• Nenhum resultado encontrado

A Spatial-Temporal Analysis of Fertility Transition and the Health Care Delivery System in Brazil

N/A
N/A
Protected

Academic year: 2021

Share "A Spatial-Temporal Analysis of Fertility Transition and the Health Care Delivery System in Brazil"

Copied!
43
0
0

Texto

(1)

A Spatial-Temporal Analysis of Fertility Transition and the Health Care

Delivery System in Brazil

Suzana M. Cavenaghi Joseph E. Potter

INTRODUCTION

An extensive literature about the sources of the fertility decline dates from long ago, and yet new findings occasionally arise in light of the diverse experiences followed by countries or regions during their fertility transition. There have been noteworthy advances in terms of new theoretical and empirical developments, yet maybe the only general agreement that can be reached in this area is that different sources and pathways lead to fertility change in different settings. Within this context, this paper has the purpose of investigating an alternative source of fertility change, namely the expansion of the health care delivery system. In Brazil between 1964 and the present, changes in the administrative structure of the system and large increases in availability of health resources allowed large segments of the population increased accessibility to services that previously had been restricted to specific groups.

But, since neither the Brazilian government nor its public health system programs ever had the objective of reducing fertility, this additional explanation of fertility decline relies on the possibility that public policies produce unintended impacts in areas that the policies were not aimed at (Rosenzweig and Wolpin 1982, Faria 1989). Hence, this is the difference of this hypothesis with the more commonly investigated idea that family planning programs. In the absence of either an explicit population policy and in spite of economic recession during the 1980’s, Brazil nevertheless progressed through a rapid decline in fertility between 1970 and early 1990’s. The explanation for the decline put forward by Brazilian social scientists generally emphasize one or more of the following: 1) the implicit or tacit anti natalist actions of the government which allowed organizations such as BEMFAM ample room to promote contraceptive practice, 2) the pressure that economic retrenchment and monetization of the economy put on the households to limit their childbearing, and 3) unintended impact of governmental policies, particularly those that led to expansion of a) the social security system, b) consumer credit, c) the mass media, and d) the health care system.

(2)

While some effort had been placed on testing empirically the mass media component of the “unintended consequences” hypothesis (Potter 1998), little has been done to confirm the unintended impact of the expansion of health care. By using a series of cross-sectional data at the municipal level, created from the censuses and from health surveys carried out by the Brazilian Census Bureau, we are able to analyze this hypothesis in depth. Therefore, in this paper, we attempt to explore the impact of the growth in the availability of health resources on limiting fertility in an effort to enrich the discussion of the dynamics of the fertility decline in Brazil.

More precisely, the objective of this paper is to assess whether when holding constant other factors related to modernization and urbanization in Brazil there is a statistically significant association between the fertility rate and indicators of the expansion of the health care system, as first contended by Faria (1983). For this purpose, first we briefly report on the methodological approach used to measure levels of health care resources at the municipal. Secondly, by applying fixed-effect models we estimate fertility models, at the municipal level, that incorporates both spatial and temporal attributes capable of measuring the net effects of health resources on fertility rates.

Links between Health Care and Fertility Decline

Health policies in Brazil have resulted in an impressive growth of the health resources available to the population, but according to Faria, more important than the growth itself are the specific characteristics that the health system achieved during the expansion. First, a primary attribute of the expansion was to place emphasis on the development of medical services based on high technology, accentuating the role of the hospital in curative care to the detriment of preventive medical care. Second, the state spent a disproportionate amount of resources on the private medical sector rather than on the public sector. Third, the health system was made available to everybody, workers and non-workers, albeit not equitably. How could the increasing availability of health resources given its characteristics have influenced fertility regulation? Faria’s answer to this question is that the health care expansion “accelerated and reinforced the growing medicalization of Brazilian society. “ (1997/98, p. 194). The author admitted being surprised that besides the effects of decreasing mortality levels on fertility decline, very little attention has been given to the role of the medicalization in society in the analyses of increasing demand for fertility control. As Faria pointed out, when sexual behavior, reproductive behavior, and childbirth are medicalized, there is a natural shift from the religious, patriarchal, and marital authority over such matters to the medical authority. This shift implies a scientific conception of childbearing; not a

(3)

traditional conception based on God’s intention or a traditional patriarchal decision about family size. In Brazil, physicians constitute important authority figures, influencing people not only on problems related to ill-health, but also in matters related to reproduction and sexual behavior.

In addition to placing women and children under medical control and surveillance, the increasing medicalization of society also allowed for increasing highly technological surgery, accentuating the belief in allopathic medicine for health therapy, and intensifying specialization in the medical field (gynecologist, obstetricians, pediatricians, etc.) (Faria 1997/98). These are all mechanisms that may have contributed to the increasing demand for and fertility regulation. Specifically in the case of Brazil, the high prevalence of female sterilization as a contraceptive method and the medicalization of childbirth, with extremely high rates of caesarian section, are entirely compatible with these assertions.

Before making an empirical attempt to link changes in the health care system with fertility decline, one has to decide on an appropriate unit of analysis. While we argue that the municipality is the most suitable unit of analysis for the questions proposed in this investigation, we must state that the argument is not that the municipality in Brazil presents internal homogeneity in all aspects. Primarily, because some of them are quite large in population and territorial extension, and secondly because Brazilian social and economic classes are fragmented even within municipalities. However, the outcomes of the public policies implemented in Brazil are more likely to be more homogeneous within the municipality than any other section of society. A second important clarification is that our argument for an aggregate model does not discard the need for understanding the micro-analytic aspects of fertility decline. The study of structural changes and its influences on normative changes is simply a part of the puzzle that must be taken into account. Finally, we note that another limitation that results from specifying the model at the municipal level is that it prevent us from directly incorporating the most import proximate determinant of fertility, contraceptive use, into our model. Reliable representative data on the practice of contraceptive methods or even of caesarian section deliveries is not available for municipalities.

DATA AND METHODS

Census Data

The censuses of 1970, 1980 and 1991 are the sources for the fertility, socioeconomic and demographic data. The data refer to a 25 percent sample of the

(4)

population in the first two years. In 1991, the sample decreased to 10 and 20 percent. The larger sample fraction is applied in small municipalities with less than 15,000 inhabitants. The variables collected from the censuses are thus weighted aggregates at the municipal level of the individual or household records.

Health Services Data

The data source to assess health resources is the periodical survey of Medical-Sanitary Assistance-AMS (Assistência Médico-Sanitária). Because the first survey put available was the 1972, it was used in this analysis; thus the 1982 and 1992 are also utilized in order to have them evenly spaced and as close as possible to the census data.

The methodology applied in the data collection is very different from the methods utilized for the censuses and household surveys. For the AMS, IBGE agents send out the questionnaire, and the health institution is held responsible, by Federal enforcement, for filling out and returning the questionnaire to IBGE. In case of delay or non-return, the agents go back to the establishment to collect the information. The distribution of the questionnaires rely on a list of health facilities’ addresses (Cadastro

de Estabelecimentos de Saúde), which is continuously updated by the Bureau of

Census’ agents with the assistance of the Health Ministry. The process of data gathering takes about three months and is generally executed from April to June of each year.

The questionnaire is applied to all health-related establishments with the exception of doctor’s offices. Hence, it contains data for all health units operating in the country and these can be identified at the municipal level. The questionnaires for the three selected years vary, but they contain a basic structure that can be followed without compromising the comparability of the data. The information included in the survey is the type of services offered by each medical specialties, personnel employed by the units, the number beds, the type of public insurance accepted, the number of inpatients and medical consultations, and the number of births and deaths that occurred during the year.

From the raw AMS data, in which an observation is the health facility, we created aggregated files at the municipal level for each year, separating the types of facilities and creating some auxiliary variables. Four types of health facilities are considered for the study: hospital, health center, health post, and health clinic. The mixed units were classified jointly with hospitals. The difference among these two is that the latter offers medical sanitary services in addition to services provided by hospitals. The health centers and health posts are analyzed separately mainly because

(5)

the former has regular professional medical assistance1. The last category of health

facility selected was the health clinic. Note that for the analyses at the municipal level, we did not select all hospitals and health clinics. Left out were units specializing in only one of the following medical specialties: geriatrics, psychiatry, odontology, physical rehabilitation, neurology, and tuberculosis treatment.

Geographic Referred Data

We needed boundary files containing information about geographical limits of the regions, states, and municipalities for 1970, 1980, and 1991, for the entire country. The 1991 boundary file was obtained from IBGE (Brazilian Census Bureau), and the 1970 and 1980 boundary files was created2 from the 1991 boundary file. The number of

municipalities for 1970, 1980, and 1991 was 3,950, 3,991, and 4,491, respectively, which represents 41 and 500 new municipalities in the respective intercensal periods.

The longitudinal analysis required the creation of the boundary files that had homogeneous areas over the two decades. That is, one unique boundary file that would contain an aggregation of municipalities or minimum municipal comparable areas (MMCA). This new boundary file was created identifying all municipalities that had border changes between 1970 and 1991. The number of MMCA's for the homogenous boundary file is 3,829 areas for the entire country. All the analyses in this investigation are based on the MMCA units, which throughout the text are interchangeably referred to as municipalities or MMCA.

Variables

The dependent variable is the total fertility rate. It was calculated from the information on children born in the last year and information on the enumerated children from zero to 4 years old. We used different procedures to correct problems of accuracy and reliability. As these corrections involve utilizing several different procedures to deal with non-response to the question of children born in the last year, underreporting, and small sample variability3. The estimation of fertility rates at the

municipal level is fully discussed in Cavenaghi (1999).

1 The permanent medical assistance is defined as the presence of a physician in the facility at least once a

week.

2 The creation of the boundary for the 1970 and 1980 files has been carried out within a project developed

at the Population Research Center: The Social Impact of Television on Reproductive Behavior in Brazil, coordinated by Joseph Potter.

3 For correction due to small sample variability it was utilized an empirical Bayesian approach (Efron and

Morris 1973, Marshall 1991) by applying a shrinkage neighborhood estimator to smooth unreliable age specific fertility rates.

(6)

The creation of the health variables to classify the municipalities according to their level of health infrastructure also involved extensive exploratory data analysis. Three dimensions of health care resources were taken into account: the total availability of hospital resources, the per capita availability of hospital and ambulatory resources, and distance to municipalities with moderate availability of hospital resources. To measure hospital resources we constructed indices by using factor analysis. To measure

total hospital resources availability the basic information utilized was the number of

small, medium and large hospitals4, hospitals that accept public health insurance,

hospitals with high medical technology, hospitals specialized in OBGYN services, specialized in surgery, and when had a blood bank. To calculate the index of per capita

availability we utilized the number of small, medium and large hospitals, and hospitals

that accept health public insurance, physicians working in the hospitals, and the number of hospital beds. The indicator of ambulatory care was calculated as the sum of the number of health posts, health centers, and clinics per 20,000 inhabitants5. The distance

to municipalities with moderate resources of hospital availability was calculated using Euclidean distance. That is, a straight-line between the coordinates (Xi, Yi) of the

municipalities’ centroid. The distance is initially calculated in decimal degree units, and then transformed to kilometers. All the indices were transformed to the scale 0 to 1, and when needed, the signs were reversed in order to have a simpler interpretation, i.e., the higher the index the better the municipalities condition with regard to the index in question6.

In order to contemplate other dimensions of social change in the municipality, we analyzed three other groups of variables. The first group indicates the level of basic infrastructure in the municipality, which relies heavily upon the information on urban infrastructure. The second dimension indicates the level of income and education of the municipality. The last set of variables is related to women’s status in the municipality.

For the municipality’s basic infrastructure we utilized the proportions of households with piped water, households with good quality sewage, households with electricity, and households in urban areas. Because Brazilian municipalities vary in size and territorial extension, and there are large differences among the regions, any one of these variables in and of itself is not sufficient to indicate the real status of the municipality’s basic infrastructure. For this reason, and because these variables are highly correlated, we combined them in one factor, called the index of basic

4 We created variables to indicate hospitals’ size based on the number of beds and physicians.

5 The aggregation was necessary because in 1972 health centers and health posts cannot be identified

because of the format of the questionnaire.

(7)

infrastructure. To maintain comparability over time, the modeling of the basic infrastructure was obtained from the pooled sample of 1970, 1980, and 1991, using the principal factor method. In addition, this index was also transformed to the 0-1 scale to facilitate interpretation.

Assessing human resources indicators at the municipal level is not a simple enterprise, and several somewhat arbitrary decisions had to be taken. First, two dimensions of human resources were considered for the analysis: the level of education and income. Although these characteristics are highly correlated within each municipality, they measure different dimensions of development. With respect to income, we decided to utilize only level and not differentiate degrees of income distribution within the municipality. In the Brazilian censuses, the question of monthly earnings, from all income sources, is asked of persons over 15 years. For the 1980 and 1991 censuses, it is possible to calculate the average income per family. In 1970, however, the information on family relationship with the head of the family is not included in the questionnaire. Therefore, it is only possible to calculate average income per household, which sometimes has more than one family living in it. Because both the currency and the minimal wage in Brazil were subject to high inflation during the 1970-1991 period, an adjustment in the data must be performed. This has been done by the Human Development Project (PNUD 1998) at the municipal level for the 1970, 1980, and 1991 censuses. The indicator we selected was an index of income estimated from the information on family income per capita, which varies from zero to one, and was adjusted by the minimal wage of September of 1991 (the census reference date). In addition, the income index had an adjustment by a factor of decreasing returns, and the international GDP was used as the limit for calculation of the decreasing returns7. Since

the information provided refers to municipalities, we estimated the values of income index for the municipal minimum comparable areas using an average among the municipalities that composed the respective MMCAs, weighted by the population of each municipality.

In regard to levels of education, two dimensions are important to consider a summary indicator. One is the schooling attainment of the adult population, which to some extent reflect the history of the municipality’s schooling resources. The second aspect is the current level of schooling, that is, the capacity a municipality has to provide schooling for the younger generation. Some fair proxies for the first dimension are the proportions illiterate of the adult population, and the proportions with different grades of schooling attainment. To reflect the current level of education, some

(8)

surrogates such as proportions of the young population not enrolled in school and the average years of delay according to expected years of schooling can be used8. Because

of the positive high partial correlation among the variables, a summarized indicator using all the information contained in each one of them, as opposed to using them as separate variables, can enhance and ease the study. On these grounds, a factor analysis was carried out to create an index of the municipality’s level of education9. The index

had the signs inverted so that the higher the value, the better the municipalities’ education, and put on a scale between zero and one.

Since women’s characteristics are known to affect fertility decline, three indicators of women’s status are considered for the analysis. The first is women’s participation in the labor market. Thus, the indicator of female labor force participation is measured as the percentage of women aged 15-49 working or looking for a job in the labor market out of all women aged 15-49. The second indicator of women’s status is the average years of education, including only women of reproductive age. The last variable used to measure women’s status in the municipality is the percentage of women

head of the family (or head of household in 1970). In Brazil, as in most Latin America

countries, the head of family is understood to be the male of the household, most often the husband. Thus, this variable is reflective in large part of the percentage of women living with her children (or alone), often as a result of increasing rates of separation and divorce (Berquó, Oliveira, and Cavenaghi 1990). In part, it also captures widowhood, although the age interval considered (15-49) does not allow for a large proportion of widows in this category.

Methods

The analysis of panel data, or as in the present case, a quasi-panel created from a series of cross-sectional surveys, gives the opportunity to study the effects of the events across time. In this case, event refers to the average number of children per woman, during her reproductive life. Since the elapsed time from one observation to the next is large, 10 years approximately, the event can be understood as the passage from the initial level to much lower levels at the end of the observation period (Allison 1994), or the reversed case. In experimental studies, at the time of the event’s occurrence, the

8 Several other proxies for both dimensions of past and current schooling were investigated, but since all

of them are highly correlated with one another, We only included variables that would indicate level of education in the same direction.

9 The variables utilized in the factor analysis were: 1) proportion of illiterate population above 15 years

old; 2) proportion of the population aged 25 and above with less than four years of schooling; 3) proportion of the population aged 25 and above with less than eight years of schooling; 4) proportion of children aged 7 to 14 not enrolled in school; and 5) average years of delay in schooling for children 10-14 years old.

(9)

researcher has controlled all possible influences, or had used a case-control experiment, to understand the process under investigation. In studying social events, this approach is not feasible. As Deaton (1997) asserted, using panel data, each individual or unit can be a control of itself, and if the characteristics of these units are known and the event is observed over time, the investigation approximates, at some extent, the experimental situations. In the modeling of fertility rates, in essence, the analysis of panel data allow one to estimate on average, how much higher or lower the event of fertility was compared to the same unit over time, and compared to other observations across units.

When analyzing the effects of public policies on any events’ outcome, there is an advantage to using panel data created from cross-sections information. As Deaton (1997) contended, the effects of public policies or governmental programs on changes in behavior may take some time to be noticed, and in the case of panel data, the time elapsed between each panel is seldom enough to allow for significant changes. Hence, the creation of panels from repeated cross-sections allows one to estimate more reliable effects than examining the effect that event’s occurrence might have on the other10.

The idea underlying the method is rather simple as it stays within the realm of least-square estimation. However, when using data for the same individual or unit repeated several times in the same model, it brings methodological problems that must to be considered in the model’s specification (Hausman 1981, Johnston 1997). Suppose that yi is the distribution of fertility rates in i municipalities, and that t is the time each

set of yi is observed. The values of yi can be estimated assuming a linear relationship of

the group of xit variables that vary in each municipality (i) and over time (t), plus the

values of the zi variables that are constant over time but vary within municipalities, plus

an error term. It follows:

it i i it it

x

z

y

=

α

+

β

+

δ

+

ν

(1)

where i=1,…,N, and t=1,…,T. If the structure of the panel data can be ignored,

that is, vit is i.i.d. (0,σ2), the observations are serially uncorrelated and units and time

errors are homoscedastic, then an ordinary least-square (OLS) estimation of the parameter is consistent using the pooled data. However, if this assumption is not valid,

10 The use of longitudinal methods with data from panels or repeated cross-sections is commonly applied

in econometrics. To only cite a few, there is the work developed by Mundlak (1978, and 1981), Hausman and Taylor (1981), Judge et all (1980), (Deaton 1985), and Hsiao (1986). In sociology and demography, some researchers already have discovered he advantages of applying this methodology (Montgomery and Casterline 1993, Galloway Hammel, D. Lee 1994, Getler and Molyneaux 1994, Rosero-Bixby and Casterline, 1994, and Frankenberg 1995). In addition, other interesting investigations, not related to the topic of this study, have been published using longitudinal data and fixed effect models, such as the papers by Liker, Augustyniak, and Duncan (1985), England et al. (1988), Jasso (1985), Firebaugh and Beck (1994), and Lichter, McLaughlin, and Ribar (1997).

(10)

and in this case it is hard to obtain validity because units are repeated across time, thus the assumption of i.i.d.(0,σ2) is not valid. In this case, OLS estimation is not efficient, and one of the solutions could be to redefine the error term, considering that information for the unit (the municipality) in different points in time is more similar to each other than to other units. To implement this, one can assume that the error term has two components (νit =miit), where mi is the municipalities’ effects that vary across units

but are constant over time, and

ε

it is pure random noise that varies independently across

unit and time. Moreover, it must be assumed that

ε

it is uncorrelated with xit. In this

case, the model rewritten in equation (2) takes into account the panel structure and can be estimated without bias depending on assumptions about the distribution of mi.

it i i i it it

x

z

m

y

=

α

+

β

+

δ

+

+

ε

(2)

According to Johnston (1997), if mi in equation (2) is assumed uncorrelated with xit, then the model is said to be a random-effects model (or mixed). On the other hand, if mi is correlated with xit, then the model is a fixed-effects (or within) model. There are

two alternatives to estimate the parameters of the random effect models efficiently. One is using a generalized least square specifying the Gaussian family distribution of the dependent variable, and the link function as identity. Another way is to use a weighted average of the between and the within effects (Johnston 1997). The between and within effects are given respectively by:

i i i i i

x

z

m

y

=

α

+

β

+

δ

+

+

ε

(3)

)

(

)

(

)

(

y

it

y

i

=

x

it

x

i

β

+

ε

it

ε

i (4) where =

t it i y T y 1 , =

t it i x T x 1 , and =

t it i T ε ε 1 .

Note that the group of variable constant over time (zi) and the effect of units (mi)

were canceled out of the differencing, or deviation-form, equation (4). This is one of the main advantages of the estimation using fixed-effects models, that is, all time-invariant variables and any unobserved variable that is correlated with the time-varying variables (xi) is controlled by an artifact in the model’s specification. On the other hand,

as Hausman has argued (1981), this can be one of its biggest defects because the effects of time-invariant variables cannot be estimated with the fixed-effect models, and sometimes the researcher is testing hypotheses based on the coefficients of these variables. This would not be a disadvantage if one is deriving a model of change, and the levels of variables that do not change with time are not relevant for the study.

(11)

However, if this is not the case, relevant and well-selected instrumental variables of the time-invariant variables can be used in their place, in order to alleviate the problem (Hausman 1981).

The fixed effects models should be preferred over the random effects since the former model assumes that the correlation between the units effect and the time varying independent variables is not zero, as it does the latter model11. However, if there are

measurement errors among the covariates, then the fixed effect estimators are not efficient (Johnston 1997, Hsiao 1986, Deaton 1997). If this is the case, then the coefficients of the variables are not only capturing effects of the covariates on the dependent variables, but also is measuring effects of the measurement errors. Since most covariates are aggregated from the individual (or household) micro data file, we assume this problem is alleviated in the present study. The information on hospital and ambulatory care refers to the aggregation of health facilities, and although some variables may present measurement errors, mainly for the small municipalities, the errors should be minimized as the indexes are not based only in one variable, but on a set of variables.

For the estimation of the models, we used a procedure for panel data analysis, with specific options to fitting fixed-effect models developed by Stata Corporation (1996). For the fixed-effects models’ estimation, the software uses a variation in the specification of equation (4), which allows for an intercept in the model, as follows:

)

(

)

(

)

(

y

it

y

i

+

y

=

α

+

x

it

x

i

+

x

β

+

ε

it

ε

i

+

ε

(5) where y y nT i t it /

∑ ∑

= , and similarly for xi and εit. The estimation of the parameters

of

equation (5) is by way of OLS regression, and a restriction across the units (individual) effects, ∑mi=0, is imposed for the estimation of α and β, since α and mi are

not separately estimable. The parameter α in this case is a mean intercept of the individual effects (unobserved), and mi can be estimated as mˆi = yi −α −ˆ βxi.

The longitudinal model, using fixed-effect estimation, is assumed as the most suitable methodology available for inspecting the validity of the hypothesis that the expansion of the health care system had or not an effect in fertility decline. While the policies and programs set up for the health expansion in Brazil did not have the explicit intention of controlling population growth or making available family planning programs, as in other developing countries, the spread of availability of health facilities

11 Specification tests can be carried out to check the validity of assuming uncorrelated errors in the

(12)

could have had unanticipated effects on fertility. Furthermore, focusing on macro-structural changes, the dynamic model proposed is apt to contribute to the investigation of the relationship between fertility decline and other structural socioeconomic changes in Brazil at the municipal scale, which might provide additional information on fertility transition in Brazil.

FERTILITY LEVEL, TRENDS, AND DIFFERENTIALS

Fertility rates have declined considerably and rapidly in the recent decades, while regional differences in terms of level and time of the initiation of the decline vary greatly across the country. The estimates of fertility rates indicate that after a long period of nearly stable fertility for the entire country, in the 1960’s the average number of children per women started to decline. At the national level, the variation was about -0.4 children per woman between 1960 and 1970. From the following decade and on, fertility dropped significantly, from a rate of 5.8 births per women in 1970 to 4.3 in 1980 to 2.8 in 1991 (Merrick and Berquó 1983, and Carvalho 1997). Due to the large heterogeneity of socioeconomic and demographic components, the fertility transition did not occur evenly in time and space in the country. The Southeast region was the pioneer in having families who lowered their fertility. Within this region, the states of Rio de Janeiro and São Paulo, which have the two largest cities in the country, were the forerunners in the trend that started long before other states and even before 1960. Between 1960 and 1970, the Southeast region, including the states of Minas Gerais and Espirito Santo, including the two states previously mentioned, is held accountable for the decline in the national average of fertility rates mentioned above. During this period, the North plus Center-West regions exhibited a small decrease (from 7.32 to 7.08) and the northeastern states supposedly had a small increase (from 7.50 to 7.58).

The 1980’s marked a general decrease in fertility rates in all regions without exception. All regions presented a drop in their rates between 32 to 36 percent in these 10 years, except the Northeast region and the state of São Paulo where fertility declined by about 24 percent in each case. The state of Rio de Janeiro had the smallest fertility rate in 1980 (2.65) and the northeastern states had the highest rate (5.71). Interestingly, the relative difference between the lowest and highest rates increased during the 1970’s. By 1991, the pattern evened out among regions, with fertility at or close to the replacement level in the more developed regions, including also the Center-West Region. The North and Northeast regions still presented average rates around four children per woman; however, it is important to point out that these rates began their declines with total fertility rates of about eight children per woman in 1970.

(13)

ASSESSING THE EXPANSION OF HEALTH IN BRAZIL

The health care system in Brazil has undergone major changes since the beginning of the century, and specially more since the 1960's. As Medici and Oliveria (1992) asserted, these transformations occurred along with a firm role played by the State in regulating and financing of the whole system. Along with changes in the management of the health policies along the years, the quantity of physical resources for hospital and ambulatory services have undergone major changes since 1972. Table 1 presents the figures for selected indicators of health resources for the entire country.

Concerning hospitals (Table 1), the largest increases occurred in the first period (72-82). The annual growth in the number of hospitals during the 1970's was 3.6 percent. Although during the 1980's the growth was not as pronounced as in the earlier decade, it still presented an increase of 1.52 percent annually. The total number of existing hospitals in 1972 was 4,498, increasing to 6,391 in 1982, and reaching 7,430 hospitals ten years later. These hospitals present a large range of variation in size, taking into account variables such as the number of beds, doctors, and patients attended. Also, they present a wide variety of medical services offered to the population, from general hospitals (that include family medicine, surgery, obstetrics, and pediatrics), to hospitals specialized in high technology medicine, such as nuclear medicine, and heart transplants.

Table 1: Evolution in the number of health resources structure and geometric growth rates, Brazil, 1972, 1982, and 1992

Health resources 1972 1982 1992 Growth

72-82(%) Growth 82-92(%) Number of Facilities Hospitals 4,489 6,391 7,430 3.60 1.52 Health Centers 4,496* 3,408 14,763 8.42 15.79 Health Clinics 2,901 6,367 9,448 8.18 4.03 Health Posts --- 6,684 8,556 2.50 Hospital Beds 256,340 410,388 422,374 4.82 0.29 Physicians Hospitals 50,429 101,463 171,308 7.24 5.38 Health Centers 10,962 56,063 17.73 Health Clinics 20,632 41,049 51,729 7.12 2.34 *

Refers to health center and health posts.

A different picture is seen in the evolution of the total number of hospital beds. It shows an annual average growth in the first decade of 3.1 percent, but is virtually

(14)

constant in the next decade. It increased from 256,340 to 410,388 from 1972 to 1982 and to 422.370 beds in 1992. On the other hand, the number of physicians working in the hospitals12 had a major transformation, almost doubling in number in each decade.

It grew from 50,429 in 1972, to 101,388 in 1982, and to 171,308 physicians in 1992, which gives a ratio around 875 inhabitants to each physician working in a hospital in the latter year. Considering all doctors, including those in private practice, Brazil has a relatively good situation as a developing country with respect to the ratio of population to doctors. According to the Human Development Report (UNDP 1994), in 1990, Brazil13 presented a ratio of 670 population-to-doctor, and the industrial countries

ranged from 210 to 710.

In addition to the change in hospital resources in the recent past, the number of health facilities aimed at the delivery of primary and preventive health care also has gone through a significant transformation. In the beginning of the 1970s, the INAMPS14 slowly decreased the financial support given to the population via the

private medical system, mainly by paying low prices for medical procedures compared to the prevailing fees in the private medical practice. Conversely, federal funding for public health was aimed at improving the delivery of public health care by building new public facilities (Medici and Oliveira 1992). The data for 1972 do not allow for the separation between health posts and health centers15, but the overall rate shows that

there was a substantial increase during the 1970’s, with an annual growth rate of 8.4 percent. The next decade, the increase in the number of health centers was extraordinary. Starting the 1980-decade with 3,408 health centers, it increased to

12 The number of physicians working in the hospitals is actually the number of jobs for physicians in all

the hospitals. Hence, this increase is not necessarily the increase in the number of doctors because it is more likely that many doctors hold more than one job, commonly one in a private and one in a public hospital. In addition, a number of doctors only work at private clinics, and therefore they are not included in this figure.

13 In this report, Brazil is classified as medium human development, with the ranking of 63 among the

developing countries.

14 INAMPS (Instituto Nacional de Assistência e Previdência Social) is the National Institute of Social

Welfare System

15 Health centers are establishments intended to provide medical sanitary and ambulatory services such as

consultations, programs for the children and women, vaccination, distribution of medicines and food, and programs of health education. They have a permanent professional staff with general physicians or specialized physicians, as well as other medical-related professions such as, nurses and dentists. Health posts are less complex establishments than the health centers and aim to provide only medical sanitary services. They do not have permanent professional personnel. Above all, they do not have physicians or other medical professionals as part of the staff. The services offered are standardized and provided by minimally educated yet trained staff. Health posts regularly render vaccinations, health education with orientation in nutrition and basic sanitary procedures, and distribution of standard medication. Health clinics are the establishments that provide ambulatory services without hospitalization, and do not render sanitary services. The polyclinics can perform general medical services such as clinical examination, pediatrics, gynecology, and obstetrics, though the majority specializes in one or more medical fields. They have permanent personnel with doctors, nurses, dentists, and other medical assistants.

(15)

14,763 units by 1992, which meant an annual growth rate of 15.8 percent (Table 1). Concerning health posts, the growth was not as pronounced as the health centers (2.5 percent annually in the 1980s); however, this may reflect the definition of health center and health post. It is likely that with the increase in the availability of health manpower, a doctor could be hired as permanent staff in these facilities, which would ease the creation of health centers rather than having to construct health posts.

The expansion of the health care system during these years also implied greater coverage across space. Not only had the number of health facilities increased, but also the number of municipalities with health care services had grown substantially, although there was an increase in the number of new municipalities during the years. As for hospital infrastructure, the relative proportion of municipalities with at least one hospital rose from 48 percent to 73 percent between 1982 and 1992. For the health center and health posts, the municipal coverage had been greater than the hospital coverage, but still it experienced substantial growth in these two decades. In 1972, 74 percent of the municipalities had at least one health center or health post, and by 1992, it reached 97 percent16.

The figures for the entire country conceal significant changes in the growth of health resources among municipalities. In order to provide a brief picture of the distribution of health resources in municipalities for the entire country in 1972 and 199217, following we present maps of the distribution of the total number of hospitals,

and the distribution of ambulatory facilities by 20,000 inhabitants (Illustrations 1, 2, 3, and 4). Although the pictures by themselves give a good sense of the changes in health resources, some numbers help in appreciating the changes occurred in Brazil from 1972 to 1992. We can observe in illustration 1 that the entire area in the west of Brazil and the entire interior of the Northeast presented a very precarious condition in terms of availability of hospitals in 1972. The majority of municipalities did not have any hospital, and with the exception of the larger cities, no municipalities had more than a few hospitals. The majority of hospitals were concentrated in the Southern region, mainly in the city of Rio de Janeiro and São Paulo, the two largest cities in the country. Comparing illustrations 1 and 2, we can observe a spatial spread of hospitals towards the west. In addition, the Northeast region experienced a notable increase in the number of hospitals. Besides the expansion in the number of hospitals in the state capitals and large cities, the decrease in shaded areas shows that many more municipalities had at least one hospital by 1992 than in 1972.

16 It is important to remark that the majority of health centers are located in wealthier municipalities in

the more developed states, and health posts in the less developed and poor municipalities.

(16)

Among the 4,491 municipalities in 1992, 3,302 of them had at least one hospital and they presented a wide range of variation in the total number of hospitals, though the majority had only a few hospitals. In half of the municipalities, there was only one hospital, while 75 percent had up to two hospitals. Also, there were two extreme outliers with 167 and 229 hospitals in São Paulo, and Rio de Janeiro, respectively. The municipalities that had large number of hospitals were the larger cities, frequently the capitals of the states, and most of them are located on, or close to, the Atlantic coast.

Both illustrations, but more the one for 1992, show a consistent pattern surrounding the large cities, where the majority of neighboring municipalities also has many hospitals. It shows a pattern of a decreasing the number of hospitals as one goes from the central municipality to the peripheries in concentric circles. However, the same picture is not seen in the municipalities with large territorial extension, most of which are in the Legal Amazon area18. In this area, the municipalities with the state

capital exhibit a large number of hospitals while the surrounding municipalities have very few or no hospital at all.

Another pattern that calls one’s attention is the seemingly equitable distribution of hospitals across all municipalities in some states. Two different processes might be generating this pattern. One can be the result of the large distances the population would have to travel to get health care. In the states with large municipalities, it is more likely that each city will have at least one hospital. Indeed, in the Northern States the majority of municipalities have at least one hospital, a fact that suggests that these municipalities are less likely to depend on their neighbors for health care. The second process is clearly not related to travel distance, and could indicate either better policy management of the distribution of health care or a more even distribution of wealth in general across municipalities within states. This pattern appears in all three states in the South region, also in Rio de Janeiro, Espirito Santo, Maranhão, Ceará, and Pernambuco. By no means do we want to imply with this examination that the availability per se of hospitals in the municipality would indicate better access in terms of quantity or quality of services. The discussion thus far serves only to suggest some processes that might play important roles in determining patterns of hospital distribution across the territory.

Some very small municipalities, in terms of population and territorial extension, do not have hospitals. Commonly, these municipalities are surrounded by or are neighbors of a larger municipality that has hospitals and will serve as the source of health resources for these small municipalities. This phenomenon can be easily identified among municipalities in the states of São Paulo, Paraná, Santa Catarina, and

18 The Legal Amazon Includes all states of the North region, and parts of the states of Maranhão,

(17)

Mato Grosso do Sul. Since the cities in these small municipalities are very close, the travel distance to a city that offers hospital services is short. Thus, it is economically worthwhile to transport patients to other cities instead of constructing and maintaining a hospital within the municipality. Therefore, rather than investing in the construction of a hospital, it has been a common procedure adopted by local governments to buy special ambulances to drive patients to cities that provide hospital care. For this reason, the distance to a municipality with health resources is an additional component to consider when analyzing level of health resources.

From the index of total availability and per capita availability of hospital resources constructed for the study, we calculated the distance to closest municipality with different level of health resources as indicated by the indices. Illustrations 3 and 4 present the maps indicating the distances to a municipality with high levels of total availability of hospital resources. The maps show that whereas there are some municipalities in the surroundings of a municipality with good hospital resources, the most municipalities are located far away from big centers with good access to health services. For instance, the average distance to a municipality with, at least, a high level of hospital resources, measured by the total availability, is around 114, 91, and 82 kilometers for 1972, 1982, and 1992, respectively. For utilization of a highly specialized health care, this may not be a very large distance to travel, but for utilization of obstetrics and gynecology services, or other general medical services, lower distances are desirable (Williams et al.1983).

In addition, the maps show more clearly the advancement in the availability of hospital resources over the years. This advancement occurs within all regions, but expansion is most evident on the west side of the country. Another effect that can be noticed in these maps is the benefit that the municipalities, mainly in the southeast and south regions, have due to their smaller territorial extension, compared to municipalities in the western regions. Therefore, besides other advantages of living in more developed zones, due to a spatial artifact the population in small municipalities in these areas have better access to health resources compared to the population living in municipalities in the less developed parts of the country.

With respect to ambulatory health care, Illustrations 5 and 6 plot the spatial distributions of the number of ambulatory facilities by 20,000 population in municipalities for 1972 and 1992, respectively. As we have observed, the growth in these types of facilities is remarkable. In 1972, Illustration 6 shows that the majority of municipalities (87 percent) had less than one ambulatory for each 5,000 population (includes all categories except four or more ambulatory for each 20,000 population). In the following decades, there was an important expansion of these facilities, and by 1992, very few municipalities had no ambulatory care installed. Municipalities that still

(18)

had few ambulatory facilities per population were those mainly in the states of Maranhão and west of Amazonas. Finally, it is important to mention that these increases occurred in the ratio to population in spite of the population growth that occurred in the period.

(19)

Illustration 2: Spatial distribution of the number of hospitals in municipalities, Brazil, 1972.

Illustration 1: Spatial distribution of the number of hospitals in municipalities, Brazil, 1992.

(20)

Illustration 3: Distance to the closest municipality with high health infrastructure availability, Brazil, 1972

Illustration 4: Distance to the closest municipality with high health infrastructure availability, Brazil, 1992

(21)

Illustration 5: Number of ambulatory facilities by 20,000 population in municipalities, Brazil, 1972

Illustration 6: Number of ambulatory facilities by 20,000 population in municipalities, Brazil, 1992

(22)

FIXED-EFFECTS MODELING RESULTS

Regional differences in Brazil are enormous. Some states, mainly the southeast part of the country are economically affluent and other states, often in the North and Northeast regions, experience very high degree of poverty. Consequently, the dynamics of socioeconomic and demographic processes are different across the country, as we have seen for the expansion in health care resources. For this reason, an initial methodological approach for the analysis of fertility is to divide the country in regions. One possible approach is to utilize the geo-political subdivision of great regions, which presents more internal similarities among them.

The three tables in the appendix (A1, A2, and A3) show the non-weighted average of

selected variables in municipalities19, according to the Brazilian geographic regions,

respectively, for each cross-sectional data set. The first group has variables related to health resources indicators, and the second group has the municipalities’ infrastructure and human resources and the set of women’s status indicators. These tables show, as already asserted, that there are large differentials by region with respect to all variables. From 1970 to 1991, fertility rates declined approximately 40 percent in all municipalities in Brazil, and regional declines varied from 30 to about 47 percent. The average levels of fertility spans from 6.55 children per woman in 1970 in the North region to 2.53 children per woman in 1991 in the Southeast region.

With respect to health indicators, the average per capita ambulatory resources was around 2 facilities for each 20,000 people in 1970, and these figures increased to 6.37 in 1991. The North region presents the lowest ambulatory per capita in 1970 among all regions, but since there was a large increase in health posts in almost all municipalities during these two decades this region reaches levels of ambulatory resources comparable to the more developed ones by 1991. Nonetheless, as we have mentioned, the growth in ambulatory for the more developed regions, as in contrast to the less developed regions, occurred for health centers and not for health posts as in the case of the North region.

For the indicators of hospital resources, the interpretation is not as straightforward as that for ambulatory resources. These indexes vary from zero to one, and each of them represents a

different aspect of health care resources20: total availability, high availability per capita, low

availability per capita (third, fourth, and fifty rows in Tables A1, A2, and A3). The total availability (row three) increases from an average level of 0.05 in 1970 to an average of 0.10 in

19 These averages did not take population at risk into account. Each municipality, regardless of size, counts as one

observation for the average’s estimations.

20 The details regarding the variables used in the construction of the indexes and the methodology applied are

(23)

23

23

1991. As for the per capita indexes, although there is an increase in population size in almost all municipalities, the average per capita availability of hospital resources increases across time, as it can be observed in the referred tables. It is important to mention that these averages are estimated among all 3,829 municipalities, the majority of which had no hospital resources (index was equal to zero in 2,026 municipalities) in the first period, though this scenario had changed substantially by 1992 (951 municipalities with zero hospital resources).

The measurements of distance to the closest municipality with some level of hospital resources further expose the changes in health resources. This can be better apprehended in light of the spatial conformation of the Brazilian municipalities. The most prevalent conformation is a large or medium municipality surrounded by several small municipalities. Thus, the change in the level of hospital resources for a large municipality implies a decreasing distance to the closest municipality with some level of hospital resources of many municipalities. Therefore, as observed in Tables A1, A2, and A3, change on the average level of distance to the closest municipality with health resource, general or per capita experienced major changes during the period analyzed. Comprehension of the role of distance in the study of health care availability is essential to the discussion, because most municipalities will have very low or no hospital availability at all, due in part to population size constraints, and therefore will depend on the hospital resources of their neighboring municipalities.

The series of aggregated indicators of municipalities’ infrastructure and human resources and women’s status (bottom of Tables A1, A2, and A3) show that the Brazilian social and economic situation in 1970 was unsatisfactory. Brazil had low average levels of income, basic infrastructure, education, women participating in the labor force, and average years of education. In addition, the regional differences were immense. Due to the heavy investments in public policies in the 1970-decade, improvements in all dimensions and across regions were notable, but regional differences persisted or even grew during the 1970’s. In the 1980-decade, on the other hand, social improvements did not keep up with the progress that occurred in the previous decade. In addition, average levels of family income decreased in municipalities. However,

basic infrastructure, education, and women’s participation in the labor force21 had important

average increases during the 1980’s. During this period, the gap in the regional differences for these variables somewhat decreased (Table A3).

In addition to regional differences, the analysis also has to consider the size of the municipality regarding timing and pace of social and demographic changes. Because the

21 The insertion of women in the labor force can be interpreted from two different points of view. It can mean that

in some cases, the woman was “coerced” to work in the labor market to maintain family income level, or in other cases, it could be due to women’s free choice. Nevertheless, in either case it indicates, or could lead to, an improvement in women’s social status.

(24)

24

municipalities in Brazil vary greatly in size, the internal dynamics of the small cities compared to those of the large cities also diverge. This mainly occurs when the large city is a state capital. Taking the example of health policies, the government in small cities will not invest in the construction of hospitals with high or even medium technological services, especially if the municipality is close to another one that already has invested in such technology. On the other hand, large municipalities do not have to invest as much in ambulatory services because many hospitals already have ambulatory functions as part of services offered to the population. Nevertheless, certain increases should be expected since the poorest population, mostly living on the fringe of large cities, will need ambulatory services if the hospital’s access is restricted.

The number of inhabitants utilized in the classification of municipalities by size is arbitrary, but it tries to capture diverse settings among municipalities and uses the smallest number of categories possible. The population in the municipalities in Brazil varies from about 700 up to more than 9,500,000 inhabitants. From the political-administrative point of view, these municipalities have equal rights before the central government. Thus, to contemplate part

of this diversity, using the enumerated population in 197022, initially we separated municipalities

with 15,000 or less inhabitants, from the medium municipalities (15,000-80,000), and from the large municipalities, 80,000 inhabitants or more. Nonetheless, because the small and medium municipalities experienced similar changes for the majority of variables, modeling on fertility rates for these municipalities were very similar, thus they were fitted together, representing municipalities with less than 80,000 inhabitants. In addition, because in some regions the largest decreases in fertility occurred during the second decade analyzed, all models share a second common characteristic: each includes the 1980-91 period in addition to the models using the entire period (1970-80-91). During the models’ specification, all variables and interactions were accounted for, but we present only the results for the most parsimonious model, which do not include the variables that were not statistically significant to .10 level (two-tail test). Consequently, each model presented in the following tables may have different sets of covariates.

Furthermore, because distance to a municipality with a good level of hospital resources could be measuring, in part, the distance to a municipality that has a good level of socioeconomic indicators in general, a new measurement of distance was included as a controlling variable. This variable appears in all models, although it may or may not present statistical significance for the specific settings. To calculate this distance, we used the index of life conditions, available in the same data set cited earlier (Fundação João Pinheiro 1996), which is derived from a

22 The individuals counted in the population are those living in permanent households, as defined by the Census

(25)

25

25

combination of several indicators taking into account aspects such as, longevity, education, income, housing, and children’s working and educational status. Using this index, we classified all municipalities into one of five groups, using the Jenk’s formula (ESRI 1995). The distance selected for the modeling was set to the closest municipality with very high levels of life conditions.

As discussed in the methods section, the longitudinal aspect of the data set is used in the modeling. In particular, random-effects models do not provide consistent results because covariates, or at least one of them, are correlated with the municipalities’ error term, as checked with Hausman’s test (1978), which tests the hypothesis that the random-effect is not appropriate in view of a correct model specification. On the premise that the model is correctly specified, the fixed-effect model is applied because it assumes that there is a different effect for each municipality --fixed for each municipality, but it is constant over time, and is an estimate of the effect for unobserved municipalities’ heterogeneity. In addition, this method of estimation assures that time-invariant covariates will not affect the coefficients in the model if the specification is correct. In the following, we present the analysis according to four settings separately: large municipalities, Northeast, Southeast and Center-West regions.

Large Municipalities

All variables included in the modeling are, directly or indirectly, associated with levels of improvement (or worsening) in all socioeconomic dimensions, and thus are likely to be correlated with each other to some extent. Likewise, fertility rates are negatively associated with socioeconomic conditions. In this situation, the modeling reflects the relative contribution that each variable has in changing fertility in the analyzed setting, when all other variables are controlled. The group analyzed here is the large municipalities, which were defined as those with at least 80,000 inhabitants in 1970. This group presents internal diversity, as it includes very large municipalities as the case of São Paulo, which has almost 10 million inhabitants. All state capitals are included in this group except for the capital of Acre and Tocantins, which did not have 80,000 inhabitants in 1970.

In these large municipalities, the model of fertility rates indicates that none of the health variables are statistically significant in their association with changes in fertility, considering the

three years period or when considering the 1980-91 period23, as shown in Table 2. The

significant variables are level of municipal education, basic infrastructure, proportion of female head of the household. As expected, each of these variables is negatively associated with

23 A model for the 1970-80 period was also estimated, but it did not show significant differences compared to the

(26)

26

fertility. The larger relative contribution to the decline in fertility in large cities, during the period analyzed, can be attributed to the level of education in the municipality, followed by the level of infrastructure. These two indicators have a significant interaction term, and it indicates that part of the effect of the municipalities’ level of education on fertility is due to high levels of basic infrastructure and vice-versa. Taking into account the interaction term between these two variables, education by itself has a high effect on fertility decline, but the effect of infrastructure is reduced by a significant amount.

The model considering only the 1980-91 period in large municipalities, in Table 2, shows that the level of income was significantly associated with fertility during this period, but had a positive effect, that is higher levels of income resulted in higher levels of fertility. This result is the opposite of that which was expected, since higher income is most often associated with low fertility. During this period, however, income had declined. If in the cross-sectional data, fertility and income are negatively correlated, and income had decreased considerably during this period, the modeling shows that if other variables were not accounted for, the decrease in income would have increased fertility rates in large municipalities from 1980 to 1991.

Table 2: Fixed-effects coefficients and t-statistic on fertility rates. Large municipalities, Brazil, 1970, 1980, 1991.

1 The units of distance are in 100 kilometers. All other variables have values between zero and one, except for

women’s years of education.

Critical t-value at 90% level of significance is ±1.64, at 95% is ± 1.96, and at 99% is ± 2.58 for d.f. →∝

With respect to health indicators, although there was an increase in the hospital resources and ambulatory facilities, the model indicates that these changes did not have a significant

Municipalities with less than 80,0000 inhabitants

Covariates 1970-80-91 1980-91 period

Coef. t Coef. t

Index level of income -- -- 1.69 3.07

Index Municipal level of education -8.42 -8.04 -10.59 -5.88

Index level of basic infrastructure -4.51 -10.72 -6.03 -7.50

Prop. female head of household -2.72 -4.59 -2.15 -2.78

Dist. to closest Mun. w/ very high ICV1 0.04 2.03 0.00 0.00

Mun. infrastructure Vs mun. education 4.37 6.76 6.82 4.43

Constant 10.62 31.99 10.46 10.61 Obs 390 260 N 130 130 T 3 2 R2-Within 0.92 0.89 corr(mi,Xß ) -0.49 -0.14

(27)

27

27

impact in changes in the fertility rate from 1970 to 1991, or from 1980 to 1991. The distance to the closest municipality with good hospital availability was not expected to have a significant effect because the large municipalities themselves were likely to have medium or better levels of hospital resources. The index of hospital resource availability was expected to have a negative association with fertility, mainly because access to sterilization and surgical birth deliveries would be facilitated in the large cities. However, these cities already had very good health resources availability by the beginning of the observation period, and the model cannot indicate if there was a significant effect when the availability of health resources was at lower levels. Therefore, the conclusion from this finding is that the changes in fertility, occurring from 1970 to 1991, or from 1980 to 1991, in cities with more than 80,000 inhabitants, was not significantly associated with the increases in health resources in these same periods, according to the indicators utilized.

Northeast Region

The models for the Northeast region, and the two others that follow have data for municipalities with 80,000 or less inhabitants in 1970. In this region, fertility rates appear to be negatively associated with ambulatory per capita and positively associated with distance to the closest municipality with at least medium level of hospital availability, at the 99 percent level of confidence. Table 3 shows these results in the modeling of fertility, for the 1970-91 and for the 1980-91 periods. The effects of ambulatory per capita and distance, though significant, are small compared to the effects of other variables, but they show an interesting variation between the 1970-91 and 1980-91 periods. The effect of ambulatory is lower when the entire period is considered than when it is only the last 11 years are used in the model. The effect of distance to hospital availability, on the contrary, is more significant and larger in the 1980-91 than for the 21-year model. These results appear to reflect the ongoing transition of fertility in the region. As observed earlier, in the Northeast the declines in fertility were small and the majority of municipalities had very high levels of fertility in the period of 1970-80, exactly the same period of higher increases in ambulatory care facilities. In the next period, 1980 to 1991, fertility rates in the Northeast manifest an important change and the effect of distance to a municipality with medium or higher level of hospital resources availability shows a significant effect on changes in fertility rates. Although the interpretation of the linear regression is straightforward, the effects of the health variables that are statistically significant for the 1970-91 and 1980-91 models for the Northeast region can be better observed in Figure 1.

(28)

28

Table 3: Fixed-effects coefficients and t-statistic on fertility rates in municipalities in the Northeast region. Brazil, 1970, 1980, 1991

1

The units of distance are in 100 kilometers. All other variables have values between zero and one, except for women’s years of education. Critical t-value at 90% level of significance is ±1.64, at 95% is ± 1.96, and at 99% is ± 2.58 for d.f. →∝

Figure 1: Effect of health variables on fertility decline. Northeast Region, Brazil, 1970, 1980, and 1991 M u n ic ip a litie s w ith le s s th a n 8 0 ,0 0 0 0 in h a b ita n ts

C o v a ria te s 1 9 7 0 -8 0 -9 1 1 9 8 0 -9 1 p e rio d C o e f. t C o e f. t A m b u la to ry (2 0 ,0 0 0 p o p u la tio n ) -0 .0 2 -4 .8 7 -0 .0 1 -2 .3 4 D is t. c lo s e s t m e d . le v e l o f h o s p . a v a ila b ility1 0 .0 8 2 .2 5 0 .2 6 5 .7 8 In d e x le v e l o f in c o m e 2 .6 2 6 .9 3 2 .3 8 4 .6 6 In d e x m u n ic ip a l le v e l o f e d u c a tio n -8 .5 0 -1 1 .7 3 -6 .2 4 -8 .0 4 In d e x le v e l o f b a s ic in fra stru c tu re -4 .7 9 -1 8 .0 3 -3 .8 3 -1 5 .3 2 P ro p . fe m a le la b o r fo rc e -- -- -0 .5 9 -2 .4 9 P ro p . fe m a le h e a d o f h o u s e h o ld -1 .8 9 -4 .9 2 -0 .9 5 -2 .0 5 A v e ra g e fe m a le e d u c a tio n (y e a rs) -0 .1 1 -5 .5 0 -- --D is t. to c lo se s t M u n . w / v e ry h ig h IC V1 0 .0 6 7 .2 4 0 .3 8 4 .5 4 M u n . e d u c a tio n V s in c o m e -1 .7 4 -5 .6 6 -0 .9 7 -2 .3 2

M u n . in fra s tru c tu re V s m u n . e d u c a tio n 3 .9 9 6 .1 1 --

--_ c o n s 8 .1 2 6 3 .2 1 6 .4 9 1 9 .7 5 O b s 2 7 8 1 1 8 5 4 n 9 2 7 9 2 7 T 3 2 R2-W ith in 0 .8 7 0 .8 6 c o rr(mi,X ß ) -0 .4 0 -0 .7 0

Effect of ambulatory care. Model 1970-91, Northeast 0 1 2 3 4 5 6 7 8 0 2 4 6 8 10 12 14 16 18 20 Ambulatory per 20,000 Fe rt ilit y r a te 1970 1980 1991

Effect of distance to availability. Model 1970-91, Northeast 0 1 2 3 4 5 6 7 8 0 100 200 300 400 500 Kilometers F ert il it y ra te 1970 1980 1991 E ffe c t o f a m b u la to ry c a re . M o d e l 1 9 8 0 -9 1 , N o rth e a st 0 1 2 3 4 5 6 7 8 0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 A m bulato ry pe r 2 0 ,0 0 0 F e rt ili ty r a te 1 9 8 0 1 9 9 1

E ffec t o f d ista n c e to a v a ila b ility . M o d e l 1 9 8 0 -9 1 , N o rth e a st 0 1 2 3 4 5 6 7 8 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 k ilo m e te rs F e rt il it y ra te 1 9 8 0 1 9 9 1

Referências

Documentos relacionados

Para tanto foi realizada uma pesquisa descritiva, utilizando-se da pesquisa documental, na Secretaria Nacional de Esporte de Alto Rendimento do Ministério do Esporte

As doenças mais frequentes com localização na cabeça dos coelhos são a doença dentária adquirida, os abcessos dentários mandibulares ou maxilares, a otite interna e o empiema da

The fourth generation of sinkholes is connected with the older Đulin ponor-Medvedica cave system and collects the water which appears deeper in the cave as permanent

The irregular pisoids from Perlova cave have rough outer surface, no nuclei, subtle and irregular lamination and no corrosional surfaces in their internal structure (Figure

The change of stage of demographic transition, when Brazil advanced to the fourth stage, with the decrease in fertility and mortality, led to changes of the

Uma das explicações para a não utilização dos recursos do Fundo foi devido ao processo de reconstrução dos países europeus, e devido ao grande fluxo de capitais no

Neste trabalho o objetivo central foi a ampliação e adequação do procedimento e programa computacional baseado no programa comercial MSC.PATRAN, para a geração automática de modelos

Ousasse apontar algumas hipóteses para a solução desse problema público a partir do exposto dos autores usados como base para fundamentação teórica, da análise dos dados