Health & Place 22 (2013) 56–67 Contents lists available at SciVerse ScienceDirect Health & Place journal homepage: www.elsevier.com/locate/healthplace Segregation and cardiovascular illness: The role of individual and metropolitan socioeconomic status Antwan Jones n Department of Sociology, The George Washington University, 801 22nd Street NW, Suite 409C, Washington, DC 20052, USA art ic l e i nf o a b s t r a c t Article history: Received 17 May 2012 Received in revised form 29 January 2013 Accepted 11 February 2013 Available online 22 March 2013 Demographic and epidemiologic research suggest that cardiovascular illness is negatively linked to socioeconomic status and positively related to racial residential segregation. Relying on 2005 data from the Behavior Risk Factor Surveillance Survey and the American Community Survey, this study examines how segregation and SES (individual and metropolitan) impact hypertension for a sample of 200,102 individuals. Multilevel analyses indicate that both segregation and hypersegregation are associated with hypertension, net of individual and spatial SES. While individual and metropolitan SES have independent effects on hypertension, these effects also differ across segregation type. In segregated and hypersegregated environments, highly educated and high-earning individuals seem to be protected against hypertension. In extremely hypersegregated areas, areas where there is very little interaction with nonblack residents, SES does not have any protective benefit. These findings reveal that SES has differential effects across segregation types and that hypertension in disadvantaged (extremely hypersegregated) areas may be a function of structural constraints rather than socioeconomic position. & 2013 Elsevier Ltd. All rights reserved. Keywords: Health Disparities Hypersegregation Hypertension Segregation Socioeconomic Status 1. Introduction Sociological and social epidemiological research on spatial inequality emphasizes how racial residential segregation and socioeconomic conditions influence health. Specifically, impoverished individuals of low socioeconomic standing may live in residentially segregated areas because of differential access to employment or educational opportunities. For this socioeconomically disadvantaged population, gaining employment or a degree could potentially propel them into higher levels of socioeconomic status (SES), and could provide opportunities to live in more racially and socioeconomically diverse areas (de Souza Briggs, 1997). Moreover, residentially segregated areas and economically disinvested communities may be gatekeepers for health-promoting resources because of a lack of social infrastructure. Research has shown that segregated areas lack convenient access to stores (Zenk et al., 2005), places to exercise (Kaczynski et al., 2010) and quality health care facilities (Williams and Collins, 2001). In addition, segregated areas have been associated with a higher risk of exposure to crime and environmental hazards, which are sources of stress on a daily basis (Acevedo-Garcia et al., 2003). Because there are large variations in the degree of racial clustering in metropolitan areas, racial differences in health also tend to be more n Tel.: þ1 202 994 0266; fax: þ1 202 994 3239. E-mail address: antwan@gwu.edu 1353-8292/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthplace.2013.02.009 pronounced in segregated cities (Subramanian et al., 2005). In this manner, residential segregation not only exacerbates socioeconomic disadvantage but also geographically accumulates healthrelated risks for minority residents in these areas. While the research on segregation and health is expansive, there are noticeable gaps that this research addresses. First, there is a paucity of research relating spatial inequality (vis-à-vis residential segregation) across metropolitan areas to socioeconomic inequality within those areas. At the individual level, social class and health (Adler et al., 1994), race and health (Williams and Collins, 2001) and race and social class (Oliver and Shapiro, 2006) are all inextricably linked, but the interconnectivity of these three key social indicators varies across space (AcevedoGarcia et al., 2003). As such, this research uses geographic heterogeneity to explore how space plays a role in determining health among people of various socioeconomic and racial backgrounds. Second, because poor socioeconomic conditions (e.g., supermarket access and quality health care facilities) and dangerous environmental conditions (e.g., crime and environmental hazards) are characteristic of segregated areas, it is unclear to what extent a person's own socioeconomic status and the socioeconomic environment in which one lives influences an individual's risk of being chronically ill. That is, could the potentially negative effects of living in an impoverished area be ameliorated by elevating a person's socioeconomic standing? The present study suggests that in some areas where there is extreme segregation, certain A. Jones / Health & Place 22 (2013) 56–67 socioeconomic indicators (such as education) may not be as predictive of health as other indicators that are more proximately related to health care (such as income). Also, while higher levels of segregation are associated with poor health, it is uncertain if extreme types of segregation (i.e., hypersegregation) have differential effects on health or if the effects of hypersegregation on health are identical to those from living in segregated environments. Thus, it is necessary to determine whether there are differences in individual and metropolitan-level health risks in segregated versus hypersegregated areas (Massey and Denton, 1989; Wilkes and Iceland, 2004), where there is an almost exclusive interaction with members of one's own race. Third, while it is assumed that segregation is negatively associated with all health outcomes, very few studies quantify the effect that segregation has on hypertension. These studies focus on racial differences in hypertension within segregated areas and assume there is socioeconomic homogeneity within segregated areas. Research on this topic suggests that whites and blacks have similar cardiovascular outcomes when they live in areas with similar levels of segregation (Thorpe Jr. et al., 2006), and higher rates of hypertension for both whites and blacks are associated with higher levels of segregation (Kershaw et al., 2011). One study finds that within the context of New York City, segregation is not associated with racial differences in hypertension (White et al., 2011), which suggests that not all segregated areas produce racial differences in hypertension. Because of the insufficiency of research on this topic, further evidence is needed to assess whether race, socioeconomic status, or both are key in understanding the role segregation has on hypertension diagnoses. Hypertension is an important health concern: it is a major risk factor for heart disease, stroke, congestive heart failure, and kidney disease (Kannel, 1996). Currently one-third of adults have been diagnosed with hypertension (Rabe-Hesketh and Skrondal, 2012). The current study uses data from a large, nationally representative sample to answer these three questions. Specifically, this research explores whether living in segregated areas is predictive of hypertension, how socioeconomic factors at the metropolitan level are related to socioeconomic factors at the individual level in predicting hypertension, and whether a person's SES buffers hypertension differently depending on the type of segregated environment in which he or she lives. In this manner, different levels and kinds of segregation in metropolitan areas may be shown to produce differential effects from individual SES on whether a person will be diagnosed with having hypertension. 2. Background 2.1. Racial residential segregation and health As a persistent feature in the US, residential segregation, or the extent to which two or more groups are physically separated in urban areas, is tied to poor health among African Americans (Williams and Jackson, 2005) through institutional racism, which is designed to protect whites from social interaction with minorities (Williams and Collins, 1995; Wilson, 1987). Further, the degree of residential segregation is much greater for blacks than for any other racial group (Massey and Denton, 1989). While individual dimensions of segregation have been tied to health, it is important to note that blacks also experience simultaneous high segregation across multiple dimensions of segregation (Osypuk and Acevedo-Garcia, 2008). The deleterious health effects of this kind of segregation, called hypersegregation, are less established in the literature. However, for both segregation and hypersegregation, related poor health outcomes range from disparities in 57 engaging in risky behaviors such as smoking or drinking alcohol (Kramer and Hogue, 2009) to mortality (Eitle, 2009; Hearst et al., 2008). It is important to disentangle segregation from hypersegregation for several reasons. First, methodologically, the two are composed of different measurements. Hypersegregation is defined as jointly high values on five segregation indices (evenness, exposure, centralization, clustering and concentration). Conversely, segregation is conventionally defined as lying on the numerical continuum of one or two segregation indices. Thus, areas can be quantified as having low or high levels of segregation, and areas can be typified as being hypersegregated. Second, conceptually, segregation and hypersegregation present different social realities for individuals who reside in the areas. By definition, segregation is the separation of one group from another. In this research, segregation is further clarified by suggesting that it occurs racially within the context of where people live. Thus, racial residential segregation is the racial separation of one group in a particular residential area from another. The division between races is physical. In contrast, hypersegregation is defined as a “multidimensional character [istic]” (Massey and Denton, 1989: 389) whereby minorities are economically, educationally, environmentally, politically, residentially, and socially isolated from whites. Accordingly, there is an exclusive daily interaction with members of one's own race within hypersegregated locales. Thus, the division between races is extreme primarily because it is both physical and social. Third, compositionally, black representation in segregated and hypersegregated metropolitan areas is distinctively different. Massey (2004) suggests that a majority of all blacks and a great majority of urban blacks experience high levels of residential segregation in metropolitan areas. Additionally, about half of all urban blacks and about 40% of all blacks experience hypersegregation, which is the kind of separation that mirrored South African apartheid (Massey, 2004). The compositional difference between segregated and hypersegregated areas is an important distinction because educational, health and social outcomes have been shown to be more detrimental for blacks in hypersegregated metropolitan areas than in segregated ones. Specifically, blacks in hypersegregated communities experience higher frequencies and more negative ramifications of school dropout (Williams and Collins, 2001), low birth weight (Osypuk and Acevedo-Garcia, 2008) and personal victimization (Eitle, 2009) than blacks in segregated communities. Thus, it is critical to compare segregated and hypersegregated metropolitan areas instead of treating them as the same phenomenon. While researchers have linked segregation to health outcomes, the literature shows some inconsistency, as other research illustrates advantageous and race-based nuances to the relationship between segregation and health. For example, Smaje (1995) suggests that the concentration of minorities in areas may mean that there is a greater level of political empowerment and community integration, which are both associated with favorable health. In addition, LeClere et al. (1997) notes that in the National Health Interview Survey-National Death Index linked files, neighborhood characteristics such as ethnic concentration lowered the risk of mortality but only for particular ethnic groups such as Mexican Americans. As evident in these studies, the community's social content (i.e., how individuals in a community are organized and relate to one another) may be a protective factor for health. Regardless, a plethora of research suggests that segregated environments are associated with deleterious health outcomes. These effects are concentrated in the spatial environment and are not artifacts of general racial differences in health. To illustrate, Fang et al. (1998) uncovered racial differences in health in racially concentrated areas. Independent of socioeconomic status, whites 58 A. Jones / Health & Place 22 (2013) 56–67 living in majority black neighborhoods in New York City had higher all-cause and coronary heart disease (CHD) mortality rates than whites living in majority white neighborhoods. However, for blacks, there was no difference in all-cause and coronary heart disease mortality rates net of SES when they lived in majority black or white neighborhoods. Moreover, recent research suggests that segregation may be positively associated with black–white differences in CHD mortality in some US counties and negatively associated in others (Gebreab and Diez-Roux, 2012). Despite the small number of studies that illustrate a nuanced relationship between race, space and health, it is generally understood that segregation restricts socioeconomic opportunities for minority individuals, which consequently affect health (Acevedo-Garcia et al., 2003). In this case, segregation can be conceptualized as the accumulated effect of poverty among its residents. There is a paucity of research that examines metropolitan socioeconomic status in tandem with individual-level SES. Metropolitan SES goes beyond individual- and neighborhood-level indicators in order to capture larger forces that help shape the socioeconomic conditions of a metropolitan area. Metropolitan SES is important to the study of health in three key ways. First, areas are generally more than the sum of individuals, and thus, metropolitan SES is more than the agglomeration of individuallevel SES. Metropolitan areas also encompass infrastructure, which would influence resources available to residents in a given area. Thus, incorporating socioeconomic indicators above and beyond individual SES measures can tap into a structural aspect to segregation. Second, neighborhoods in metropolitan areas provide conditions for healthy (and conversely, unhealthy) behaviors. Research on built environment and health suggests that segregated and poor neighborhoods tend to have fewer healthy food options while having greater access to corner stores which provide foods high in calories, sugar, and trans fat (Kramer and Hogue, 2009). In addition, neighborhoods that have parks, gyms, and bike/pedestrian trails are also found in less segregated metropolitan areas with higher affluence. Indeed, these community perks are represented in white, upper middle-class neighborhoods, which may indicate that businesses and urban planners take into account individual SES when strategically building up metropolitan SES (Morland et al., 2002). Thus, the conditions and the amenities associated with living in a particular neighborhood may not be accurately captured through the use of individual SES measures alone. Third, metropolitan areas also have histories tied to access to advantage. There is a semi-stabilizing aspect to these areas, in that, historically (and longitudinally), certain communities within metropolitan areas have been racially and/or socioeconomically exclusive. As a consequence, low-income, minority groups had poor access to affluent residential areas and to the often highquality resources and services available in those areas, limiting socioeconomic achievement. In turn, denial of homeownership and other mechanisms of wealth accrual created barriers for racial minorities to have the means to live in established areas of affluence (Massey and Denton, 1993), creating de facto racially segregated neighborhoods. The disproportionate amount of wealth that whites have accrued, and continue to accrue (Taylor et al., 2011) has translated into a health advantage. Findings from recent research suggest that wealth is predictive of health, even after adjusting for income (Pollack et al., 2007). In addition, the black/white wealth gap is partially responsible for the black/white health gap, independent of differences in income between blacks and whites (Oliver and Shapiro, 2006). While the mechanisms used to exclude certain groups out of neighborhoods are not as visible as in the past (Massey and Denton, 1993), both historical disadvantage and new methods of exclusion (e.g., Freeman (2005)) result in sustained uneven development. In turn, uneven development translates to areas having differing levels of SES (Ellen et al., 2001). In all, segregation is uniquely tied to socioeconomic conditions of metropolitan areas which could relate to racial differences in health. 2.2. Theoretical perspectives to segregation and health Current perspectives regarding segregation and health stress the importance of the built environment in facilitating healthy lifestyles. In racially-segregated residential areas, often characterized as being of low socioeconomic standing, there are limited facilities that would serve to promote healthy living. For instance, research has suggested that low-income and majority–minority neighborhoods are associated with a lack of parks and open spaces for leisure and exercise (Adler et al., 1994; Freeman, 2005), supermarkets and other healthy food-serving outlets (Morland et al., 2002; Zenk et al., 2005), amenities associated with a middle-class lifestyle such as fitness centers (de Souza Briggs, 1997), and environmentally-sound neighborhoods without chemical hazards or pollution that could lead to a decreased life expectancy (Do et al., 2012; Oliver and Shapiro, 2006). Indeed, lower SES individuals who tend to live in areas where there are limited health facilities often have an increased likelihood of being in poor health (Acevedo-Garcia et al., 2003), as they do not have the resources (e.g., cars and public transportation) to compensate for living in areas that lack these opportunities. This disadvantage manifests as "double jeopardy" for low-income racial and ethnic minorities, who are structurally disadvantaged by both their residential location and their relative positions in the hierarchies of race and class in US society (Taylor et al., 2011). As such, research has suggested that low-income minorities fare worse in segregated areas than minorities in the same area who have greater social standing (Pollack et al., 2007). Moreover, the built environment alone cannot fully explain racial differences in health outcomes. Recent research suggests that the health risks are similar for many outcomes for individuals who live close to these amenities and those persons who live far from them (Raudenbush and Byrk, 2002). Access and utilization are often discussed in tandem when discussing the built environment. That is, places for exercise highly curb the risk of being hypertensive (Messer et al., 2010), high-crime, fairly-dense metropolitan areas, then residents may not feel compelled to exercise since their immediate and tangible risk of being a victim of crime would outweigh the long-term risk of having unmanageable blood pressure (Maldonado and Greenland, 2002). Another mechanism which affects the health of individuals living in segregated areas is environmental stress. The early work of Harburg et al. (1973) and the subsequent work of James et al. (1984) on the John Henryism hypothesis contains insight on the relationship between stress and health in the urban environment. They argued that persons, African American or white, living in high “socioecologic stress” areas (characterized by low SES and high rates of social instability, as measured by crime) had a high risk for stressful experiences on a daily basis, increasing the likelihood of high blood pressure (Thorpe Jr. et al., 2006). For African Americans, and especially darker skinned African American men (Dressler, 1991; Dressler et al., 2003), there was the added stressor of racist interactions with police or other white persons in positions in authority. These racist interactions are likely to provoke hostility on the part of the African American participant in the interaction, who may then suppress that hostility to avoid negative repercussions. Thus, John Henryism hypothesis predicts that darker-skinned African American men who lived in high stress areas and who have suppressed hostility would have the highest blood pressure. Research results have been A. Jones / Health & Place 22 (2013) 56–67 generally consistent with these predictions, although the strength of anger expression and suppression effects have been found to be modest (Schum et al., 2003). While individual and societal level explanations remain prevalent in this literature, there is an emerging research on biosocial mechanisms relating race to health. The need for a new model has emerged because racial differences in health are resistant to full explanation using the usual array of socioeconomic and demographic control variables. Massey (2004) argues that long-term exposure to social disorder and violence in segregated environments produces a high allostatic load (Leal et al., 2011; Mair et al., 2011) among African Americans, which leads to a variety of deleterious health outcomes. He specifies a biosocial model of racial stratification and uses well-documented literature to help support his argument that health is indirectly associated with segregation. This biosocial model takes advantage of the community context in which a person resides. However, lacking empirical evidence, it is important not to assume that segregated and hypersegregated environments explain health outcomes in the same way. Based on the work that has been reviewed, this research attempts to provide clarity on whether individual characteristics or the environment in which a person lives is more important to health. Specifically, this project is concerned with whether the potentially negative effects of living in poor or segregated environments can be potentially negated when SES is high. Past research suggests a tenuous connection between individual and area-level measures in predicting health, particularly with regards to socioeconomic characteristics (Lee and Ferraro, 2007). However, this connection has not been explored at a large scale with all metropolitan areas. This work is also concerned with whether the type of segregation has differential effects on cardiovascular risk. Prior research assumes that increases in segregation levels would produce linear increases in poor health (Acevedo-Garcia et al., 2003). However, it is unclear whether segregated environments are more detrimental to cardiovascular health than hypersegregated areas. As such, this research addresses whether living in segregated areas is predictive of hypertension, how socioeconomic factors at the metropolitan level are related to socioeconomic factors at the individual level in predicting hypertension, and whether a person's own SES buffers hypertension differently, depending on the type of segregated environment in which he or she lives. 3. Data and methods This project relies on two complimentary data sources. Individuallevel information is derived from the 2005 Behavioral Risk Factor Surveillance System (BRFSS) data. The BRFSS is a collaborative project between the Centers for Disease Control and Prevention, and U. S. states and territories that measures behavioral risk factors in the adult (at least 18 years old) population living in households (Centers for Disease Control and Prevention, 2006). Its objective is to collect uniform, state-specific data on preventive health practices and risk behaviors that are linked to chronic disease, injuries and preventable infectious diseases. Respondents are identified through telephonebased methods. Although around 95% of all households have telephones, coverage ranges from 87 to 98% across states and varies for subgroups (Mokdad et al., 2003). Post-stratification weights are used to correct for any bias caused by non-telephone coverage. One of the highlights for using this particular dataset is that the publicly available dataset has geocoded information at the metropolitan level. That is, information on the metropolitan area in which the respondent lives is included in this dataset. A unique strength of this data is its large population sample, where missing data accounts for only 7% of each 59 measure used in this analysis. Unless otherwise specified, mean replacement is used to handle missing data. The analytic sample contains 200,102 persons. To supplement the respondent's information captured in the BRFSS, metropolitan measures come from the 2005 American Community Survey (ACS). The ACS is a monthly household survey being developed by the US Census Bureau to provide data users with annual estimates of household, social, and economic characteristics for geographies and populations of at least 65,000 people. In addition, the ACS annually updates their multi-year demographic estimates for geographies down to the block group. The ACS data will be used to supplement the BRFSS with area-level indicators that affect health. 3.1. Outcome measure The primary focus of the analysis is on how individual and metropolitan measures differentially influence cardiovascular illness. The outcome of interest is a self-reported measure of hypertension. Hypertension is measured by a question that asks if respondents have ever been told by a health professional that they have high blood pressure. Approximately 38% of the total sample had been told that they have high blood pressure. 3.2. Individual-level independent measures A key indicator to health is the socioeconomic standing of the respondent. Three variables are used in order to operationalize socioeconomic status. The respondent's highest level of completed education is divided into four categories: less than high school (the reference category), high school graduate, some college, and graduated college. Income is also assessed and includes the household's total income, excluding income from interest, dividends, and other investment, as well as other income for each person such as disability assistance, social security, and public assistance, which are contributed to the household but are not necessarily from earnings. Poisson iterative regression replacement (Agresti, 2002; Rencher, 2002) was used to substitute any missing data in the original income variables and sample selection tests (Allison, 1995; Miranda and Rabe-Hesketh, 2006) via maximum likelihood were performed to assess whether there was a selection effect of income in the sample. There was no support for a sample selection bias with regards to income. Sensitivity analyses (not shown) indicated that because of the far right skewness of income in this data, categorical breakdowns of income best fit the data. As such, income is represented by a series of dichotomous variables represented by these categories: less than $15,000 (the reference category), $15,000 to $20,000, $20,000 to $25,000, $25,000 to $35,000, $35,000 to $50,000, $50,000 to $75,000, and more than $75,000. Lastly, for all respondents, there is a variable that tracks employment status. This is a binary variable indicating if the respondents worked during the previous calendar year. Other health-related variables at the individual level are used as control measures that have been used in research because they are related to hypertension (Dressler et al., 2003; Eitle, 2009). Body Mass Index (BMI) is a continuous measure of body fat based on height and weight. Smoker status indicates if respondents are current smokers (the reference category), former smokers, or have never smoked at survey date. Alcohol consumption is measured by the number of days per month that the respondents had at least one alcoholic drink. Physical activity is measured by the number of days per week that respondents performed at least 10 minutes of physical activity excluding work-related activities. Insured is a binary variable indicating if the respondent has public or private health insurance. 60 A. Jones / Health & Place 22 (2013) 56–67 Demographic controls are also included in the analysis. This project controls for race, gender, age (in years), and marital status. Race is defined by a series of categorical variables. Blacks serve as the reference category in all analyses. There are additional variables for white and other racial groups. The majority of individuals included in the “other” category are of Asian/Pacific Islander ancestry. However, due to sample size limitation, this particular racial group cannot be separated into a different category. Also, a probable limitation of the dataset is that the sample is overwhelmingly white, which may create constraints in sample variability within the analyses (Agresti, 2002). Marital status is coded with several categorical variables indicating if the respondent is married (the reference category), separated, never married, or coupled with an unmarried partner. 3.3. Metropolitan-level independent measures The focal metropolitan measure used in this research is segregation type. The BRFSS data includes metropolitan statistical area (MSA) codes (also present in the ACS data) that correspond to a particular metropolitan area. The work of Wilkes and Iceland (2004) was used to operationalize the different types of black–white segregation that exist across metropolitan areas. Wilkes and Iceland identified 29 MSAs as hypersegregated areas in their research. Using the assumption that hypersegregated communities are semi-stable throughout time, this research assumes that the hypersegregated areas in 2004 would be hypersegregated in 2005. Wilkes and Iceland (2004) also found that five of the 29 hypersegregated MSAs scored high (0.60 or greater) on five indices of segregation, indicating extreme hypersegregation. Those five indices are evenness, exposure, centralization, clustering, and concentration, which are detailed in Massey and Denton's (1989) work on segregation. Based on the previous work on segregation and hypersegregation, this research distinguishes four unique types of segregation: extreme hypersegregation, hypersegregation, segregation, and nonsegregated. Extremely hypersegregated MSAs correspond to the five MSAs that score high (0.60 or greater) on all five indices of segregation according to Wilkes and Iceland's research. Hypersegregated MSAs correspond to the MSAs who scored high on four of the five indices of segregation, as outlined by the researchers (i.e., the remaining 24 of the 29 originally mentioned in Wilkes and Iceland's research). Segregated MSAs are thus, by definition, those remaining MSAs that scored high on one, two, or three of the five segregation measures. Lastly nonsegregated MSAs (the reference category) are those MSAs that did not score high on any of the segregation indices. Also important to account for is the socioeconomic standing in these metropolitan areas, in order to fully exhaust the spatial nature of hypersegregation. Four measures are used to account for metropolitan SES: first is population size, which is logged since there is univariate non-normality in the measure (i.e., the population sizes are not normally distributed across all metropolitan areas). Second, the percentage of black residents in the MSA is used to control for racial concentration. Third, the percentage of residents in poverty is also used as a control measure. Lastly, the percentage of female-headed households, a measure of concentrated poverty determined for each MSA, is employed. These covariates are based on the 2005 ACS estimates and were merged onto the BRFSS data file using the MSA codes. 3.4. Analytic strategy Hierarchical linear modeling (HLM) was used to assess the independent effects of individual and metropolitan measures on an individual's hypertension diagnosis. Because of the binary nature of the outcome measure, the dependent variable is modeled in a logistic regression run in HLM. Also, to evaluate the fixed effects of the variables as well as the explained and unaccounted variation in the measures, mixed effects HLM modeling was employed. Analyses were performed in Stata 12.1 (StataCorp, 2011) using the XTMELOGIT command. To ensure the most accurate results, ten adaptive Gaussian quadrature (AGQ) points were used. The more points that are used, the more accurate the approximation to the log likelihood (Rabe-Hesketh and Skrondal, 2012). The model development is as follows: Model 1 is the zeroorder model with individual SES measures, Model 2 adds the individual-level controls, Model 3 adds segregation type, and Model 4 adds metropolitan SES measures. It is important to note that the first two models are first-order models, while the last two models incorporate level-2 measures. Thus, for the last two models, random effects will also be estimated as part of variance analysis. Lastly, stratified models will be estimated to see the effect of these measures in nonsegregated, segregated, hypersegregated, and extremely hypersegregated metropolitan areas. 4. Results 4.1. Sample description Table 1 presents an overview of the variables used for this study. With a large, representative sample size of 200,102, the data suggest that 37.8% of the respondents were told by a medical professional that they have high blood pressure (i.e., hypertension). Approximately 8.7% of the sample is black and 7.2% are classified as other. The sample is overwhelmingly white, representing about 84.1% of the entire sample. Turning to the socioeconomic characteristics of the individuals in the sample, over half of the respondents (60%) have either some college training or have completed college. Conversely, less than 10% never finished high school and approximately 30% graduated high school. For income, the sample is somewhat skewed, in that percentages increase as income categories increase. Over half of the sample is at or below the $35,000–$50,000 range, indicating that a sizable proportion of the sample makes above $50,000. In addition, an overwhelming majority of the respondents are employed (88%). The measures associated with health suggest a moderately healthy sample of individuals in the data. The average BMI score is 27.2, which indicates a slightly overweight sample. Also, over half of the sample (52.3%) has never smoked, and 28.1% has quit smoking at the time of the survey. On average, respondents drank some alcohol about 10 days in the previous month, suggesting a low to moderate amount of alcohol consumption. Individuals in the survey engaged in physical activity at least 4 days of the week, on average. A large majority of the respondents have some kind of health insurance (88%). Demographically, the sample is overwhelmingly female, with about 38.2% of the sample comprised of men. The average age of the respondent is around 51.2 years old, although there are sizable representations of age groups in the sample. The predominant union type in this sample is marriage, with about 55.7% of the respondents being married and 2.6% cohabiting with their partner. About 28.1% of the sample has been married, but are legally separated, divorced, or widowed. Close to 14% of the sample has never been married. Table 1 indicates that the respondents overwhelming live in nonsegregated metropolitan areas, with only 30% living in areas with some degree of racial segregation. Moreover, 22.2% of the sample lives in segregated areas, while 6.4% of the sample lives in hypersegregated areas. Comprising only a small number of areas, extremely hypersegregated areas are least represented in this data—less than 2% of the sample resides in this type of segregated metropolitan area. The socioeconomic characteristics of the areas suggest low levels of spatial disadvantage. The mean population is around 848,000 residents. Areas in this sample have, on average, 10.5% of A. Jones / Health & Place 22 (2013) 56–67 61 Table 1 Distribution of individual- and metropolitan-level characteristics in sample. Source: 2005 BRFSS and 2005 ACS. Outcome measure Hypertension diagnosis Individual (level 1) variables Educational level Less than high school High school graduate Some college Graduated college Household income Less than $15,000 $15,000 to $20,000 $20,000 to $25,000 $25,000 to $35,000 $35,000 to $50,000 $50,000 to $75,000 More than $75,000 Employed Body mass index Smoker status Current smoker Former smoker Never smoked Alcohol consumption (days per month) Physical activity (days per week) Insured Race White Black Other Male Age Marital status Never married Cohabiting Married Separated Metropolitan (level 2) variables Segregation type Nonsegregated Segregated Hypersegregated Extremely hypersegregated Population size Percentage of black residents Percentage of residents in poverty Percentage of female-headed households N Mean or % Standard deviation Range 37.8% – – 9.9% 30.1% 27.0% 33.0% – – – – – – – – 9.6% 7.9% 9.6% 12.7% 17.5% 18.8% 23.8% 88.0% 27.2 – – – – – – – – 5.7 – – – – – – – – 8.4–42.1 19.6% 28.1% 52.3% 10.4 4.4 88.0% – – – 5.3 2.4 – – – – 0–30 0–7 – 84.1% 8.7% 7.2% 38.2% 51.2 – – – – 17.1 – – – – 18–99 13.6% 2.6% 55.7% 28.1% – – – – – – – – 69.9% 22.2% 6.4% 1.5% 848,169.8 10.5% 9.7% 12.8% 200,102 – – – – 941,084.2 – – – – – – – 65,076–5,193,448 - Table 2 Number and percentage of hypertensive and non-hypertensive respondents by segregation type and race. Source: 2005 BRFSS and 2005 ACS. Nonsegregated MSA Segregated MSA Hypersegregated MSA Extremely hypersegregated MSA Total N % N % N % N % N % White Hypertensive Non-Hypertensive Total 47,906 70,794 118,700 40.4% 59.6% 11,587 26,405 37,992 30.5% 69.5% 2,924 6,506 9,430 31.0% 69.0% 656 1,561 2,217 29.6% 70.4% 63,073 105,266 168,339 37.2% 62.8% Black Hypertensive Non-Hypertensive Total 5,609 4,090 9,699 57.8% 42.2% 1,675 2,500 4,175 40.1% 59.9% 1,078 1,711 2,789 38.7% 61.3% 276 389 665 41.5% 58.5% 8,638 8,690 17,328 49.9% 50.1% Other Hypertensive Non-Hypertensive Total 3,262 8,240 11,502 28.4% 71.6% 455 1,751 2,206 20.6% 79.4% 120 445 565 21.2% 78.8% 42 120 162 25.9% 74.1% 3,879 10,556 14,435 26.9% 73.1% its residents in poverty. Similarly, 10.5% of residents in the sample are black. Lastly, the average percentage of female-headed households in the metropolitan areas represented in this data is roughly 13%. In order to see if race plays a role in the distribution of hypertension across metropolitan areas, a separate cross-tab is presented in Table 2 that includes a racial breakdown in hypertensive 62 A. Jones / Health & Place 22 (2013) 56–67 diagnoses within each segregation type. As a whole, 37.2% of whites, compared to 49.9% of blacks and 26.9% of other races, have been diagnosed with hypertension. The racial disparity in hypertension is greatest within nonsegregated areas, where there is a 17.4 percentage-point difference between whites and blacks. The black– white gap in hypertension is lowest in hypersegregated areas, where 38.7% of blacks and 31% of whites have been diagnosed with hypertension. 4.2. Multivariate analyses Table 3 presents the multilevel coefficients for predicting hypertension diagnosis for the entire sample. Model 1 includes all measures of individual SES. In Model 1, education is statistically associated with hypertension. Having higher levels of education is protective against being diagnosed with hypertension. This association is also present in both income and employment status. However, for income, a positive relationship between the two lowest levels of income emerges. Compared to individuals who make less than $15,000, individuals who earn between $15,000 and $20,000 have significantly higher odds of being diagnosed with hypertension. This result appears to illustrate that the “working poor,” i.e., individuals who fall short of being below the poverty line and do not qualify for Medicare benefits, are more likely to be diagnosed with hypertension than persons who fall below the poverty line (less than $15,000). Even though the threshold difference between working poor and individuals below the poverty line is compelling, the effect diminishes in Model 2. The remaining individual-level controls related to the demographics of the respondents and their health are added to this model. With the exception of one category of income, the other SES coefficients remain significant, in the same direction and in approximately the same magnitude as seen in the previous model. Supplemental analyses suggest that the inclusion of insurance status creates nonsignificance in the $15,000–$20,000 income category. The health and demographic controls also independently affect hypertension diagnosis net of the socioeconomic status of the respondent. Increasing levels of BMI and alcohol consumption are Table 3 Multilevel analyses of hypertension diagnosis, combined sample (N ¼200,102). Source: 2005 BRFSS and 2005 ACS. Intercept Individual (level 1) variables Educational level (less than high school) High school graduate Some college Graduated college Household income (less than $15,000) $15,000 to $20,000 $20,000 to $25,000 $25,000 to $35,000 $35,000 to $50,000 $50,000 to $75,000 More than $75,000 Employed Body mass index Smoker status (current smoker) Former smoker Never smoked Alcohol consumption (days per month) Physical activity (days per week) Insured Race (black) White Other Male Age Marital status (married) Never married Cohabiting Separated Model 1 Model 2 Model 3 Model 4 −0.03nnn −5.60nnn −5.59nnn −5.67nnn −0.08nnn −0.21nnn −0.36nnn −0.04nnn −0.11nnn −0.25nnn −0.05nn −0.10nnn 0.25nnn −0.03 −0.09nnn −0.23nnn 0.05nnn −0.04nnn −0.05nnn −0.12nnn −0.22nnn −0.32nnn −0.88nnn 0.03 −0.06nnn −0.07nnn −0.07nnn −0.10nnn −0.14nnn −0.19nnn 0.10nnn 0.03 −0.06nnn −0.07nnn −0.07nnn −0.10nnn −0.14nnn −0.19nnn 0.10nnn 0.03 −0.05nnn −0.07nnn −0.06nnn −0.09nnn −0.13nnn −0.19nnn 0.10nnn −0.05n −0.16nnn 0.00nnn −0.03nnn 0.18nnn −0.05n −0.16nnn 0.00nnn −0.03nnn 0.18nnn −0.04 −0.16nnn 0.00nnn −0.03nnn 0.19nnn −0.57nnn −0.56nnn 0.12nnn 0.05nnn −0.57nnn −0.57nnn 0.12nnn 0.05nnn −0.57nnn −0.50nnn 0.12nnn 0.05nnn −0.02 −0.08 0.06nnn −0.02 −0.08 0.06nnn −0.01 −0.06 0.06nnn 0.01n 0.01nnn 0.02nnn 0.01 0.08nnn 0.00 −0.01 0.09 0.66nnn 1.32nnn Metropolitan (level 2) variables Segregation type (nonsegregated) Segregated Hypersegregated Extremely hypersegregated Log population size Percentage of black residents Percentage of residents in poverty Percentage of female-headed households AIC Pseudo R2 (level 1) 2,135 0.06 Notes: Contrast categories are in parentheses. AIC ¼Akaike's information criterion. n p o 0.05. p o0.01. nnn po 0.001. nn 1,983 0.28 1,976 0.29 1,970 0.34 A. Jones / Health & Place 22 (2013) 56–67 positively associated with hypertension. Not smoking at the survey date and engaging in physical activity are associated with not being diagnosed with hypertension. Having insurance is associated with being medically diagnosed with hypertension. Age, gender, and marital status are all positively associated with hypertension in the expected ways. That is, individuals who are older, male, or separated are statistically likely to be diagnosed with hypertension. Also, net of socioeconomic status, whites and others are less likely to be diagnosed with hypertension compared to blacks. Model 3 introduces the higher-order (level 2) metropolitan measures. This model assesses the effect of segregation on hypertension diagnosis. An auxiliary zero-order model containing only the four segregation measures revealed all measures to be positive, significant, and increasing monotonically. This unshown analysis suggests that there is an independent effect of segregation on hypertension. Model 3, which includes the individual measures, shows a somewhat consistent pattern. After adjusting for individual measures (which in this model changed slightly from the previous model but retained their significance), residence in a segregated, hypersegregated, or extremely hypersegregated areas continue to indicate a higher probability of being diagnosed with hypertension, relative to living in a nonsegregated area. Model 4 adds measures relating to metropolitan-level SES to explore how different kinds of socioeconomic status indicators from various different levels affect hypertension. In a zero-order model with only the four metropolitan SES measures, all measures except the proportion of black residents were significant (results not shown). In this model, which controls for individual variables and segregation measures, only the proportion of MSA residents in poverty and the proportion of female-headed households remain significant. In both cases, these variables, net of all controls, are positively related to hypertension diagnosis. Indeed, the more people in poverty and the more female-headed households in an area, the greater the chance of being diagnosed with hypertension. Adding the metropolitan SES measures attenuates the effect that living in an extremely hypersegregated area has on hypertension. The proportion of female-headed households and the proportion in poverty both contribute to explaining away the effect of living in extreme hypersegregation. However, residence in a segregated or hypersegregated area is still significant in predicting hypertension, net of individual-level characteristics including socioeconomic status. Lastly, most of the individual effects of socioeconomic status retain their significance from Models 3 to 4, which indicates that both metropolitan and individual SES impact hypertension. The only category to become statistically insignificant is high school graduate. In the previous models, being a high school graduate is associated with a lower likelihood of being diagnosed with hypertension relative to having less than a high school education. However, in the final model, the likelihood of having hypertension is the same for high school graduates and those with less than a high school education. Supplemental analyses indicate that the proportion of poverty contributes to this change in statistical significance across Models 3 and 4. 4.3. Variance components Because both fixed and random effects are estimated in the models, a discussion of the variance components is warranted. Table 4 displays the random effects estimates for the higher order (i.e., level 2) measures. Across the models, the intercept is significant, indicating that there is some unexplained variance among metropolitan areas when variables are entered into the models. To apply, the intercept for Model 2 (which includes all of the individual, or level 1, variables) suggests that 80% (0.04/0.05) of the variance in hypertension diagnosis that is explainable by 63 Table 4 Variance components of multilevel analyses of hypertension. Source: 2005 BRFSS and 2005 ACS. Model 1 Model 2 Model 3 Model 4 Intercept 0.05n 0.04n Segregation type (nonsegregated) Segregated Hypersegregated Extremely hypersegregated Log population size Percentage of black residents Percentage of residents in poverty Percentage of female-headed households Level 1 error 0.45nnn 0.41nnn 0.02n 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.01 0.07nnn 0.49nnn 0.01 0.41nnn 0.40nnn Notes: Contrast categories are in parentheses. n p o 0.05. nn po 0.01. p o 0.001. nnn individual-level factors is accounted for by including health and demographic information. Likewise, Model 3 (which includes all of the individual variables and segregation type) suggests that 50% (0.02/0.04) of the variance in hypertension diagnosis that is explainable by metropolitan-level factors is accounted for by including information on segregation type. However, most notably, in Model 4, which includes all variables and metropolitanlevel SES controls, the intercept loses significance. This suggests that the addition of the metropolitan-level SES measures explains the variance in hypertension diagnosis that could have been explained by incorporating individual-level measures and segregation type. As such, there is no statistical difference in hypertension diagnosis across MSAs. Statistically, a zero variance is not uncommon for hierarchical linear models and indicates that the MSA variance is not statistically different from zero (Lee and Ferraro, 2007). Two of the MSA-level variables have significant random effects as well. Between MSAs, there is significant variation in the proportion of black residents and the proportion of individuals in poverty that affects being diagnosed with hypertension, which is unaccounted for by including the individual and metropolitan measures in the analyses. The variance components for the segregation types are not statistically different from zero, suggesting that the variation in hypertension across all types of segregation is accounted for by including individual measures. 4.4. Stratified analyses Table 5 is an abridged table that presents the multilevel analyses of hypertension diagnosis for each area type. Each model presented is a full model, with all measures controlled for simultaneously. However, only the coefficients of the individual SES measures appear in the table. Because the segregation variables were significant in the final model of Table 3, one would expect to find area-specific differences in the effects that individual measures have on being diagnosed with hypertension. In unshown analyses, there were significant cross-level interactions between segregation type and individual SES measures, which suggest that stratified analyses are methodologically appropriate to explore the variable effects within each segregation context (DeMaris, 2004; Raudenbush and Byrk, 2002). In nonsegregated metropolitan areas, the individual-level SES effects are similar to those of the combined sample. That is, being highly educated (i.e., a college graduate), having more than $15,000 in annual income, or being employed at least part-time are all negatively related to hypertension diagnosis. Moreover, this 64 A. Jones / Health & Place 22 (2013) 56–67 Table 5 Selected coefficients for multilevel analysis of hypertension stratified by metropolitan segregation. Source: 2005 BRFSS and 2005 ACS. Individual SES Educational level (less than high school) High school graduate Some college Graduated college Household income (less than $15,000) $15,000 to $20,000 $20,000 to $25,000 $25,000 to $35,000 $35,000 to $50,000 $50,000 to $75,000 More than $75,000 Employed Pseudo R2 (Level 1) N Nonsegregated MSA Segregated MSA Hypersegregated MSA Extremely hypersegregated MSA 0.01 −0.03 −0.16nnn 0.05 0.03 −0.08 −0.07 −0.16 −0.27 0.00 −0.12 −0.19 −0.14nnn −0.22nnn −0.25nnn −0.29nnn −0.36nnn −0.40nnn −0.15nnn −0.12nnn −0.27nnn −0.22nnn −0.26nnn −0.37nnn −0.37nnn −0.15nnn 0.01 −0.21 −0.37nn −0.25nnn −0.19nnn −0.31nnn −0.22nnn 0.05 −0.29 −0.51 −0.33 −0.51 −0.43nnn −0.14 0.37 139,901 0.29 44,373 0.31 12,784 0.32 3,044 Notes: Contrast categories are in parentheses. All models control for all measures used in previous analyses. n po 0.05. nn p o0.01. po 0.001. nnn model suggests that education is influential in predicting hypertension diagnosis in nonsegregated areas only if you are a college graduate. Thus, the monotonic nature of education is not present within nonsegregated metropolitan areas. In segregated areas, an interesting pattern emerges—individual SES loses some statistical strength in predicting hypertension. While income and employment maintain their relationships as discussed in the model for nonsegregated areas, education is not significant in predicting hypertension diagnosis. Degree attainment (or levels of education), which previously has been protective against hypertension, appears not to have a significant protective effect in areas that are racially segregated. Also, some of the coefficients are smaller in magnitude, which provides some evidence that the disparities among the income groupings are smaller than in nonsegregated MSAs. However, because these coefficients are not standardized, it is important to interpret them with caution. In hypersegregated areas, the pattern of declining SES significance and magnitude continues. In hypersegregated areas (i.e., areas where minority residents may not encounter members outside of their own race) the effects of individual-level SES become sparse. In this kind of environment, education and lower levels of income (i.e., $15,000 to $25,000) are not significant in predicting hypertension diagnosis. Thus, only through employment and earning higher levels of income are individuals in these areas more likely to have lower odds of being diagnosed with hypertension. Lastly, in extremely hypersegregated areas, neither education nor employment matter in predicting hypertension. In fact, only high levels of income (i.e., more than $75,000) result in lower odds of being diagnosed with hypertension. 5. Discussion This research used nationally representative data to assess how socioeconomic status and metropolitan area residential segregation jointly affect hypertension. It sought out to answer three questions. First, was segregation associated with hypertension? This research suggested that there is an association between segregation and hypertension. Bivariate results indicated that hypertension diagnoses were represented most in nonsegregated metropolitan areas, where nearly 70% of people in the sample reside. Turning to racial differences in hypertension across segregation types, the black-white gap in hypertension is greatest in nonsegregated areas and lowest in hypersegregated areas. This research provided some evidence that, although there is a clear racial component to the distribution of hypertension in metropolitan areas, the gap is narrowest in areas that are characteristically the poorest in infrastructure. Thus, extreme forms of segregation may shape hypertension risk by affecting all people with a given area similarly due to the lack of access to health-promoting structures (such as safe, open spaces, and high-quality hospitals). Multivariate results suggested that living in segregated and hypersegregated metropolitan areas increased hypertension risk compared to living in nonsegregated areas. However, when controlling for metropolitan-level socioeconomic status indicators, there was no statistical difference in hypertension diagnosis for individuals living in nonsegregated and extremely hypersegregated areas. This finding could be due to the low sample size of individuals who live in this extreme type of segregation. In either case, it is important to note that the risk of being diagnosed with hypertension is enhanced with more extreme forms of segregation. Prior work has indicated that segregation is tied to various health outcomes (Gebreab and Diez-Roux, 2012; Kannel, 1996; Mair et al., 2011). This research has added strong support in this finding by suggesting that segregation is also related to hypertension. Second, how did metropolitan SES relate to individual SES in predicting hypertension? The findings indicated that individual SES measures (education, income, and employment) maintained statistical significance even after controlling for metropolitan SES measures. This result indicated that individual SES is very powerful in predicting hypertension. Specifically, college graduates, individuals who are in higher income levels, and those who are employed had low risks of being diagnosed with hypertension. Also, metropolitan SES measures significantly and independently predicted hypertension diagnosis. That is, increasing numbers of blacks and individuals in poverty within a metropolitan area was associated with an increased risk of an individual in a metropolitan area being diagnosed with hypertension. Thus, context matters in predicting hypertension diagnosis. In addition, metropolitan SES uniquely explained variation in hypertension. The variance in hypertension explainable by MSA factors was accounted for by including metropolitan SES measures. So, while A. Jones / Health & Place 22 (2013) 56–67 segregation was important in predicting hypertension, both the material conditions and racial composition inequalities that exist within segregated areas were more powerful in explaining hypertension. Recent work suggested various pathways through which segregation affects health (Acevedo-Garcia et al., 2003; Gebreab and Diez-Roux, 2012). This research has suggested two potential mechanisms that future research can use to investigate a causal link between segregation and hypertension. Third, did the effect of an individual's SES differ across different types of segregation? While the robustness of the data is compromised due to the low representation of people in the sample residing in extremely hypersegregated areas (3,044 out of 200,102 individuals lived in this type of area), the research overwhelmingly supports the notion that individual SES mattered less in more segregated areas. Specifically, in segregated areas, education was not protective against hypertension diagnosis. In hypersegregated areas, education and lower-to-mid levels of income were not protective. In areas where interactions are exclusively within one's minority group (i.e., extremely hypersegregated areas), only at high levels of income (4$75,000) were people protected against hypertension diagnosis. However, these results should be interpreted with caution, since the low number in extremely hypersegregated areas is statistically problematic to generalize to all individuals within this type of segregated context. This finding adds to our understanding of how social and economic inequalities in health are context-dependent. In extremely hypersegregated areas, where there are poor living conditions, unreliable or poor quality medical care, and a high concentration of poverty, only those individuals who are affluent have the means to seek proper medical/health care and engage in diversions from the stressors of living in these areas, which would relatively improve their health. Indeed, healthy lifestyles are most effective for people living in positive social circumstances and least effective in the converse (Blaxter, 1990). This research provided some support for that argument. It is important to recognize that individual and metropolitan SES are interrelated but affect health in different ways. Individual SES corresponds to a social raking based on the indictors used in this project, while metropolitan SES relates to the objective social organization of individuals within a space with various different physical structures and living conditions (Diez-Roux et al., 1997). In prior work, neighborhood disadvantages negatively affect the health of its residents above the effects of personal disadvantages (Ross and Mirowsky, 2001). Some research have suggested that because of this interplay, more research needs to be done in order to capture the intersectionality of individual and spatial SES (Acevedo-Garcia et al., 2003), which was the main goal of this research. Prior research on segregation and hypertension has used only one measure of segregation such as an index of isolation (Kershaw et al., 2011; Thorpe Jr. et al., 2006) or composite proxies such as economic deprivation (White et al., 2011), and these studies have focused on racial differences in hypertension. Based on these measures, they have suggested that whites and blacks have similar cardiovascular outcomes when they live in areas with similar levels of segregation (Thorpe Jr. et al., 2006), but higher rates of hypertension for both whites and blacks are found in higher levels of segregation (Kershaw et al., 2011). This research has used multiple indicators of segregation to create typologies of segregated areas, and this study has included numerous economic indicators at the individual and metropolitan level. This research has some similarities and divergences from prior research. Consistent with past research, this research found that in all segregation types, blacks comprise the largest percentage of hypertensive cases than whites, but the gap narrows in more extreme forms of segregation. However, unlike prior research that suggests that race 65 differences are statistically nonsignificant in segregated areas, this research found that race and SES were dually and differentially important in predicting hypertension diagnosis depending on the type of segregated area in which one resides. Thus, this research contributes to the larger body of work on segregation and hypertension by empirically demonstrating the independent and joint roles that both race and socioeconomic standing have on hypertension in various different kinds of segregated environments. An interesting finding with direct policy implications deals with insurance. Intuitively, insurance was associated with a lower risk of being diagnosed with hypertension. Moreover, when insurance status is controlled for in the model, individuals who made less than $15,000 had the same risk of high blood pressure as those who could be considered working poor (i.e., income between $15,000 and $20,000), who had a higher risk of being diagnosed with hypertension in a previous model. Thus, having insurance seems to be a normalizing force for individuals who do not get insurance from their work, do not qualify for public insurance (e.g., Medicaid) because they work, and who cannot afford private insurance. Expanding health insurance coverage through universal health care could combat health disadvantage by reducing hypertension risk for individuals who are unable to afford insurance. In addition, as many researchers suggest (Rosenbaum, 1995; Stuart and Sorenson, 2003), the negative effects of segregation could be combated if US policy is shifted from focusing on desegregation to a pointed effort to provide equal opportunity to reside in neighborhoods and communities. Racial integration is not an appropriate goal for curbing health disparities. Rather, giving people chances to live in different areas would correct SESrelated health disparities. This study is not without its limitations. First, the sample size for the BRFSS was large, indicating that perhaps the estimates were capturing small differences in a large sample as being significant. The reader is warned that the same results may not be guaranteed if using a smaller sample size. In addition, the sample was collected via telephone interviews, which indicates some class bias on who can be selected to participate in the study. While no sample selection effects were found with regards to income, it is important to note that by virtue of the sampling method, severely impoverished persons may not be adequately represented in this sample. However, coupled with a class bias, there is also a smaller representation of blacks in this large sample. With less than 9% in the analytic sample being African American, the sample may be skewed to more affluent blacks, a demographic that is rather small in representation in the US. Future studies may want to improve on ways to estimate those who are least represented in the BRFSS. Second, this research could not analyze racial groups other than whites and blacks due to sample size limitations. Having the ability to look at racial differences and effects of covariates would have strengthened this research because it could have pointed to mechanisms that affect cardiovascular disease across racial groups. Future research should use this work to situate why racial and ethnic differences in hypertension continue to persist. Recent research by Do et al. (2012) used propensity score matching and found that the racial differences in self-rated health could be attenuated by the inclusion of family poverty levels (FPL) and household wealth. They argued that prior researchers have not completely measured or accounted for SES in their models. Unfortunately, the 2005 BRFSS does not include wealth or FPL data, so this project is unable to replicate Do's research. Third, hypertension is self-reported and not based on medical examinations or records. The reporting of health depends on whether patients choose to consult their general practitioner and is based on their own decisions. The self-reporting of health on surveys is conceptually problematic for two reasons. Bias could be 66 A. Jones / Health & Place 22 (2013) 56–67 embedded in the study by the use of self-reported data. If this research relied on the assumption that people accurately report their morbidities based on a doctor's evaluation, then the assumption fails to acknowledge those people who are unable to seek medical advice. Also, self-reported health status could be self-diagnosed, which could lead to error in misdiagnosis. Misdiagnosis through self-reports could result in significant variability in self-reported health, which problematizes the validity of the measurement. Fourth, with cross-sectional data, it is nearly impossible to disentangle causal ordering of events because all information is gathered at one specific point in time. This research presented associations between certain processes (e.g., SES) and health. However, it is unclear, for instance, if SES directly impacts health, if SES is impacted by health, or if SES and health are both impacted by each other. Thus, a longitudinal study on hypertension and SES would illuminate how direct associations (or causal ordering of events) impact health. Finally, structural confounding is a concern in multilevel analysis of segregation and health. Specifically, people who reside in vastly different areas are not exchangeable because various factors have allowed these individuals to select differentially into their specific area (Maldonado and Greenland, 2002). Even if these individuals are matched on numerous factors that are present in the data, there are still unobserved characteristics (such as individual motivations for moving into a particular area or experiences of housing discrimination) that would make these individuals incomparable and therefore not exchangeable. This concern is relevant for this study, since structural conditions have dually shaped minorities' access to certain neighborhoods and access to opportunities tied to individual socioeconomic status, such as entrance into institutions of higher education or employment (Wilson, 1987). As such, the people who live in extremely hypersegregated areas may not be exchangeable with those living in less segregated areas because of unobserved factors that influence and constrain their residential decisions. Methodologically, this kind of selection would mean that “certain covariate strata will contain only subjects who could never be exposed, a violation of the positivity or experimental treatment effect assumption” (Messer et al., 2010: 664). Outside of a few studies that use propensity score matching to address this problem (e.g., Leal et al., 2011; Messer et al., 2010), structural confounding has largely been ignored in the literature. Future research should be aware of this concern when trying to make more causal inferences from observational data (Messer, 2007). Despite the aforementioned limitations, this research is not without its strengths. With recent data, this study has expanded our understanding of segregation and hypertension in two key ways. This research conceptualized segregation as a multidimensional characteristic in metropolitan areas that relies on five unique indicators to assess it. As such, this research was able to focus on different types of segregation, namely hypersegregation. As a conceptually different phenomenon from segregation, hypersegregation was found to be a unique predictor of hypertension diagnosis. Lastly, this research also conceptualized socioeconomic status as having different effects at different levels of analysis. That is, an individual's SES and the corresponding socioeconomic context in which that individual lives are both unique predictors of hypertension. However, these two levels of SES matter less in segregated areas. As such, the linkages between race, socioeconomic context, segregation and hypertension could prompt for a new line of research that directly examines the specific ties between the individual, community, and society on health. References Acevedo-Garcia, D., Lochner, K.A., Osypuk, T.L., Subramanian, S.V., 2003. Future directions in residential segregation and health research: a multilevel approach. American Journal of Public Health 93, 215–221. Adler, N.E., Boyce, T., Chesney, M.A., Cohen, S., Folkman, S., Kahn, R.L., Syme, S.L., 1994. Socioeconomic status and health. American Psychologist 49, 15–24. Agresti, A., 2002. Categorical Data Analysis, second ed. John Wiley & Sons, Hoboken, NJ. Allison, P.D., 1995. Survival Analysis Using the SAS System: A Practical Guide. SAS Institute, Cary, NC. Blaxter, M., 1990. Health and Lifestyles. Routledge, New York, NY. Centers for Disease Control and Prevention, 2006. Behavioral Risk Factor Surveillance System Survey Data. U.S. Department of Health and Human Services, Atlanta, Georgia. de Souza Briggs, X., 1997. Moving up versus moving out: neighborhood effects in housing mobility programs. Housing Policy Debate 8, 195–234. DeMaris, A., 2004. Regression with Social Data: Modeling Continuous and Limited Response Variables. Wiley, Hoboken, NJ. Diez-Roux, A.V., Nieto, F.J., Muntaner, C., Tyroler, H.A., Comstock, G.W., Shahar, E., Cooper, L.S., Watson, R.L., Szklo, M., 1997. Neighborhood environments and coronary heart disease: a multilevel analysis. American Journal of Epidemiology 146, 48–63. Do, D.P., Frank, R., Finch, B.K., 2012. Does SES explain more of the black/white health gap that we thought? Revisiting our approach toward understanding racial disparities in health. Social Science & Medicine 74, 1385–1393. Dressler, W., 1991. Social class, skin color and arterial blood pressure in two societies. Ethnicity and Disease 1, 60–77. Dressler, W., Oths, K.S., Gravlee, C.C., 2003. Race and ethnicity in health research: models to explain disparities. Annual Review of Anthropology 34, 231–252. Eitle, D., 2009. Dimensions of racial segregation, hypersegregation, and black homicide rates. Journal of Criminal Justice 37, 28–36. Ellen, I.G., Mijanovich, T., Dillman, K.N., 2001. Neighborhood effects on health: exploring the links and assessing the evidence. Journal of Urban Affairs 23, 391–408. Fang, J., Madhavan, S., Bosworth, W., Alderman, M.H., 1998. Residential segregation and mortality in New York City. Social Science & Medicine 47, 469–476. Freeman, L., 2005. Displacement or succession?: Residential mobility in gentrifying neighborhoods. Urban Affairs Review 40, 463–491. Gebreab, S.Y., Diez-Roux, A.V., 2012. Exploring racial disparities in CHD mortality between blacks and whites across the United States: a geographically weighted regression approach. Health & Place 18, 1006–1014. Harburg, E., Erfurt, J.C., Hauenstein, L.S., Chape, C., Schull, W.J., Schork, M.A., 1973. Socio-ecological stress, suppressed hostility, skin color, and African Americanwhite male blood pressure: Detroit. Psychosomatic Medicine 35, 276–296. Hearst, M.O., Oakes, J.M., Johnson, P.J., 2008. The effect of racial residential segregation on black infant mortality. American Journal of Epidemiology 168, 1247–1254. James, S.A., Hartnett, S.A., Kalsbeek, W.D., 1984. John Henryism and blood pressure differences among African American Men II: The role of occupational stressors. Journal of Behavioral Medicine 7, 259–275. Kaczynski, A.T., Johnson, A.J., Saelens, B.E., 2010. Neighborhood land use diversity and physical activity in adjacent parks. Health & Place 16, 413–415. Kannel, W.B., 1996. Blood pressure as a cardiovascular risk factor. Journal of the American Medical Association 275, 1571–1576. Kershaw, K.N., Roux, A.V.D., Burgard, S.A., Lisabeth, L.D., Mujahid, M.S., Schulz, A.J., 2011. Metropolitan-level racial residential segregation and black–white disparities in hypertension. American Journal of Epidemiology 174, 537–545. Kramer, M.R., Hogue, C.R., 2009. Is segregation bad for your health? Epidemiologic reviews 31, 178–194. Leal, C., Bean, K., Thomas, F., Chaix, B., 2011. Are associations between neighborhood socioeconomic characteristics and body mass index or waist circumference based on model extrapolations? Epidemiology 22, 694–703. LeClere, F., Rogers, B., Peters, K.D., R.G., 1997. Ethnicity and mortality in the United States: individual and community correlates. Social Forces 76, 169–198. Lee, M.-A., Ferraro, K.F., 2007. Neighborhood residential segregation and physical health among Hispanic Americans: good, bad, or benign? Journal of Health and Social Behavior 48, 131–148. Mair, C.A., Cutchin, M.P., Peek, M.K., 2011. Allostatic load in an environmental riskscape: the role of stressors and gender. Health & Place 17, 978–987. Maldonado, G., Greenland, S., 2002. Estimating causal effects. International Journal of Epidemiology 31, 422–429. Massey, D.S., 2004. Segregation and stratification: a biosocial perspective. Du Bois Review 1, 7–25. Massey, D.S., Denton, N.A., 1989. Hypersegregation in US metropolitan areas: African American and Hispanic segregation along five dimensions. Demography 26, 373–391. Massey, D.S., Denton, N.A., 1993. American Apartheid: Segregation and the Making of the Underclass. Harvard University Press, Cambridge, MA. Messer, L.C., 2007. Invited commentary: beyond the metrics for measuring neighborhood effects. American Journal of Epidemiology 165, 868–871. Messer, L.C., Oakes, J.M., Mason, S., 2010. Effects of socioeconomic and racial residential segregation on preterm birth: a cautionary tale of structural confounding. American Journal of Epidemiology 171, 664–673. Miranda, A., Rabe-Hesketh, S., 2006. Maximum likelihood estimation of endogenous switching and sample selection models for binary, ordinal and count variables. The Stata Journal 6, 285–308. Mokdad, A.H., Stroup, D.F., Giles, W.H., 2003. Public health surveillance for behavioral risk factors in a changing environment: recommendations from the behavioral risk factor surveillance team. Morbidity and Mortality Weekly Report 52, 1–12. Morland, K., Wing, S., Diez-Roux, A., Poole, C., 2002. Neighborhood characteristics associated with the location of food stores and food service places. American Journal of Preventive Medicine 22, 23–29. A. Jones / Health & Place 22 (2013) 56–67 Oliver, M.L., Shapiro, T.M., 2006. Black Wealth/White Wealth: A New Perspective on Racial Inequality. Routledge, New York, NY. Osypuk, T.L., Acevedo-Garcia, D., 2008. Are racial disparities in preterm birth larger in hypersegregated areas? American Journal of Epidemiology 167, 1295–1304. Pollack, C.E., Chideya, S., Cubbin, C., Williams, B., Dekker, M., Braveman, P., 2007. Should health studies measure wealth?: a systematic review. American Journal of Preventive Medicine 33, 250–264. Rabe-Hesketh, S., Skrondal, A., 2012. Multilevel and Longitudinal Modeling Using Stata, third ed. Stata Press, College Station, TX. Raudenbush, S.W., Byrk, A.S., 2002. Hierarchical Linear Models: Applications and Data Analysis Methods, second ed. Sage Publications, Thousand Oaks, CA. Rencher, A.C., 2002. Methods of Multivariate Analysis, second ed. WileyInterscience, New York, NY. Rosenbaum, J.E., 1995. Changing the geography of opportunity by expanding residential choice: lessons from the Gautreaux program. Housing Policy Debate 6, 231–269. Ross, C.E., Mirowsky, J., 2001. Neighborhood disadvantage, disorder, and health. Journal of Health and Social Behavior 42, 258–276. Schum, J.L., Jorgensen, R.S., Verhaeghen, P., Sauro, M., Thibodeau, R., 2003. Trait anger, anger expression and ambulatory blood pressure: a meta-analytic review. Journal of Behavioral Medicine 26, 395–415. Smaje, C., 1995. Ethnic residential concentration and health: evidence for a positive effect. Policy and Politics 23, 251–269. StataCorp, 2011. Stata Statistical Software: Release 12. StataCorp LP, College Station, TX. Stuart, T., Sorenson, O., 2003. The geography of opportunity: spatial heterogeneity in founding rates and the performance of biotechnology firms. Research Policy 32, 229–253. 67 Subramanian, S.V., Acevedo-Garcia, D., Osypuk, T.L., 2005. Racial residential segregation and geographic heterogeneity in black/white disparity in poor self-rated health in the US: a multilevel statistical analysis. Social Science & Medicine 60, 1667–1679. Taylor, P., Kochhar, R., Fry, R., Velasco, G., Motel, S., 2011. Twenty-to-One: Wealth Gaps Rise to Record Highs between Whites, Blacks and Hispanics. Pew Research Center Social & Demographic Trends, Washington, DC. Thorpe Jr., R.J., Brandon, D.T., LaVeist, T.A., 2006. Social context as an explanation for race disparities in hypertension: findings from the exploring health disparities in integrated communities (EHDIC) study. Social Science & Medicine 67, 1604–1611. White, K., Borrell, L.N., Wong, D.W., Galea, S., Ogedegbe, G., Glymour, M.M., 2011. Racial/ethnic residential segregation and self-reported hypertension among US- and foreign-born blacks in New York City. American Journal of Hypertension 24, 904–910. Wilkes, R., Iceland, J., 2004. Hypersegregation in the twenty-first century. Demography 41, 23–36. Williams, D., Collins, C., 2001. Racial residential segregation: a fundamental cause of racial disparities in health. Public Health Reports 116, 404–416. Williams, D.R., Collins, C., 1995. US socioeconomic and racial differences in health: patterns and explanations. Annual Review of Sociology 21, 349–386. Williams, D.R., Jackson, P.B., 2005. Social sources of racial disparities in health. Health Affairs 24, 325–334. Wilson, W.J., 1987. The Truly Disadvantaged: The Inner City, the Underclass and Public Policy. University of Chicago Press, Chicago, Illinois. Zenk, S.N., Schulz, A.J., Israel, B.A., James, S.A., Bao, S., Wilson, M., 2005. Neighborhood racial composition, neighborhood poverty, and the spatial accessibility of supermarkets in metropolitan Detroit. American Journal of Public Health 95, 660–668.