The Civil Rights Act of 1964, Hospital Desegregation and Black Infant Mortality in Mississippi*

Transcription

1 The Civil Rights Act of 1964, Hospital Desegregation and Black Infant Mortality in Mississippi* Douglas V. Almond Kenneth Y. Chay Michael Greenstone * We thank Richard Rogerson, two anonymous referees, and many seminar participants and colleagues for helpful comments. We are indebted to Dick Johnson and Harold Armstrong of the Mississippi State Department of Health, Joan Exline of the University of Southern Mississippi, Sarah Haldeman of the Lyndon B. Johnson Library, and Gary Kennedy of the Bureau of Economic Analysis for invaluable assistance with some of the data. Chuan Goh, Heather Royer, Stacy Sneeringer, Paul Torelli, Paulina Oliva Vallejo, Carolina Caetano, Juan Sui, and Elizabeth Greenwood provided excellent research assistance. The Center on the Economics and Demography of Aging at UC-Berkeley provided funding.

2 The Civil Rights Act of 1964, Hospital Desegregation and Black Infant Mortality in Mississippi Douglas V. Almond, Kenneth Y. Chay, and Michael Greenstone NBER Working Paper No. March 2008 JEL No. J15, I18, I11, I38, N32 ABSTRACT We examine the role of Title VI of the Civil Rights Act of 1964 in the reduction of the black infant mortality rate (IMR) in the United States after Black-white IMR convergence was greatest in the South where access to hospital care among blacks increased the most and was driven by a reduction in black infant death occurring 1 to 12 months after birth (post-neonatal mortality). Using a county-level database for Mississippi the state with the largest decline in racial inequality we find hospital desegregation to be the leading cause of these improvements. The gains in black access to hospitals coincide with a striking reduction in black post-neonatal death for causes considered preventable with timely hospital care. The location and timing of the racial convergence across Mississippi s counties correspond with when their hospitals received certification for reimbursement from the Medicare program, which required compliance with Title VI. We calculate that this came at a small increase in costs to Mississippi s hospitals and conclude that the financial incentives of Medicare were crucial to achieving hospital desegregation. Douglas V. Almond Kenneth Y. Chay Department of Economics Department of Economics Columbia University Brown University International Affairs Building, MC 3308 Box B 420 West 118th Street Providence, RI New York, NY and NBER and NBER kenneth_chay@brown.edu da2152@columbia.edu Michael Greenstone Department of Economics MIT 50 Memorial Drive E Cambridge, MA and NBER mgreenst@mit.edu

3 Introduction In 1965, over 40 African American infants died within a year of birth for every 1,000 born in the United States. Over the next ten years, the black infant mortality rate fell to 24-per-thousand, with roughly 7,000 more black babies born in 1975 surviving to the age of one than if the trend before 1965 had continued. The relative infant mortality rate (IMR) of blacks also fell significantly in these years with 1966 to 1971 comprising the only period of sharp reductions in the black-white IMR gap after World War II (Almond, Chay and Greenstone 2003). Nearly all of this convergence was driven by a reduction in the racial gap in the post-neonatal mortality rate, defined as the death rate in the period from 28 days to 1 year after birth (Collins and Thomasson 2002). In this paper, we find wide variability in the decline in black infant mortality after 1965 across American states, with the greatest improvement occurring in Mississippi. The reduction in Mississippi was driven by a striking decline in post-neonatal death, mostly among causes of death that could be prevented with a timely medical intervention. Figure 1, for example, shows the black and white postneonatal death rates due to diarrheal dehydration (gastroenteritis) and pneumonia in Mississippi from 1955 to In the early 1960s, black rates were 10-times higher than the rates for whites, and showed no trend toward a relative improvement. After 1965, however, the black mortality rate plummets by 65 percent by 1971, while the white rate remains constant. This dramatic racial convergence for these causes of infant death is particularly noteworthy for two reasons. First, it accounts for a significant share of the overall reduction in black infant mortality in Mississippi the largest share of any state in the entire country during this period. Second, effective treatments for both diagnoses were well known and widely available in hospitals at the time. Rather than representing an innovation in medical technology, a change in behavior, or an improvement in the home environment of black infants, we show that this trend shift was primarily driven by federal desegregation efforts that opened up access to hospital care for black infants. Specifically, two key pieces of legislation Title VI of the Civil Rights Act of 1964 and Title XVIII of the Social Security Amendments of 1965 (Medicare) prompted Southern hospitals to integrate for both legal and economic reasons. We place an emphasis on Mississippi since the evidence suggests that its hospitals were the

4 most segregated in the United States before the two Acts, and lagged other states hospitals in reaching compliance with the desegregation mandates of Title VI of the Civil Rights Act. We establish that, before the Civil Rights Act, differences across states in racial inequality in hospital access and infant mortality were strongly correlated with the racial compositions of states, and this correlation fell substantially after the Act. The states with the largest black population shares i.e., in the Deep South had both the greatest inequality before the Act and the greatest convergence afterward. Using a unique county-level database for Mississippi, we find hospital desegregation to be the leading cause of these improvements. The location and timing of the racial convergence across Mississippi s counties correspond with when their hospitals received certification for reimbursement from the Medicare program, which required compliance with Title VI. Indeed, the staggered receipt of Medicare certification explains nearly all of the staggered reduction in black mortality due to preventable causes illustrated in Figure 1. None of the other, detailed covariates which include the characteristics of women giving birth; income; and payments, by source, of government transfers match these patterns. Based on detailed hospital-level data, we further find that the Mississippi hospitals that received early Medicare certification served areas with more people eligible for Medicare and with smaller black population shares. This suggests that hospitals contrasted the financial benefits of integrating to receive Medicare dollars with the monetary and psychic costs of serving the black population. The hospital data also imply that the costs of the care provided to reduce black infant mortality were low. After desegregating, hospitals utilized pre-existing hospital services with excess capacity; they did not increase personnel or capital. From this, we conclude that the financial incentives of Medicare were decisive in achieving the integration of Mississippi s hospitals. Section I provides background on federal efforts to desegregate hospitals. Section II documents the variation across states in racial inequality and racial convergence in hospital access and infant death. Section III provides evidence on the impact of hospital desegregation in Mississippi. Finally, Section IV presents calculations of the costs of the additional hospital care provided to black infants, and discusses the mechanisms through which black improvements were achieved. 2

5 I. The Federally Mandated Desegregation of Hospitals Here, we describe the discrimination in hospital practices that existed before 1965; detail federal efforts to desegregate hospitals during the mid-1960s; and discuss the causes of infant death that would have responded most to increased access to the medical care available in hospitals. A. Segregation in Southern Hospitals Prior to 1965 Hospitals and other health facilities, which should be the institutions of healing, often because of discrimination become the means of shortening life. (Robert Nash, Director of Office of Equal Health Opportunity, 1966) Throughout the 1950s and early 1960s, racial segregation in Southern society extended to all areas of life, including health services. A Journal of the National Medical Association survey of 523 hospitals in 1955 to 1956 found that while 87 percent of hospitals in the North had a policy of integration, only 10 percent of hospitals in the South had one (Cornely 1957). A 1956 survey of 2,400 Southern hospital administrators found that only 17 percent were willing to accept desegregation, with 62 percent adamantly opposed and 10 percent favoring separate hospitals (Seham 1964). These policies continued into the early 1960s for example, a 1963 survey found that 85 percent of Southern hospitals reported some type of racial segregation or exclusion compared to less than 2 percent in Northern and Western states (U.S. Commission on Civil Rights 1963:141). In single-building hospitals, the pattern of discrimination usually consisted of a separate wing or floor for Negro patients or the outright exclusion of a race (U.S. Commission on Civil Rights 1966:10). A relatively unrecognized aspect of hospital segregation in the South was the federal government s role in supporting it both legally and financially through the Hospital Survey and Construction Act of 1946, also known as the Hill-Burton Act. Two decades after its enactment, 350,000 hospital beds approximately half of all existing beds in the United States had been constructed under the Hill-Burton program. As Hill-Burton funding was more generous in poorer areas, the federal government was the dominant force in postwar hospital construction in the South (DHEW 1966). A stipulation for a facility to receive funding was the provision of medical services to the indigent and poor. The Hill-Burton Act is the only federal legislation of the 20 th century to codify racial segregation 3

6 and exclusion (Smith 1999). In its original form, the 1946 Act explicitly invoked the principle of separate but equal with regard to access of different races to hospitals built using Hill-Burton funds. This clause remained after the 1954 Brown vs. Board of Education decision that separate but equal was unconstitutional, despite several amendments and extensions to other parts of the legislation. 1 Moreover, the Hill-Burton program funded health care facilities that specifically excluded individuals on the basis of race. 2 In a 1964 issue of the New England Journal of Medicine, Max Seham states, To 20,000,000 citizens, discrimination and segregation followed from the separate-but-equal clause of Hill-Burton. 3 B. The End of Legalized Discrimination in Health Care Facilities Three landmark events in the mid-1960s in turn involving the judicial, legislative and executive branches of the U.S. government were instrumental in ending racial segregation in the American healthcare system. First, in Greensboro, North Carolina in 1963, Dr. George Simkins joined with black patients and other physicians in suing two hospitals which denied admission and staff privileges to blacks. These hospitals, although private, had received Hill-Burton construction funds. In November 1963, the Fourth Circuit Court of Appeals decided Simkins v. Moses H. Cone Memorial Hospital in favor of the plaintiffs. The court declared that the section of the Hill-Burton legislation permitting separate-but-equal health facilities was unconstitutional. The majority opinion states: Racial discrimination in medical facilities is 1 A February 1963 memo from the General Counsel at the Department of Health, Education and Welfare to the Assistant Surgeon General states, The acceptance of internal segregation is also, I believe, in accord with the intent of Congress at the time when the legislation was enacted, and in this respect also Congress has made no change to the law (DHEW memo February 12, 1963:1). The same memo states that Hill-Burton grants should continue to be awarded to segregated facilities despite any doubts [administrative officers] individually may entertain regarding the constitutional validity of the statutes. In August 1963, the U.S. Senate rejected a proposal that would have required grants under the Hill-Burton Act to only go to hospitals that did not discriminate on the basis of race (U.S. Court of Appeals, Fourth Circuit, 1963:23). 2 A hospital constructed with Hill-Burton money could exclude a race outright if it [indicated in its application that certain persons in this area will be denied admission to the proposed facilities as patients because of race, creed, or color and] reported to the Surgeon General that other facilities for the excluded race were available (U.S. Court of Appeals, Fourth Circuit, 1963:7). Ninety percent of the funds for one-race only facilities went to all-white facilities (U.S. Commission on Civil Rights 1963). 3 Many states outside of the South had laws in place rendering discrimination in hospitals illegal, including those built with Hill-Burton funds. Ohio s requirement for receiving Hill-Burton funds is typical of these states: The applicant hereby assures the State Agency that it will make its facilities available to all persons residing in the area to be served without discrimination as to race, creed, or color. (Ohio Department of Health, October 1949) 4

7 at least partly responsible for the fact that in North Carolina the rate of Negro infant mortality is twice the rate for whites... (U.S. Court of Appeals, Fourth Circuit 1963:14). In March 1964, the Supreme Court declined to review the Appeals Court decision, leaving it in place. The immediate result of this decision was that new applications for Hill-Burton funds and pending projects were required to be nondiscriminatory with respect to admissions, staff privileges, and access to all portions of the facilities (Smith 1999). Since half of all Hill-Burton grantees were private non-profits (DHEW February 12, 1963), the ruling s impact was not limited to public health facilities. However, the ruling did not affect segregated hospitals that had previously received Hill-Burton funding. Indeed, lawmakers cited the Simkins case as demonstrating the need for definitive federal legislation. The legislative intervention that soon followed was Title VI of the Civil Rights Act of 1964 signed on July 2, 1964, with the regulations requiring hospital desegregation effective in January 1965 which prohibits discrimination and segregation in any institution receiving federal funds. 4 According to the U.S. Commission on Civil Rights (1966:1), Seldom has any piece of legislation been so broad in scope, sweeping across departmental, geographical and political lines, as Title VI of the Civil Rights Act of The same report states that while Title VI by itself led to the integration of some hospitals in the South by the end of 1965, In nearly two-thirds of the hospitals surveyed, there were discernible patterns of noncompliance (1966:14). This was despite the fact that several of these hospitals had claimed compliance with Title VI. The final prong of federal efforts to end discrimination in hospitals came with the inception of the Medicare program on July 1, Following directions from President Johnson, hospitals were made eligible for Medicare reimbursement only if they were found to be in full compliance with Title VI of the Civil Rights Act through a certification process. 5 Thus, Title XVIII of the Social Security Amendments 4 Title VI stipulates that Patients must be admitted to facilities without regard to their race, creed, color, or national origin. Once admitted, patients must have access to all portions of the facility and to all services without discrimination. They may not be segregated within any portion of the facility, provided a different service, restricted in their enjoyment of any privilege, or treated differently because of their race, creed, color, or national origin (Reynolds 1997:1852). 5 In a speech to the American public the night before Medicare went into effect, Johnson said, Medicare will succeed if hospitals accept their responsibility under the law not to discriminate against any patient because of race (Reynolds 1997). See Quadagno (2000) for more details on the interaction between Title VI and Medicare. 5

8 of 1965, which established Medicare, provided Title VI with a significant financial lever to end racial discrimination in hospitals. 6 One observer has noted, Medicare payments to hospitals promised to be generous, and thereby essential. In essence, hospitals had to choose between affluence through compliance and bankruptcy (Smith, 2005a:320). Further, in the months leading up to July 1966, the Office of Equal Health Opportunity sent a Children s Crusade of 750 federal employees to perform on-site inspections of hospitals, verifying compliance with Title VI (Smith 1999:113). 7 The hospital integration guidelines for Medicare certification included that: 1) hospitals provide inpatient and outpatient care without regard to race, color, or national origin; 2) all patients be assigned to rooms, wards, floors, sections and buildings without regard to race, color, or national origin; 3) employees and medical staff be notified in writing of the hospital s compliance with the Civil Rights Act; and 4) hospitals which end discriminatory practices notify those persons previously excluded from services (Reynolds 2004). By June 30, 1966, 92 percent of all hospital beds in the United States were found to be in compliance with Title VI (Smith 1999:141). In July and through the Fall of 1966, the Office of Equal Health Opportunity worked to secure compliance in the remaining hospitals. This quiet revolution meant that in Southern hospitals, Negroes are being admitted and treated as anyone else for the first time (Nash 1968:246). 8 C. The Mississippi Context [I]n all important areas of citizenship, a Negro in Mississippi receives substantially less than his due consideration as an American and as a Mississippian. This denial extends from the time he is denied the right to be born in a nonsegregated hospital, through his segregated and inferior school 6 The private insurance sector applied additional pressure. Blue Cross informed administrators of noncompliant hospitals that individuals aged 65 and over would not be covered by Blue Cross-Blue Shield insurance since they were now eligible for Medicare (Reynolds 1997). It also made coverage available to fill gaps in Medicare benefits for patient services at compliant hospitals (Hospitals, JAHA, June 16, 1966: ). 7 On the process of enforcing integration of health care facilities, Robert Nash (Director of the Office of Equal Health Opportunity, Public Health Serice) wrote, Some people have compared our administration of Title VI to the practices of some of the world's outstanding dictatorships (Nash 1968:251). 8 Reynolds (1997) quotes Wilbur Cohen, who as the Secretary of HEW was a principal architect of the Medicare legislation: There is one other important contribution of Medicare and Medicaid which has not yet received public notice the virtual dismantling of segregation of hospitals, physicians offices, nursing homes, and clinics as of July 1, 1966 If Medicare and Medicaid had not made another single contribution, this result would be sufficient to enshrine it as one of the most significant social reforms of the decade [if not the century]. 6

9 years and his productive years when jobs for which he can qualify are refused, to the day he dies and is laid to rest in a cemetery for Negroes only. (Mississippi Advisory Committee to the U.S. Commission on Civil Rights 1963:4) Prior to these events, Mississippi s hospitals were among the most intransigent discriminators. Through 1964, Mississippi State law required that in hospitals maintained by the State for treatment of white and colored patients hospital administrators must provide separate entrances for whites and blacks and that the entrances shall be used by the races only for which they are prepared. The Governor of Mississippi was empowered to remove any administrator failing to comply with this law (DHEW memo, June 26, 1964:6). Mississippi was the only state in the nation that required segregation in state-run hospitals (DHEW memo, January 31, 1956). As a result, before 1965 the Mississippi Commission on Hospital Care s annual list of hospitals divides the beds in each hospital (both public and private) into those intended for whites and those intended for the colored. The floor plans for all Diagnostic Health Centers in Mississippi (published in the Mississippi State Board of Health Biennial Report) clearly indicate racial segregation in these facilities, with separate entrances and treatment rooms. Further, very few new facilities constructed with Hill- Burton money were located in the predominantly black counties in the Mississippi Delta region (see Almond, Chay and Greenstone 2003). 9 Mississippi had the highest black infant mortality rate in the country. While the black IMR in the Southern states outside of Mississippi was 42 per thousand births in 1965, it was 55-per-thousand in Mississippi. Further, 70 percent of this gap was due to the difference in black post-neonatal mortality rates, which were 26 in Mississippi and 17 in the rest of the South in 1965 (authors calculations). Regarding the complete hospital desegregation required for Medicare certification, Mississippi s hospitals lagged behind those in other states. As of June 23, 1966, only 20 percent of Mississippi s hospitals had been found to be in compliance with Title VI (Smith 1999:140). Less than half of its (nonmilitary) hospitals received Medicare certification by February 1967, with 64 percent certified by 9 From 1948 to 1956, Mississippi received more Hill-Burton funds per-capita than any other state over twice the national average. Nearly all of the hospitals and health facilities in Mississippi had been built using Hill-Burton money, even as racial segregation and exclusion were permitted by Hill-Burton. 7

10 February 1968, and 71 and 78 percent receiving certification by October 1968 and October 1969, respectively (data from the American Hospital Association Guide issues). In contrast, in Alabama which, after Mississippi, had the second-lowest Title VI compliance rate of 15 percent as of June 3, 1966 (Reynolds 1997) hospital compliance jumped to 51 percent by June 23, 1966 (Smith 1999:140); 90 percent of Alabama s hospitals received Medicare certification by February 1967 and close to 100- percent by October Reports at aggregated geographic levels published by the Department of Health, Education, and Welfare suggest that black access to hospital care in Mississippi improved dramatically after the Civil Rights Act. For example, the 84 percent black-white gap in hospital discharge rates for children under age 15 in (DHEW June 1966, Table 2) was cut by a factor of five by 1969 (DHEW April 1972, Number 70, Table E). We use the staggered certification of Title VI compliance in Mississippi hospitals to look more closely at the effects of hospital desegregation. D. Gastroenteritis, Pneumonia, and the Black-White Gap in Infant Mortality Prior to the diffusion of technologies benefiting premature and low-weight infants during the 1970s and 1980s (Cutler and Meara 1999), medical care was most successful in preventing deaths during the post-neonatal period i.e., 1 to 12 months after birth. Post-neonatal deaths are predominantly caused by negative events that occur after birth, such as infectious diseases and accidents (Grossman and Jacobowitz 1981). For pneumonia and gastroenteritis infections in particular, death could be prevented with timely medical treatment in a hospital, in part due to the greater availability of antibiotics following World War II (Shapiro, et al. 1968). For example, with regard to gastroenteritis, Emery (1976:204) writes that mortality rates relate to the earliness [or] lateness of diagnosis and admission to hospital. 10 Similarly for pneumonia, a 1964 pediatrics textbook states, The outcome is dependent upon early 10 Gastroenteritis often results in diarrheal dehydration in the infant a leading cause of death historically and in the developing world today. While today oral rehydration salt packets are effective in preventing death, in the period of our study the most effective treatments required hospital admission. Due to the rapid progression of dehydration, fluid resuscitation is critical and at that time would have been carried out intravenously. As IV s were only available in hospitals, hospital care was critical to an infant s recovery. 8

11 diagnosis and the appropriateness of [medical] treatment (Nelson 1964: ). 11 In 1965, black infants were four times as likely as whites to die from gastroenteritis and pneumonia during the post-neonatal period throughout the entire United States, and over 11 times as likely in Mississippi (see Figure 1). A primary factor for the racial divergence in infant mortality before 1965 was the divergent experiences of black and white infants with these two eminently treatable afflictions after World War II. We hypothesize that the increase in access to hospitals for black infants should strongly impact racial convergence in these causes of death. II. Regional Trends in Infant Mortality Here, we test the hypotheses that: black-white infant mortality rates (particularly for treatable causes) should converge more after 1965 in the states with the most hospital segregation before the Civil Rights Act; there should be similar patterns in racial differences in hospital access; and the racial patterns in infant mortality and hospital access should be correlated both across states and within the South, with the greatest convergence occurring in Mississippi. The data used in this section come from the annual publications of the Vital Statistics of the United States (see the Appendix for more details). To quantify the degree of segregation and racism in hospitals before the Civil Rights Act, we use racial differences in the proportion of births occurring in a hospital with a physician present the only measure of hospital access available in the Vital Statistics. Becker s (1971) theories of labor market discrimination imply that racial discrimination will be higher in imperfectly competitive markets where blacks are a higher share of the population. Thus, we first examine how our measures of racial inequality in infant care and health outcomes correspond with the share of all births that are black across states. 12 Figure 2 demonstrates this idea. It presents black-white differences in the percentage of births 11 In the more serious case of bacterial (cf viral) pneumonia, antibiotics, including penicillin, are critical to recovery. Infants with either type of pneumonia often required oxygen, intravenous fluids, and aerosols to assist with breathing. In the case of bacterial pneumonia, antibiotics were administered intravenously in the 1960s. 12 This is the first study to use the implications of Becker s theory to quantify discrimination in hospital care. Though not addressed in detail in Becker (1971), the proportion of a local population that is black may be the result of conditions (e.g., fertility of land, factor inputs and prices, industrial labor demand) that affect both the location and mobility of African-Americans and the (discriminatory) institutions that may arise in response (e.g., Jim Crow laws after the abolition of slavery). 9

12 that occur in a hospital (Panels A and B) and in post-neonatal mortality rates (Panels C and D), by the percentage of births that are black in the state. 13 Panel A shows that between 1958 and 1960 before the Civil Rights Act there is a strong association between racial inequality in hospital births and the proportion of births that are black. Non-Southern states have much lower black birth shares and much less racial inequality in hospital births than most of the South. However, Missouri and Illinois the non- Southern states with the highest black birth shares had levels of inequality similar to the Southern states with the lowest black birth shares. Alabama, South Carolina and Mississippi have the greatest black birth shares and hospital inequality between 1958 and 1960 the birth share and gap in hospital births in Mississippi were 54 and 51 percent, respectively. A bivariate linear regression of hospital gaps on birth shares for all 24 states results in a regression coefficient (standard error) of (0.131), with birth shares explaining 79 percent of the variation in hospital gaps across states. Panel B shows a significant narrowing in the hospital birth gap after the Civil Rights Act. By , while the black share of births has changed little since the end of the 1950s, most states have no racial gap in hospital birth rates. The greatest convergence occurs in the states with the greatest inequality before the Civil Rights Act, with Mississippi s gap falling from 51 to 6 percent. The linear association between hospital gaps and birth shares is statistically significant regression coefficient (standard error) of (0.039); but is seven-times smaller in magnitude, with birth shares now explaining 49 percent of the overall variance in hospital gaps. Panels C and D show similar patterns in the black-white gap in post-neonatal mortality rates (PNMR). Before the Civil Rights Act, the states with the highest black birth shares also have the largest racial gaps in PNMR. Among non-southern states, Missouri and Illinois have the highest PNMR gaps in 1958 to 1960 ( per 1,000 births) percent higher than the gap in Michigan. In the South, Mississippi and South Carolina have black-white PNMR gaps of 17.8 (per 1,000 births), which is nearly 13 For most of the analysis, we use data on nonwhites as an accurate proxy for blacks since we only examine states in which over 95 percent of nonwhite births are black (over 98 percent for most of these states). 10

13 double the gaps in Tennessee and Texas. The linear regression of PNMR gaps on birth shares results in a regression coefficient of (0.029), with birth shares explaining 79 percent of the variance in PNMR gaps across states. After the Civil Rights Act (Panel D), there is substantial convergence in racial PNMR gaps. The states with the greatest inequality at the end of the 1950s experience the greatest PNMR convergence e.g., Mississippi s PNMR gap falls by 55 percent by the early 1970s. The corresponding regression coefficient is now (0.021) three times smaller than in Panel C with birth shares accounting for 52 percent of the variation in PNMR differences. The patterns in Panels C and D correspond with those in Panels A and B. We conclude that: 1) discrimination in hospital care before the Civil Rights Act was greater in states with larger black populations; 2) discrimination in access after the Act fell the most in these states; and 3) racial gaps in causes of infant death amenable to hospital care (post-neonatal mortality) responded to this increase in access for blacks. On the last point, before the Act black birth shares are a strong predictor of the racial difference in PNMR due to gastroenteritis and pneumonia explaining 65 percent of its variation but their predictive power falls to 47 percent after the Act. In contrast, birth shares have little correlation with neonatal mortality differences both before and after the Act, with R-squared s of 7 and 6 percent. To further investigate the timing of the racial convergence in infant mortality, we examine annual trends in three different regions of the country labeled Deep South (Alabama, Georgia, Mississippi, North Carolina, South Carolina); Rest of South (Arkansas, Delaware, Florida, Kentucky, Louisiana, Maryland, Tennessee, Texas, Virginia, West Virginia); and North (Illinois, Indiana, Michigan, Missouri, New York, Ohio, Pennsylvania). This division of states is based on both geographic proximity and the findings in Figure We do not include Connecticut and New Jersey in this analysis since both states are missing data on nonwhites in some of the years considered. The summary statistics in Table A1 show: Deep South states have the highest share of black births occurring in rural, non-metro counties (70% in ) relative to the Rest of the South (41%) and North (5%); racial inequality in hospital birth rates and PNMRs before the Civil Rights Act is greatest in the Deep South, particularly in non-metro counties; inequality is much lower in metro counties in all three areas; the greater inequality in the Deep South before the Civil Rights Act is due to both greater inequality within metro and non-metro counties and a higher proportion of black births occurring in non-metro counties; and the black-white gaps in hospital birth rates and post-neonatal mortality fell substantially more in the Deep South (both non-metro 11

14 Figure 3 presents annual data on black and white neonatal (NMR) and post-neonatal (PNMR) mortality rates in the three regions. For white infants, the levels and patterns of NMR and PNMR are similar in all three regions, with both mortality rates declining over the period and at a faster rate after For black infants, however, there are significant, cross-regional differences in the mortality level and change, particularly for the PNMR. Panel B shows that black PNMRs are highest in the Deep South 21.6 per-thousand in 1965, which is 38 (85) percent higher than in the rest of the South (North) and relatively stable before the Civil Rights Act. The Deep South also exhibits the greatest reduction in black PNMRs after 1965, declining by over 10-per-thousand by 1972, though there are accelerated reductions in all three regions. Black NMRs are highest in the North and lowest in the Deep South in Panel A possibly due to selective survival to live birth or reporting error (David and Siegel 1983) but their timeseries patterns are similar, with a slightly larger decline in the North after the mid-1960s. Table 1 presents the results of fitting trend-break linear regression models to the patterns in Figure 3. Using data from 1960 to 1974, we separately estimate the following equation in each area: (1) y srt = γ r t + β r (t 1965) 1(t > 1965) + α sr + ε srt, where y is a measure of infant mortality; s, r and t index state, race and time; α sr are state-by-race fixed effects; t is a pre-1965 time trend that is allowed to vary by race; (t 1965) 1(t > 1965) allows for a shift in trend after 1965 (1( ) is an indicator function equal to one after 1965), which is allowed to vary by race; and ε srt is an idiosyncratic error term that is allowed to be heteroskedastic and correlated over time at the state-level. 15 The regressions are weighted by the number of births in each state-race-year cell. The table contains estimates of black-white differences in pre-1965 trends (γ b γ w ) and post-1965 trend breaks (β b β w ), for four different infant mortality measures in each region. The first two columns and metro counties) after the 1964 Act than in the corresponding areas of the Rest of the South and the North. Almond, Chay and Greenstone (2003) provide an analysis of racial trends in the rural (non-metro counties) South and urban (metro counties) South and North. 15 Here and below we allow for state- and county-level clustering over time in the errors whenever we condition on location fixed effects. We also calculated confidence intervals using the wild cluster bootstrap approach that Cameron, Gelbach, and Miller (2007) find can lead to finite-sample improvements over standard clustering methods when the number of clusters is small. The resulting bootstrapped confidence intervals are similar to those implied by the clustered standard errors presented in this paper. We thank Jonah Gelbach for providing the code for the wild cluster bootstrap. 12

15 show no acceleration in racial convergence after 1965 in either neonatal death or neonatal death due to premature birth and low birth weight in the Deep South. In the other two regions, there is greater convergence in NMRs after 1965, with the NMR gap in the rest of the South and North falling an additional 0.50 to 0.64 (per 1,000 births) per-year relative to pre-existing trends. Columns 3 and 4 show significant reductions in the Deep South in the gaps for post-neonatal death and post-neonatal death due to gastroenteritis and pneumonia (and influenza). While there is no detectable trend from 1960 to 1965, the PNMR gap falls by 1.16 per-year after 1965, implying a convergence of 10.4 (per-thousand births) by The post-1965 PNMR convergence in the Deep South is 2.8 and 3.5 times greater than the convergence in the other two regions; the Rest of the South experienced relative improvements in black PNMRs before 1965 as well. In the Deep South, the trend shift in the diarrhea and pneumonia death gap accounts for 62 percent of the PNMR trend break. In the other two regions, the estimated trend breaks for these two causes are not significant. The convergence in the overall infant mortality gap after 1965 (relative to pre-existing trends) is 30 to 37 percent greater in the Deep South than in the other two regions. This is entirely due to a greater reduction in the PNMR gap, particularly among causes of death for which effective treatments were only available in hospitals. In a pooled analysis, the differences in the size of the trend breaks in the PNMR and treatable PNMR gaps between the Deep South and the other two regions are highly significant, with t-ratios of 4.24 and 5.11 for PNMR, and 4.66 and 2.59 for preventable PNMR. For the NMR gap, the across-area differences in trend breaks are not significant at conventional levels. We now perform an annual analysis of across-state racial inequality in the spirit of Figure 2. We estimate the following regression equation using data from 1960 to 1975: (2) y srt = λ t fracblack st + β t Black srt fracblack st + γ rt + α sr + ε srt, where y is the outcome of interest; γ rt are race-specific year effects; α sr are state-by-race fixed effects; fracblack st is the fraction of births in the state-year that are black; Black srt is an indicator equal to one if the cell is for black infants; ε srt is an error that is allowed to be heteroskedastic and correlated over time at the state-level; and the regressions are weighted by the number of births in each of the 704 state-race-year cells. The parameters of interest are the β t s, as they measure changes over time in the association 13

16 between black-white differences in outcomes and black birth shares. Figure 4 presents the results for hospital birth rates, neonatal mortality rates, and post-neonatal mortality rates attributable to diarrhea and pneumonia ( Preventable PNMR). It shows the estimated differences (β t β 1960 ) measuring the change in coefficients relative to 1960 where β 1960 has been normalized to zero. The statistical significance of the estimated differences is also indicated. The preventable PNMR series shows little change in the correlation between state-level racial gaps and black birth shares before 1966, but a sharp reduction after which is significant at the fivepercent level by 1968 and the one-percent level by The correlation between the NMR gap and birth shares, by contrast, is constant over the entire period. Preventable PNMR and all-cause PNMR account for 88 and 100 percent, respectively, of the overall infant mortality convergence by At the same time, the negative correlation between the hospital birth gap and black birth shares falls in magnitude, with an accelerated convergence after 1968 that, along with preventable PNMR, is significant at the onepercent level in the years 1972 to Figure 2 established that Mississippi was the state with the greatest racial inequality in both hospital access and infant mortality before the Civil Rights Act and the greatest convergence afterward. Figure 5 investigates this further by contrasting Mississippi s annual patterns with those of Alabama, Illinois and New York. Panel A shows that, while white post-neonatal mortality was similar in all four states, the black PNMR was substantially higher in Mississippi before the Act. As of 1965, Mississippi s black PNMR was 25.6 per thousand births, which is 30-percent higher than the rate in Alabama, two and 2.5-times the rates in Illinois and New York, and five-times the white rates. Mississippi has the greatest decline in black PNMRs after the Act, falling by a factor of two from 1966 to There is evidence that hospital desegregation played a role: black PNMRs in Mississippi exhibit a staggered decline after 1965, with the sharpest reduction occurring between 1966 and This matches hospitals staggered receipt of Medicare certification in Mississippi, which required hospital integration per Title VI. Alabama had a sharp decline in black PNMRs between 1965 and 1967 and little reduction afterward, which corresponds with the fact that 90 percent of its hospitals had received Medicare certification by February In Illinois and New York states with laws in place forbidding 14

17 discrimination in publicly-funded hospitals before the 1964 Act there is only a slightly accelerated decline in black PNMRs after Panel B presents differences between Mississippi and Alabama and between Mississippi and Illinois in black post-neonatal mortality and hospital birth rates. Black PNMRs in Mississippi converged toward those in Alabama and Illinois after the mid-1960s, and their relative levels and timing of changes are highly correlated with the patterns of black hospital birth rates between 1959 and Hoynes and Schanzenbach (2008) show that the Food Stamps program rolled out faster in Illinois than in Mississippi (and Alabama) between 1964 and 1975; suggesting that it cannot explain the patterns in Figure III. The Impact of Hospital Integration in Mississippi Mississippi stands out as the state with the most racial inequality (and greatest black population shares) before the Act and the largest decline in inequality after the Act. We now use county-level data for Mississippi to establish evidence of the causal role of hospital desegregation. We show that: prior to the Act, the counties with larger black population shares had greater racial inequality in infant access and mortality; and the timing of the gains for black infants correspond with the timing of hospitals receipt of Medicare certification our measure of Title VI compliance by county. This is not the case for any of our other, observable variables. We link several sources to construct a database for Mississippi s counties. From Mississippi Department of Health (MSDH) administrative data, the annual Vital Statistics of the United States, and the Natality and Mortality Detail Files, we have information on numbers of births by race, gender, mother s age and marital status, county of mother s residence, county of the birth s occurrence, whether the birth occurred in a hospital with a physician present; and infant deaths by race, county, timing of death (neonatal or post-neonatal) and detailed cause-of-death. Data on the allocation of hospital beds-by-race 16 Figure A1 contrasts infant access and mortality in Mississippi and South Carolina the state with the secondhighest racial inequality before the Act in Figure 2. It shows a strong association between Mississippi-South Carolina differences in black (and black-white differences in) infant mortality and differences in hospital birth rates. Mississippi s black infants lose ground in both access and health from 1960 to 1965, and catch up significantly to South Carolina s infants between 1970 and 1972, when the remainder of Mississippi s uncertified hospitals received Medicare certification (all of South Carolina s hospitals had been previously certified). 15

18 for 1960 and 1962 to 1964 come from the List of Mississippi Hospitals with License Status and Governing Bodies. The 1972 Mississippi Statistical Abstract and the Regional Economic Information System of the Bureau of Economic Analysis provide per-capita income and government transfer payments, by type, for 1959, 1962 and 1965 to Data on each Mississippi hospital, including their dates of certification for Medicare, come from the annual issues of the American Hospital Association Guide. The Appendix provides more details on each data source. A. Empirical Results We first examine the association of racial inequality in access and mortality with black birth shares across counties. Table 2 presents the results before the Civil Rights Act (Panel A) and after (Panel B) from the following regression equations: (3a) y c = α + X cr β r + θ fracblack c + ε c, and (3b) y cr = α + X cr β r + γ Black cr + λ fracblack c + θ Black cr fracblack c + ε cr, where c and r index county and race; X is a vector that includes race-specific proportions by mother s age categories, marital status, and infant gender, with effects that vary by race; fracblack is the fraction of births in the county that are black; Black is an indicator equal to one if the cell is for black infants; ε is an error that is allowed to be heteroskedastic; and the regressions are weighted by the number of births in each cell. Equation (3a) is used for the county-level outcomes of per-capita income and the black-white ratio of hospital beds per birth; and equation (3b) for black-white differences in hospital birth rates and post-neonatal death due to gastroenteritis and pneumonia. The coefficient of interest is θ, as it measures the correlation between racial inequality and birth shares across counties. Columns (1a) and (1b) in Panel A show that counties with higher black birth shares between 1962 and 1964 have lower per-capita income in 1962, with a correlation that is unaffected by controlling for the characteristics of the women giving birth. Columns (2a) to (2c) show that black-white differences in hospital birth rates are also highly correlated with birth shares e.g., a one-standard deviation change in birth shares (0.2) is associated with an increased hospital birth gap of 14-per-100 births. This correlation is unaffected by controlling for mother s characteristics and per-capita income, and income has no effect 16

19 on the hospital gap. Similar results for the black-white ratio of hospital beds per-birth are shown in columns (3a) to (3c), with a one-standard deviation increase in birth shares associated with a 0.18 decrease in the ratio. Income per-capita is uncorrelated with the hospital bed ratio. Thus, before the Civil Rights Act, hospital access for black infants was lower in counties that were more black. Almond, Chay and Greenstone (2003) document similar evidence for other measures of access e.g., we find a negative association between black birth shares and the likelihood of a county receiving a new diagnostic health center in the late 1950s and early 1960s. Columns (4a) to (4c) show that the racial gaps in access are correlated with large differences in post-neonatal death rates due to diarrhea and pneumonia. A one-standard deviation increase in black birth shares is associated with a 2.2 (per 1,000 births) larger gap in preventable PNMR. The racial gap in neonatal mortality has no correlation with black birth shares (results available from the authors), implying that only the causes of death more responsive to hospital admission vary by the racial composition of a county. County-level income has no effect on the preventable PNMR gap. After the Civil Rights Act (Panel B), there remains a strong, negative correlation between percapita income in 1970 and black birth shares between 1970 and However, the racial gap in hospital birth rates is now weakly correlated with black birth shares after adjusting for mother s characteristics. The correlation is 3.5 to 5.5-times smaller in magnitude than it is between 1962 and Columns (4a) to (4c) show a weak association between birth shares and the gap in preventable PNMR, which is 3 to 5- times smaller than before the Civil Rights Act. These findings mirror those in Figure 2. We summarize the annual changes across Mississippi s counties by fitting trend-break regression models similar to equation (1). Using data from 1960 to 1970, we estimate the regression: (4) y crt = γ r t + β r (t 1966) 1(t > 1966) + X crt β r + α cr + ε crt, where X contains race-specific proportions of mother s age and marital status and infant gender; and we allow for a shift in trend after The regressions are weighted by the number of births in each cell. Table 3 presents the results for black-white differences in (Panel A), and white levels of (Panel B), four measures of infant mortality, as well as the percentage of births in a hospital with a physician present. Columns (1a) and (1b) show that while the black-white gap in infant mortality changed little 17

20 from 1960 to 1966, it fell at an annual rate of 3-per-thousand after Columns (2a) to (3b) show that 70 percent of this accelerated convergence can be attributed to infant deaths due to diarrhea, pneumonia and accidents or to post-neonatal mortality. Further, 87 percent of the racial convergence in PNMRs is attributable to two causes of death diarrhea and pneumonia [columns (4a) and (4b]. 17 Adjusting for mother s characteristics has little effect on the estimates. In fact, the relative characteristics of the black women who gave birth worsened over the decade e.g., black-white differences in the proportions of births occurring among teenage or unmarried women rose steadily between 1960 and Columns (5a) and (5b) show a corresponding acceleration in the racial convergence of hospital birth rates after 1966, with the black-white gap falling at an annual rate of 4-per-100 births more than before In Panel B, white infants exhibit improvements in hospital birth rates and in deaths due to preventable causes between 1960 and Their estimated trend breaks imply similar rates of improvement after Thus, white outcomes were unaffected by the greater hospital access of black infants after (In Section IV, we document an explanation for this finding; Mississippi s hospitals were operating well below capacity before integration.) We now examine the association of this racial convergence with the staggered timing of Title VI compliance in Mississippi s hospitals, as measured by their receipt of Medicare certification. The Guide Issues of the AHA are used to determine when a county s black residents received access to a desegregated hospital. Appendix Table 1 shows the dates when each hospital in each county received Medicare certification, and the resulting year of hospital integration assigned to the county. In general, we assigned to each county an integration year of: 1967 if it contained a hospital that received certification by February 1967; 1968 if its earliest hospital certification was in February or October of 1968; and 1969 (or 1970) if its earliest certification was in October 1969 (or November 1971) Separate estimation results for gastroenteritis and pneumonia imply that about 45 percent of their combined racial convergence is due to the former and 55 percent to the latter. 18 Below, we discuss evidence that all of Mississippi s hospitals were in compliance with Title VI by the beginning of 1970 in order for Mississippi to quality for federal matching funds for Medicaid. In the AHA Guides, there is a gap in the dates of Medicare certification between October 1969 and November