Frankenberg2 Doc - CiteSeerX

0 downloads 0 Views 122KB Size Report
Jakarta: MOH. ———. 1994. “Pedoman pembinaan teknis bidan di desa.” Report, ... “Menjaga mutu pelayanaan bidan desa.” Popula- tion Studies Center, Gajah ...
WOMEN'S HEALTH AND PREGNANCY OUTCOMES

253

WOMEN’S HEALTH AND PREGNANCY OUTCOMES: DO SERVICES MAKE A DIFFERENCE?* ELIZABETH FRANKENBERG AND DUNCAN THOMAS

We use data from the Indonesia Family Life Survey to investigate the impact of a major expansion in access to midwifery services on health and pregnancy outcomes for women of reproductive age. Between 1990 and 1998 Indonesia trained some 50,000 midwives. Between 1993 and 1997 these midwives tended to be placed in relatively poor communities that were relatively distant from health centers. We show that additions of village midwives to communities between 1993 and 1997 are associated with a significant increase in body mass index in 1997 relative to 1993 for women of reproductive age, but not for men or for older women. The presence of a village midwife during pregnancy is also associated with increased birthweight. Both results are robust to the inclusion of community-level fixed effects, a strategy that addresses many of the concerns about biases because of nonrandom program placement.

ecline in mortality is among the most fundamental deD mographic changes experienced by developing countries over the past half-century. Today, individuals are leading longer and healthier lives than did their parents and grandparents. In part these changes reflect investments in human resources by both individuals and governments. In virtually every developing country, governments have built, stocked, and staffed schools, health facilities, and family planning clinics, albeit with varying degrees of success. Although clinical studies have demonstrated that some health interventions in fact improve health, researchers have long debated about the contribution of public health investments to health improvements and mortality decline. Most macro-level studies conclude that the effect of public spending on health is small (Filmer and Pritchett 1999; Musgrove 1996). At the micro level, some studies have concluded that investments in providing public health services have a positive causal effect on health outcomes (Caldwell 1986; Jamison et al. 1993). The majority of studies, however, indicate that increases in public spending have little or no impact on health; in some cases, public-sector investments are even associated with poorer health outcomes. (For a discussion, see Strauss and Thomas 1995.) * Elizabeth Frankenberg, RAND, 1700 Main Street, Santa Monica, CA 90407; E-mail: [email protected]. Duncan Thomas, RAND and University of California at Los Angeles; E-mail: [email protected]. This work was supported by NICHD grants P50HD12639, R29HD32627, and P01HD28372, by NIA grant P30AG12815, and by the POLICY Project. We gratefully acknowledge the comments of Bondan Sikoki, Wayan Suriastini, the editors, two anonymous reviewers, and participants at seminars at the University of California at Los Angeles, the Gadjah Mada University, the University of Maryland, the University of Michigan, the University of Pennsylvania, and the University of Washington.

Demography, Volume 38-Number 2, May 2001: 253–265

At least two critical problems have plagued this literature. The first, and perhaps the more difficult to address, is that public health investments are not likely to be located at random with respect to health outcomes. For example, if programs are carefully targeted they will be placed where health outcomes are poor and/or utilization of services is low. If all program placement decisions are based on observable characteristics that are controlled in an evaluation of the program, such targeting poses no conceptual difficulty. Yet insofar as program placement is associated with characteristics that are not observed, failure to take account of nonrandom placement will generally lead to biased estimates of the impact of the investment (Angeles, Guilkey, and Mroz 1998). Rosenzweig and Wolpin (1986), for example, show that in a cross-section regression, children’s nutritional status is negatively associated with exposure to public health programs in Laguna, The Philippines. In contrast, these authors find a positive and significant effect when they examine how changes in nutritional status respond to changes in exposure to public health programs. They attribute the negative correlation in the cross-section estimates to the nonrandom placement of programs. A second major stumbling block in this literature is the lack of adequate data, on several dimensions. Measurement of health investments is not straightforward; this surely contributes to the weakness of evidence in the macro literature. In the micro literature, the shortcomings of community-level data on the accessibility and quality of health services that can be linked to individual-level information are well known (Akin, Guilkey, and Denton 1995; Pullum 1991; Thomas and Maluccio 1996), although recent advances in geographical information systems have facilitated the combination of administrative data with sociodemographic surveys (Entwisle et al. 1997). Detailed community-level data linked to individual-level data are not always sufficient: the application of methods that control community- or individual-specific unobservables requires repeated observations on health outcomes, and very few longitudinal surveys contain that information on respondents as well as on the health services and other services to which they have access. We use data from a new, extremely rich longitudinal survey from Indonesia to evaluate whether government efforts to provide health care have an impact on the populations targeted by the programs. Specifically, we consider the Village Midwife program, which was initiated in the 1990s and is estimated to have posted some 50,000 midwives 253

254

throughout the country (Gani 1996; Kosen and Gunawan 1996; Sweet, Tickner, and Maclean 1995). Our goal is to provide evidence on the effectiveness of this large and important community-based public health service intervention that is targeted explicitly to reproductive-age women in underserved communities. Our results are of general interest because these types of programs have been implemented in many developing countries. To measure the effect on health status of the introduction of a new health worker in a community, we draw on the “quasi-experiment” that occurred in Indonesia by comparing changes in health status in communities that gained a health worker with such changes in communities that did not. We recognize that unobserved factors may influence the introduction of a health worker to a community, which would cause bias in these “fixed-effects” estimates of the impact of health workers on health outcomes; thus we take an additional step in the analysis. Because the health workers are midwives who were trained primarily to serve women of reproductive age, we contrast the impact on the health of these women (the “treated”) with that of other adults (the “controls”) who live in the same community into which the midwife was introduced. Our main results focus on the effects of introducing a village midwife on a general measure of adults’ health, the body mass index (BMI). After controlling community-level heterogeneity, we find that among reproductive-age women, BMI increases significantly in communities that gained a village midwife and that the increase is substantively important. In contrast, men and older women (our “control” groups) do not experience as large an increase in BMI. For women of reproductive age, the benefits of access to midwives extend to pregnancy outcomes: we also find that the introduction of a midwife is associated with increases in birthweight. We conclude that the expansion of the Village Midwife program has yielded significant improvements in health, particularly for women of reproductive age. BACKGROUND Notwithstanding the economic crisis of the late 1990s, socioeconomic development in Indonesia has improved substantially over the past three decades. From 1967 to 1997 Indonesia’s per capita gross domestic product (GDP) increased by almost 5% per year. At the same time, Indonesia achieved nearly universal enrollment in primary school and substantial increases in secondary-school enrollment. Since the early 1960s, several indicators of health status in Indonesia also have shown major improvements. The infant mortality rate has declined steadily, and by the mid-1990s life expectancy surpassed 60 years. Maternal mortality, however, had not shown such impressive gains as of the early 1990s, and the Indonesian government expressed considerable concern about this dimension of health outcomes. At 390 to 650 deaths per 100,000 live births, this rate was the highest in any of the ASEAN nations (Handayani et al. 1997; Mukti 1996; UNICEF 2000a, 2000b). In fact, for much of the 1990s Indonesia’s statistics

DEMOGRAPHY, VOLUME 38-NUMBER 2, MAY 2001

for maternal mortality were on a par with those in India and Bangladesh, even though the per capita GDP in Indonesia was about 50% higher than in India and about twice as high as in Bangladesh (Sarwono, Mundiharno, and Fortney 1997). To address poor maternal health, the Ministry of Health (MOH) embarked on an ambitious program to make midwifery services more widely available by training midwives and posting them to villages throughout Indonesia (Handayani et al. 1997; Kosen and Gunawan 1996; MOH 1994). Between 1990 and 1996 the Government of Indonesia planned to provide a midwife in every nonmetropolitan village or township (MOH 1994). Midwives typically were recruited from three-year nursing academies and received one additional year of midwifery training (Sweet et al. 1995). By 1998, 54,000 midwives had been trained; between 1986 and 1996 the number of midwives per 10,000 population increased more than tenfold from 0.2 to 2.6 (Hull et al. 1998; MOH 2000; Reproductive Health Focus 2000). Once assigned to a community, the midwives are paid a salary by the Government of Indonesia for three to six years (Hull et al. 1998). They maintain a public practice during normal working hours and are allowed to practice privately after hours. It is expected that midwives will build up a client base while working for the government; thus, when their contract ends, they can maintain their practice in the village without a government salary (Gani 1996; MOH 1994). The role of the village midwife, as described by the Indonesian MOH, suggests that she will affect health status, particularly of reproductive-age women. Her duties include promoting community participation in health, providing health and family planning services, working with traditional birth attendants, and referring complicated obstetric cases to health centers and hospitals. She is to serve as a health resource in her community, actively seeking out patients and visiting them in their homes rather than waiting passively until they come to her (MOH 1994). These activities bring a village midwife into contact with a wide array of community residents in a variety of settings, and provide her with opportunities to advise clients on nutrition, food preparation, sanitation, and other health-promoting behaviors. Village midwives provide general services in addition to those oriented toward maternal and well-baby care, as supported by research in central Java (Mukti et al. 1997). On the basis of interviews, record abstraction, and client observations with 19 village midwives, the study finds that although reproductive-age women are the primary clients, midwives also treat nonobstetric patients, including men. Additional evidence about the role of village midwives comes from interviews with 157 village midwives, which were conducted as part of the Community and Facility component of the Indonesia Family Life Survey (IFLS) in 1997 (described further below). Table 1 summarizes some of the results from those interviews. In regard to service provision, the village midwives offer their communities more than prenatal care, delivery assistance, and family planning; about half also provide child immunizations. The great majority of village midwives provide more general curative care, and

WOMEN'S HEALTH AND PREGNANCY OUTCOMES

255

TABLE 1. SERVICE PROVISION BY VILLAGE MIDWIVES Type of Service

Percentage Who Provide

Maternity Care Prenatal exam Delivery assistance Tetanus toxoid injection

98 91 66

Family Planning Oral contraceptives IUD insertions Injectable contraceptives

91 41 94

Children’s Immunizations

48

General Care Curative care: exams, medicine Stitch wounds Incise and drain abscesses

97 76 34

Drugs Antibiotics Cough medicine Oral rehydration solution Iron tablets Vitamin A

96 94 93 95 84

Note: Based on data from 157 village midwives interviewed as part of the IFLS2 Community-Facility Survey.

stitch wounds. About one-third say they can incise and drain abscesses. Almost all village midwives dispense medications such as antibiotics, cough medicine, vitamins, and supplements of micronutrients such as iron and Vitamin A. The comprehensiveness of services offered by village midwives suggests some of the pathways through which availability of a village midwife may improve health. For example, if a village midwife provides curative care, her presence may reduce durations of illness from diarrheal and respiratory diseases and thus may limit the weight loss associated with such illnesses. Because of the midwife’s years of health training and her ability to offer an array of curative and preventive services, coupled with nutrition education and distribution of vitamins and micronutrients, her arrival in a community may well lead to improvements in her clients’ nutritional status. The Village Midwife program builds on the public health system of clinics and outreach activities established in Indonesia during the 1970s and 1980s. The backbone of this system is the community health center (puskesmas). The health center provides an array of services and is a basic source of subsidized outpatient care. Health centers generally are headed by a doctor, who oversees a midwife and various paramedical workers (MOH 1990). In better-off areas the center’s staff may include several doctors, as well as one or two dentists. Each subdistrict (kecamatan), consisting of 20 to 40 villages or townships, has one or more health centers. Staff members of the health center, in conjunction with family planning fieldworkers, are responsible for conducting

outreach activities, such as supervision of posyandus (neighborhood health posts), within the villages and townships in their catchment area. The posyandu is held monthly and is attended by children under five and their mothers. It is staffed by neighborhood volunteers and (if possible) by staff members from the health centers or by family planning fieldworkers. (The latter also provide contraceptive supplies to workers from the health centers and to posyandus.) When health workers attend, the posts generally provide prenatal care, immunization, and contraceptive injections (Kosen and Gunawan 1996). When health workers do not attend, services are limited to provision of vitamins and oral rehydration solution, nutritional screening, and oral contraceptives. Private practitioners also are an important source of health care in Indonesia. Private services are more widely available in urban than in rural areas, but because employees of the health center generally offer private services in offhours, private practitioners are found in rural areas as well (Brotowasisto et al. 1988; Gani 1996; World Bank 1990). CONCEPTUAL FRAMEWORK In Indonesia as in other countries, improvements in health outcomes and expansion in health services have occurred simultaneously. This fact, however, does not tell us whether the investments in services caused the improvements in health. It is plausible that other factors that have changed, including economic growth, are correlated both with improvements in health and with greater access to services. In an effort to isolate the role of health services, a number of studies have contrasted spatial variation in program availability or strength with spatial variation in health outcomes. Yet a correlation between access and health outcomes at a point in time does not identify the direction of causality. Services may be provided in a particular location in response to demand for those services, or people who want services may move to places where they are provided (Rosenzweig and Wolpin 1986, 1988). Either scenario yields a spurious correlation between access to services and health outcomes because the relationship is governed by a common (unobserved) factor. It is also possible that governments target particular types of communities for interventions. Targeting will not bias estimates of the effects of the intervention if it is based on characteristics that are observed and controlled in a regression context. If targeting is based on unobserved characteristics, however (or if the full set of characteristics used for targeting is not controlled in the regression), and if those unobserved characteristics are correlated with the outcome of interest, estimated effects of the intervention will be biased. The direction of that bias is ambiguous. To illustrate, imagine that government services are provided in communities that are underserved by private providers and that health status in those communities is relatively poor, everything else held equal. Unless all characteristics that underlie the placement of the program are controlled, the estimated impact of the intervention will be biased negatively, and the bias will be greatest for the interventions targeted to

256

DEMOGRAPHY, VOLUME 38-NUMBER 2, MAY 2001

the people who need them most. This issue of selective program placement is important in the context of health policies in Indonesia (Frankenberg 1992; Gertler and Molyneaux 1994; Pitt, Rosenzweig, and Gibbons 1993). In theory, these complicating issues are sidestepped by social experiments involving random assignment of subjects to treatment and control groups. Although such experiments have produced valuable findings regarding some policy questions (see, for example, Berggren, Ewbank, and Berggren 1981; Dow et al. 1999; Faveau et al. 1991; Newhouse 1994), they have their own drawbacks. They tend to be small in scale and to involve homogeneous populations; thus their generalizability is limited (Ewbank 1994). They are typically expensive, take a long time to complete, and can be difficult to implement. In some instances, experiments induce behavioral responses (such as migration to areas that are served in the trial) that substantially complicate evaluation of the intervention. In our view, observational data are an important complement to evaluations of interventions based on randomized trials. Of course, studies based on observational data cannot ignore the complicating issues discussed above. We adopt a quasi-experimental approach to evaluate the effects of an expansion in access to midwifery services and health outcomes in Indonesia. Using longitudinal household survey data, we compare an individual’s health before the introduction of a midwife in a community with the same individual’s health after the intervention. In doing so, we sweep out of the model all factors that are fixed at the individual and community level and enter the model additively, including any fixed characteristics that are correlated with the placement of midwives. This “fixed-effects” model has been used extensively in the program evaluation literature (for a discussion, see Heckman and Robb 1985). We are contrasting changes in health of the “treated” with changes in health of a control group, namely respondents in communities where midwives were not introduced: ∆θi = α + βMc + εic, where ∆θi is the change in health of individual i and Mc is an indicator variable for whether or not a village midwife was introduced in community c. Time-varying unobserved heterogeneity that affects changes in health is captured in εic. The intercept, α, reflects changes in health of the population between the two waves of the survey that are not related to the introduction of a midwife. β measures the difference in changes in health status of those living in communities where a midwife was introduced relative to other communities. This is an “average treatment effect,” calculated over all people living in the “treated” communities. The Village Midwife program was conceived out of concern for maternal mortality. Because reproductive-age women are likely to benefit most from the introduction of a midwife, we refine the treatment group to include only those women in the treated communities. We compare the effect of introducing a midwife on their health with the effect on the health of men of the same age living in the same communities: ∆θi = α1Iipf + α2Iipm + β1Mc × Iipf + β2Mc × Iipm + εic ,

where Iipf is an indicator variable for prime-age females and Iipm is defined analogously for prime-age males. The coefficient on the interaction between the prime-age female and midwife indicator variables, β1, is an estimate of the change in the health of a prime-age woman in a “treated” community relative to the change in health of a similar woman in a community where a midwife was not introduced. If the introduction of a midwife in a village is uncorrelated with time-varying unobserved heterogeneity, εic, then this model will provide an unbiased estimate of the effect of the program. Below, however, we show that midwives are more likely to be introduced in poorer communities with little infrastructure. If changes in health differ between poorer and better-off communities, β1 will be a biased estimate of the effect of the program. We can gain some sense of the extent of that bias by comparing changes in health of men in communities where a midwife was introduced with changes in health of men in other communities. Under the assumption that midwives have no effect on males’ health, this difference, β2, will be a measure of the “program placement” effect. The “difference-in-difference” between the effect on females and the effect on males, β1 – β2, nets out the “program placement” effect and thus provides an estimate of the “midwife” effect. It may be that midwives do in fact influence males’ health—directly (through providing services to men, for example) or indirectly (through spillovers such as nutrition education to women, which in turn affects men’s health). In this case, the “difference-in-difference” will be a biased estimate of the impact of introducing a midwife. The empirical importance of this concern can be probed by expanding the control groups to include older females, Iof , and older males, Iom : ∆θi = α1Iipf + α2Iipm + α3Iiof + α4Iiom + β1Mc × Iipf + β2Mc × Iipm + β3Mc × Iiof + β4Mc × Iiom + εic .

(1)

Older men are the least likely to benefit directly from the introduction of a midwife. If we assume that midwives are not detrimental to older men’s health, the difference-indifference, β1 – β4, provides a lower-bound estimate of the effect of a midwife.1 Older women’s health, on the other hand, has more in common with that of prime-age women; thus older women may well benefit from the introduction of a midwife. Therefore we expect that β1 – β3 is likely to understate the effect of a midwife. If the survey measures all the correlates of changes in health status that affect the allocation of midwives, it is possible to directly estimate the effect of a midwife by controlling those characteristics in the regression. We will experiment with this approach by drawing on the rich array of community-level information contained in our data source. In addition, the inclusion of individual- and communitylevel observables will increase the efficiency of the regression estimates. 1. Midwives might encourage families to reduce their investments in older men’s health, which would bias upward the difference-in-difference results. This strikes us as unlikely, however.

WOMEN'S HEALTH AND PREGNANCY OUTCOMES

257

It is possible, however, that even with controls for observed differences across communities, the introduction of a midwife is correlated with unobserved heterogeneity, εic, which would bias estimates of the program’s effect. Thus we include a community-specific fixed effect, µc; this effect, in a regression of changes in health, ∆θ, serves as a community-specific time trend and sweeps out all changes that are common across adults in each community that gained a midwife. The conceptual experiment that we have in mind is to contrast changes in health of reproductive-age women with changes in health of other adults living in the same community. Bias due to program placement will be absorbed in the community effect, and we can estimate the effect of the midwife program. Clearly, in this case, we can estimate only the difference-in-differences. We exclude the term for prime-age males from the regressions,

DATA The data we use for this study come from two rounds of the IFLS, an ongoing panel survey of individuals, households, communities, and facilities. The first round of data (IFLS1, collected in 1993) included interviews with 7,224 households (Frankenberg and Karoly 1995). The IFLS conducted interviews in 321 enumeration areas in 13 of Indonesia’s 26 provinces, and represents about 83% of the Indonesian population.2 In 1997 we constructed a resurvey (IFLS2) in which we sought to reinterview all IFLS1 households (and all members of these households in 1997), as well as a set of target members of IFLS1 households in 1993 who had migrated out by 1997 (Frankenberg and Thomas 2000). IFLS2 succeeded in reinterviewing 94.5% of IFLS1 households and 92% of the individuals who were age-eligible for this study. When we condition on observable characteristics (measured in 1993), recontact is slightly higher (0.7%, t = 1.3) in communities that gained a village midwife than in those that did not. We conclude that attrition is not likely to be a source of contamination in our results. The IFLS questionnaire covers a broad array of topics. A trained anthropometrist recorded the height and weight of

each household member in both IFLS1 and IFLS2—a central consideration for this study. Our primary indicator of adults’ health will be body mass index (BMI), which is weight (in kilograms) divided by height (in meters) squared. BMI is more directly interpretable than weight (which varies systematically with height); extreme values of BMI are associated with elevated risk of morbidity, difficulties in activities of daily living, and mortality (Fogel 1998; Strauss and Thomas 1998; Waaler 1984). BMI also is associated with physical capacity as indicated by maximal oxygen uptake (Spurr 1983) and labor productivity (Thomas and Strauss 1997). Table 2 presents summary statistics of BMI levels for four groups: reproductive-age women (age 20 to 45 in 1993), men of the same age, older women, and older men. On average, BMI has increased for prime-age men and women but has remained constant for older respondents. The table also reports the percentage of each group whose BMI is below 18.5, a cutoff below which elevated risks of morbidity and mortality are well documented. About 10% of prime-age adults fall below this cutoff; this percentage declined between 1993 and 1997. Some 30% of older adults are below the cutoff; the percentage has increased for older men. In a tiny percentage of Indonesians, the BMI is high enough to suggest that they are at risk of health problems from being overweight.3 The regression models are specified in terms of change in BMI for each respondent; this can be regarded as change in weight for prime-age adults (for whom height is fixed). We interpret change in BMI as indicating a change in general health status. Because increases in BMI in the normal range do not have the same implications for health as do increases among those with low BMI, we also present results that focus on respondents of the latter type. In part, the changes in BMI reflect changes over the life course and changes in diet or energy expenditure due to changes in availability of household resources. The regressions control each respondent’s age and education (which are displayed in Table 2) along with household per capita expenditure (PCE) at the time of the survey. PCE is considered to be a reliable measure of resource availability in the household. In this paper we focus on the impact of expanding the Village Midwife program. As clarified in the discussion above, it is important to control for community-level characteristics that might be correlated both with changes in health and with the introduction of a midwife. The IFLS is a particularly rich resource in this regard. Each wave of the survey contains a detailed set of community questionnaires administered in the IFLS enumeration areas. Extensive interviews are conducted with the head of the village or township (or a designated staff member), with the head of the commu-

2. The 321 IFLS enumeration areas are small survey-defined clusters of households located in 312 administrative areas known as desa (village) or keluruhan (township), of which there are more than 62,000 in Indonesia. We refer to desa and keluruhan collectively as “villages.” For the remainder of this paper we use the term community to designate both an IFLS enumeration area and the larger administrative area (“village”) in which it is located.

3. In 1993 only 4.5% of the sample had a BMI of 28 or higher, the level above which morbidity and mortality have been shown to rise (Fogel 1998; Waaler 1984). Rates are low for each of the demographic groups as well. Among women of reproductive age, 6.7% had a BMI of 28 or higher, as did 6.4% of women 46 and older. Among men, rates were 2% for younger men and 1.9% for older men. In 1997 a total of 6% of respondents had a BMI of 28 or higher.

∆θi = α1Iipf + α3Iiof + α4Iiom + β1Mc × Iipf + β3Mc × Iiof + β4Mc × Iiom + Xiγ + µc + εic,

(2)

but include individual characteristics, Xi, to improve efficiency. The difference-in-differences will be biased if program placement is based on the health of reproductive-age women relative to the health of other adults in a particular community. We will explore the evidence for this sort of targeting in the analyses below.

258

DEMOGRAPHY, VOLUME 38-NUMBER 2, MAY 2001

TABLE 2. ADULTS’ BODY MASS INDEX, 1993 AND 1997 Sex and Age, 1993 (Sample Size) Women 20–45 (3,030) BMI % BMI < 18.5 Age Education Men 20–45 (2,232) BMI % BMI < 18.5 Age Education Women 46 and Older (1,913) BMI % BMI < 18.5 Age Education Men 46 and Older (1,649) BMI % BMI < 18.5 Age Education

1993

1997

Change

22.1 (0.06) 12.7 (0.6) 37 5.4

22.8 (0.07) 9.6 (0.5)

0.7a (0.03) –3.1a (0.5)

21.2 (0.06) 12.1 (0.7) 36 6.6

21.6 (0.06) 10.7 (0.7)

0.3a (0.04) –1.4 (0.6)

21.0 (0.09) 29.1 (1.0) 61 2.4

21.0 (0.09) 28.8 (1.0)

20.4 (0.07) 26.9 (1.0) 62 4.7

20.4 (0.08) 30.4 (1.0)

0.0 (0.04) –0.3 (0.7)

0.0 (0.03) 3.5a (0.8)

Notes: Sample includes 8,824 individuals who were interviewed and measured in both IFLS1 and IFLS2 and were at least 20 years old in IFLS1. Standard errors are in parentheses. a

Difference between 1993 and 1997 levels is significant at < .05.

nity women’s group (typically the wife of the head of the village), and with knowledgeable informants in a sample of up to 12 health providers and up to eight schools in the community. Drawing on those data, we construct measures of other dimensions of the health service environment and of levels of infrastructure for each wave of the survey. Table 3 summarizes aspects of the health service environment and the physical infrastructure environment, as measured by the IFLS1 and IFLS2 community-facility surveys. Access to the Village Midwife program is measured with an indicator of whether a village midwife was present in the community in each of the two survey years. Access to health services is measured as the distance to the health center and to the private practitioner that are closest to the village leader’s office. With respect to outreach efforts by health centers, we construct a variable indicating whether or not the

community’s posyandus receive monthly visits from health center staff members. Physical infrastructure is measured by whether a public phone is located in the community and whether the community’s main roads are paved. The IFLS reflects the dramatic expansion of the Village Midwife program documented in the literature on the Indonesian health system. In 1993 just under 10% of IFLS communities had a village midwife; by 1997 this percentage had increased to 46% (Table 3). Over the four-year period between survey waves, more than one-third of IFLS communities gained a village midwife. The data also suggest that one aspect of health centers’ outreach to communities declined somewhat between 1993 and 1997, as reported by the head of the village women’s group. The percentage of communities reporting that health center staff members visited posyandus in the community monthly decreased from 96% in 1993 to 88% in 1997. Only about 3% of communities gained monthly visits to posyandus from health center staff members, whereas 11% of communities lost such visits. Possibly in these communities village midwives now attend the posyandu, rendering supervisory visits from health center staff less necessary. The basic measures of access to public and to private services—distances to the closest public and private facilities as reported by the village leader—changed little between 1993 and 1997. In 1993 the mean distances to public and to private facilities were 1.0 and 0.6 kilometers respectively. In 1997 the mean distances were 1.1 and 0.5 kilometers. Neither change is statistically significant. The distance to a health center probably did not change because most of the

TABLE 3. ACCESS TO HEALTH CARE AND THE HEALTH OUTREACH PROGRAMS, 1993 AND 1997 1993

1997

9.4

45.8a 36.4

95.6

87.9a 2.8 10.6

Mean Distance (km) to Closest Health Center

1.0

1.1

Mean Distance (km) to Closest Private Practitioner

0.6

0.5

% With Public Telephone Gained public telephone Lost public telephone

44.2

52.0a 12.8 5.0

% With Main Roads Paved Gained paved (main) roads

70.7

84.4a 13.7

% With Village Midwife Gained village midwife % Receiving Monthly Visits to Posyandu From Health Center Staff Gained monthly visits Lost monthly visits

Note: Level of observation is IFLS enumeration area; sample size is 321. a

Difference between 1993 and 1997 levels is significant at < .05.

WOMEN'S HEALTH AND PREGNANCY OUTCOMES

expansion in fixed-site government health facilities took place before the 1990s. This fact is helpful in identifying the effect of an expansion in the midwife program. With respect to physical infrastructure, about half the communities had a public phone in 1997, up from 44% in 1993. Between 1993 and 1997 the fraction of communities in which most roads are paved increased by 14 percentage points, bringing the total percentage to 84%. The descriptive statistics indicate a substantial increase in access to village midwives between 1993 and 1997. In examining how these midwives were allocated across communities, we use the IFLS data from 1993 to explore how aspects of socioeconomic development and health status, measured at the community level in 1993, are associated with expansion in access to midwives between 1993 and 1997. The dependent variable in the regressions is a dichotomous indicator of whether the community gained a village midwife between 1993 and 1997. The results are presented in Table 4. In the first model, we include only average per capita expenditure levels of households in the community (measured in 1993). This model tests whether gaining a village midwife varies with the community’s wealth. Expenditure is specified as a spline with a knot at the 25th percentile. For communities in the lowest quartile of the expenditure distribution, higher household expenditure does not affect the probability that a village midwife will be assigned to the community between 1993 and 1997. In contrast, for mean expenditure level in communities with expenditures in the top three quartiles of the distribution, the coefficient is large, negative, and statistically significant. The results provide strong evidence that among the IFLS communities, the poorest as of 1993 were most likely to gain a village midwife by 1997. In the second specification, we introduce controls for province (coefficients not shown) and for other aspects of community infrastructure. The introduction of these additional controls produces almost a threefold increase in the R2 of the model, from 0.08 to 0.22. Moreover, the results reveal that the greater a community’s distance from a health center in 1993, the more likely that community was to gain a village midwife by 1997. Distance from a private practitioner also has a positive but only marginally significant effect. In addition, communities with a public phone in 1993 were significantly less likely to gain a village midwife by 1997. In the third specification we add controls for per capita expenditure levels in 1997 and for whether the community’s posyandus received monthly visits from health center staff members in 1997. Because we control simultaneously for these characteristics in 1993, the 1997 characteristics can be regarded as reflecting change since 1993. On the basis of the coefficients for the 1997 characteristics, it does not appear that the communities that were becoming poorer over time were more likely to gain a midwife, or that health centers reduced their outreach activities in communities that gained a midwife. In the fourth specification, we introduce a control for the average body mass index of adults in the community in 1993,

259

as a means of assessing whether health status in the community is correlated with subsequent introduction of a midwife. The coefficient on this variable is not statistically significant. Possibly the BMI of certain demographic groups (rather than of all adults) is correlated with the allocation of village midwives. For example, midwives may be targeted toward communities in which women were particularly disadvantaged. In the fifth model we add variables measuring the average BMI of men, of women age 50 and above, and of men 50 and above. The coefficient on mean BMI captures the correlation between the BMI of prime-age females in 1993 and the introduction of a midwife. Midwives were more likely to be introduced in communities in which men were heavier than women, and less likely to be introduced where older women were lighter than prime-age women. On the margin, the presence of men who are heavy is positively associated with gaining a village midwife, whereas the presence of older women who are light is negatively associated with gaining a village midwife. Neither of these correlations, however, is significant, and as a group, the BMI variables are not statistically significant. In the sixth specification we add measures of the percentage of adults (by age and sex group) whose BMI is less than 18.5, to ascertain whether the addition of a village midwife responds to the prevalence of poor health in 1993 (rather than to an indicator of average health). None of the coefficients on these variables is statistically significant, nor are the health status measures jointly significant. We also tested for a correlation between mean level of children’s nutritional status in 1993 and receipt of a village midwife by 1997, and found no significant relationship between the two. The community-level measures of health status in 1993 do not appear to predict gaining a village midwife by 1997. Nor does their presence in the models change the relationships of economic status and of access to infrastructure to gaining a village midwife. In sum, it appears that the increase in the number of village midwives between 1993 and 1997 was not a direct response to levels of nutritional status in 1993. Nor was the allocation of midwives to communities random, however. Instead, the empirical evidence suggests that the communities into which village midwives were introduced between 1993 and 1997 were those that in 1993 were relatively poor and located far from public health services. RESULTS The results presented in Table 4 suggest that a community’s levels of poverty and remoteness influence whether it received a village midwife. If the characteristics that influence receiving a midwife also influence health status, as seems likely, cross-sectional estimates of the relationship between presence of a village midwife and health status will be biased unless the specifications include controls for all the factors that affect both health status and allocation of village midwives. We address this issue with the strategies described in the conceptual framework, estimating four models that relate change in BMI to exposure to a village midwife. An increase in BMI over time is interpreted as health-improving.

260

DEMOGRAPHY, VOLUME 38-NUMBER 2, MAY 2001

TABLE 4. COMMUNITY-LEVEL CORRELATES OF GAINING A VILLAGE MIDWIFE BY 1997

1993 per Capita Expenditure < 25th Percentile (Spline) 1993 per Capita Expenditure ≥ 25th Percentile (Spline) 1997 per Capita Expenditure < 25th Percentile (Spline) 1997 per Capita Expenditure ≥ 25th Percentile (Spline) Mean BMI, 1993

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

0.27 (0.78) –1.90* (0.39)

1.05 (0.92) –1.28* (0.48)

1.53 (1.09) –1.00† (0.51) –0.78 (1.09) –0.51 (0.37)

1.32 (0.94) –1.12† (0.48)

1.32 (0.96) –1.08* (0.49)

1.07 (1.01) –1.00* (0.49)

–0.16 (0.10)

–0.24 (0.21) 0.37† (0.21) –0.17† (0.09) –0.09 (0.11)

–0.21 (0.36) 0.27* (0.13) 0.24† (0.13) 0.22 (0.74)

–0.14 (0.37) 0.25* (0.12) 0.28* (0.14) 0.08 (0.74)

–0.08 (0.26) 0.27 (0.24) –0.18 (0.12) –0.02 (0.14) 3.82 (2.94) –0.28 (1.93) –0.22 (1.10) 0.62 (1.10) –0.09 (0.37) 0.25* (0.12) 0.26† (0.14) 0.13 (0.76)

–0.87† (0.38) 0.01 (0.33) 0.33 (0.35) 10.03 96.97 (0.00) 0.23

–0.85* (0.39) –0.01 (0.34) 0.44 (0.36) –11.73 103.98 (0.00) 0.25 9.13 (0.06)

–0.76† (0.39) 0.03 (0.35) 0.39 (0.37) –12.07 106.58 (0.00) 0.25 11.81 (0.16)

Mean BMI, Males, 1993 Mean BMI, Females > 50 Mean BMI, Males > 50 % With BMI < 18.5 % Males With BMI < 18.5 % Females ≥ 50, With BMI < 18.5 % Males ≥ 50, With BMI < 18.5 Urban Residence

–0.27 (0.36) 0.27* (0.12) 0.25† (0.13) 0.15 (0.73)

Distance to Nearest Health Center Distance to Nearest Private Practice Monthly Visit by Health Ctr. Staff (1993) Monthly Visit by Health Ctr. Staff (1997) Public Phone in Community Market in Community Main Roads Paved Constant F Prob (F) R2 Chi-Square (Health Indicators)

–2.79 35.05 (0.00) 0.08

–0.92* (0.38) –0.03 (0.33) 0.37 (0.35) –11.00 93.52 (0.00) 0.22

–0.16 (0.37) 0.32 (0.13) 0.31* (0.14) 0.11 (0.74) 0.79 (0.52) –0.84* (0.38) 0.03 (0.33) 0.36 (0.36) –8.67 98.86 (0.00) 0.23

Notes: Logistic regressions. Level of observation is IFLS enumeration area; sample size is 321. Standard errors reported in parentheses. †

p ≤ .10; *p ≤ .05

WOMEN'S HEALTH AND PREGNANCY OUTCOMES

The independent variable of primary interest is whether the individual lived in a community that gained a village midwife between 1993 and 1997. Midwives and Adult BMI Table 5 presents the main regression results. Estimates of β are reported in panels A and B. In Models 3 and 4 we include controls for individual and household observables. These include respondent’s education, age, and (at the household level) per capita expenditure. All are specified as spline functions with several knots to allow flexibly for nonlinearities. Model 3 also includes other community-level measures such as urban/rural status, gain or loss of monthly visits from health center staff, changes in distances to health centers and private practices, gain of paved roads, and gain or loss of a public phone. We begin with the correlation between the change in an adult’s BMI measured in 1993 and in 1997 and whether a midwife was introduced into the village between 1993 and 1997. As shown in the first column, that correlation is essentially zero. Following the discussion above, in the second specification (column 2) we refine the treatment and control groups. Women of reproductive age (20–45 years) are considered the “treatment” group and are contrasted with three “control” groups: men in the same age group, women over 45, and men over 45. This specification allows us to examine the correlations between gaining a village midwife and change in BMI for the four demographic groups, and to test whether the correlations differ across the groups. (These “difference-in-difference” tests are presented in panel C.) The results from this specification indicate that the addition of a village midwife to a community is associated positively with change in BMI for women of reproductive age but negatively for the other demographic groups. The negative correlation is significant for older men. We do not interpret the negative effects as indicating that midwives hurt everyone except young women, but rather that these results capture the “program placement” effects; thus they reflect the fact that midwives are allocated to communities where improvements in health status are unlikely. As discussed above, the difference-in-differences address this concern. The pertinent estimates, reported in panel C, indicate that the presence of a midwife is associated with significantly improved health in women of reproductive age relative to the health of other demographic groups. This result persists when we include observable characteristics of the respondents and their communities (Model 3), although the differential effect on older and younger women is slightly smaller (and significant only at 10%). The fact that residence in a community that gained a village midwife is associated with improved BMI only among prime-age women suggests that the relationship is causal. If placement of village midwives occurred in communities where nutritional status improved for other reasons, one would expect a positive correlation with the introduction of a village midwife for all demographic groups.

261

Our final specification (Model 4) goes one step further. We include a community-specific time trend to ask whether, within communities that gained a village midwife, the health of reproductive-age women improved more than that of other adults. The difference-in-difference estimates (panel C) indicate that the answer is affirmative in regard to men: BMI improved by about 0.20 more for reproductive-age women in these communities than for older or younger men, and these differences are significant. Although the difference is slightly larger for older men, in keeping with our expectation that spillover benefits of a midwife would be smallest for this group, the difference between the effect on younger and older men is small and not significant. Midwives, however, apparently are associated with spillover benefits for older women: although the latter benefit less than women of reproductive age, that difference-in-difference is not significant. Inferences drawn from the difference-in-difference results are remarkably consistent across the three empirical specifications. The evidence suggests that unobserved heterogeneity contaminates the estimates, particularly among older respondents; thus we are inclined to place the greatest weight on the estimates in Model 4. Because we are using observations on individuals at two points in time, we cannot explore dynamics underlying the effect of a midwife in a community. Rather, in a linear and additive framework, we are measuring the cumulative effect, by 1997, of a midwife introduced to the community between 1993 and 1997. If all the gains in BMI associated with the introduction of a village midwife accrued to people with a BMI in the normal range, the benefits of expansion of the village midwife would not be obvious. Therefore we reestimated the models, restricting the sample to respondents whose BMI was 21 or below in 1993 (roughly half the sample). Panel D of Table 5 reports the estimated difference-in-differences, which are larger than for the entire sample.4 These results indicate that individuals with lower BMI benefit more from the introduction of a village midwife.5 The results indicate that increased access to village midwives between 1993 and 1997 has had a positive impact on

4. We prefer this to an alternative specification that focuses on whether a respondent is above or below a particular cutoff point. The 1993 IFLS data contain evidence of a positive association between BMI and greater functioning, better health, and reduced morbidity among people with BMI below 21 (Strauss and Thomas 1998). Moreover, for reproductive-age women at risk of becoming pregnant, a low BMI is a particular disadvantage because it increases the amount of weight they must gain to achieve a healthy pregnancy (Krasovec and Anderson 1991). A discrete outcome would discard much of the information about improvements in health and would tend to bias the estimates toward not finding program effects that exist. 5. We also explored whether gaining a village midwife particularly benefits women who are similar to the midwife in age and education, as suggested by Rogers and Solomon (1975) with respect to traditional midwives. Our results show that the midwife’s age relative to her client’s has no impact on her effectiveness. Midwives, who themselves are quite well educated, appear to exert a slightly larger effect on the BMI of women with little education; this point suggests that socioeconomic similarity between a midwife and her potential clients is not the force governing her effectiveness.

262

DEMOGRAPHY, VOLUME 38-NUMBER 2, MAY 2001

TABLE 5. CHANGE IN BMI BETWEEN 1993 AND 1997 AND GAINING A VILLAGE MIDWIFE

Covariates Individual observables Community observables Community-specific time trend A. Gain Village Midwife by 1997

Model 1 OLS

Model 2 OLS

Model 3 OLS

Model 4 FE

.–– .–– .––

.–– .–– .––

.Yes .Yes .––

.Yes .–– .Yes

0.120† (0.065) –0.075 (0.062) –0.105 (0.076) –0.173* (0.066)

0.112† (0.066) –0.087 (0.064) –0.059 (0.077) –0.148* (0.067)

0.174† (0.087) .–– 0.040 (0.100) –0.032 (0.103)

0.195† (0.090) 0.225* (0.092) 0.293* (0.098)

0.199* (0.089) 0.171† (0.093) 0.261* (0.098)

0.174* (0.087) 0.134 (0.093) 0.206* (0.097)

0.212† (0.114) 0.289* (0.122) 0.265* (0.111)

0.224* (0.112) 0.302* (0.120) 0.274* (0.110)

0.241* (0.109) 0.263* (0.115) 0.210† (0.114)

–0.043 (0.035)

B. Gain Village Midwife if Female 20–45 Male 20–45 Female > 45 Male > 45 C. Difference-in-Difference, Whole Sample, Controlsa Male 20–45 Female > 45 Male > 45 D. Difference-in-Difference, BMI ≤ 21, Controlsa Male 20-45 Female > 45 Male > 45

Notes: The dependent variable is the difference between BMI in 1997 and 1993 (BMI in 1997 – BMI in 1993). Difference-indifference is the differential effect of gaining a midwife on the treatments (females age 20–45) relative to the controls. In panel C it is based on estimates from panel B; in panel D it is based on the subsample with low BMI in 1993. Individual observables include per capita expenditure in 1993 (splines with knots at the quartiles), education (splines with knots at 6 and 9 years), and age (splines with knots at 25, 35, 45, 55, and 65 years) all measured at baseline in 1993. Community-level observables include whether the community’s posyandus gained or lost monthly visits from health center staff, whether the community gained paved roads, whether the community gained or lost a public phone, and changes in distances to the closest health center and private practitioner. Sample size is 8,824 adults age 20 and older in 1993 for the entire sample. For panel D, where the results are restricted to individuals with BMI ≤ 21, sample size is 4,718. Robust standard errors that permit within-community correlations are reported in parentheses. a

Females age 20–45 are the treatment group.



p < .10; *p < .05

women’s health, particularly for women of reproductive age. These effects are greater for women whose BMI was low in 1993. Because no similar effect occurs for males, we conclude that the effect for women does not reflect placement of midwives in communities where health would have improved in any case.

We cannot rule out the possibility that midwives were placed in communities where young women’s health would have improved relative to men’s. Although that scenario strikes us as unlikely, we can explore its relevance by assessing whether the timing of the introduction of a midwife to a community affects women’s reproductive health. There-

WOMEN'S HEALTH AND PREGNANCY OUTCOMES

fore we now contrast the birthweight of babies born before and after a midwife is introduced into a community. Midwives and Birthweight We use birthweight as a measure of pregnancy outcome. Birthweight is not only a marker of a successful pregnancy; it also affects the child’s subsequent health. Data from the Philippines have shown that birthweight is correlated both with survival during the neonatal period and with the risk of stunting in the first two years of life (Adair and Guilkey 1997; Popkin et al. 1993). In both rounds of the IFLS, women were asked to provide detailed accounts of all pregnancies that occurred in the five years before the survey, including birthweight (if the baby was weighed). We pool the data from IFLS1 and IFLS2 to obtain information on 5,155 pregnancies (reported by 3,445 women) that occurred between 1988 and 1997 and ended in live births. The mothers reported birthweights for a total of 3,315 births (64% of all births). Mean birthweight was 3,162 grams; 8.5% of infants were reported as weighing less than 2,500 grams (the standard cutoff for low birthweight). Another 6.3% were reported as weighing exactly 2,500 grams. The distribution of reported birthweights in the IFLS data does not suggest unusually high or low proportions of lowbirthweight babies relative to those in other developing countries or relative to other data from Indonesia (Boerma et al. 1996). We observe heaping on weights (in kilograms) that end in .0 or .5, as has been observed in other data sets from developing countries with retrospectively reported birthweight data (Robles and Goldman 1999). The heaping indicates measurement error in the reported birthweights; such error, for our purposes, will inflate standard errors but will not bias the estimated effect of a midwife. We also examined the correlates of reporting a birthweight (results not shown). The probability that a birthweight is reported increases with the mother’s age (up to age 35) and, as one might expect, with level of education and with household per capita expenditure. Birthweights also are much more likely to be reported for first births and for infants delivered either in a medical setting or at home with the attendance of a biomedically trained assistant than for infants delivered at home with the assistance of traditional birth attendants. Birthweight is more likely to be reported for more recent births, but there is no association between the presence of a village midwife in the community during the pregnancy and whether a birthweight was reported. (This finding holds across all communities and in only those communities that had a village midwife by 1997.) In analyzing the relationship between birthweight and access to a village midwife, we used data from the IFLS2 Community-Facility Survey on the number of years a village midwife had been present in the community, combined with information on the time of conception, to construct a variable indicating whether a village midwife was present in the community during the pregnancy. In communities that had received a village midwife by 1997, 63% of pregnancies occurred before the village midwife arrived; 37% occurred af-

263

ter her arrival. This within-community variation in exposure to the program can be used to estimate the effect of the village midwife’s presence on birthweight, net of aspects of the community that are fixed over time and affect both the allocation of midwives and pregnancy outcomes. Table 6 presents results from these fixed-effects analyses of birthweight. The first column provides the coefficients for the relationship between birthweight and the presence of a village midwife during the pregnancy, with no controls. Column 2 adds a variety of pregnancy-specific, mother-specific, and community-specific controls. For each pregnancy we include markers for whether the pregnancy was the woman’s first and for the infant’s sex, as well as an indicator of year of birth. We also include measures of the mother’s height, her educational level, and the (log of) per capita household expenditure. At the community level, we include distance to public and private health services, whether roads are paved, presence of a public phone, and monthly visits from health center staff members. Children born before October 1995 were matched to the 1993 community data; those born in October 1995 or later were matched to the 1997 community data. In both specifications, birthweights are significantly greater in a community after a midwife is introduced than before. To attribute this finding to a program placement effect, one would have to argue that midwives were allocated to areas where birthweight would have improved even in the absence of midwives; this seems very unlikely.

TABLE 6. RELATIONSHIP BETWEEN BIRTHWEIGHT AND PRESENCE OF VILLAGE MIDWIFE DURING PREGNANCY

Covariates Community fixed effect Individual observables Community observables Village Midwife Present During Pregnancy Year of Birth

Model 1

Model 2

Yes –– ––

Yes Yes Yes

67.65† (34.49)

79.50* (40.47) 1.89 (4.85)

Notes: The dependent variable is birthweight measured in grams. Both models include a community-level fixed effect. Presence of a village midwife during pregnancy is based on information on timing of pregnancy and on arrival date of village midwife. Individual and community controls are introduced in Model 2. These include first birth, sex, maternal height, maternal age (splines with knots at 25 and 35 years), maternal education (splines with knots at 6 and 9 years), per capita expenditure (splines with knots at the 20th, 50th, and 80th percentile), year of birth, distance to the closest health center, distance to the closest private practitioner, presence of a phone, whether the community’s posyandus receive monthly visits from health center staff, and whether the main roads in the community are paved. Sample size is 3,315 births. Standard errors are in parentheses. p ≤ .10; *p ≤ .05



264

To capture any time trends in birthweight, we also include in Model 2 a term for the year when the baby was born. It is potentially difficult to disentangle an effect of time on birthweight from an effect of the presence of village midwife because village midwives were phased into communities over time. Thus, as year of birth increases, so does the probability that a village midwife was present in the community. The coefficient on year of birth does not indicate evidence of a significant time trend in birthweights. We also estimated the time trend for birthweight separately by whether a village midwife was ever posted to the community, but the time trends were not statistically significant for either type of community; nor did the trends differ from one another. CONCLUSION Both the results for change in body mass index and the results for birthweight suggest that gaining access to a village midwife is associated with improvements in health outcomes for women of reproductive age, and for their babies. The impact of the midwife’s presence on adult health status is limited to women, primarily those between ages 20 and 45. In communities that gained a village midwife, the change in reproductive-age women’s BMI is significantly larger than men’s. For reproductive-age women whose 1993 BMI was 21 or lower, the difference-in-difference estimates suggest that the addition of a village midwife was accompanied by an increase in BMI equaling at least 0.2. If 0.2 is added to the 1993 BMI of women of reproductive age, the percentage whose BMI is less than 21 declines from 44% to 41.3% (a decrease of 6%), while the percentage whose BMI is less than 18.5 declines from 12.8% to 10.9% (a decrease of nearly 15%). The estimated effect of gaining a village midwife is to increase birthweight by about 80 grams. The percentage of infants who benefit by a gain of 80 grams depends on the range of weights for which a gain of 80 grams is assumed to improve health. About 8.5% of the babies for whom weights are reported weighed less than 2,500 grams. It is likely that all of these infants would have been at least somewhat better off if they had been 80 grams heavier at birth, even if they remained below the 2,500-gram threshold for normal birthweight. In addition, a gain of 80 grams is likely to improve the health of the babies whose weight was reported as exactly 2,500 grams (6.3%) and who therefore were at the threshold of normal birthweight, and for babies just above the threshold but still relatively light. In this paper we have focused on developing and implementing a statistical strategy for estimating the size, direction, and statistical significance of the association between access to village midwives and health outcomes. Our results reveal that gaining a village midwife has a effect on the body mass index of women of reproductive age. This effect is larger for women whose BMI was low in 1993. We also find a small effect on birthweight. These estimates are robust to several strategies in which we attempt to correct for unobservable characteristics that might govern both access to

DEMOGRAPHY, VOLUME 38-NUMBER 2, MAY 2001

midwives and health outcomes; thus they increase our confidence that a causal mechanism underlies the relationships we observe in the data. For both body mass index and birthweight, the effects of gaining a village midwife are health-improving and statistically significant. It is likely that they presage positive effects of the Village Midwife program on a wider array of health behaviors and outcomes. REFERENCES Adair, L. and D. Guilkey. 1997. “Age-Specific Determinants of Stunting in Filipino Children.” Journal of Nutrition 127:314–20. Akin, J., D. Guilkey, and E. Denton. 1995. “Quality of Services and Demand for Health Care in Nigeria: A Multinomial Logit Estimation.” Social Science and Medicine 40:1527–37. Angeles, G., D. Guilkey, and T. Mroz. 1998. “Purposive Program Placement and the Estimation of Family Planning Program Effects in Tanzania.” Journal of the American Statistical Association 93(443):884–99. Berggren, W., D. Ewbank, and G. Berggren. 1981. “Reduction of Mortality in Rural Haiti Through a Primary-Health Care Program.” New England Journal of Medicine 304:1324–30. Boerma, J., K. Weinstein, S. Rutstein, and A. Sommerfelt. 1996. “Data on Birthweight in Developing Countries: Can Surveys Help?” Bulletin of the World Health Organization 74:209–16. Brotowasisto, O. Gish, R. Malik, and P. Sudharto. 1988. “Health Care Financing in Indonesia.” Health Policy and Planning 3(2):131–40. Caldwell, J.C. 1986. “Routes to Low Mortality in Poor Countries.” Population and Development Review 12(2):171–220. Dow, W., P. Gertler, J. Strauss, and D. Thomas. 1999. “Health Care Prices, Health and Labor Outcomes: Experimental Evidence.” Santa Monica: RAND. Entwisle, B., R. Rindfuss, S. Walsh, T. Evans, and S. Curran. 1997. “Geographic Information Systems, Spatial Network Analysis, and Contraceptive Choice.” Demography 34:171–87. Ewbank, D. 1994. “Evaluating Large Scale Health Interventions: Methodological Issues.” Pp. 277–96 in Evaluation of the Impact of Health Interventions, edited by H. Rashad, R. Gray, and T. Boerma. Liége: International Union for the Scientific Study of Population. Faveau, V., K. Stewart, S. Khan, and J. Chakraborty. 1991. “Effect on Mortality of a Community-Based Maternity-Care Programme in Rural Bangladesh.” The Lancet 338:1183–86. Filmer, D. and L. Pritchett. 1999. “The Impact of Public Spending on Health: Does Money Matter?” Social Science and Medicine 49:1309–23. Fogel, R. 1998. “Economic Growth, Population Theory, and Physiology: The Bearing of Long-Term Processes on the Making of Economic Policy.” Pp. 257–83 in Economic Demography, Vol. 1, edited by T.P. Schultz. Cheltenham, UK and Northampton, MA: Elgar. Frankenberg, E. 1992. “Infant and Early Childhood Mortality in Indonesia: The Impact of Access to Health Facilities and Other Community Characteristics on Mortality Risks.” PhD dissertation, Department of Sociology and Graduate Group in Demography, University of Pennsylvania. Frankenberg, E. and L. Karoly. 1995. “The 1993 Indonesian Fam-

WOMEN'S HEALTH AND PREGNANCY OUTCOMES

ily Life Survey: Overview and Field Report.” Santa Monica: RAND. Frankenberg, E. and D. Thomas. 2000. “The Indonesia Family Life Survey (IFLS): Study Design and Results From Waves 1 and 2.” Santa Monica: RAND. Gani, A. 1996. “Improving Quality in Public Sector Hospitals in Indonesia.” International Journal of Health Planning and Management 11:275–96. Gertler, P. and J. Molyneaux. 1994. “How Economic Development and Family Planning Programs Combined to Reduce Indonesian Fertility.” Demography 31:33–63. Handayani, L., L. Wilujeng, S. Sukirno, S. Pranata, and Daryadi. 1997. “Menuju pelayanan persalinan terpadu.” Report, Population Studies Center, Gajah Mada University and the Ford Foundation. Heckman, J.J. and R. Robb. 1985. “Alternative Methods for Evaluating the Impact of Interventions.” Pp. 156–245 in Longitudinal Analysis of Labor Market Data, edited by J.J. Heckman and B. Singer. New York: Cambridge University Press. Hull, T., Widayatun, A. Raharto, and B. Setiawan. 1998. “Village Midwives in Maluku.” Policy Paper 1, Center for Population and Manpower Studies, Indonesia Institute of Sciences, Jakarta. Jamison, D.T., W.H. Mosley, A.R. Meashem, and J.-L. Bobadilla, eds. 1993. Disease Control Priorities in Developing Countries. New York: Oxford University Press. Kosen, S. and S. Gunawan. 1996. “Health Services in Indonesia.” Medical Journal of Australia 165:641–44. Krasovec, K. and M. Anderson. 1991. “Maternal Anthropometry for Prediction of Pregnancy Outcomes: Memorandum From a USAID/WHO/PAHO/MotherCare Meeting.” Bulletin of the World Health Organization 69:523–32. Ministry of Health (MOH). 1990. Primary Health Care in Indonesia. Jakarta: MOH. ———. 1994. “Pedoman pembinaan teknis bidan di desa.” Report, Directorate General of Community Health, Republic of Indonesia, Jakarta. ———. 2000. “Indicators for the Third Evaluation of HFA 2000.” Report, Republic of Indonesia, Jakarta. Mukti, A. 1996. “Menjaga mutu pelayanaan bidan desa.” Population Studies Center, Gajah Mada University, Yogyakarta and the Ford Foundation. Mukti, A., A. Radyowiyati, M. Hakimi, and R. Lamsudin. 1997. “Prescribing Pattern of Village Midwives for Non-Obstetric Cases in Purworejo, Central Java, Indonesia.” Reprint 24, Community Health and Nutrition Research Laboratory, Yogyakarta. Musgrove, P. 1996. “Public and Private Roles in Health: Theory and Financing Patterns.” Discussion Paper 339, World Bank, Washington, DC. Newhouse, J. 1994. Free for All? Lessons From the RAND Health Insurance Project. Cambridge, MA: Harvard University Press.

265

Pitt, M., M. Rosenzweig, and D.M. Gibbons. 1993. “The Determinants and Consequences of the Placement of Government Programs in Indonesia.” World Bank Economic Review 7:319–48. Popkin, B., D. Guilkey, B. Schwarz, and W. Flieger. 1993. “Survival in the Perinatal Period: A Prospective Analysis.” Journal of Biosocial Science 25:359–70. Pullum, T. 1991. The Relationship of Service Availability to Contraceptive Use in Rural Guatemala. Columbia: IRD/Macro International. Reproductive Health Focus. 2000. “Indonesia: South Kalimantan Is on the Road to Healthier Mothers and Newborns.” Issue 8, Project MotherCare, Washington, DC. Robles, A. and N. Goldman. 1999. “Can Accurate Data on Birthweight Be Obtained From Health Interview Surveys?” International Journal of Epidemiology 28:925–31. Rogers, E. and D. Solomon. 1975. “Traditional Midwives and Family Planning in Asia.” Studies in Family Planning 6(5):126–33. Rosenzweig, M. and K. Wolpin. 1986. “Evaluating the Effects of Optimally Distributed Public Programs.” American Economic Review 76:470–82. ———. 1988. “Migration Selectivity and the Effects of Public Programs.” Journal of Public Economics 37:265–89. Sarwono, S., H. Mundiharno, and J. Fortney. 1997. “Reducing Maternal Mortality.” Report, Demographic Institute, Jakarta. Spurr, G. 1983. “Nutritional Status and Physical Work Capacity.” Yearbook of Physical Anthropology 1983:1–35. Strauss, J. and D. Thomas. 1995. “Human Resources: Empirical Models of Household Decisions.” Pp. 1885–2023 in Handbook of Development Economics, Vol. IIIA, edited by J.R. Behrman and T.N. Srinivasan. Amsterdam: North Holland. ———. 1998. “Health, Nutrition, and Economic Development.” Journal of Economic Literature 36:766–817. Sweet, B., V. Tickner, and G. Maclean. 1995. “Midwifery in Indonesia: A Professional Snapshot.” Modern Midwife 5(6):8–13. Thomas, D. and J. Maluccio. 1996. “Contraceptive Choice, Fertility, and Public Policy in Zimbabwe.” World Bank Economic Review 10(1):189–222. Thomas, D. and J. Strauss. 1997. “Health and Wages of Men and Women in Urban Brazil.” Journal of Econometrics 77(1):159– 86. UNICEF. 2000a. “Revised 1990 Estimates of Maternal Mortality: A New Approach by WHO and UNICEF.” Retrieved November 28, 2000 (www.unicef.org/reseval/mattab). ———. 2000b. “A Network of Community Midwives in Indonesia.” Retrieved November 28, 2000 (www.unicef.org/ programme/health/women/safe_mth/ids_cs). Waaler, H. 1984. “Height, Weight, and Mortality: The Norwegian Experience.” Acta Medica Scandinavia, supp. 679:1–56. World Bank. 1990. “Health Sector Priorities Review: Overview.” Washington, DC: World Bank.