also had fewer health care providers, less medical equipment and fewer drugs. Utilization of selected primary care services was also found to be lower in high poverty concentration areas than in areas of low poverty concentration. Although, we cannot derive a definitive conclusion from the tabulations, it appears that the quality of services delivered by health facilities is an important factor in explaining utilization differences.
We compared the physical availability to and quality of primary health care services for the 10 best and 10 worst health status clusters among the high poverty concentration clusters of the sample. The results indicate that the average distances to health facilities between these two sets of poor clusters were almost identical, but the two groups differed significantly in terms of the quality of services delivered from the health facilities, the educational status of women and their husbands, women's knowledge about family planning and other medical care services, and the percentage of women widowed or divorced. These findings probably imply that health status in poor areas is affected by a wide variety of social and economic factors.
Policy implications
A number of studies in recent years have found that many health, nutrition and population programmes are less effective in reaching the poor than the better-off, and that the programmes thus may well be contributing to rather than alleviating poor–rich disparities in health status (i.e. Gwatkin et al. 2003). By linking nationally representative household-level data with facility-level data from the same sample areas, this study allowed us to investigate the role of health care availability and quality in explaining these disparities. The results point to two policy implications. First, lowering the physical distance between high poverty concentration areas and health centres is likely to be effective in improving the health status of relatively poor households. Secondly, the analysis of health outcomes of women and children living in high poverty concentration clusters also implies that the quality of services offered from health centres may be more important than the distance to the health centre. Therefore, if lowering the distance to health centres from high poverty concentration areas is not feasible, relatively poor households can also be reached more effectively by improving the quality of care delivered from the health centres. Clearly, more effective systems to finance and deliver health services are central to efforts to reduce rich–poor inequities.
Limitations to the analysis
There are a number of limitations to the study. First, because the TDHS did not collect information on cluster-level population density, it was not possible to investigate whether population density, rather than poverty concentration, could explain the higher distance of health facilities from poorer areas. This is important, as an alternative explanation for the observed relationship between geographic concentration of poverty and health could be the higher prevalence of poverty in remote areas of a country, as found by Minot and Baulch (2002) for Vietnam. Our emphasis on cluster-level poverty without controlling for the density of relatively poor households (per square kilometre) in the cluster may have created some bias. Secondly, while the overall findings presented in this paper are consistent with the hypotheses that community-level poverty has an independent effect on health status and that raising the availability and quality of health care services to high poverty concentration areas can be effective in reducing inequities that currently disadvantage the poor, the study is descriptive in nature. Our purpose is to assess the differences in health outcomes, service utilization and the health care supply environment between high poverty concentration areas and less poor areas. Further research is needed that investigates the causal role of these factors in the relationship between poverty and health. One promising approach is to apply multi-level regression techniques along the lines suggested by Duncan et al. (1993, 1996, 1999), Stephens (1995) and Rice and Jones (1997) in order to shed further light on the relative effects of household-level vs. community-level factors on service utilization and health outcomes.
|
Biographies |
|---|
M Mahmud Khan, PhD, Tsai and Kung Professor of Health Systems Management, received his BA and MA in Economics from the University of Dhaka, Bangladesh, and completed two MA degrees and one PhD from Stanford University, USA. After his graduation from Stanford in 1987, he joined the Economics Department at the University of Washington, Seattle, as a visiting assistant professor. At Tulane, Dr Khan teaches econometrics for health services research, cost-benefit and cost-effectiveness analysis, Comparative Health Systems, and Health and Economic Development. He has also taught in various academic and research institutions in Taiwan, Thailand, Bangladesh and South Africa. He has published more than 40 research articles and received best teacher of the year award in 1994 and the teaching scholar award in 2003 from the School of Public Health of Tulane University. [Address: 1440 Canal Street, Suite 1900, New Orleans, LA 70112, USA]
David Hotchkiss, PhD, is an associate professor of health economics in Tulane University's Department of International Health and Development. He is currently serving as a senior advisor for the PHRplus Project, a 5-year USAID-funded project that aims to strengthen national health systems in low- and middle-income countries. Dr Hotchkiss has carried out many studies on issues that relate to demand for health care and health care financing. He received his MA in Demography from Georgetown University and his PhD in Economics from the University of North Carolina.
Andrés A Berruti, MA, is a professor of International Economics I and Health Economics at the Department of Economics of the Universidad Nacional de Cuyo in Argentina. He has also worked as a professor of Macroeconomics I and Microeconomics I and II at the Universidad de Congreso in Argentina, as well as an instructor at Tulane University's Department of Economics. He is currently involved in a series of research on health system reform with PHRplus and Tulane; and the Argentine pension system with Conferencia Interamericana de Seguridad Social in Mexico City. His research focuses on health and economic development, economic determinants of health status, pension reform in Argentina and dynamic economics in chaotic models. He holds degrees in economics from the Universidad Nacional de Cuyo and from Tulane University. [Address: Department of International Health and Development, Tulane University, 1440 Canal Street, Suite 2200, New Orleans, LA 70115, USA]
Paul Hutchinson, PhD, is an assistant professor in Tulane University's Department of IHD, specializing in health economics. Dr Hutchinson's research has focused on the determinants of health care utilization in numerous country contexts, including Uganda, South Africa, Bangladesh, Sri Lanka and China. He has headed studies examining equity in health, decentralization reforms, and the cost-effectiveness of health communications programmes. He is also the PI for a study of the incidence of vasectomies in the United States. Prior to joining the faculty at Tulane, Dr Hutchinson was a Research Associate at the Carolina Population Center at the University of North Carolina. Dr Hutchinson has also spent time working for the World Bank at their Resident Mission in Kampala, Uganda. Dr Hutchinson received his PhD in economics from the University of North Carolina at Chapel Hill. [Address: 1440 Canal Street, Suite 2200, New Orleans, LA 70112, USA]
|
Appendix |
|---|
Following Filmer and Pritchett (2001)
, the information on dwelling conditions (i.e. source of water, sanitation facilities, type of floor, availability of electricity) and asset ownership (i.e. radio, refrigerator, bicycle, car) were used to construct a wealth score by applying the principal component approach. The purpose of this technique is to discover or to reduce the dimensionality of the data set to identify new meaningful underlying variables. This procedure transforms a number of potentially correlated variables into a smaller number of uncorrelated variables called principal components. At the first stage, this approach identifies the weighted linear combination of living conditions and ownership variables to explain the largest amount of total variation in the data (first principal component). A second linear combination of the variables, perpendicular to the first one, tries to explain the maximum of the remaining variance, and so on. In order to calculate the principal components, we first compute the eigenvectors of the correlation matrix of the household ownership item variables. The first eigenvector is the vector corresponding to the largest characteristic root of the correlation matrix. There are as many eigenvectors as there are items, say n. Let 
nxn be the correlation matrix, then 
is called an eigenvector of corresponding to the eigenvalue 
if:
 |
where In is the identity matrix (nxn) and the elements of are the factor loadings. Thus the first principal component accounts for as much of the variability in the data as possible, and each succeeding component accounts for as much of the remaining variability as possible.
In this paper, we have used only the first principal component. The categorical responses were converted into dichotomous variables and standardized values were estimated for each of the dichotomous variables defined. The wealth score was defined as
 |
where, Ws is the wealth index for a household, vi is the loading for asset or dwelling category i, Si is the standardized value of that variable and n is the number of assets or dwelling conditions considered in calculating the wealth scores.
|
Acknowledgments |
|---|
This study was supported by the Partners for Health Reformplus (PHRplus) Project with funding from the US Agency for International Development under Contract No. HRN-C-00–95–00024. We greatly appreciate comments from Sara Bennett and two anonymous reviewers on previous versions of the paper.
|
References |
|---|
Akin JS, Guilkey DK, Hutchinson PL, McIntosh M. 1998. Price elasticities of demand for curative health care with control for sample selectivity on endogenous illness: an analysis for Sri Lanka. Health Economics 7: 509–31.[CrossRef][ISI][Medline]
Alderman H, Lavy V. 1996. Household responses to public health services: cost and quality tradeoffs. The World Bank Research Observer 11: 3–22.
Backlund E, Sorlie PD, Johnson NJ. 1996. The shape of the relationship between income and mortality in the United States: evidence from the National Longitudinal Mortality Study. Annals of Epidemiology 6: 12–20.[Medline]
Beegle K. 1995. The Quality and Availability of Family Planning Services and Contraceptive Use in Tanzania. Living Standards Measurement Study Working Paper No. 114. Washington DC: World Bank.
Bigman D, Dercon S, Guillaume D, Lambotte M. 2000. Community targeting for poverty reduction in Burkina Faso. The World Bank Economic Review 14: 167–94.
Booth B, Verma M. 1992. Decreased access to medical care for girls in Punjab, India: the roles of age, religion and distance. American Journal of Public Health 82: 1155–7.
Brabyn L, Skelly C. 2002. Modeling population access to New Zealand public hospitals. International Journal of Health Geographics 1: 3.
Buor D. 2003. Analyzing the promacy of distance in the utilization of health services in Ahafo-Ano South District, Ghana. International Journal of Health Planning and Management 18: 293–311.[CrossRef][ISI][Medline]
Bureau of Statistics [Tanzania] and Macro International Inc. 1997. Tanzania Demographic and Health Survey 1996. Calverton, MD: Bureau of Statistics and Macro International.
Chen S, Guilkey DK. 2002. Determinants of contraceptive method choice in rural Tanzania between 1991 and 1999. MEASURE Evaluation Working Paper WP-02–52. Chapel Hill, NC: University of North Carolina at Chapel Hill.
De Savigny D, Kasale H, Mbuya C et al. 2002. Tanzania Essential Health Interventions Project, TEHIP "Interventions" – an overview. Dar es Salaam: Tanzania Ministry of Health. TEHIP Discussion Paper No. 2, version 1.3, March.
Duncan C, Jones K, Moon G. 1993. Do places matter? A multilevel analysis of regional variations in health-related behaviour in Britain. Social Science and Medicine 37: 725–33.[CrossRef][ISI][Medline]
Duncan, C, Jones K, Moon G. 1996. Health-related behaviour in context: a multilevel modelling approach. Social Science and Medicine 42: 817–30.
Duncan C, Jones K, Moon G. 1999. Smoking and deprivation: are there neighbourhood effects? Social Science and Medicine 48: 497–505.
Filmer D, Pritchett L. 1999. The effect of household wealth on educational attainment: evidence from 35 countries. Population and Development Review 25: 85–120.[CrossRef][ISI]
Filmer D, Pritchett L. 2001. Estimating Wealth Effects without Expenditure Data – or Tears: An Application to Educational Enrollments in the States of India. Demography 38: 115–32.[ISI][Medline]
Gilson L, Kilima P, Tanner M. 1994. Local government decentralization and the health sector in Tanzania. Public Administration and Development 14: 451–477.
Gray A. 1993. World health and disease. Chapter 2. Buckingham: Open University Press.
Gwatkin D, Rutstein S, Johnson K, Suliman E, Wagstaff A. 2003. Socio-economic differences in health, nutrition and population. Second Edition, Volumes I and II. Washington DC: World Bank.
Hutchinson P. 2002. Decentralization in Tanzania: the view of District Health Management Teams. MEASURE Evaluation Working Paper WP-02–48. Chapel Hill, NC: University of North Carolina at Chapel Hill.
Marmot M. 1999. Introduction. In: Marmot M, Wilkinson RG (eds). Social determinants of health. Oxford: Oxford University Press.
Minot N, Baulch B. 2002. The spatial distribution of poverty in Vietnam and the potential for targeting. Washington DC: International Food Policy Research Institute (IFPRI), April, unpublished report.
Montgomery M, Gragnolati M, Burke K, Paredes E. 2000. Measuring living standards with proxy variables. Demography 37: 155–74.[ISI][Medline]
Muller I, Smith T, Mellor S, Rare L, Genton B. 1998. The effect of distance from home on attendance at a small rural health center in Papua New Guinea. International Journal of Epidemiology 27: 878–84.
Mwabu G, Ainsworth M, Nyamete A. 1994. Quality of medical care and choice of medical treatment in Kenya: an empirical analysis. Journal of Human Resources 28: 838–62.[CrossRef]
Nemet GF, Bailey AJ. 2000. Distance and health care utilization among the rural elderly. Social Science and Medicine 50: 1197–208.[CrossRef]
Noorali R, Luby S, Rahbar MH. 1999. Does use of a government service depend on distance from the health facility? Health Policy and Planning 14: 191–7.
Ofcansky F. 1997. Historical dictionary of Tanzania. Lanham, MD: Scarecrow Press.
Polednak AP. 1997. Segregation, poverty and mortality in urban African Americans. Oxford: Oxford University Press.
Rice N, Jones A. 1997. Multilevel models and health economics. Health Economics 6: 561–75.[CrossRef][ISI][Medline]
Sahn DE, Younger SD, Genicot G. 2003. The demand for health care services in rural Tanzania. Oxford Bulletin of Economics and Statistics 65: 241–59.[CrossRef][ISI]
Schellenberg JRMA, Nathan R, Abdulla S et al. 2002. Risk factors for child mortality in rural Tanzania. Tropical Medicine and International Health 7: 506–11.[CrossRef]
Schellenberg JRMA, Victora C, Mushi A et al. 2003. Inequities among the very poor: health care for children in rural southern Tanzania. The Lancet 361: 561–6.
Sen A. 2000. Development as freedom. New York, NY: Anchor Books.
Shaw M, Dorling D, Smith G. 1999. Poverty, social exclusion and minorities. In: Marmot M, Wilkinson RG (eds). Social determinants of health. Oxford: Oxford University Press.
Stephens C. 1995. The urban environment, poverty and health in developing countries. Health Policy