Impact of augmented prenatal care on birth outcomes of Medicaid recipients in New York City

Journal of Health Economics 18 Ž1999. 31–67

Impact of augmented prenatal care on birth outcomes of Medicaid recipients in New York City Theodore Joyce

)

Baruch College and Graduate Center, City UniÕersity of New York and National Bureau of Economic Research, New York, NY, USA Received 30 April 1997; revised 31 December 1997; accepted 15 January 1998

Abstract I examine whether New York State’s Prenatal Care Assistance Program ŽPCAP. is associated with greater use of prenatal services and improved birth outcomes. PCAP is New York State’s augmented prenatal care initiative that became a part of the Medicaid program after expansion in income eligibility thresholds in January, 1990. Data are from the linkage of Medicaid administrative files with New York City birth certificates Ž N s 23 249.. For women on cash assistance, I find PCAP is associated with a 20% increase in the likelihood of enrollment in WIC, an increase in mean birth weight of 35 g and a 1.3 percentage point drop in the rate of low birth weight. Associations between PCAP and improved birth outcomes for women on medical assistance are similar, but appear contaminated by selection bias. Reductions in newborn costs associated with PCAP participation are modest, between US$100–300 per recipient, and are insufficient to offset program expenditures. q 1999 Elsevier Science B.V. All rights reserved. JEL classification: I18 Keywords: Prenatal care; Medicaid; Birth outcomes

) Corresponding author. National Bureau of Economic Research, 50 East 42nd Street, 17th Floor, New York, NY 10017-5405, USA. Tel.: q1-212-953-0200 ext. 111; fax: q1-212-953-0339; e-mail: [email protected].

0167-6296r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 7 - 6 2 9 6 Ž 9 8 . 0 0 0 2 7 - 7

32

T. Joyce r Journal of Health Economics 18 (1999) 31–67

1. Introduction Evaluations of the Medicaid changes in the late 1980s have been primarily concerned with the impact of income eligibility expansions on enrollment, prenatal care utilization and birth outcomes ŽPiper et al., 1990, 1994; Currie and Gruber, 1996; Cole, 1995; Kenney and Dubay, 1995.. Changes in Medicaid, however, were not limited to increases in income eligibility thresholds. Legislation also permitted states to offer enhanced reimbursement to qualified providers that provided comprehensive prenatal services to Medicaid recipients. By 1992, 37 states reimbursed qualified obstetric providers for augmented prenatal services ranging from case management to health education ŽFrost et al., 1993.. In this paper, I examine whether women on Medicaid in New York City who receive enhanced prenatal care services have better birth outcomes than women who do not receive such services. I use the rapid expansion in the availability of enhanced prenatal care services associated with the Medicaid eligibility expansion in New York to identify the effect of the receipt of augmented prenatal care on newborn health. Three studies based on single-state analyses show that augmented prenatal care programs for Medicaid recipients improve birth outcomes ŽBuescher et al., 1991; Korenbrot et al., 1995; Reichman and Florio, 1996.. There is also evidence from studies that randomized women between comprehensive and routine prenatal care that birth outcomes improve with more intensive prenatal services ŽOlds et al., 1986; McLaughlin et al., 1992.. The evidence, however, is far from robust. An analysis from Tennessee found no association between case management for pregnant Medicaid recipients and infant health ŽPiper et al., 1996.. Moreover, potential selection bias is addressed in only one of the studies based on secondary data ŽReichman and Florio, 1996.. Nor do the studies based on randomized designs provide the long-sought ‘gold standard’ against which other prenatal evaluations should be measured. Olds et al., 1986, for instance, report that comprehensive prenatal care is associated with an increase of 395 g among teens but has no effect among older women. Not only are such gains very large, but they are based on births to only 45 adolescents. The findings, therefore, are so selective and based on such small samples as to call into question their relevance to large, state initiatives under Medicaid. In this study, I examine whether New York State’s Prenatal Care Assistance Program ŽPCAP. is associated with greater use of prenatal services and improved birth outcomes. PCAP is New York State’s augmented prenatal care initiative that became a part of the Medicaid program after expansion in income eligibility thresholds in January 1990. Unlike most previous analyses of comprehensive prenatal care initiatives at the state level, I use several strategies in an attempt to minimize potential bias from unobserved differences between participants and non-participants of PCAP that are related to infant health. Specifically, I stratify

T. Joyce r Journal of Health Economics 18 (1999) 31–67

33

the analysis by year and Medicaid category since women on cash assistance are primarily US citizens whose eligibility was unaffected by the expansions. By contrast, women on medical assistance are predominantly foreign-born and potentially more heterogeneous than women on cash assistance given the doubling of income eligibility thresholds. I further stratify analyses by the timing of the first prenatal care visit so as to compare participants of PCAP who initiate prenatal care early with non-participants who also began care promptly. The rationale is that women who initiate prenatal early, regardless of participation in PCAP, share similar concerns for the health of the fetus that are difficult to measure with observable characteristics available from vital statistics. Finally, I use the rapid rise in PCAP enrollment prompted by the Medicaid eligibility expansion in January 1990 to instrument for PCAP participation. Previous analyses have been limited to a single cross-section of women and have been unable to exploit the quasi-experimental nature of the Medicaid expansion. I find that in cross-sectional regressions, Medicaid recipients served by PCAP providers have more prenatal care visits, are more likely to be enrolled in the Special Supplemental Program for Women Infants and Children ŽWIC. and to have infants of greater mean birth weight than Medicaid recipients who receive prenatal care from non-PCAP providers. 1 The increases in utilization of prenatal services and birth outcomes, although statistically significant, are modest. I then use measures for the growth in local area PCAP providers between 1989 and 1991 as instruments for PCAP participation. Treatment effects are eliminated for the sub-category of Medicaid recipients on medical assistance, but persist in some specifications for women on cash assistance, although statistically insignificant. I conclude that PCAP may be responsible for a modest improvement in birth weight of infants to women on cash assistance, but that the short-term savings measured as the reduction in newborn inpatient costs do not offset program expenditures.

2. Description of PCAP The Prenatal Care Assistance Program ŽPCAP. is New York State’s comprehensive prenatal care initiative. In order to be designated a qualified PCAP provider, facilities must demonstrate the capability to offer a prescribed range of comprehensive prenatal services or have established subcontracts with other qualified providers. The general categories of comprehensive care include risk assessment, nutritional services, health education, as well as prenatal diagnostic 1 The Special Supplemental Program for Women, Infants and Children ŽWIC. is a federal program initiated in 1971. The program’s objective is to improve the nutritional status of pregnant women, lactating mothers and children with nutritional education and supplemental foods.

34

T. Joyce r Journal of Health Economics 18 (1999) 31–67

and treatment services. Providers must also offer after-hour access and emergency consultation ŽNew York State Department of Health, 1989a.. Reimbursement for prenatal care visits under PCAP are substantially greater than payment under Medicaid. In 1988, Medicaid paid approximately US$65 per obstetric outpatient visits in New York City municipal hospitals. PCAP, by contrast, paid US$127 for an initial obstetric visit in New York City hospitals and US$76 per subsequent visit ŽNew York State Department of Health, 1989b.. Although the PCAP protocol has remained stable, the program was financed and administered differently between 1989 and 1991. Prior to 1990, PCAP was a state-funded grant program separate from Medicaid that reimbursed providers on a fee-for-service basis for prenatal and postpartum visits. PCAP did not cover inpatient hospital costs, and physician or nurse midwife delivery fees were low and only allowable under specific circumstances. Eligibility for PCAP was also more generous than Medicaid prior to 1990. Women with incomes less than 185% of the federal poverty level were eligible for PCAP, whereas the Medicaid threshold was approximately 100% of poverty. Despite the relatively generous reimbursement as compared with Medicaid, many obstetrical providers did not participate in PCAP prior to 1990. To be classified as a PCAP provider, institutions had to demonstrate the capacity to offer the requisite range of services. Small providers or free-standing clinics may have viewed such requirements as too costly. This does not explain why many hospitals were not PCAP providers in 1989, since most hospitals could have easily established the necessary access to outreach and ancillary services. One possible explanation for the relatively low participation by hospitals in 1989 as compared to 1991 is that PCAP did not pay for delivery services. Designation as a PCAP provider might attract indigent patients that although covered for prenatal care, might not be covered for more expensive delivery and newborn care services. In January 1990, PCAP became part of the Medicaid program. The change in financing and administration along with the expansion in eligibility and streamlined enrollment procedures increased the benefits to both the state as well as its obstetrical providers. Under Medicaid, all PCAP expenditures were eligible for federal sharing, which lowered costs to the state of providing these services. For obstetrical providers, the incentives to participate were also enhanced. Reimbursement for prenatal and postpartum visits continued to exceed fees received by non-PCAP obstetric providers that served Medicaid recipients. 2 More impor-

2 New York City hospitals and free-standing clinics received about US$229 in 1991 for an initial prenatal visit and US$119 per visit thereafter. Reimbursement for ancillary services was included in these fees, which obviated more complex billing under the previous system. By design, the number of visits that could be billed was unlimited, which encouraged providers to err on the side of overutilization ŽNew York State Department of Health, 1989b..

T. Joyce r Journal of Health Economics 18 (1999) 31–67

35

tantly, however, the Medicaid eligibility expansion and enrollment reforms eliminated the risk of uncompensated delivery services that existed for PCAP providers in 1989. In 1991, any woman eligible for Medicaid was automatically eligible for PCAP. This guaranteed that all delivery and newborn costs of women in PCAP were covered by Medicaid. In addition, PCAP providers had authority to declare a client presumptively eligible for Medicaid, and were to serve as a client’s advocate in the Medicaid application process. Clients no longer had to pursue medical assistance on their own. Nor was a client’s immigration status relevant. With a minimum of organization, a PCAP provider after January 1990 could indemnify against losses from serving poor pregnant women. One indication of how much Medicaid reforms facilitated PCAP enrollment comes from the growth in deliveries covered by medical assistance and the changing characteristics of PCAP participants. The bottom row of Table 1 shows the number of births to women by year, PCAP status and Medicaid eligibility. In the third quarter of 1989, there were 4167 births Ž2354 q 1813. to women on medical assistance in New York City. By 1991, that figure had risen to 6588 Ž1190 q 5398., an increase of 58%. Even more striking is the increase in PCAP participation among women on medical assistance. Between 1989 and 1991, PCAP enrollment tripled among women on medical assistance. 3 By the third quarter of 1991, 80% of PCAP participants on medical assistance were foreign-born Žcolumn 8.. The important observation for this analysis is that the Medicaid expansion in January, 1990 stimulated a large increase in the number of women served by PCAP. If the increase in PCAP participation was exogenous to individual assessments of program benefits and individual health behaviors, it provides a useful instrument with which to identify effects of PCAP on prenatal care utilization and infant health. 2.1. Determination of PCAP participation I assume that any woman who received prenatal care from a designated PCAP provider received the appropriate PCAP services. The assumption, however, is strictly true only for 1991, since a PCAP provider was under contractual obligation to provide PCAP services to all Medicaid recipients after January 1990. In 1989, before the Medicaid eligibility expansion, a PCAP provider could only bill for

3 This is somewhat of an overstatement, since I do not know the number of women who were enrolled in PCAP, but not on Medicaid in 1989. The number is probably small since only 12% of PCAP participants on medical assistance in 1991 would have been ineligible for Medicaid in 1989.

36

Cash assistance

Medical assistance

1989

1991

1989

1991

Non-PCAP

PCAP

Non-PCAP

PCAP

Non-PCAP

PCAP

Non-PCAP

PCAP

Ž1.

Ž2.

Ž3.

Ž4.

Ž5.

Ž6.

Ž7.

Ž8.

3116 12.6 2.1 16.4

3183) ) 10.0 ) ) 1.4 13.2 ) )

3074 15.1 2.8 19.1

3209) 8.6 ) ) 1.2 ) ) 11.4 ) )

3204 9.9 1.9 12.5

3297) ) 6.1) ) 0.7 ) ) 10.3 )

3195 11.0 2.4 13.5

3287) ) 6.1) ) 0.9 ) ) 9.5 ) )

Prenatal care No. of visitsb Month care beganb No care b WIC

7.3 3.4 12.0 36.3

7.8 ) ) 3.8 ) ) 6.4 ) ) 46.6 ) )

6.8 3.2 17.1 37.3

8.3 ) ) 3.7 ) ) 5.7 ) ) 49.5 ) )

7.7 3.9 9.2 32.4

8.6 ) ) 3.9 3.7 ) ) 46.9 ) )

7.6 3.5 11.5 31.3

8.8 ) ) 3.7 ) ) 4.7 ) ) 50.1) )

Demographic (%) Whiterother non-Hispanic Black non-Hispanics Puerto Rican Dominican

10.6 44.7 26.1 11.9

5.3 ) ) 43.3 35.7 ) ) 10.3

9.4 50.0 21.4 12.0

7.9 39.8 ) ) 34.7 ) ) 11.6

19.2 31.1 9.8 14.1

12.1) ) 38.2 ) ) 12.1) 18.0 ) )

26.6 29.3 8.9 17.5

21.7 ) ) 27.8 8.9 13.3 ) )

Birth outcomes Birth weight Žg. Low birth weight Ž%. Very low birth weight Ž%. Preterm births Ž%b .

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Table 1 Mean characteristics of mothers and infants by category of Medicaid, year, and enrollment in the Prenatal Care Assistance Program ŽPCAP. a

3.8 ) 67.6 80.0 ) ) 20.0 ) ) 6.6 )

4.5 70.7 79.3 18.9 6.6

4.1 66.0 ) ) 79.2 20.3 5.2 )

22.1 31.6 52.9 18.9 13.3

16.3 ) ) 30.4 59.6 ) ) 20.8 ) ) 16.3

11.4 41.8 51.7 17.9 12.7

25.2 ) ) 20.4 ) ) 52.1) ) 14.6 ) 13.2 ) )

BehaÕioral (%) Cocaine G1r2 pack cigarettesrday

6.3 13.5

4.0 ) ) 12.8

5.8 18.1

2.4 ) ) 10.3 ) )

2.2 4.3

0.3 ) ) 3.2

3.9 7.1

0.3 ) ) 2.8 ) )

Schooling Less than high school College or more

44.2 1.5

52.8 ) ) 1.0

42.6 1.8

48.3 1.2

35.3 4.1

39.9 ) ) 3.6

27.6 6.5

35.6 ) ) 5.4

N a

2322

3676

1919

4577

2354

1813

1190

5398

Cash assistance refers to women who were eligible for Medicaid because they receive cash assistance through either AFDC, SSI, or Home Relief; Medical Assistance refers to women who were not receiving cash assistance but who were enrolled in Medicaid through the Medical Assistance Program. PCAP indicates that the woman was enrolled in the Prenatal Care Assistance Program at some point during her pregnancy. b Means based on known values only. ) p- 0.05 based on pair-wise test by PCAP status. )) p- 0.01.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

5.0 68.6 77.2 16.9 8.1

Other Hispanic Born in US Unmarried Mother’s age - 20 Mother worked during pregnancy

37

T. Joyce r Journal of Health Economics 18 (1999) 31–67

38

PCAP services if the woman had no health insurance and her family income was less than 185% of the federal poverty level. PCAP providers who treated women on Medicaid in 1989 had to bill Medicaid for prenatal services. Features of the Medicaid program in 1989, however, suggest that many women whose delivery eventually was financed by Medicaid were eligible for PCAP at the time they initiated prenatal care. One reason is because New York State did not have presumptive eligibility in 1989, and thus, a woman’s eligibility for Medicaid might not be determined until well after the initiation of prenatal care, or even after birth. Poor women who sought prenatal care from a PCAP provider could be serviced as PCAP participants until determination of their Medicaid eligibility was completed. Second, expenditures for birth could trigger ‘spend down’ eligibility for PCAP clients 4 ŽNew York State Department of Health, 1989b.. There is no information on how many Medicaid recipients received PCAP services in this manner. What is important for this analysis is that a substantial portion of women whose delivery was financed by Medicaid were probably eligible for PCAP when they initiated prenatal care. Another reason to believe that PCAP providers did not differentiate between Medicaid recipients and poor uninsured women in 1989 is that PCAP could well be described as a quality assurance initiative. The services mandated under PCAP for the most part follow American College of Obstetricians and Gynecologists ŽACOG. standards for prenatal care ŽAmerican College of Obstetricians and Gynecologists, 1989.. PCAP providers have greater incentive to follow these guidelines than do obstetric providers that have large Medicaid case loads, since the PCAP program has a designated group of auditors that oversee compliance by PCAP providers. It seems reasonable, therefore, to assume that once clinicians and administrators at PCAP facilities have established the reporting and referral infrastructure necessary to comply with PCAP, that these same procedures would be followed for other poor women whose Medicaid eligibility was uncertain and even for women already covered by Medicaid. The upshot is that I will have misclassified women as PCAP participants who in fact were non-participants if women on Medicaid in 1989 did not receive PCAP services, even though they obtained prenatal care at PCAP providers. This will bias results towards the null estimates of program impact in the 1989 cross-section, but may lead to an overestimation of program effects in the instrumental variable analysis as described in more detail below.

4

In the 1988 PCAP annual report, the authors write, ‘‘Inpatient hospital services are not reimbursed by PCAP. However, the inpatient costs for PCAP clients frequently trigger Medicaid ‘spend-down’ eligibility and facility reimbursement’’ ŽNew York State Department of Health, 1989b, p. 4.. The comment appears directed at PCAP providers concerned with the lack of coverage for delivery costs.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

39

3. Data Data used in this study are from the linkage of Medicaid administrative files with New York State birth certificates performed by the New York State Department of Social Services, Office of Health and Long Term Care. Ninety-three percent of all women on Medicaid who delivered a baby in New York State hospitals between July 1 and September 30, 1989 and July 1 through September 30, 1991 were linked to birth certificates Ž N s 43 503.. In this study, I use the subset of singleton births to New York City residents Ž N s 23 249. as provided to me by the New York State Department of Social Services, Office of Health and Long-Term Care. The data distinguish women on Medicaid, but not cash assistance, from women whose Medicaid status is linked to their participation in Aid to Families with Dependent Children ŽAFDC., Supplemental Security Income ŽSSI., and Home Relief ŽHR.. Over 85% of women on cash assistance are in AFDC. The data also distinguish women on medical assistance who became eligible because of the expanded income threshold Ž100 to 185% of federal poverty level. from women whose income would have made them eligible for medical assistance prior to the expansion. All other characteristics of mothers and infants are from the birth certificate. 3.1. Prenatal care As I discuss in more detail below, one strategy to minimize unobserved heterogeneity is to stratify women by the timing of their first prenatal care visit. Following Kotelchuck, 1994, I use four primary groupings for when care began: women who initiate care in the first four months of pregnancy; women who initiate care in months 5 and 6; women who initiate care after the sixth month; and women who have no prenatal care. Women who have no prenatal care are distinct from women for whom both the number of visits and the timing of the first visit are missing. The latter were eliminated from the analytical data set w n s 1084 Ž4.7%.x. Finally, there are women for whom the timing of the first visit is missing, but who report more than zero prenatal visits. These women are excluded from analyses based on the timing of the first prenatal visit, but are included in all other analyses. The distribution of women by the timing of the first prenatal visit cross-classified by year, PCAP status, and birth outcomes is presented in table of Appendix A. 3.2. Outcomes I examine three birth outcomes: birth weight, a continuous measure, and three dichotomous indicators, low birth weight Žless than 2500 g., very low birth weight

T. Joyce r Journal of Health Economics 18 (1999) 31–67

40

Žless than 1500 g. and term low birth weight births Žless than 2500 g and greater than 37 weeks gestation.. 5 I also examine the association between PCAP and two measures of prenatal care utilization, WIC enrollment and prenatal care visits, as well as the association between PCAP and smoking during pregnancy.

4. Hypotheses and methods The primary question of interest is the effect of PCAP participation on birth outcomes. I view the outcome equations as reduced form infant health production functions. I examine birth weight because it is a continuous, well-measured and frequently used indicator of infant health. I also use dichotomous measures of infant health because of their strong association with morbidity and mortality and their frequent use in the public health and epidemiological literature. In addition, I use several dichotomous measures of birth weight to distinguish plausible from implausible treatment effects associated with PCAP participation. Low birth weight, for example, can be broadly divided between infants born too soon and infants who grow too slow in utero. The general consensus in the clinical literature is that relatively little is known about the causes of preterm delivery ŽCollaborative Group on Preterm Birth Prevention, 1993; Gibbs et al., 1992; Hack and Merkatz, 1995.. Clinicians estimate that no more than 25% of preterm births are even theoretically preventable ŽTucker et al., 1991.. The epidemiology of fetal growth, however, offers more scope for intervention, since smoking has such a well-documented link to fetal growth ŽLi et al., 1993; Alexander and Korenbrot, 1995.. Inadequate maternal weight gain is also related to fetal growth retardation, although links between weight gain and nutrition are less consistent ŽScholl et al., 1991.. A reasonable interpretation of the clinical literature, therefore, is that one should be skeptical of associations between PCAP and Õery low birth weight, since almost all infants less than 1500 g are preterm. A more plausible association is likely to be found between PCAP participation and fetal growth given PCAP’s emphasis on health education and nutritional counseling. As a proxy for fetal

5

Gestational age is the difference between date of birth and date of the mother’s last menstrual period ŽLMP. as computed by the New York City Department of Health. If gestational age were missing or implausible, I used the clinician’s estimate of gestation. Otherwise I coded gestational age as missing Ž ns 318.. There was substantial heaping of gestational age at 40 weeks, especially in the clinician’s estimate. For this reason, I did not use gestational age as a separate outcome and limited its use to the dichotomous distinction between term and preterm births. I view very low birth weight as a good proxy for extreme prematurity since over 89% of all very low birth weight births in my sample have recorded gestations of less than 37 weeks and 68% have gestations less than 32 weeks.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

41

growth, therefore, I will test whether PCAP is associated with birth weight and the rate of low birth weight among term births only Žbirths 37 weeks gestation or greater.. In an attempt to understand the mechanisms by which PCAP might impact on infant health, I test whether women in PCAP receive more prenatal services than non-participants based on three measures: the number of prenatal care visits conditional on when care was initiated; whether the women was enrolled in WIC; and whether the woman smoked during pregnancy. I expect women in PCAP to have more prenatal visits, on average, than non-participants. The New York State Department of Health requires that PCAP providers follow-up on missed appointments and make appropriate referrals. Moreover, New York State reimburses PCAP providers on a fee-for-service basis. As noted above, fees are relatively generous compared with payments received by non-PCAP obstetric providers who serve Medicaid recipients. The upshot is that PCAP’s service protocol and financial incentives should encourage visits. I also expect women in PCAP to have higher rates of WIC participation than women not in PCAP. By contract, PCAP providers are to offer nutritional counseling and to assist clients, when appropriate, with enrollment in WIC. Another reason to assess the impact of WIC on PCAP is that, should I find a positive association between PCAP and fetal growth, then a plausible mechanism would be differences in nutritional services as proxied by WIC ŽRush et al., 1988; Devaney et al., 1992.. Finally, I test whether PCAP participants are less likely to have smoked at least half a pack of cigarettes per day during pregnancy than non-PCAP participants. This is a crude measure of smoking and any association with PCAP should be interpreted with caution. For instance, I cannot distinguish between women who never smoked at all or women who smoked less than half a pack per day during pregnancy. In addition, the smoking indicator does not differentiate between women who smoked heavily for some portion of the pregnancy, but then stopped, from women who smoked more than half a pack per day throughout pregnancy. Despite these limitations, smoking is a robust risk factor for fetal growth retardation and any association with PCAP participation would be suggestive of a potentially causal mechanism. 4.1. Selection bias The main concern in the assessment of PCAP is selection bias. Do women who participate in PCAP have superior pre-pregnancy health, experience less stress, have fewer vaginal infections and are they less likely to engage in unhealthy behaviors than non-participants of similar income, schooling, race, age, marital status and reproductive history? If hard to measure factors that affect infant health

42

T. Joyce r Journal of Health Economics 18 (1999) 31–67

also affect PCAP participation, then I may attribute gains to PCAP that in fact reflect these unobserved differences. I take several approaches to selection bias. On the demand side, I seek a set of controls that are likely to be similar in both observed and unobserved characteristics as women served by PCAP providers, the treatment group. I begin by stratifying all analyses by Medicaid category. I separate women on Medicaid through their enrollment in a cash assistance program ŽAFDC, SSI and Home Relief. from women who receive Medicaid coverage through the medical assistance program. Previous work has ignored this distinction. As noted above, there are large observed differences between women who receive cash assistance and those whose benefits are limited to medical assistance, the primary difference being nativity Žsee Table 1.. The superior birth outcomes of foreign as compared to US-born women of similar socioeconomic status is a consistent empirical finding that is not well understood ŽSingh and Yu, 1996.. In addition, neither income nor family structure guidelines for AFDC were altered between 1989 and 1991, whereas the Medicaid expansion increased the income eligibility threshold for medical assistance from approximately 100 to up to 185% of the federal poverty level. Newly eligible recipients may exacerbate heterogeneity among women on medical assistance. I also stratify the analysis by year. In previous studies of enhanced prenatal care services, researchers have used a single cross-section of births to assess effectiveness. I have two cross-sections, separated by a major expansion in PCAP participation. The treatment protocol for PCAP participants, however, did not change between 1989 and 1991, and thus, all else constant, I would not expect average treatment effects associated with PCAP to vary statistically between the two years. If treatment effects differ between 1989 and 1991, then there may have been a non-random response to the eligibility expansion. PCAP, for instance, may have attracted women at relatively low risk for adverse outcomes among the newly eligible, which, all else constant, would increase average treatment effects relative to 1989. This is more likely to occur among women on medical as compared to cash assistance given the rise in eligibility thresholds for the former. Treatment effects may also vary between 1989 and 1991 because of misclassification of PCAP participants in 1989. I would expect treatment effects associated with PCAP to be greater in 1991 than 1989, if in fact, women who received prenatal care from PCAP providers in 1989 did not receive the appropriate PCAP services. In sum, I interpret a lack of temporal homogeneity of treatment effects as a warning of potential biases. A third strategy to control selection is to interact PCAP participation with the timing of the first prenatal care visit. I am particularly interested in comparisons between PCAP participants who initiate care in the first four months of pregnancy with non-participants who also begin care early. I believe that the timing of the first prenatal care visit reflects concern for the pregnancy and a positive attitude toward primary care. More generally, comparisons based on the timing of first

T. Joyce r Journal of Health Economics 18 (1999) 31–67

43

visit may distribute other hard-to-measure determinants of birth outcomes such as stress and unhealthy behaviors more equally between PCAP participants and non-participants. 6 4.2. Instrumental Õariables I also attempt to correct for demand side bias using an instrumental variables approach. I use growth in PCAP providers associated with the Medicaid eligibility expansion as an exogenous rise in availability that increased PCAP participation. 7 I proxy changes in PCAP availability in three ways. First, I use the number of PCAP provider sites in 1988 and 1992 by health area. 8 I have identified 66 PCAP sites in 1988 and 80 in 1992, although this may be an undercount. 9 As additional instruments, I use interactions between health district and year to capture general changes in PCAP availability. As a final strategy, I use interactions between year and hospital of delivery as another proxy for the change in availability of PCAP services following the Medicaid eligibility expansion in January, 1990. By using hospital of delivery, I implicitly include their affiliated outpatient sites. In addition, 13 out of 41 hospitals with newborn nurseries became PCAP providers between 1989 and 1991 and as I show below, there was dramatic growth in the percentage of PCAP participants who delivered at these hospitals over this period. For the IV procedure to be valid, there must be no direct effect of increases in the availability of PCAP providers on infant health. This would be violated, for example, if outreach by providers and the State Department of Health induced favorable selection among pregnant women towards PCAP providers. It must be the case, therefore, that the distribution of unobserved risk factors within groups of women being compared is unaffected by the growth in PCAP sites. This would be true if women continue to access their local or regular obstetric provider regardless

6 Reichman and Florio, 1996, for example, control for the timing of the first visit in regression of birth weight on receipt of comprehensive prenatal services. The specification assumes, however, that effects of augmented prenatal care are the same regardless of when care begins. An interactive specification obviously relaxes this restriction. As I show below, the largest treatment effects associated with PCAP are for women who begin care after the seventh month of pregnancy, a dubious result highly suggestive of selection bias. 7 In other words, the IV strategy assumes no supply-side selection. Reichman and Florio, 1996, make a similar argument with respect to Healthstart in New Jersey. But theirs is a single cross-section and thus, relies on a static measure of availability to identify participation. 8 New York City is divided into 352 health areas which aggregate up into 30 health districts. 9 The list of provider sites comes from several sources. The New York State Department of Social Services, Division of Health and Long Term Care provided me with a list of PCAP proÕiders based on claims for PCAP services in 1989 and 1991. Each provider was listed only once regardless of how many outpatient sites it maintained. I obtained separate lists of PCAP sites from the New York State and New York City Departments of Health in 1988 and 1992. There were differences among the lists.

44

T. Joyce r Journal of Health Economics 18 (1999) 31–67

of a change in its PCAP status, a reasonable assumption since women are enrolled by PCAP providers at the first prenatal care visit. Unlike AFDC, for example, women do not apply for PCAP at a central office. A woman on Medicaid automatically receives PCAP services by virtue of her provider’s status as a PCAP facility. 10 Furthermore, by 1991, PCAP providers had authority to make decisions regarding presumptive eligibility for Medicaid. Thus, any uninsured woman who initiated care with a PCAP provider could be deemed presumptively eligible for Medicaid and enrolled in PCAP at the initial visit. The fact that PCAP grew so much between 1989 and 1991 among foreign-born women, most of whom would have been eligible prior to the expansion, suggests that increased availability of PCAP providers and streamlined enrollment procedures were important to the rise in participation. Another potential difficulty with the IV strategy is the possibility of omitted variable bias. As Moffitt Ž1995. points out, IV estimates identified by interactions between health district and year, or hospital and year are equivalent to coefficients obtained from aggregate regressions of health district changes or hospital-specific changes in mean birth weight on aggregate changes in PCAP participation, if no individual-specific covariates are included. If, for example, growth in PCAP providers occurred in health districts with diminishing drug use, or if the hospitals that chose to participate were of higher quality than those that did not, then validity of the estimates is potentially compromised. Finally, I must demonstrate that interactions between year and health district or year and hospital have appreciable explanatory power for PCAP participation, or I increase the likelihood of finite sample bias ŽBound et al., 1995.. Given these considerations, I interpret IV estimates cautiously.

5. Results Selected characteristics of women and infants by Medicaid status, PCAP participation and year are shown in Table 1. Women who receive cash as opposed to only medical assistance differ substantially along a host of demographic and behavioral measures. Women on cash assistance are more likely to be black non-Hispanic or Puerto Rican, less likely to be married, more likely to have been born in the US, have less schooling, and are more likely to have smoked or to have used cocaine during pregnancy than women who receive medical assistance only. Not surprisingly, there are large differences in birth outcomes. Women who

10

As discussed above, a woman on Medicaid in 1989 may have received prenatal care at a PCAP facility, but not have received the full range of PCAP services.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

45

receive cash assistance have lower mean birth weight and higher rates of low birth weight Žcolumns 1–4. than women on medical assistance Žcolumns 5–8.. Differences in prenatal care are more muted. WIC participation and timing of the first prenatal visit are similar whereas mean visits differ; the latter can probably be explained by the greater proportion of women with no prenatal care among cash recipients. When I examine differences by PCAP status within Medicaid categories over time, I observe that differences in birth outcomes between participants and non-participants grew substantially in two years. In 1989, the rate of low birth weight was 10.0% among PCAP participants on cash assistance and 12.6 among non-participants Žcolumns 1 and 2.. Statistically significant differences also exist for mean birth weight. By 1991, these differences had increased. The rate of low birth weight rose from 12.6 to 15.1% between 1989 and 1991 among infants of cash recipients not in PCAP Žcolumns 1 and 3., and fell among infants of PCAP participants Žcolumns 2 and 4.. Shifts are so dramatic that by 1991 there is a 6.5 percentage point difference in the rate of low birth weight, and a 1.8 percentage point difference in the rate of very low birth weight between PCAP and non-PCAP participants on cash assistance in 1991 Žcolumns 3 and 4.. The same pattern exists for women on medical assistance Žcolumns 5–8.. As PCAP participation increased between 1989 and 1991, rates of low and very low birth weight increased among non-participants Žcolumns 5 and 7.. The worsening of outcomes among the shrinking pool of eligible non-participants appears explicable to some extent by changes in observable characteristics. The proportion of US-born women, the proportion of infants exposed to tobacco, and the proportion of women with no prenatal care all rose among non-participants of PCAP between 1989 and 1991. I turn, therefore, to the multivariate analysis in order to adjust for measured differences. 5.1. MultiÕariate estimates of PCAP and birth outcomes Average treatment effects of PCAP participation on three outcomes are displayed in Table 2: birth weight measured in grams and dichotomous indicators of low birth weight and very low birth weight. 11 Each figure is from a separate regression estimated for each year and category of Medicaid. 12 Estimates are

11 The average treatment effect in the birth weight regression is simply the coefficient on the PCAP indicator. For the three binary outcomes estimated as probits, the average treatment effect is the mean of the difference in predicted probabilities for each woman assuming that she participated and then did not participate in PCAP. I used the delta method to obtain standard errors. 12 An F-test decisively rejected the null of homogeneity of coefficients between Medicaid categories Ž F24,A s 4.23; p- 0.01.. I also reject homogeneity by year for women on cash assistance Ž F23,A s1.72; p- 0.025., but not for women on medical assistance Ž F24,A s1.00..

46

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Table 2 Adjusted differences in birth weight Žg. and rates of low birth weight, very low birth weight and preterm birth Žin% pts. associated with participation in the Prenatal Care Assistance Program ŽPCAP. by Medicaid status and year of delivery, New York City 1989 and 1991a,b PCAP specification

Dependent variable: birth weight Žg. Cash assistance 1989

Sample includes women 69) Ž16. with no prenatal care Sample excludes women 47) Ž17. with no prenatal care Term births only Ž ) 36 wks., 31)) Ž15. excludes no care

Medical assistance 1991

1989

1991

123) Ž16.

101) Ž18.

78) Ž19.

83) Ž17.

78) Ž18.

33))) Ž19.

51) Ž16.

54)) Ž16.

14 Ž17.

Dependent Õariable: low BW rate (in % pts.) Cash Assistance 1989

Medical Assistance 1991

1989

Sample includes women y2.3) Ž0.9. y5.7) Ž0.9. y3.9) Ž0.8. with no prenatal care Sample excludes women y1.5))) Ž0.8. y3.7) Ž1.0. y2.9) Ž0.8. with no prenatal care Term births only Ž ) 36 wks., y0.1 Ž0.6. y1.8)) Ž0.8. y1.8) Ž0.6. excludes no care

1991 y3.8) Ž1.0. y2.0)) Ž0.9. y1.5))) Ž0.9.

Dependent Õariable: Õery low BW rate (in % pts.) Cash assistance

Sample includes women with no prenatal care Sample excludes women with no prenatal care

Medical assistance

1989

1991

1989

y0.5 Ž0.3.

y1.1) Ž0.4.

y1.6) Ž0.50. y1.6) Ž0.50.

1991

y0.4 Ž0.3.

y0.4 Ž0.3.

y0.4 Ž0.3.

y1.3)) Ž0.5.

a

For birth weight, figures represent differences in grams; for low birth weight and very low birth weight, figures reflect average treatment effects in percentage points. Standard errors are in parentheses. We estimate birth weight regressions by ordinary least squares and all dichotomous outcomes by probit. Regressions include controls for maternal age, racerethnicity, nativity, previous fetal loss, maternal schooling, paternal schooling, marital status, sex of the infant and census tract poverty rates. b ) p- 0.01; )) p- 0.05; ))) p- 0.10.

adjusted for maternal age, racerethnicity, mother’s schooling, father’s schooling, marital status, parity, infant’s sex and census tract poverty rate. For births that occurred in 1991 among women on medical assistance, I use an additional indicator of whether the woman qualified for Medicaid because her family income

T. Joyce r Journal of Health Economics 18 (1999) 31–67

47

was between 100 and 185% of the federal poverty level. Standard errors are in parentheses. For each outcome I show, by row, estimates obtained from different samples. The first sample includes all births, the second excludes women with no recorded prenatal care, and the third sample includes term births only. 13 Within Medicaid category and year, improvements in all birth outcomes associated with PCAP are greatest in samples that include women with no prenatal care. Moreover, the relative fall in treatment effects from exclusion of those with no prenatal care is roughly similar across Medicaid categories and year. For instance, among women on cash assistance in 1989, the adjusted mean difference in birth weight associated with PCAP is 69 g Žcolumn 1.. This falls to 47 g when those with no care are excluded, a decline of 33%. Among women on cash assistance in 1991, the average treatment effect associated with PCAP is 123 g, but only 83 g when I eliminate women with no care, again a fall of almost a third. One explanation for the decline is that women in PCAP but with no prenatal care are probably misclassified, since PCAP designation indicates that the State received a claim for prenatal care from a PCAP provider. On the other hand, women not in PCAP and with no recorded prenatal care are more likely to be women who actually received no care during pregnancy. By including women with no prenatal care, I add to the controls—the non-PCAP participants—a disproportionately large group of women at risk for adverse outcomes. 14 Since the objective of the study is to assess enhanced prenatal care services, I view women with no prenatal care as poor controls for women in PCAP. Several patterns emerge when I focus on outcomes from samples that exclude women with no prenatal care. First, within Medicaid categories, average treatment effects do not vary statistically by year, although some differences are large in magnitude. 15 Thus, averaged across columns, I find that PCAP is associated with a statistically significant effect on birth weight of about 55 g and a decrease in the rate of low birth weight of 2.0 percentage points. Second, when I exclude preterm births, treatment effects fall by approximately half. Third, with one exception, I find no effect of PCAP on rates of very low birth weight. To this point, therefore, findings are roughly equivalent to those reported in the literature. I view the lack of a consistent relationship between PCAP and very low

13 Women enrolled in PCAP but with no recorded prenatal visits on the birth certificate are clearly misclassified. To be designated a PCAP enrollee, administrators in the New York State Department of Social Services would have to have received a claim for payment by a PCAP provider. Given the well-documented recording errors among birth certificates, the misclassification probably occurred among vital records ŽPiper et al., 1993.. 14 Seventeen percent of infants to women not in PCAP and with no prenatal care, for instance, were exposed to cocaine as compared to 9.5% of infants to mothers in PCAP but also with no prenatal care. 15 This is based on a z-score in which the standard error is the square root of the sum of the variances for each estimate.

48

T. Joyce r Journal of Health Economics 18 (1999) 31–67

birth weight favorably since I am skeptical that PCAP could have a meaningful impact on this subgroup of preterm infants. The lack of a statistical difference in average treatment effects by year is some evidence that misclassification of PCAP participants in 1989 might not be substantial. I now turn to results based on specifications in which I interact PCAP participation with the timing of the first prenatal visit. 5.2. Effects of PCAP stratified by when care began I now examine differences by PCAP status and timing of the first visit in a multivariate context. Specifically, I run the following regressions: W s X b q d 1T 1 q d 2 Ž PT1 . q d 3T 2 q d4 Ž PT2 . q d 5 Ž PT3 .

Ž 1.

Let BW be the birth outcome and X the demographic and obstetrical determinants of birth used previously. Let T1 stand for women who begin prenatal care in the first four months of pregnancy and who are not in PCAP; let PT1 be women who also begin prenatal care in the first four months of pregnancy and who are enrolled in PCAP. Similarly, let T2 represent women who begin care in months five and six of pregnancy, and let PT2 be those who begin care also in months five and six but who also participate in PCAP. Let PT3 be women who begin care in months 7–10 and who are enrolled in PCAP. The reference category is women who begin care late, months 7–10 and who do not participate in PCAP. Lastly, b and d i are coefficients. I exclude women with no prenatal care, as well as women with positive prenatal care visits, but for whom the timing of the first visit is unknown. Table in Appendix A shows birth outcomes stratified by payor status, year, timing of the first prenatal care visit and PCAP participation. I am not interested in individual coefficients per se since differences with respect to the reference category are uninformative. Instead, I focus on mean differences and average treatment effects between women who initiate care around the same time during pregnancy. 16 Specifically, I am interested in the magnitude and significance of d 2 – d 1 , d4 – d 3 , and d 5, using the notation from Eq. Ž1.. I present these differences for birth weight in Table 3 by Medicaid status, year and PCAP participation. Panel A shows results from the basic specification and Panel B includes only term births Žgreater than 36 weeks gestation.. The first column of Panel A shows results for women on cash assistance in 1989. I find that women in PCAP who begin prenatal care in the first four months of pregnancy have infants 53 g heavier on average than women who also begin 16 As before, average treatment effects for dichotomous outcomes are the percentage point differences in predicted probabilities obtained by averaging individual predictions across all women. Standard errors are obtained by the delta method.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

49

Table 3 Adjusted differences in birth weight Žg. associated with participation in the Prenatal Care Assistance Program ŽPCAP. based on comparisons of women who initiate care at the same point during pregnancy by Medicaid status and year of delivery, New York City 1989 and 1991a,b Panel A

Basic specification Cash assistance

Medical assistance

1989 ns 4713

1991 ns 5038

1989 ns 3475

1991 ns 5533

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

Prenatal care began Months 1–4 Months 5–6 Months 7))

53)) Ž21.2. 27 Ž36.7. 20 Ž48.2.

83) Ž22.2. 70 Ž37.2. 144) Ž53.7.

55)) Ž24.0. 124) Ž37.9. 84 Ž50.5.

40))) Ž24.1. y25 Ž41.6. 74 Ž55.1.

Panel B

Term births only Ž37 or more weeks gestation. Cash assistance

Prenatal care began Months 1–4 Months 5–6 Months 7))

Medical assistance

1989 n s 4113

1991 ns 4445

1989 ns 3114

1991 ns 5015

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

46)) Ž20.0. y1 Ž33.7. 52 Ž44.0.

36 Ž21.0. 14 Ž34.4. 116)) Ž49.3.

34 Ž22.3. 112)) Ž34.7. 84))) Ž45.4.

y3 Ž22.4. y34 Ž37.3. 58 Ž49.2.

a

Figures in panel A and B show birth weight differences in grams for infants born to women enrolled in PCAP vs. infants born to women not in PCAP both of whom began prenatal care at approximately the same point of pregnancy adjusted for maternal age, racerethnicity, nativity, previous fetal loss, maternal schooling, paternal schooling, marital status, sex of the infant and census tract poverty rates. b Standard errors in parentheses. ) p- 0.01; )) p- 0.05; ))) p- 0.10.

care early, but who are not in PCAP. Looking across the top row of Panel A, 50 g is about the average difference in adjusted mean birth weight. The exception is the comparison of women on cash assistance in 1991, in which the mean difference is 83 g. Comparisons between women who begin care in the fifth or sixth month of pregnancy, or those who begin care after the sixth month are less consistent and measured imprecisely. The largest effect, 144 g, is for women on cash assistance in 1991 who begin care late. Estimates in Panel B show differences in birth weight by PCAP participation among term births only, a proxy for fetal growth. Specifically, in 1989 women on cash assistance who initiated prenatal care in the first four months of pregnancy had infants at term that weighed 46 g heavier, on average, than similar women who also began care early but who were not served by a PCAP provider. Differences in adjusted mean birth weight for other women who began care early are substantially smaller and statistically insignificant. Thus, for women who initiate care early, the primary difference in mean birth weight appears related to a decreased incidence of preterm birth. This is not true for women who begin care

50

T. Joyce r Journal of Health Economics 18 (1999) 31–67

after the fourth month; differences in adjusted mean birth weight by PCAP participation are essentially unaffected by elimination of preterm births. In Table 4, I repeat the analysis from Table 3, but with the rate of low birth weight, the rate of term low birth weight and the rate of very low birth weight as

Table 4 Average treatment effects in percentage points on low birth weight, term low birth weight and very low birth weight associated with participation in the Prenatal Care Assistance Program ŽPCAP. based on comparisons of women who initiate care at the same point during pregnancy by Medicaid status and year of delivery, New York City 1989 and 1991a,b Panel A

%Low birth weight: basic specification Cash assistance 1989 ns 4713

1991 ns 5038

1989 ns 3475

1991 ns 5533

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

y3.2) Ž1.2. y4.1)) Ž2.1. y11.3) Ž3.4.

y3.4) Ž1.1. y2.7 Ž1.9. y3.4))) Ž1.8.

y2.7)) Ž1.9. y0.7 Ž2.0. y0.7 Ž2.2.

Prenatal care began Months 1–4 y2.4)) Ž1.1. Months 5–6 y1.0 Ž1.9. Months 7)) y0.6 Ž2.1. Panel B

Medical assistance

%LBW: Term births only Ž37 or more wks gestation. Cash assistance 1991 ns 4445

1989 ns 3114

1991 ns 5015

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

y0.4 Ž1.0. y0.5 Ž1.6. y8.8 Ž3.2.

y1.7)) Ž0.8. y2.2 Ž1.6. y2.9 Ž1.5.

y0.8)) Ž0.9. y0.7 Ž1.6. y0.6 Ž1.9.

Prenatal care began Months 1–4 y1.7 Ž0.9. Months 5–6 y2.7))) Ž1.6. Months 7)) y2.2 Ž2.0. Panel C

Medical assistance

1989 ns 4113

%Very low birth weight: basic specification Cash assistance 1989 PCAP–nonPCAP

Prenatal care began Months 1–4 y0.5 Ž0.5. Months 5–6 y0.2 Ž0.5. Months 7)) y0.1 Ž0.6. a

Medical assistance 1991 PCAP–nonPCAP

1989 PCAP–nonPCAP

1991 PCAP–nonPCAP

y0.7 Ž0.5. y2.0)) Ž1.0. y1.2 Ž1.2.

y0.6 Ž0.4. y2.6) Ž0.8. y0.7 Ž0.5.

y1.8) Ž0.7. 0.1 Ž0.6. 0.1 Ž0.5.

Figures in panel A show average treatment effects in percentage points for infants born to women enrolled in PCAP vs. infants born to women not in PCAP both of whom began prenatal care at approximately the same point during pregnancy as estimated by probits. Models include controls for maternal age, racerethnicity, nativity, previous fetal loss, maternal schooling, paternal schooling, marital status, sex of infant and local area poverty rates. Panel B displays average treatment effects for term births only and Panel C is for very low birth weight births. b Standard errors are in parentheses. ) p- 0.01; )) p- 0.05; ))) p- 0.10.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

51

outcomes. Overall, results are concordant with the general conclusions drawn from the analysis of birth weight as a continuous variable. There is no consistent association between PCAP participation and very low birth weight among women who initiate care in the first four months of pregnancy ŽPanel C.. I do show differences of approximately 2.8 percentage points in the rate of low birth weight averaged across year and payor among women who begin care early ŽTable 4, Panel A, row 1., but these estimates are reduced by about half when I exclude preterm births ŽTable 4, Panel B, row 1.. There is, however, an anomalous finding. The difference in low birth weight by PCAP status between women on cash assistance in 1991 who begin prenatal care in the seventh month or later is unreasonably large, 11.3 percentage points, and remains large even after elimination of preterm births ŽPanel B, column 2, row 3.. Summary data in table of Appendix A indicate that outcomes for this subgroup are extreme, even for women who begin prenatal care late. For instance, the rate of low birth weight among non-PCAP women on cash assistance who begin care in the seventh month or later is 15.6% in 1991—2.3 times greater than the rate of low birth among the same category of women in 1989, and almost three times greater than the rate for infants of non-PCAP women on medical assistance who also began care late in 1989 and 1991 Žsee table of the Appendix A, column 6.. Moreover, there are relatively few observations in the late care cells and therefore, treatment effects for this subgroup should be interpreted cautiously. In sum, I have tried to lessen differences between PCAP and non-PCAP women due to unobserved heterogeneity by interacting PCAP participation with the timing of the first prenatal visit. My preferred specification is the comparison by PCAP status of women who begin care early, since they are arguably the most motivated or most risk-averse women in the sample. Based on these comparisons, PCAP is associated with a gain of approximately 55 g in mean birth weight and 30 g in the birth weight of term infants. There is little association between PCAP and the probability of a very low birth weight birth. I now assess whether differences by PCAP participation in prenatal care visits, WIC enrollment and smoking during pregnancy can explain the increase in birth weight 5.3. How PCAP affects outcomes Table 5 displays differences in the average treatment effects of PCAP on WIC enrollment, smoking during pregnancy, and prenatal care visits. All estimates have been adjusted for demographic and socioeconomic characteristics of women. I show that women in PCAP, as compared to non-participants, are between 8 and 14 percentage points more likely to enroll in WIC, a relative increase of about 20%. There are no differences in smoking during pregnancy associated with PCAP except for women on cash assistance in 1991. Finally, women in PCAP obtain approximately 0.5 visits more than women not in PCAP, a difference of about 6% based on a mean of eight visits.

52

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Table 5 Average treatment effects associated with participation in the Prenatal Care Assistance Program ŽPCAP. on WIC enrollment, smoking during pregnancy and prenatal care visits by Medicaid status and year of delivery, New York City 1989 and 1991a,b Cash Assistance

Dependent Õariable WIC Enrollment Ž% pts.. Smoked 1r2 packrday Ž% pts. Prenatal care visits

Medical Assistance

1989

1991

1989

1991

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

8.2) Ž1.4.

10.5) Ž1.5.

9.4) Ž1.5.

14.0) Ž1.3.

.0.4 Ž0.9.

y5.1) Ž1.2.

y.0.3 Ž0.6.

y0.7 Ž0.5.

0.23)) Ž0.11.

0.80) Ž0.12.

0.41 Ž0.12.

0.67) Ž0.13.

a

Estimates of average treatment effects for WIC and smoking were obtained from probit analysis and prenatal visit by ordinary least squares. All results have been adjusted for maternal age, racerethnicity, nativity, previous fetal loss, maternal schooling, paternal schooling, marital status, sex of infant, local area poverty rates, and health district of residence. Average treatment effects for prenatal visits have also been adjusted for the month in which prenatal care began. b Standard errors are in parentheses. ))) p- 0.10; )) p- 0.05; ) p- 0.01.

To understand whether observed differences in prenatal behaviors by PCAP status explain differences in birth weight, I re-ran the regressions in Table 2 for the sample that excludes women with no prenatal care Žsecond row in each panel of Table 2.. A fall in average treatment effects of PCAP on birth weight with the inclusion of these inputs would be consistent with the interpretation that PCAP operates on birth outcomes by modifying such behaviors. Table 6 presents average treatment effects associated with PCAP obtained from specifications identical to those in Table 2, but with smoking, prenatal visits, and WIC participation included. With one exception, coefficients on all three behaviors have the expected sign and are statistically significant Žresults not shown.. The coefficient on smoking is between 143 and 202 g which is consistent with epidemiological studies ŽStein and Kline, 1983.. The coefficient on WIC is between y5 and 70 g, substantially less than effects reported by Devaney et al., 1992. Despite the inclusion of these behaviors, effects on birth weight associated with PCAP participation are essentially unchanged from those obtained in Table 2. The exception is women on cash assistance in 1991; here, I find that average treatment effects associated with PCAP fall by about one-third when smoking, prenatal visits, and WIC are included. In sum, women in PCAP are more likely to be enrolled in WIC and receive somewhat more prenatal care visits than do non-participants of PCAP. These differences, however, are largely unrelated to the gains in birth weight associated with PCAP. These regressions, therefore, fail to suggest potential mechanisms by

T. Joyce r Journal of Health Economics 18 (1999) 31–67

53

Table 6 Average treatment effects associated with participation in the Prenatal Care Assistance Program ŽPCAP. on birth weight, low birth weight and very low birth adjusted for smoking, WIC, and prenatal care visits by Medicaid status and year of delivery, New York City 1989 and 1991a,b Dependent variable

Birth weight Žg. Low birth weight Ž% pts.. Very low birth weight Ž% pts..

Cash assistance 1989 ns 5181

1991 ns 5597

Medical assistance 1989 ns 3713

1991 ns 5993

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

PCAP–nonPCAP

45) Ž16.

54) Ž17.

69) Ž18.

25 Ž19.

y1.3 Ž0.8.

y2.4) Ž0.9.

y2.5) Ž0.8.

y1.8))) Ž0.9.

y0.4 Ž0.3.

y.0.7))) Ž0.4.

y0.9) Ž0.3.

y1.2)) Ž0.5.

a

Figures are average treatment effects of PCAP participation on birth outcomes. Estimates were obtained from probit analysis for low and very low birth weight. In addition to smoking, WIC and prenatal care visits, all results have been adjusted for maternal age, racerethnicity, nativity, previous fetal loss, maternal schooling, paternal schooling, marital status, sex of infant and local area poverty rates. b Standard errors are in parentheses. ) p- 0.01; )) p- 0.05; ))) p- 0.10.

which to explain adjusted mean differences in birth outcomes associated with PCAP participation. I turn, therefore, to estimates of the effect of PCAP participation on birth weight obtained by instrumental variables. 5.4. Instrumental Õariables To this point, I have treated PCAP participation as exogenous. I have tried to limit unobserved heterogeneity between treatment and control groups by stratifying analyses by year and Medicaid category and interacting PCAP participation with timing of the first prenatal visit. I have also included measures of prenatal behavior. In this section, I present results based on instrumental variables. I pool data by year but not Medicaid category in order to exploit growth in PCAP availability as a result of Medicaid reforms in January 1990. A general description of the IV model is as follows: PCAPs X d 1 q H d 2 q d 3YR q Ž H = YR . d4 q d 5 PS q u

Ž 2.

BW s X b 1 q H b 2 q b 3YR q a 1 PCAPq Õ

Ž 3.

PCAP is a dichotomous indicator of participation; BW represents a birth outcome and X represents demographic and obstetrical characteristics. H is a matrix of either health district or hospital fixed effects; YR is a year dummy and ŽH = YR. is the set of interactions, and PS is the number of PCAP provider sites

54

T. Joyce r Journal of Health Economics 18 (1999) 31–67

by year and health area. As in Eq. Ž1., I focus on two outcomes, birth weight and low birth weight. I use two-stage least squares when birth weight is the outcome and Murphy and Topel Ž1985., two-step estimator for low birth weight. For the latter, I estimate separate probits for Eqs. Ž2. and Ž3., use the predicted probability of PCAP participation from Eq. Ž2. as a regressor in Eq. Ž3., and then adjust the standard errors in Eq. Ž3. to allow for cross-equation residual correlation. 17 I present results from two separate IV models. In the first, I instrument PCAP participation with a set of health districtsryear interactions wH = YRx and the number of PCAP provider sites ŽPS. by health area and year. 18 As noted previously, interactions between health district and year serve as a proxy for increases in the availability of PCAP services and are meant to augment the number of provider sites which I believe are underestimated Žsee footnote 9 .. For women on medical assistance, there was large growth in PCAP participation, and it varied substantially by health district. Among women on cash assistance, growth was more modest given the relatively high rates of AFDC women served by PCAP providers in 1989. The second IV specification uses interactions between year and hospital of delivery to instrument for PCAP participation. Again, the objective is to proxy changes in the availability of PCAP services. I assume that many poor women who deliver at a hospital receive prenatal services at the facility’s outpatient or satellite clinics. 19 Between 1989 and 1991, New York City hospitals could be categorized in three ways: those that never obtained PCAP status; those that were always PCAP providers; and those that changed PCAP status. Table 7 displays key characteristics by hospital type, Medicaid category, and year. In general, PCAP participation rises in hospitals that changed status and is stable in hospitals whose status did not change. 20 Despite the large increase in PCAP participation, there is

17 The estimated treatment effects obtained by linear two-stage least squares when low birth weight is the outcome were very close to those obtained using Murphy and Topel’s two-step estimator: y.0.3 percentage points Žses 2.4. for women on cash assistance and y2.7 percentage points Žses 2.2. for women on medical assistance. 18 New York City subdivides health districts into health areas. There are 30 health districts and 352 health areas. 19 For reasons of confidentiality, I was not given information as to where a woman received her prenatal care. Birth certificates, however, indicate the hospital of delivery. Thus, I knew whether a woman received PCAP services and whether the hospital she delivered at was a PCAP provider. From data in Table 7, it is clear that a woman could have received prenatal services from a PCAP provider but delivered at a non-PCAP hospital. From the perspective of an ‘experiment’, as long as referral patterns for deliveries from free-standing clinics did not appreciably change with the Medicaid expansion, comparisons between treatment and control hospitals should be useful. Appendix C lists the hospitals by PCAP status. 20 For women on medical assistance, there is a substantial rise in PCAP participation among hospitals that were always PCAP providers. This is due primarily to the growth in Medicaid financed births resulting from the eligibility expansion in 1990.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

55

Table 7 Changes in PCAP participation, birth weight and rates of low birth weight by hospital of delivery, year and Medicaid category, New York City, 1989–1991a Cash assistance

a of births % in PCAP Mean birth weight % low BW

a of births % in PCAP Mean birth weight % low BW a

Never a PCAP hospital

Always a PCAP hospital

Changed to a PCAP hospital

1989

1991

1989

1991

1989

1991

757 31.3 3207

761 19.5) 3202

3120 80.5 3183

3117 84.2) 3183

1040 37.4 3239

1423 86.5) 3248

8.6

10.4

10.1

9.3

7.4

7.9

Medical assistance Never a PCAP Hospital

Always a PCAP Hospital

Changed to a PCAP Hospital

1989

1991

1989

1991

1989

1991

352 22.4 3261

588 25.9 3321

2243 58.1 3270

3037 91.0) 3297)))

1028 25.8 3289

2187 87.6) 3272

7.1

5.3

6.9

6.0

7.4

6.4

) p- 0.01; )) p- 0.05; ))) p- 0.10.

no change in mean birth weight or the rate of low birth weight between 1989 and 1991 by type of hospital. Nor does a simple difference-in-difference analysis comparing changes in birth outcomes among hospitals that changed PCAP status with outcomes among hospitals whose PCAP status remained unchanged reveal any effects associated with PCAP. 21 Single equation and IV estimates of a 1 from Eq. Ž3. are displayed in Table 8. Columns headed by IV-A use health districtryear and health area PCAP provider sites as instruments. Columns headed as IV-B use interactions of year and hospital of delivery as instruments. For women on medical assistance, there is no associa-

21 Selection by either providers or women would undermine inferences based on the figures in Table 7. As to the quality of providers, there is no obvious evidence of selection. Of the 40 hospitals listed in Appendix C, 11 or just under a quarter are municipal hospitals. Of the hospitals that became PCAP providers between 1989 and 1991, 3 out of 13, or less than a quarter are municipal facilities. There is, however, an increase of 383 births Ž37%. to women on cash assistance among hospitals that changed 2 PCAP status ŽTable 7, top panel, last two columns.. The change is statistically significant w xŽ1. s 45.2x. Sixty-one percent of the increase, however, occurred among two facilities that both more than tripled the number of deliveries to women on cash assistance. I re-estimated the change in low birth weight in Table 7 excluding these two facilities and the results changed inconsequentially. The lack of differences in birth outcomes between 1989 and 1991 among hospitals that changed PCAP status appears unrelated to the rise in births to women on cash assistance in a few facilities.

56

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Table 8 Differences in birth weight Žg. and rates of low birth weight Žin % pts. associated with participation in the Prenatal Care Assistance Program ŽPCAP. as estimated by ordinary least squares ŽOLS., probit, and instrumental variables ŽIV. by Medicaid status in New York City, pooled 1989 and 1991a,b Dependent variable: birth weight Žg. Cash assistance

PCAP Žyess1.

Medical assistance c

c

OLS

IV-A

IV-B

62) Ž11.8.

88 Ž82.7. 0.2 41.0 8.1)

30 Ž47.6. 0.9 47.4 51.3)

Hausman test Ž t-statistic. x 2 over-id test d F-test of excluded instrumentsd

OLS

IV-Ac

IV-B c

43) Ž12.0.

y38 Ž67.8. y2.3 33.0 12.9)

y31 Ž48.7. y2.9 58.2)) 36.3)

Dependent variable: LBW rate Žin % pts.. Cash assistance

PCAP Žyess1. N

Medical assistance

Probit

2-StepIV-Ac

2-StepIV-B c

Probit

2-StepIV-Ac

2-Step IV-B c

y2.5) Ž0.6. 10 778

y4.1 Ž4.2. 10 778

0.4 Ž2.4. 10 778

y2.4) Ž0.6. 9706

y0.7 Ž4.9. 9706

y2.2 Ž2.3. 9706

a

Figures for birth weight are the coefficients on the indicator variable for PCAP participation. For low birth weight, figures are percentage points changes associated with PCAP participation. All regressions include controls for maternal age, racerethnicity, nativity, previous fetal loss, maternal schooling, paternal schooling, marital status, sex of the infant and census tract poverty rates, and year. Birth weight regressions are estimated by two-stage least squares. I use Murphy and Topel’s two-step estimator when low birth weight is the outcome. b Standard errors are in parentheses; ) p- 0.01; )) p- 0.05; ))) p- 0.10. c IV-A uses health districtsryear interactions plus health area PCAP providers as instruments; IV-B uses interactions between year and hospital of delivery. d 2 2 For IV-A, critical values for xŽ29. is 42.6 Ž p- 0.05.; for IV-B, xŽ39. critical value is 54.6 Ž p- 0.05.; Critical values for FŽ30,A . and FŽ40,A . at p- 0.05 are 1.47 and 1.39, respectively.

tion between PCAP participation and improved birth outcomes regardless of which set instruments are used Žlast two columns.. In the birth weight regression, the IV coefficients have the wrong sign and are statistically insignificant. In the low birth weight models, the coefficients are negative, but measured imprecisely. A Wu– Hausman test rejects homogeneity between IV and OLS estimates at the 0.05 level. A test of over identification supports the validity of the health districtryear interactions as instruments, but not the hospitalryear interactions. Moreover, F-tests on the set of excluded instruments indicate that finite sample bias should not be a problem ŽBound et al., 1995.. In short, OLS estimates of treatment effects for women on medical assistance appear contaminated by selection bias. For women on cash assistance, inferences are less definitive because the IV estimates are sensitive to the set of instruments employed Žcolumns 2 and 3..

T. Joyce r Journal of Health Economics 18 (1999) 31–67

57

When I instrument PCAP participation with health districtryear interactions ŽIV-A., estimates are larger than those obtained by OLS, although statistically insignificant. In contrast, estimates obtained with hospitalryear interactions as instruments ŽIV-B. are smaller than those obtained by OLS, and in the case of low birth, the coefficient has the wrong sign. Adverse selection among PCAP participants into PCAP hospitals within health districts could explain why the two sets of instruments yield different estimates. In other words, women at relatively high risk for adverse outcomes who would have delivered at a non-PCAP facility had the Medicaid expansion not occurred, delivered at PCAP facilities after the expansion. As long as this ‘switching’ remains within district, one could observe improvements in birth outcomes based on districtryear instruments, while obtaining an attenuation of effects based on hospitalryear instruments. The summary statistics in Table 7, however, are not consistent with this interpretation. Among women on cash assistance Žtop panel., there was a drop in the percent of births to PCAP participants in hospitals that were never PCAP providers from 31.3 to 19.5% between 1989 and 1991, a decline that is suggestive of migration from non-PCAP to PCAP facilities. Yet, the rate of low birth weight rose from 8.6% in 1989 to 10.4% in 1991 in the non-PCAP hospitals ŽTable 7, columns 1 and 2., suggesting that women who ‘switched’ facilities were at relatively low risk for adverse outcomes compared to those who stayed—a change more consistent with favorable selection. In summary, results based on instrumental variables suggest that the expansion in PCAP participation had relatively little impact on birth outcomes. For women on medical assistance, the IV estimates in Table 8 are consistent with the summary data in Table 7: there are no meaningful changes in birth outcomes between 1989 and 1991 despite large increases in women serviced by PCAP. For women on cash assistance, inferences are more equivocal because estimated treatment effects are sensitive to the set of instruments employed. I prefer the IV estimates based on hospitalryear interactions for two reasons: first, variation in PCAP participation by hospital is a more direct measure of changes in PCAP availability than the less-specific changes associated with health districtryear interactions; and second, summary data provide no obvious evidence of adverse selection among women from non-PCAP to PCAP facilities, a response that would undermine the validity of hospital yearrinteractions as instruments. The lack of a consistent effect of PCAP participation on birth weight as reported in Tables 7 and 8 is unlikely to have been the results of misclassification of PCAP participation in 1989 as discussed in Section 2. If PCAP were in fact effective, and if many women who obtained prenatal care from PCAP providers in 1989 did not receive PCAP services, then I would expect to observe improvements in outcomes among women who delivered at hospitals that were always PCAP providers, and even greater improvements among hospitals that became PCAP providers. In addition, a simple IV estimator based on the ratio of the change in low birth weight by year and hospital to the change in PCAP participation by year

58

T. Joyce r Journal of Health Economics 18 (1999) 31–67

and hospital would overestimate improvements in low birth weight associated with PCAP by underestimating growth in PCAP participation. 5.5. Reduction in newborn costs associated with PCAP I conclude the analysis of PCAP with an estimate of reductions in newborn delivery costs associated with participation in PCAP. I want to emphasize that reduction in newborn delivery costs is a limited view of benefits and represents only a component of a more complete cost-effectiveness or cost–benefit analysis. 22 To convert improvements in birth outcomes to reductions in newborn costs, I use discharge abstracts linked to birth certificates of newborns whose mothers were on Medicaid and who delivered in a New York City public hospital in 1991 Ž n s 22 123.. 23 I regress the natural logarithm of newborn costs on birth weight and birth weight squared adjusted for the same set of demographic and socio-economic variables as in the birth outcome regressions. In another newborn cost regression, I use a dichotomous indicator of low birth weight instead of a quadratic specification of birth weight. 24 Savings in 1991 dollars associated with decreases in birth weight and rates of low birth weight are displayed in Table 9. I assume that PCAP improves birth outcomes of all participants equally and evaluate savings from two points. The first point is at the geometric mean of all discharges ŽUS$2165.; the second is at the predicted cost of an infant that weighs 2500 g ŽUS$3696.. Based on the analysis of birth outcomes, I evaluate savings for three changes in birth weight and rates of low birth weight that vary from conservative to more optimistic. Thus, if PCAP improves birth weight on average by 30 g, I estimate a reduction in newborn costs of US$189 per PCAP participant. 25 If I assume that the 30-g increase raises an infant’s weight from 2500 to 2530 g, then savings rise to US$322. The lower half of Table 9 shows results from a similar exercise with changes in the rate of low birth weight also evaluated at two cost points. Finally, if I assume that PCAP has no effect on birth weight among women on medical

22

I limit savings to newborn costs because I have data from another recently completed project in which we linked birth certificates to discharge abstracts for all deliveries in a New York City municipal hospital in 1991. I do not estimate lives saved because I do not assess infant mortality ŽSee Currie and Gruber, 1996.. For the same reason, I do not incorporate possible savings due to long-term costs of low birth weight ŽSee Lewit et al., 1995.. 23 Details as to the data can be found in Joyce et al., 1996. 24 Coefficients from the newborn cost regressions are presented in the Appendix of Joyce, 1997. 25 The US$189 dollars in savings was obtained as follows: we lnŽ2165. ye lnŽ2165.q b Ž30.q a Ž900. x where b is the coefficient on the linear birth weight term and a is the regression coefficient on the squared birth weight term Žsee Appendix Table 2..

T. Joyce r Journal of Health Economics 18 (1999) 31–67

59

Table 9 Reduction in 1991 newborn costs associated with participation in the Prenatal Care Assistance Program, New York City 1989 and 1991 Evaluated at

Change in newborn costs associated with an increase in birth weight of 30 g 50 g 70 g

US$2165a US$3696 b

yUS$189 yUS$323

Evaluated at

Change in newborn costs associated with a reduction in the rate of low birth weight of 1.0% pts.

2.0% pts.

3.0% pts.

US$2165 US$3696

yUS$61 yUS$104

yUS$122 yUS$208

yUS$183 yUS$313

yUS$305 yUS$521

yUS$414 yUS$706

a

The geometric mean of newborn costs for infants of women on Medicaid who delivered in a New York City public hospital in 1991 Ž ns 22, 285.. b Predicted cost Žin 1991 dollars. for a male infant who weighs 2500 g assuming mother is US-born, Black, non-Hispanic, between 21 and 34 years of age, unmarried, with less than a high school education and with father’s education unknown based on newborn costs regressions. See tables in Appendices B and C for regression estimates.

assistance, then savings per participant are reduced by 47%, which is the proportion of PCAP participants on medical assistance in 1989 and 1991. What are the incremental costs of PCAP? The preferred specification compares effects of PCAP for women who begin care in the first four months of pregnancy. The mean number of prenatal care visits for women who begin care early is approximately 10. PCAP paid US$229 for an initial visit in a public hospital outpatient clinic in 1991 and US$119 for each prenatal visit thereafter. 26 Outpatient fees under regular Medicaid were approximately US$65 in 1991 for public hospitals in New York City. 27 Based on these figures, PCAP expends US$651 more than what would be paid under regular Medicaid for prenatal care initiated in the first four months of pregnancy. This is probably an overestimate of the difference since PCAP’s fees include routine ancillary services such as ultrasound and regular Medicaid fees do not. According to figures in Table 9, reductions in the rate of low birth weight of three percentage points do not generate sufficient savings to recover prenatal costs under the assumption that PCAP affects all participants; PCAP must realize gains in birth weight of between 50 and 70 g for

26 From the Appendix of the New York State Department of Health Prenatal Care Assistance Program Comprehensive Provider Agreement as provided by Nancy Cuddihy, then Director of the Perinatal Health Unit. 27 Personal communications with Joanne Marks, Finance Department, Bronx Municipal Hospital and Medical Center.

60

T. Joyce r Journal of Health Economics 18 (1999) 31–67

infants that weigh 2500 g in order to recoup prenatal outlays. If I assume that PCAP has no effect on the birth outcomes of women on medical assistance, then even increases of 70 g in birth weight or reductions in low birth weight of three percentage points among women on cash assistance would not generate sufficient reduction in newborn costs to offset increased fees to providers for all PCAP participants.

6. Discussion In this study, I test whether participation in New York State’s Prenatal Care Assistance Program, PCAP, is associated with healthier infants. Based on adjusted means, I report a consistent and positive association between participation in PCAP and improved infant health. Specifically, I find women in PCAP have infants that weigh approximately 50 g more and have rates of low birth weight 2.5 percentage points less than infants not enrolled. These results are similar to those reported by Reichman and Florio Ž1996. and Buescher et al. Ž1991.. If, however, I eliminate preterm infants, gains associated with PCAP participation fall to about 35 g in mean birth weight and 1.3 percentage points in rates of low birth weight. Improvements in birth weight and rates of low birth weight of these magnitudes would not lower newborn costs sufficiently to offset program expenditures. In fact, I estimate that a 110-g increase in mean birth weight of infants to PCAP participants would be necessary to offset program outlays for prenatal services under the assumption that PCAP affects infants of women on cash assistance and medical assistance equally. Results obtained from IV methods, however, offer much less support for an association between receipt of PCAP services and improved infant health, especially among women on medical assistance, and thus, even a break even point of 110 g is optimistic. My conclusions, therefore, are more reserved than those reached by other researchers. For readers not persuaded by IV estimates, the most credible evidence of a positive association between PCAP participation and improved birth outcomes occurs among term births for women on cash assistance. Support for this finding comes from the robust epidemiological relationship between increased fetal growth and reductions in smoking, and a less consistent but potential relationship between maternal weight gain during pregnancy and fetal growth. PCAP providers are under contractual obligation to promote health education and to offer nutritional counseling, both of which have the potential to affect smoking and weight gain. The finding that women in PCAP are more likely to enroll in WIC is at least consistent with estimated improvements in the birth weight of term infants, but I show little association between heavy smoking and PCAP participation and relatively small differences in the number of prenatal care visits by PCAP status. Thus, the data do not provide convincing evidence of possible mechanisms that might explain adjusted mean differences in birth outcome associated with PCAP.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

61

Evidence of selection bias among PCAP participants is inferred by comparing results based on IV methods that show no statistically significant effect of PCAP on birth outcomes, to the estimated gains obtained by single-equation methods. The IV estimates suggest that infants of women ‘moved’ by the Medicaid expansion, especially those covered by medical assistance, were unaffected by participation in PCAP ŽAngrist et al., 1996.. The validity of the IV estimates relies on whether the Medicaid expansion stimulated an exogenous increase in PCAP participation by both clients and providers. In this regard, I find the simple comparison by hospitals in Table 7 compelling. First, there is no evidence to suggest large switching by women from non-PCAP to PCAP hospitals. Nor is there evidence that the quality of PCAP providers was greater in one group of hospitals or another based on the proportion of municipal facilities. The lack of any improvement in birth outcomes in hospitals that changed to PCAP from non-PCAP providers, despite the huge increase in PCAP participation, is consistent with the IV estimates and undermines results obtained by single-equation methods. Finally, there is a need to move beyond simple adjusted means or odd ratios based on binary indicators of participation in future assessments of comprehensive prenatal services associated with the Medicaid expansions. Researchers should exploit variations in time, eligibility groups and geographic boundaries in an effort to create more plausible control groups or to generate instruments. Finally, health policy analysts should be able to defend the biological or epidemiological plausibility of reported associations. The general lesson emerging from analyses of the Medicaid expansions of the late 1980s is that we may have overstated the effectiveness of prenatal care to improve infant health for many of the reasons just enumerated. Analyses of comprehensive prenatal services should avoid these mistakes so that more effective policies can be designed and implemented.

Acknowledgements This work was support by a grant from the Agency for Health Care Policy and Research ŽAHCPR. to the National Bureau of Economic Research ŽNBER. wGrant No. R03 HS08424-01x. I have benefited greatly from discussions with Robert Kaestner. I owe thanks to Barbara Brustman, Lawrence Clark, and Georgie DiFerdinando from the New York State Department of Health, Steve Schwartz, Director of Vital Statistics, New York City Department of Health and from John Black, New York City Bureau of Medicaid Analysis for help with data and Ewa Wojas for research assistance. Finally, I am grateful to Michael Grossman, Andrew Racine and Howard Minkoff and to participants in the Boston University, Harvard and MIT health economics seminar. I thank two anonymous referees and especially the editor for their comments. As always, any error is mine.

62

Appendix A

No prenatal care PCAP non-PCAP Ž7. Ž8.

Month care began unknown PCAP non-PCAP Ž9. Ž10.

2957 21.9 4.5 224

2814q 26.0 ) 4.6 262

3152 12.0 1.3 308

3063 15.6 2.5 160

Cash assistance, 1991 LBW % Preterm %b Very LBW% N

8.2 11.7 1.1 2492

11.3 ) 16.8 ) 1.7 841

7.9 9.0 0.8 847

11.7q 12.7 a 2.7q 298

5.7 7.6 0.5 419

15.6 ) 10.6 1.4 141

12.9 15.1 3.2 249

27.8 ) 33.0 ) 4.7 299

14.2 17.4 0.0 176

22.9q 26.2 a 6.9) 175

Medical assistance, 1989 LBW % Preterm %b Very LBW% N

5.5 11.0 0.9 1019

8.3q 12.5 1.4 1104

7.1 7.9 0.0 378

8.8 10.5 2.1) 467

3.2 5.9 0.0 187

5.9 6.6 0.6 320

12.1 12.5 4.5 66

22.0 14.1 3.3 205

6.1 20.0 4.4 115

7.3 13.8 2.4 123

Medical assistance, 1991 LBW % Preterm %b Very LBW% N

5.9 10.3 0.8 3130

8.8q 13.6q 2.2) 589

6.0 9.0 0.7 1068

6.8q 7.5 0.5 190

4.1 3.6 0.0 442

5.3 6.1 0.0 114

10.6 13.4 3.3 245

31.8 ) 35.0 ) 8.5q 129

6.4 9.2 1.7 358

8.8 9.9 3.9 102

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Birth outcomes by payor status, year, when prenatal care began, and PCAP participation Prenatal care initiated in Months 1–4 Months 5–6 Months 7q PCAP non-PCAP PCAP non-PCAP PCAP non-PCAP Ž1. Ž2. Ž3. Ž4. Ž5. Ž6. Cash assistance, 1989 Birth weight Žg. 3213 3170 a 3192 3181 3230 3227 LBW % 8.1 10.5q 10.6 9.2 6.5 6.7 Very LBW% 0.9 1.9 0.8 0.5 0.5 0.5 N 1842 1166 714 380 403 208

Appendix B

Parity Second or higher birth First birth Parity unknown Previous fetal loss Žyess1. Unmarried Žyes s1.

y 0.02 Ž0.01. 0.05 ) Ž0.02. y0.02 Ž0.03. 0.01 Ž0.01.

y 0.02 Ž0.02. 0.05 Ž0.04. y0.02 Ž0.04. 0.01 Ž0.01.

y 0.02 ) ) Ž0.01. 0.04 ) ) Ž0.02. y0.01 Ž0.04. 0.04 ) Ž0.01.

y 0.02 ) ) Ž0.01. 0.03 Ž0.02. y0.01 Ž0.04. 0.03 Ž0.01.

RacerEthnicity Whiterother Black non-Hispanic Puerto Rican Dominican Other Hispanic Racerethnicity unknown

y 0.09 ) Ž0.02. 0.13 ) Ž0.02. 0.13 ) Ž0.02. 0.09 ) Ž0.03. 0.06 Ž0.04.

y 0.08 Ž0.07. 0.13 Ž0.10. 0.12 Ž0.10. 0.08 Ž0.07. 0.06 Ž0.06.

y 0.05 ) Ž0.01. 0.09 ) Ž0.02. 0.07 ) Ž0.02. 0.07 ) Ž0.01. y0.02 Ž0.02.

y 0.05 ) ) Ž0.01. 0.08 ) ) Ž0.02. 0.07 ) Ž0.02. 0.07 ) Ž0.01. y0.02 Ž0.02.

NatiÕity Foreign born Born in US Birth place unknown

y y0.04 ) Ž0.01. y0.14 ) ) ) Ž0.08.

y y0.04 Ž0.04. y0.12 Ž0.12.

y y0.11) Ž0.01. y0.04 Ž0.12.

y y0.10 Ž0.01. y0.04 Ž0.11.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Marginal effects and standard errors Žin parentheses. from first-stage regression of PCAP participation estimated by ordinary least squares and maximum likelihood probit by medicaid status for pooled years 1989 and 1991a Dependent variable: PCAP participation Ž1s yes. Cash assistance Medical assistance OLS Probit OLS Probit Mother’s age Age 20–34 y y y y Age - 20 y0.01 Ž0.01. y0.01 Ž0.02. y0.02 Ž0.01. y0.02 Ž0.01. Age ) 34 y0.02 Ž0.02. y0.01 Ž0.02. 0.01 Ž0.02. 0.01 Ž0.02. Male infant Žyess1. y0.01 Ž0.01. y0.01 Ž0.01. 0.01 Ž0.01. 0.01 Ž0.01.

63

64

Father’s schooling Less than high school High school Some college Schooling unknown % poor in census tract Income 100–180 FPL Year s1991 PCAP providers in health area R 2 or pseudo R 2 2 FŽ29,A . or xŽ29,. for health districtsb 2 FŽ29,A . or xŽ29,. for health districts = yr N )

y y0.02 ) ) Ž0.01. y0.05 ) Ž0.01. 0.02 Ž0.03.

y y0.02 Ž0.02. y0.05 Ž0.04. 0.02 Ž0.03.

y y0.02 ) ) ) Ž0.01. y0.03 ) ) Ž0.01. y0.02 Ž0.02.

y y0.02 ) ) Ž0.01. y0.03 ) ) Ž0.01. y0.02 Ž0.02.

y y0.03 Ž0.01. y0.002 ) ) ) Ž0.02. y0.02 Ž0.01. 0.001) Ž0.000. na 0.05 Ž0.05. 0.02 Ž0.01.

y y0.04 Ž0.03. y0.003 Ž0.02. y0.02 Ž0.02. 0.001 Ž0.001. na 0.09 Ž0.09. 0.02 Ž0.02.

y y0.02 ) ) ) Ž0.01. y0.03 Ž0.02. y0.02 ) ) ) Ž0.01. 0.001) ) ) Ž0.000. 0.05 ) ) Ž0.02. 0.47 ) Ž0.07. 0.05 ) ) Ž0.02.

y y0.03 Ž0.01. y0.03 Ž0.02. y0.03 Ž0.01. 0.001) ) ) Ž0.000. 0.05 Ž0.02. 0.45 ) Ž0.07. 0.04 Ž0.02.

0.11 19.5 )

0.09 436 )

0.23 15.1)

0.20 288 )

8.2 )

202 )

12.9 )

280 )

10778

10778

9706

9706

p- 0.01. )) p- 0.05. ))) p- 0.10. b 2 Critical value for FŽ29,A . is 1.47 and xŽ29,. is 42.56.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

Mother’s schooling Less than high school High school Some college Schooling unknown

T. Joyce r Journal of Health Economics 18 (1999) 31–67

65

Appendix C. New York City Hospitals by PCAP Status 1989–1991a

Hospitals that changed from a non-PCAP provider to a PCAP provider between 1989–1991

Hospitals that were always PCAP providers between 1989–1991

Hospitals that were never PCAP providers between 1989–1991

Long Island College Elmhurst b New York Hospital Beth Israel NY Downtown ŽBeekman. Bellevueb Metropolitanb Interfaith Brooklyn Maimonides Methodist Flushing Jamaica

Harlemb Booth Memorial St John’s Episcopal Lenox Hill Albert Einstein Columbia Presbyterianc Lutheran New York University Lincolnb Long Island Jewish Union LaGuardia Hospital St Vincent’s St. John’s Queens St LukesrRoosevelt North Central b Bronx Lebanon Coney Islandb Kings b Montefiore Our Lady of Mercy St Mary’s Bronx Municipal b Mount Sinai Queens Hospital Center b Staten Island Hospital Woodhull b

a

This list has been constructed from memos from the New York State Department of Health, Office of Perinatal Health Unit to PCAP providers as well as from provider lists based on claim records from PCAP providers to the New York State Department of Social Services, Division of Health and Long Term Care. I eliminated a branch of Saint John’s Episcopal hospital, the Interfaith Medical Center, because I could not determine from any lists whether the Interfaith Center served PCAP recipients in 1989. b Indicates that the facility is a municipal hospital. c Includes the Allen Pavilion and Sloane Hospital for Women.

References Alexander, G., Korenbrot, C., 1995. The role of prenatal care in preventing low birth weight. The Future of Children 5, 103–120.

66

T. Joyce r Journal of Health Economics 18 (1999) 31–67

American College of Obstetricians and Gynecologists, 1989. Standards for obstetric-gynecologic services. Am. Coll. Obstetr. Gynecol., Washington, DC. Angrist, J., Imbens, G., Rubin, D., 1996. Identification of causal effects using instrumental variables. J. Am. Stat. Assoc. 91, 444–455. Bound, J., Jaeger, D., Baker, R., 1995. Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak. JASA 90, 443–450. Buescher, P., Roth, M., Williams, D., Goforth, C., 1991. An evaluation of the impact of maternity care coordination on medicaid birth outcomes in North Carolina. Am. J. Pub. Health 81, 1625–1629. Cole, N., 1995. Increasing access to health care: The effects of medicaid expansions for pregnant women. Report prepared by ABT Associates for Health Care Financing Administration, Contract No. 500-89-0063. Collaborative Group on Preterm Birth Prevention, 1993. Multicenter randomized, controlled trial of a preterm birth prevention program. Am. J. Obstet. Gynecol., 169, 352–366. Currie, J., Gruber, J., 1996. Saving babies: the efficacy and cost of recent expansions of medicaid eligibility for pregnant women. J. Pol. Econ. 104, 1263–1296. Devaney, B., Bilheimer, L., Schore, J., 1992. Medicaid costs and birth outcomes: the effects of prenatal WIC participation and the use of prenatal care. J. Policy Anal. Manage. 11, 573–592. Frost, J. et al., 1993. State implementation of the medicaid eligibility expansions for pregnant women. Alan Guttmacher Institute, New York. Gibbs, R.S., et al., 1992. A review of premature birth and subclinical infection. Am. J. Obstetr. Gynecol. Ž166. 1515–1528. Hack, M., Merkatz, I., 1995. Preterm delivery and low birth weight—a dire legacy. NE J. Med. 333, 1772–1773. Joyce, T., 1997. Impact of augmented prenatal care on birth outcomes of medicaid recipients in New York city. National Bureau of Economic Research Working Paper No. 6029, May. Joyce, T., Racine, A., Grossman, M., 1996. Neonatal costs of maternal substance abuse. Final Report of Grant No. 5R01 DA07557 from the National Institute on Drug Abuse to the National Bureau o of Economic Research, October. Kenney, G., Dubay, L., 1995. A national study of the impacts of medicaid expansions on pregnant women. Manuscript, The Urban Institute. Korenbrot, C. et al., 1995. Evaluation of California’s statewide implementation of enhanced perinatal services as medicaid benefits. Public Health Rep. 110, 125–134. Kotelchuck, M., 1994. An evaluation of the Kessner adequacy of prenatal care index and a proposed adequacy of prenatal care utilization index. Am. J. Public Health 84, 1414–1420. Lewit, G. et al., 1995. The direct cost of low birth weight. The Future of Children 5, 35–56. Li, C.Q. et al., 1993. The impact on infant birth weight and gestational age of Cotinine-validated smoking reduction during pregnancy. J. Am. Med. Assoc. 269, 1519–1524. McLaughlin, F.J. et al., 1992. Randomized trial of comprehensive prenatal care for low-income women: effect on birth weight. Pediatre 89, 128–132. Moffitt, R., 1995. Selection bias adjustment in treatment-Effect models as a method of aggregation. Proceedings of the American Statistical Association, pp. 234–238. Murphy, K., Topel, R., 1985. Estimation and inference in two-step econometric models. J. Bus. Econ. Stat. 3, 370–379. New York State Department of Health, 1989a. Department of Health Memorandum. Health Care Facilities Series: H-60, D and TC-40, Series 89–91, November 11. New York State Department of Health, 1989b. Prenatal Care Assistance Programs: 1988 Annual Report. New York State Department of Health, Albany. Olds, D.L., Henderson, C.R., Tatelbaum, R., Chamberlin, R., 1986. Improving the delivery of prenatal care and outcomes of pregnancy: a randomized trial of nurse home visitation. Pediatre 77, 16–28.

T. Joyce r Journal of Health Economics 18 (1999) 31–67

67

Piper, J.M., Ray, W.A., Griffen, M.R., 1990. Effects of medicaid eligibility expansion on prenatal care and pregnancy outcome in Tennessee. JAMA 264, 2219–2223. Piper, J.M., Mitchel, E.F., Ray, W.A., 1994. Expanded medicaid coverage for pregnant women to 100 percent of the federal poverty line. Am. J. Prev. Med. 10, 97–102. Piper, J.M. et al., 1993. Validation of 1989 Tennessee Birth Certificates using maternal and newborn hospital records. Am. J. Epidemiol. 137, 758–768. Piper, J.M., Mitchel, E.F., Ray, W.A., 1996. Evaluation of a program for prenatal care case management. Fam. Plann. Perspect. 28, 65–68. Reichman, N., Florio, M., 1996. The effects of enriched prenatal care services on Medicaid birth outcomes in New Jersey. J. Health Econ. 15, 455–476. Rush, D. et al., 1988. National WIC evaluation: evaluation of the special supplemental food program for women, infants, and children: V. Longitudinal Study of Pregnant Women. Am. J. Clin. Nutri. 48, 439–483. Scholl, T. et al., 1991. Maternal weight gain, diet and infant birth weight: correlations during adolescent pregnancy. J. Clin. Epidemiol. 44, 423–428, Supplement. Singh, G.K., Yu, S.M., 1996. Adverse pregnancy outcomes: differences between US- and foreign-born women in major US racial and ethnic groups. Am. J. Public Health 86, 837–843. Stein, Z., Kline, J., 1983. Smoking, alcohol and reproduction. Am. J. Public Health 73, 1154–1155. Tucker, J.M. et al., 1991. Etiologies of preterm birth in an indigent population: is prevention a logical expectation?. Obstet. Gynecol. 77, 343–347.