Residential Mobility Across Local Areas in the United States and the Geographic Distribution of the Healthy Population

Abstract

Determining whether population dynamics provide competing explanations to place effects for observed geographic patterns of population health is critical for understanding health inequality. We focus on the working-age population—the period of adulthood when health disparities are greatest—and analyze detailed data on residential mobility collected for the first time in the 2000 U.S. census. Residential mobility over a five-year period is frequent and selective, with some variation by race and gender. Even so, we found little evidence that mobility biases cross-sectional snapshots of local population health. Areas undergoing large or rapid population growth or decline may be exceptions. Overall, place of residence is an important health indicator; yet, the frequency of residential mobility raises questions of interpretation from etiological or policy perspectives, complicating simple understandings that residential exposures alone explain the association between place and health. Psychosocial stressors related to contingencies of social identity associated with being black, urban, or poor in the United States may also have adverse health impacts that track with structural location even with movement across residential areas.

Introduction

The possibility that the relationships between socioeconomic position or race/ethnicity and health are influenced by geographic and community conditions that structure and contextualize them has garnered great interest over the past decade or longer (Cozier et al. 2007; Davey Smith et al. 1998a; Diez Roux 2001; Freedman et al. 2011; Geronimus 2000; Hayward et al. 1997; Nordstrom et al. 2004). A resurgence of interest in the health impact of residence in high-poverty urban areas, in particular, has been notable among public health researchers (Galea and Vlahov 2005; Fitzpatrick and LaGory 2011), and also among sociologists who study residential areas characterized by urban decay, concentrated poverty, and racial segregation (Massey and Denton 1993; Wilson 1996). After reviewing the growing evidence on poverty, race, place, and health, Fitzpatrick and LaGory (2011:x) concluded, “until the public’s view of health encompasses a basic understanding that where we live affects how we live and die, the health status of certain segments of the population cannot improve.”

Geronimus et al. (1996, 1999, 2001, 2011) analyzed differences in adult disability and mortality rates across a regionally diverse set of black and white, urban and rural, high-poverty and more-affluent local populations, finding great variation in these rates across population types. Cause-of-death analyses from 1980 through 2000 documented a substantial and entrenched burden of chronic disease among working-age black men and women living in high-poverty areas (Geronimus et al. 2011). Black adults in high-poverty urban areas faced extremely disadvantageous mortality schedules through middle and old age, not only in comparison with white or black national averages or with black residents of more-affluent communities but also compared with black residents of high-poverty rural communities. For example, age-standardized mortality ratios in 2000 for 18- to 64-year-old black men or women in Southside Chicago and Delta Louisiana—two study populations in which 46 % of families lived below the poverty line—were greater than three times the national average in Chicago yet only about two times the national average in Louisiana. This was the case even though high school and college graduation rates were substantially lower in the population of Louisiana than in that of Chicago.

Standardized mortality rate differences between high-poverty urban and high-poverty rural areas suggest that place characteristics may have impacts on health above and beyond the effects of the sociodemographic characteristics of local populations. It may be reasonable to hypothesize that characteristics of urban areas undermine longevity or that aspects of rural areas confer protection. However, because most studies of place effects on health use cross-sectional data (Beard et al. 2009; Diez-Roux 2007; Osypuk and Acevedo-Garcia 2008; Shih et al. 2011),¹ this interpretation implicitly assumes static local populations. Yet, local populations are in fact dynamic, and health may be associated with population shifts over time. Thus, observed geographic patterns of health could also reflect the influence of health-related migration on the geographic distribution of the healthy population. These patterns underlie the health-selective migration hypothesis: namely, that differences in health profiles across distinct areas may arise, in part, because the residentially mobile may be more healthy or less healthy than those who do not move. For example, the 50 % higher excess mortality in the poor black South Chicago population compared with the poor black Louisiana Delta population could be explained by differences in place characteristics between the areas (including environmental exposures), by stable differences in populations between the two areas, or by a net out-migration of the healthy from Chicago or a net in-migration of the healthy into the Delta population (selective migration).

Gaining purchase on whether selective migration is an important competing explanation to place effects or to static compositional effects for health differences observed across local areas is critical for identifying the most promising future directions for research on racial/ethnic and socioeconomic inequalities in health and for policy to eliminate them. Considerable work has explored the impact of selective migration on health among immigrants to and from the United States, particularly in the context of the Latino health paradox² (Palloni and Arias 2004). Some UK studies have documented the important impact of selective migration on the health profiles of more- and less-distressed areas in that country (Brimblecombe et al. 1999, 2000; Davey Smith et al. 1998b; Norman et al. 2005). To our knowledge, we are the first researchers to address the health-selective migration hypothesis in the context of residential mobility within the United States.

The few investigators who have considered the relationship between local population health and residential mobility patterns within the United States have asked a different question: whether health influences the mobility decisions of those at or near retirement (Findley 1988; Halliday and Kimitt 2008; Longino et al. 1991, 2008; Speare et al. 1991). They have not focused on health disparities or the younger working-age population, for whom health disparities are greater (Elo and Preston 1996; Geronimus 1992; Geronimus et al. 2001, 2006, 2007), and they have not considered the impacts of selective migration for the association between place and population health.

Health and Migration in the United States

There is reason to postulate that health and residential mobility may be associated. Several theoretical perspectives on residential mobility across neighborhoods in the United States highlight the importance to mobility decisions of the demographic and life-cycle characteristics of nuclear household heads, including age, sex, and marital status (South et al. 2005; Tolnay 2003). To the extent that the demographic and life-cycle characteristics influencing residential moves are associated with current or future health status, it is reasonable to hypothesize that residential moves are systematically related to health.

Few data sets link residential population health and migration. Investigators of intraurban mobility per se (i.e., without a health and place component) have often analyzed data from the Panel Study of Income Dynamics (PSID). The strength of the PSID is its longitudinal dimension. However, because of the nature and size of its sample, researchers need to aggregate respondents residing in high-poverty areas across the nation and, thus, cannot focus on geographically defined local environmental exposures. Recent investigations on the longitudinal effects of neighborhood stratification on educational or developmental outcomes in children have used this aggregated approach (Jackson and Mare 2007; Wodke et al. 2011).

Although its time horizons are proscribed—and, like the PSID, its health data are limited—the U.S. census yields large sample sizes, making estimates highly reliable. The census also provides residential population data, uniquely allowing local geographic places to be delineated in order to observe the nature and extent of moves within, into, and out of specific local areas and their net effect on areal population health.

Research Questions

Considering blacks and whites separately and (where population size allows) men and women separately within each race, we explore whether observed geographic patterns of health reflect the influence of health-related migration through these research questions:

1. What is the extent of residential mobility flows into or out of a diverse set of local areas? If we find little evidence of residential mobility, or in mobility between types of areas with very different health profiles, these findings alone would cast doubt on the health-selective migration hypothesis.

2. To what extent is residential mobility between local areas associated with either health-induced functional limitations or with sociodemographic characteristics associated with health more broadly, such as age and education? If the associations are insignificant, this would also weaken the plausibility of the health-selective migration hypothesis.

3. If residential mobility and health-induced functional limitation are associated, does such selection affect cross-sectional estimates of local area prevalence of health-induced functional limitation? By comparing local area health profiles in 2000 with what they would have been had no migration occurred, we will garner evidence on the quantitative impact of health-selective migration on cross-sectional snapshots of the health of local populations.

Methods

The 2000 census was the first to ask respondents their previous residence down to the ZIP (postal) code level, potentially allowing exploration of short-distance moves or moves between locally defined geographic areas.³ Although this ZIP code information is not included in census data that are usually available for research, we obtained authorization to analyze it at the Michigan Census Research Data Center (MCRDC). Reporting of our findings is limited to those the census permits after disclosure review.

Overview and Characteristics of the Study Populations

Our analytic sample comprises the entire non-Hispanic black or non-Hispanic white populations aged 16 to 64 in 20 geographic aggregates of economically similar ZIP codes (urban areas) or counties (rural areas). Although we focus on these focal areas as points of migration departure or arrival, our analysis includes all moves for these areas from or to anywhere in the United States.

Our focal areas are those selected by Geronimus et al. (1996, 1999, 2011) to explore diversity in the consequences of areal poverty for urban and rural black and white mortality.⁴ For this earlier work, we originally selected areas of concentrated poverty of sufficient size for the reliable calculation of mortality schedules, then supplementing them with geographically proximate nonpoor areas. Using census data, we could replicate these areas exactly across decades. By studying focal areas whose mortality profiles have been previously described, our findings can inform understanding of mortality differences as well as in census health measures.

These areas represent regional and racial diversity throughout the United States and provide useful comparisons across race, geographic location (i.e., northern vs. southern) and urbanicity. Table 1 lists acronyms, economic characteristics, and standardized mortality ratios (SMRs)—directly standardized with reference to the age distribution of the white population nationwide in 2000 according to gender—for the focal populations in the year 2000 (Geronimus et al. 2011). The focal black populations generally are less economically advantaged and have higher SMRs than the focal white populations. However, among either black or white populations, residents of nonpoor areas generally have lower SMRs than residents of high-poverty areas and those in high-poverty urban areas generally have the highest SMRs, including in comparison with same-race populations from rural areas with similar or higher poverty levels and lower educational levels.

Table 1.

Economic characteristics and age-standardized death rates,^a local black or white populations, aged 16–64 years, 2000^b

				Standardized Mortality Ratios
Area Type/Area	Focal Area Acronym	Median Income	% Families < Poverty	Men	Women
Blacks		44,173	25	1.84	1.90
High-poverty urban
Harlem	HAR	35,896	35	2.25	2.36
East Detroit	EDT	36,811	35	2.80	2.80
Southside Chicago	SSC	33,961	46	3.32	3.15
Watts	WTS	33,993	38	2.89	2.52
High-poverty rural
East North Carolina	ENC	34,054	35	1.96	1.87
Black Belt Alabama	BBA	27,642	42	2.21	2.16
Delta Louisiana	DLA	24,987	46	2.00	2.10
Nonpoor urban
Bronx	BRX	55,205	15	0.96	1.07
Queens	QNY	72,181	7	0.68	0.69
Northwest Detroit	NWD	55,206	16	1.90	1.77
Southwest Chicago	SWC	50,268	20	2.13	2.04
Crenshaw/Baldwin Hills	CBH	55,759	20	1.79	1.65
Whites		70,526	8	1.00	1.00
High-poverty urban
Cleveland	CLD	43,017	18	1.75	1.43
Detroit	DET	43,952	23	2.72	2.36
High-poverty rural
Appalachian Kentucky	AKY	31,035	34	1.59	1.56
West North Carolina	WNC	48,137	14	1.05	1.02
Northeast Alabama	NAL	46,861	14	1.43	1.24
South Central Louisiana	CLA	45,238	18	1.32	1.34
Nonpoor urban
Western Cleveland	WCL	79,025	5	0.75	0.82
Sterling Heights, MI	SHD	101,860	5	0.70	0.77

^aRates are directly standardized with reference to the age distribution of the white population nationwide in 2000 according to gender.

^bThe 20 local population groups included black or white residents of the following neighborhoods, counties, or health center districts. Blacks, Poor Urban Areas: Southside Chicago—Near Southside, Douglas, Oakland, Fuller Park, Grand Boulevard, and Washington Park community areas located south of the Chicago Loop. Eastside Detroit—Central, University, Central Business District, Foch, Jefferson Mack, Airport, St. Jean, Chene, and Jeffries subcommunities of Detroit. Harlem—Central Harlem Health Center District. Watts—Watts and adjacent areas to the south and west. Poor Rural Areas: East North Carolina—Pitt, Northampton, Halifax, and Edgecomb counties in the Coast Plains region of North Carolina. Black Belt Alabama—rural counties in and around the Black Belt region including Dallas, Fayette, Greene, Bibb, Sumter, Hale, Lamar, Marengo, Marion, Perry, and Pickens counties. Delta Louisiana—Caldwell, East Carroll, Franklin, Jackson, Madison, Morehouse, Richland, Tensas, Union, West Carroll, Avoyelles, Catahoula, Concordia, Grant, La Salle, Vernon, and Winn parishes in the Mississippi River Delta region. Nonpoor Urban Areas: Queens—African Americans living in eastern Queens in sections of the Jamaica East and Flushing Health Center Districts. Bronx—African Americans living in the North Bronx sections of the Pelham Bay Health Center District. Southwest Chicago—African Americans living in the Roseland, Pullman, and West Pullman community areas. Northwest Detroit—African Americans living in the Palmer Park, Pembroke, Bagley, Redford, Rosedale Park, Evergreen, and Greenfield neighborhoods. Crenshaw/Baldwin Hills—African Americans living in the Crenshaw, Ladera Heights, Leimert Park, Baldwin Hills, and Windsor Hills communities. Whites, Poor Urban Areas: Central Cleveland—Brooklyn-Centre, Clark-Fulton, Cudell, Detroit-Shoreway, Edgewater, Ohio City-Near West Side, Old Brooklyn, Stockyards, and Tremont statistical planning areas on the near west side of Cleveland. Detroit—whites living in Delray, Clark Park, Chadsey, Condon, Springwells, Jeffries, State Fair, Burbank, Denby, Finney, Mt. Olivet, Grant, Davison, Pershing, and Nolan subcommunities on the northeastern and southern periphery of Detroit. Poor Rural Areas: Appalachian Kentucky—Lee, Leslie, Owsley, Wolfe, Breathitt, Knott, Letcher, and Perry counties. West North Carolina—Alleghany, Ashe, Avery, Mitchell, Watauga, Wilkes, and Yancey counties. Northeast Alabama—whites living in DeKalb, Jackson, and Marshall counties. South Central Louisiana—whites living in Acadia and Vermillion parishes. Nonpoor Urban Areas: Cleveland—Bay Village, Fairview Park, Lakewood, North Olmsted, Rocky River, and Westlake communities. Sterling Heights—Sterling Heights, a northern suburb of Detroit. See Geronimus et al. (2011).

In our earlier work (Geronimus et al. 1996, 1999, 2011) we studied only these areas and, for ease of classification, dichotomized them as “poor” or “nonpoor.” We retained this classification when studying the focal areas alone. However, because our current focus is on moves into or out of these focal areas to or from anywhere in the United States, we were able to classify places across the United States as low poverty (< 10 % of adults in poverty), moderate poverty (10 % to 20 %), high poverty (20 % to 30 %), and very high poverty (more than 30 %), in addition to disaggregating them in terms of urbanicity and region.

Study Data

Study data were drawn from the Sample Edited Detail File (SEDF) of the 2000 decennial census. The SEDF represents roughly 16 % of the population and provides average sample sizes for the monoracial⁵ black or white working-age men and women in our populations of just under 5,500 men and just over 6,100 women. The SEDF includes all the long-form records and in greater detail than the public-use files. For example, the public-use files have confidentiality edits, such as top-coding on income or collapsing of detailed categories into broader categories.

In Item 15, the 2000 census long form asked detailed questions about residential mobility within the United States:

• (15a).
  Did this person live in this house or apartment 5 years ago (on April 1, 1995)?

  1. Yes, this house. Skip to 16.
  2. No, outside the United States. Print name of foreign country, below; then skip to 16.
  3. No, different house in the United States.
• (15b).
  Where did this person live 5 years ago?

  • Name of City, Town or Post Office

    • Did this person live inside the limits of the city or town?
  • Name of county; Name of state; ZIP Code

Response rates in focal areas were high (92 % to 97 %). A trivial fraction of movers into our study areas migrated from another country (ranging from 0.3 % to 0.9 %).

We limited our analysis to those who were identified as living in the same dwelling unit five years previous—or, if not, who reported a valid 1995 ZIP code. We divided our sample into three mobility categories: (1) stayers, (2) movers into our local areas between 1995 and 2000, and (3) movers out of our local areas between 1995 and 2000. Stayers are residents who did not report a move between 1995 and 2000 or those residents who reported a move but whose reported 1995 ZIP code or county fell within the 2000 focal area. Movers in are those whose 1995 ZIP code/county fell outside the 2000 focal area. Movers out are those whose 1995 ZIP code/county fell within the focal area but who are no longer living in the area in 2000.

The 2000 census long form asked respondents two sets of health-related questions probing functional limitations induced by chronic health conditions.⁶ We counted a respondent as limited if they answered yes to any of these questions⁷:

• (16)
  Because of a physical, mental, or emotional condition lasting 6 months or more, does this person have any difficulty in doing any of the following activities:

  1. Learning, remembering or concentrating?
  2. Dressing, bathing or getting around inside the home?
  3. Going outside the home alone to shop or visit a doctor’s office?
  4. Working at a job or business?
• (17)
  Does this person have any of the following long-lasting conditions:

  1. Blindness, deafness or severe vision or hearing impairment?
  2. A condition that substantially limits one or more basic physical activities such as walking, climbing stairs, reaching, or lifting?

Measurement Issues

ZIP Code Data

As the first researchers to use ZIP code data to identify residential mobility patterns, we tried to assess their validity (Geronimus et al. 2008). We found that when respondents reported previous ZIP codes of residence, they generally reported these items accurately: only 2 % of reported ZIP codes were invalid, and an additional 2 % were inconsistent with other data in the file (e.g., county of residence). Although the census imputed values when 1995 county or state data were missing, no imputations were made for ZIP codes.

Table 2 shows response and imputation or missing rates for each of the migration questions by area type. The vast majority of those who moved dwelling units reported both their previous state and county of residence. However, 20 % to 30 % of movers did not report ZIP code data for previous residence. Moreover, when we estimated logit regressions on the sample of individuals who would have been asked to report their ZIP codes of previous residence based on their answers to earlier questions, we found that men, the unemployed, and the disabled were each about 20 % less likely to report their previous ZIP code than were their counterparts. We also found that those who made moderate- or long-distance moves were about 60 % less likely to report previous ZIP codes than were short-distance movers. However, long-distance moves were substantially less common than short-distance moves; and when they did occur, we could identify them using state or county of previous residence when ZIP code was missing.⁸ Despite nonrandomness, sensitivity analyses described in the results suggest that our conclusions are robust to excluding those with missing ZIP code information.

Table 2.

Fraction imputed or missing, by total sample of valid movers, by local areas

		Imputed Previous County		Imputed Previous State		Missing ZIP Codes
Types	Imputed 15aa	Fraction of Totalb	Fraction of Moversc	Fraction of Totald	Fraction of Moverse	Fraction of Totalf	Fraction of Moversg
Urban Poor Blacks	0.08	0.02	0.08	0.02	0.06	0.10	0.32
Rural Poor Blacks	0.07	0.01	0.04	0.01	0.04	0.06	0.20
Urban Nonpoor Blacks	0.09	0.02	0.06	0.01	0.05	0.08	0.31
Urban Poor Whites	0.05	0.01	0.03	0.01	0.03	0.08	0.20
Rural Poor Whites	0.04	0.01	0.03	0.01	0.03	0.06	0.17
Urban Nonpoor Whites	0.03	0.01	0.03	0.01	0.03	0.06	0.17

^aRespondents were asked whether they had moved households within the past five years. Numerator is people who did not respond to 15a. Denominator is total population.

^bNumerator is respondents answering that they moved dwelling units in 15a who did not respond to the question about county in 15b. Denominator is the total enumerated population.

^cNumerator is respondents who answered that they moved dwelling units in 15a who did not respond to the question about county in 15b. Denominator is people who answered that they moved dwelling units in 15a.

^dNumerator is respondents answering that they moved dwelling units in 15a who did not respond to the question about state in 15b. Denominator is the total enumerated population.

^eNumerator is respondents who answered that they moved dwelling units in 15a who did not respond to the question about state in 15b. Denominator is people who answered that they moved dwelling units in 15a.

^fNumerator is respondents answering that they moved dwelling units in 15a who did not respond to the question about ZIP code in 15b. Denominator is the total enumerated population.

^gNumerator is respondents who answered that they moved dwelling units in 15a and who did not respond to the question about state in 15b. Denominator is people who answered that they moved dwelling units in 15a.

Health Indicator

The health-induced functional limitation data are the only health indicators available in the census. Although they are neither sufficient for medical diagnosis of specific conditions ⁹ nor fully representative of all aspects of health that might be associated with either place or residential migration, there is reason to believe that the census measures are sufficient for the analytic objectives of this study (Brimblecombe et al. 1999, 2000; Hayward and Heron 1999; Geronimus et al. 2001; Norman et al. 2005).¹⁰ Self-reported measures of health are highly correlated with clinical measures of morbidity and predictive of subsequent death, health care use, and labor market behavior (Idler and Benyamini 1997; Manning et al. 1982). Researchers have also found that conceptually distinct self-reported health and chronic condition measures—including global health measures such as these census measures, or Activities of Daily Living (ADLs) and Instrumental Activities of Daily Living (IADLs)—are highly correlated with one another (Johnson and Wolinsky 1993). Regarding population contrasts, we note that general patterns of health-induced limitations by age and race in the 2000 census are consistent with what is known about age and racial patterns of disability from other sources.¹¹

Statistical Analysis

We took a two-step approach to address research Question 1 about the extent, nature, and types of moves made within, into, and out of the focal areas to or from anywhere in the nation. First, we tracked the percentage of each focal population who moved dwelling units between 1995 and 2000. Second, we classified movers as those who moved within the focal area, those who moved out of the area after 1995, and those who moved into the area by 2000. We tracked whether movers stayed within a specific focal area, moved between similar local areas anywhere in the United States (e.g., from one poor urban area to another), or moved across area types in the United States (e.g., from poor to nonpoor areas; from poor urban to poor rural areas).

To address research Questions 2 and 3 on the potential association between residential mobility and health limitations, we first compared the health-induced limitation rates of those who moved into or out of our focal areas with those who remained in them. Additionally, we estimated the probability that those living in each of our focal areas in 1995 moved out by 2000 as a function of their health-induced functional limitation status in a series of logistic regression models that ultimately also controlled for age and education. In a similar fashion, we estimated the probability of moving into one of our focal areas between 1995 and 2000. These estimates allowed us to determine whether health-induced limitation and residential mobility are associated as well as the extent to which age and education affect this association.

The net effect that migration might have on resident populations will depend on both the characteristics of movers and the frequencies of moves. To measure the extent to which differential residential mobility affected the geographic distribution of the healthy and unhealthy populations, we calculated the geographic distribution of health status of focal area residents by (1) residence as of 1995 and (2) current residence in 2000. A comparison between the two provides a direct measure of the effect of residential mobility on the geographic distribution of the healthy population over the five-year time horizon.

Results

Migration

Figure 1 displays that between one-quarter and almost one-half of the 2000 residents of each focal area had moved dwelling units since 1995. The vast majority of the movers relocated short distances. For example, of the Harlem residents who moved between 1995 and 2000, 47 % stayed within the same Harlem focal area, 64 % moved into Harlem from another part of the same county (Manhattan), and 85 % of those who moved into Harlem came from another part of New York State (Fig. 1). Similarly, 76 % of those moving out of Harlem stayed within New York state: 38 % stayed in Harlem, 47 % stayed in Manhattan, and 65 % stayed in the same metropolitan statistical area (MSA).

Fig. 1.

Percentage of the 2000 local area populations who moved between 1995 and 2000

Moves Between Urban and Rural or Between Northern and Southern Areas

Tables 3 (movers in) and 4 (movers out) provide information on the nature of moves. We found little evidence of migration between northern urban and southern rural local areas. The vast majority of movers into or out of the black northern urban poor areas moved within the same MSA; most others moved from or to another northern MSA, or to a southern MSA. For example, 87 % of movers in to Eastside Detroit and 88 % of movers out of the area moved within the same MSA. Although we found some evidence of movement between rural and urban areas within the same region (North or South), the number who moved across regions to or from a high-poverty rural area is often too small to disclose according to the census disclosure review (e.g., Harlem movers to a rural area). Similarly, the vast majority of white residents of high-poverty urban areas moved within the same MSA, and virtually all stayed in a northern MSA. For example, 78 % of Cleveland movers out went to the same MSA, and only 4 % went to a southern MSA.

Table 3.

Local area population who moved, 2000

	Urban High-Poverty (%)				Urban Nonpoor (%)					Rural High-Poverty (%)
	Harlem NYC	Eastside Detroit	Southside Chicago	Watts, Los Angeles	Bronx	Queens	NW Detroit	Southwest Chicago	Crenshaw/Bald win Hills	East North Carolina	Black Belt Alabama	Delta Louisiana
Blacks
All Movers
Stayed in same local area	47	47	42	47	33	32	37	40	38	76	83	80
Movers From Outside Local Area
From same MSA	81	87	84	92	93	92	89	90	88
From other non-southern MSA	9	7	8	4	3	4	6	5	7	23	41	33
From southern MSA	8	5	6	3	3	3	3	2	4	42	38	35
From rural non-South	NA	NA	NA	NA	NA	NA	NA	NA	NA	2	NA	3
From rural South	NA	NA	NA	NA	NA	NA	NA	NA	NA	33	NA	28
From low-poverty area	7	5	7	8	7	20	11	8	13	10	10	9
From moderate-poverty area	22	30	35	37	22	48	29	39	37	32	24	25
From high-poverty area	38	49	24	32	34	20	39	29	27	37	41	27
From very high-poverty area	33	15	32	22	37	12	21	24	23	21	25	38

	Urban High-Poverty (%)		Urban Nonpoor (%)		Rural High-Poverty (%)
	Cleveland	Detroit	West Cleveland	Sterling Heights	Appalachian Kentucky	West NC	Northeast Alabama	South Central Louisiana
Whites
All Movers
Stayed in same local area	49	51	47	27	83	64	75	71
Movers From Outside Local Area
From same MSA	74	755	62	86
From other non-southern MSA	17	13	28	9	27	13	12	4
From southern MSA	4	5	5	NA	22	62	63	59
From rural non-South	3	NA	4	2	9	6	3	2
From rural South	2	NA	1	NA	41	19	22	35
From low-poverty area	81	70	76	87	40	77	60	38
From moderate-poverty area	11	17	12	8	30	19	35	56
From high-poverty area	6	7	9	3	19	3	3	NA
From very high-poverty area	2	5	3	2	11	1	2	NA

Notes: When cell size is too small for the census to permit disclosure, the cell is labeled “NA.” Low poverty = <10 % of adults in poverty; moderate poverty = 10 % to 20 % of adults in poverty; high poverty = 20 % to 30 %; very high poverty = over 30 %.

Among movers into or out of high-poverty southern rural areas, the largest share moved to or from a southern MSA. For example, in East North Carolina, 42 % of movers in came from and 70 % of movers out went to a southern MSA. The stream between southern rural and urban areas was substantial in both directions (to and from) and for whites and blacks. Small percentages of black populations moved between southern rural areas, and virtually none moved between a southern rural and a northern rural area. White residents of high-poverty rural areas moved into or out of other rural areas at a greater rate than blacks, but they primarily moved between southern rural areas.

The vast majority of movers into and out of nonpoor urban areas, black or white, moved within the same MSA. Although small percentages moved between a northern and southern MSA, virtually none (undisclosed - <5 %) moved between an urban focal area and a rural area (North or South).

Areal Economic Mobility

Regarding moves between areas with different poverty levels, patterns vary considerably for black and white populations. Among black residents of high-poverty urban areas who moved, about one-half or more moved between high-poverty urban areas, and up to 95 % moved to or from moderate- or high-poverty areas. The percentage moving to or from low-poverty areas never exceeded 19 % (movers out of Harlem) and was 10 % or less in many cases (movers into Harlem, Eastside Detroit, Southside Chicago, and Watts). In stark contrast, the vast majority of whites who moved to or from urban high-poverty areas moved to or from low-poverty areas. Roughly 85 % of those who moved out of white urban high-poverty areas moved to a low-poverty area, and 70 % to 80 % of movers into the high-poverty white urban areas came from low-poverty areas.

The economic patterns of residential mobility for the rural areas were similar: most of the movers in and out of the high-poverty black areas made moves to and from similar poor areas, and the white rural areas saw more areal social mobility in their migration patterns. The majority of black movers out from rural high-poverty areas went to moderate-poverty or other high-poverty areas, and no more than 20 % moved to a low-poverty area. For example, 75 % of East North Carolina movers out went to moderate- or high-poverty areas, and only 19 % went to a low-poverty area. In contrast, a substantial proportion of whites in rural high-poverty areas (37 % to 74 %) moved to low-poverty areas, and a very small percentage moved to high-poverty areas, except for those moving out of the Appalachian Kentucky area, among whom about 22 % moved to another high-poverty area (still a smaller percentage than found for blacks moving from high-poverty rural areas).

Only 10 % of black rural movers in to high-poverty rural areas came from low-poverty areas, whereas up to 80 % of white rural movers in to high-poverty rural areas came from low-poverty areas. Most black rural movers in to high-poverty rural areas came from other high-poverty areas, and the percentage of white movers in from high-poverty areas is so small that we are not permitted to disclose it.

Among movers to or from the nonpoor urban areas, black-white variation in origin and destination is also apparent. Unlike black movers from high-poverty areas, who tended to make socioeconomically lateral moves, the majority of black movers out of nonpoor areas in 1995 (60 % to 82 %) moved to higher-poverty areas by 2000, with a smaller percentage moving to another low-poverty area. In striking contrast, the vast majority of white movers out from nonpoor areas (83 % to 92 %) moved to other low-poverty areas, with only 8 % to 17 % moving to areas that were less well-off.

With the exception of movers in to the Northwest Detroit area, black movers in to the nonpoor urban areas came primarily from moderate- or high-poverty areas. In the case of Northwest Detroit, 70 % came from low-poverty areas, and about 30 % came from high-poverty areas. Unlike most black movers to urban nonpoor areas, the vast majority of white movers into urban nonpoor areas came from other low-poverty areas (78 % to 90 %). Thus, unlike their black counterparts, the vast majority of white movers into or from nonpoor urban areas moved to or from other low-poverty areas.

What do the findings addressing research Question 1 say about the potential for health-selective migration as a contributor to the area-specific mortality differences found in previous research? We found considerable residential mobility, especially among blacks in northern urban areas. This mobility appears to be migration from one area of concentrated poverty to another, which would seem to argue against the plausibility of health-selective migration being an important determinant of observed geographic differences in health among poor blacks. We found at least as much upward mobility among blacks living in poor rural communities as in poor urban areas. For selective migration to explain the large mortality difference found between focal areas in the urban North and rural South, the northern upward out-migration would have to be substantially more health-selected than the rural southern upward out-migration. The similar mobility patterns we observed between the black urban high-poverty areas and black rural high-poverty areas suggest that documented urban-rural differences in mortality are unlikely to be importantly influenced by health-selective migration—either urban-rural or North-South. Still, both socioeconomic and geographic mobility occurred often enough that we cannot use migration streams alone to dismiss the possibility that health-selective migration might affect some geographic health profiles.

Migration and Health

Table 5 presents sex-specific age-standardized health-induced disability prevalence for each local area stratified by residential mobility status—stayers, movers out, and movers in—between 1995 and 2000. With two exceptions—Watts movers in and Bronx movers out—black men who moved in or out of any focal area had the same or lower disability prevalence than those who stayed. Differences between black male movers and stayers were generally larger in the high-poverty focal areas than in the nonpoor ones, with the largest disability rate difference found in Eastside Detroit: 8 percentage points higher for stayers than for movers out. Differences between black male movers out and stayers were statistically significant in four areas—Harlem, East Detroit, Northwest Detroit, and Southwest Chicago—and marginally significant (p < .10) in Black Belt Alabama. Disability differences between black male movers in and stayers were marginally significant in East North Carolina, Black Belt Alabama, and Delta Louisiana.

Table 5.

Age-standardized disability prevalence by residential mobility status, selected local areas, 2000

	Men			Women
Area Type/Area	Stayers	Movers Out	Movers In	Stayers	Movers Out	Movers In
Blacks
High-poverty urban
Harlem	30	24*	26	32	30	27
East Detroit	34	26***	30	32	32	29
Southside Chicago	27	27	23	29	24*	19***
Watts	31	28	33	31	28†	32
High-poverty rural
East North Carolina	30	26	24†	28	24	24
Black Belt Alabama	32	24†	25†	31	35	33
Delta Louisiana	28	26	21†	26	22	22
Nonpoor urban
Bronx	24	26	22	21	17	23
Queens	21	18	20	19	21	19
Northwest Detroit	26	23*	25	26	20***	29*
Southwest Chicago	26	19***	22	23	22	24
Crenshaw/Baldwin Hills	24	22	21	22	25	24
Whites
High-poverty urban
Cleveland	25	23	20*	23	23	24
Detroit	30	21***	21*	27	18***	30
High-poverty rural
Appalachian Kentucky	38	34	32	30	28	35
West North Carolina	22	16*	17***	19	15*	16*
Northeast Alabama	22	23	20	20	20	20
South Central Louisiana	25	25	21	19	17	17
Nonpoor urban
Western Cleveland	11	10	12	10	10	12
Sterling Heights, MI	13	10*	12	12	9*	13

Note: Asterisks below refer to statistically significant differences between the rates for movers rate and stayers.

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001

In most cases, black women movers in or movers out of poor areas also had the same or lower disability rates than stayers. Unlike for men, black women movers in nonpoor areas tended to have lower disability rates than stayers when they were movers out but higher disability rates when they were movers in, although the differences were small for both move types.

White men movers in or out generally had lower disability rates than stayers in all areas, although the differences were insignificant except for Detroit, where white male movers had a 9 percentage point lower disability rate than white male stayers. White women showed a less consistent pattern. In most cases, we found no statistically significant differences between movers and stayers; where we did, movers had lower disability rates than stayers.

Residential Mobility, Age, and Education

Table 6 presents sex-specific age-standardized high school graduation rates¹² for each focal area stratified by residential mobility status between 1995 and 2000.¹³ For every focal area, the same or a higher percentage of black male movers (in or out) were high school graduates than were stayers. This difference was generally large and statistically significant in the high-poverty areas but not in the nonpoor areas. In several cases in high-poverty areas, graduation rates were about 10 percentage points higher among movers than among stayers, and movers out of poor rural areas had graduation rates that were 11 to 20 percentage points higher than those of stayers.

Table 6.

Percentage high school graduates (age-standardized) by residential mobility status, selected local areas, 2000

	Men (%)			Women (%)
Area Type/Area	Stayers	Movers Out	Movers In	Stayers	Movers Out	Movers In
Blacks
High-poverty urban
Harlem	64	72*	74*	70	75*	74
East Detroit	60	70***	69***	67	73***	74***
Southside Chicago	70	72	80***	74	75	85***
Watts	65	73***	68	68	73*	71
High-poverty rural
East North Carolina	61	72***	64	68	76*	76*
Black Belt Alabama	61	73*	68	68	68	75*
Delta Louisiana	54	74***	69***	63	77***	71*
Nonpoor urban
Bronx	77	83	81	83	86	80
Queens	79	83	79	86	87	87
Northwest Detroit	75	78	78	82	85*	79
Southwest Chicago	72	73	73	79	82*	80
Crenshaw/Baldwin Hills	77	81	81	82	81	80
Whites
High-poverty urban
Cleveland	67	76***	81***	71	77*	84***
Detroit	63	75	81***	68	79***	82***
High-poverty rural
Appalachian Kentucky	56	71***	64*	66	72*	59
West North Carolina	71	83***	85***	77	83*	89***
Northeast Alabama	68	81***	77***	73	81***	79*
South Central Louisiana	67	80***	80***	73	85***	78
Nonpoor urban
Western Cleveland	88	88	89	89	91*	91
Sterling Heights, MI	87	89	84†	88	90	88

Note: Significance indicators below refer to statistically significant differences between the rates for movers rate and stayers.

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001

The patterns for black women in poor local areas were similar to black men’s, although more muted. In nonpoor urban areas, black women stayers were generally as likely to be high school graduates as movers in, and black women movers out were often more likely to be high school graduates than stayers, with the difference being statistically significant in two areas (Northwest Detroit and Southwest Chicago).

White men in urban or rural high-poverty areas who were movers had substantially greater high school graduation rates than did stayers. In nonpoor urban areas, differences between movers and stayers were small, and the direction was inconsistent. Patterns for white women were generally similar in all areas, but education differences between stayers and movers were smaller than for white men and less likely to be statistically significant.

Logistic Regression Results

In Table 7, we report the results of our race- and sex-specific regressions of the odds of moving out of and the odds of moving in to high-poverty urban, high-poverty rural, and nonpoor urban between 1995 and 2000. Explanatory variables include dummy variables for age (referent = 16–24 years), high school graduate, college graduate (referent = less than high school education), focal area of residence (in 1995, for odds of moving out; in 2000, for odds of moving in), and disability status.

Table 7.

Log odds of moving by type of area, disability status, and education, 1995–2000

	Men		Women
	Local Area/Disability/Age	Local Area/Disability/Age/Education	Local Area/Disability/Age	Local Area/Disability/Age/Education
Moving Out
Blacks
High-poverty urban
Disabled	–0.28 [0.07]***	–0.22 [0.07]***	–0.21[0.05]***	–0.15 [0.05]***
High school graduate		0.41 [0.07]***		0.23 [0.10]***
College graduate		0.76 [0.10]***		0.58 [0.08]***
High-poverty rural
Disabled	–0.27 [0.10]***	–0.15 [0.10]	–0.27 [0.09]***	–0.14 [0.09]
High school graduate		1.09 [0.11]***		0.73 [0.09]***
College graduate		1.97 [0.18]***		1.63 [0.13]***
Nonpoor urban
Disabled	–0.22 [0.06]***	–0.19 [0.06]***	–0.15 [0.05]***	–0.12 [0.05]*
High school graduate		0.13 [0.08]†		0.33 [0.07]***
College graduate		0.16 [0.11]		0.39 [0.08]***
Whites
High-poverty urban
Disabled	–0.36 [0.08]***	–0.24 [0.08]***	–0.33 [0.08]***	–0.22 [0.08]***
High school graduate		0.56 [0.08]***		0.51 [0.08]***
College graduate		0.91 [0.11]***		0.96 [0.11]***
High-poverty rural
Disabled	–0.25 [0.07]***	0.04 [0.07]	–0.20 [0.07]***	0.01 [0.07]
High school graduate		0.73 [0.08]***		0.56 [0.07]***
College graduate		2.11 [0.09]***		1.61 [0.08]***
Nonpoor urban
Disabled	–0.21 [0.09]*	–0.13 [0.09]	–0.17 [0.09]*	–0.08 [0.09]
High school graduate		0.68 [0.11]***		0.81 [0.11]***
College graduate		0.87 [0.12]***		0.99 [0.12]***
Moving In
Blacks
High-poverty urban
Disabled	–0.18 [0.08]***	–0.08 [0.08]	–0.24 [0.06]***	–0.14 [0.06]*
High school graduate		0.41 [0.09]***		0.32 [0.07]***
College graduate		1.17 [0.12]***		1.01 [0.09]***
High-poverty rural
Disabled	–0.40 [0.11]***	–0.32 [0.11]***	–0.21 [0.10]*	–0.13 [0.10]
High school graduate		0.48 [0.11]***		0.50 [0.11]***
College graduate		1.16 [0.19]***		0.99 [0.15]***
Nonpoor urban
Disabled	–0.19 [0.07]***	–0.18 [0.07]*	0.01 [0.06]	0.02 [0.06]
High school graduate		0.13 [0.08]†		–0.02 [0.07]
College graduate		0.16 [0.11]		0.02 [0.09]
Whites
High-poverty urban
Disabled	–0.50 [0.10]***	–0.23 [0.11]*	–0.34 [0.10]***	0.20 [0.11]†
High school graduate		0.94 [0.12]***		0.87 [0.12]***
College graduate		2.01 [0.14]***		2.03 [0.14]***
High-poverty rural
Disabled	–0.29 [0.06]***	–0.08 [0.06]	–0.22 [0.06]***	0.06 [0.06]
High school graduate		0.65 [0.66]***		0.51 [0.07]***
College graduate		1.60 [0.08]***		1.12 [0.08]***
Nonpoor urban
Disabled	–0.11 [0.09]	–0.05 [0.10]	–0.25 [0.08]***	0.11 [0.09]
High school graduate		0.31 [0.12]***		0.77 [0.13]***
College graduate		0.47 [0.13]***		1.09 [0.13]***

Notes: All models include age dummy variables and area-specific effects. For example, the models for poor urban blacks include controls for whether the individual lived in Harlem, Detroit, Chicago, or Los Angeles. The high school graduate category includes college graduates. Numbers in brackets are standard errors.

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001

As a general rule, we found that the odds of moving between 1995 and 2000 were higher in the younger and the more-educated groups than in the counterparts. Net of age, those who did not complete high school usually had the lowest odds of moving, and those who completed only high school had lower odds of moving than those with a college education, with the odds of moving for college graduates significantly larger in many cases. Exceptions to this were seen among black women who, for example, exhibited little difference by education in their odds of moving into a nonpoor urban area.

Odds of moving were in many cases lower among those with health-induced disability; however, the differences were smaller than seen for age or education, and they were attenuated by controls for age and education. In fact, in high-poverty rural populations and in white nonpoor areas, the odds of moving out were insensitive to disability status net of age and education. Overall, selection on education is larger and stronger than on health-induced disability per se, fully accounting for the age-adjusted relationship between migration and disability status in these cases.

Women showed no difference, net of age and education, in their odds of moving into focal areas by disability status, with the exception of urban poor black women, for whom the odds were 14 % lower among the disabled (compared with 32 % and 101 % higher for high school and college graduates, respectively). Moving into urban nonpoor areas was insensitive to education or disability, net of age. Among white women in urban high-poverty areas, the odds of moving in were 19 % higher for the disabled (but 97 % and 200 % higher for high school and college graduates, respectively).

For men, many groups showed no difference by disability status in the odds of being movers in, net of age and education, with slightly reduced odds found for disabled black men in rural high-poverty areas, disabled black men in urban nonpoor areas, and disabled white men in urban high-poverty areas. Here, too, the odds of moving in associated with health-induced disability were generally dwarfed in size by the impact of education.

Using multinomial logit models, we found no evidence that disability status affected the nature of either the origin from which or the destination to which an individual moved (not shown).

Impact of Selective Migration on Estimates

The findings to this point suggest that residential mobility is selected on factors associated with health (age and education) and sometimes on health-induced disability net of age and education, with some signs of variation in the nature and degree of selectivity by gender, area type, and race. Does such selection affect cross-sectional estimates of local area disability prevalence?

In Table 8, we present estimates of health-induced disability prevalence by sex in each focal area in 1995 and 2000 age-standardized to the U.S. black or U.S. white age distributions. The “1995” values represent the fraction suffering health-induced disabilities among those who were living in each focal area as of 1995: that is, as if no residential mobility had occurred between 1995 and 2000. The “2000” values represent the fraction suffering health-induced disabilities among the actual population living in each focal area in 2000.

Table 8.

Estimates of age-standardized disability prevalence (percentages), based on 1995 and 2000 local area populations

	Disability Prevalence (%)
	Men		Women
	1995	2000	1995	2000
Blacks
High-poverty urban
Harlem	29.2	30.0	31.3	31.6
Eastside Detroit	31.9	33.0	31.9	31.5
Southside Chicago	27.1	26.3	27.1	26.1
Watts	30.4	31.2	30.0	31.3
High-poverty rural
East North Carolina	29.0	29.1	27.0	27.3
Black Belt Alabama	31.4	31.9	30.7	30.9
Delta Louisiana	28.4	27.7	25.6	25.8
Nonpoor urban
Bronx	24.1	23.3	20.4	21.4
Queens	20.6	20.7	19.1	18.5
Northwest Detroit	25.5	25.9	24.5	25.7**
Southwest Chicago	24.4	25.2	22.5	22.9
Crenshaw/Baldwin Hills	23.8	23.3	22.6	22.4
Whites
High-poverty urban
Cleveland	23.9	23.3	23.1	23.3
Detroit	26.9	28.7 †	23.5	26.8***
High-poverty rural
Appalachian Kentucky	37.9	38.1	30.0	30.3
West North Carolina	21.4	21.3	19.0	19.2
Northeast Alabama	21.9	21.7	19.7	19.8
South Central Louisiana	24.4	24.2	18.5	18.4
Nonpoor urban
Western Cleveland	10.8	10.8	9.5	9.4
Sterling Heights, Michigan	11.9	12.4	10.7	11.4

Notes: Bolded, italicized contrasts are statistically significantly different from one another. Significance indicators below refer to statistically significant differences between 1995 and 2000; p values were calculated using the method proposed by Bonferroni (Weisstein 2012).

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001

Differences in age-standardized estimates of disability prevalence based on the 1995 versus the 2000 population of each focal area are small. For black men or black women, the two estimates are within about 1 percentage point in all areas. This is also the case for white men and white women; the notable exception is Detroit, where estimated disability prevalence is higher for the 2000 than the 1995 population, by 2 points for men and 3 points for women. The only statistically significant differences are those for black women in Northwest Detroit and for white women and white men in Detroit.

In most areas, the lower disability rates for the movers in and the movers out compared with the stayers seem to approximately cancel each other out, leaving estimates based on the 1995 and 2000 populations roughly equivalent. In the three Detroit cases where disability rates differed by year, the rates were always higher when using the 2000 definition of the population. A combination of factors can account for this. First, in the case of the white Detroit populations, the groups moving out were substantially larger than those moving in. Thus, for example, 37 % of the white women residing in Detroit in 1995 had moved out by 2000. In contrast, only 13 % of white women living in our Detroit area in 2000 had moved in between 1995 and 2000. For white men, 33 % of 1995 Detroit residents had moved out, but only 13 % of 2000 Detroit residents had moved in during the previous five years. Given these disparate flows, even if the movers in had lower disability rates than the movers out, the flows would not have canceled each other out, and the disability rate would still have been higher in the 2000 Detroit population. In addition, for black women in Northwest Detroit and white women in Detroit, the limitation rates for movers in were, if anything, marginally higher than the limitation rates for the stayers.

One concern about using two snapshots in a five-year period might be that health differences between those who had and had not moved out of areas might diverge outside the window studied. This would be more likely if net out-migration favored higher- or lower-SES groups. To check the plausibility of this possibility, we looked at the educational attainment of our 20 populations defined in terms of their 1995 and 2000 residential locations (not shown). Differences in education between the populations defined in terms of their 1995 and 2000 populations are small. Where meaningful and statistically significant differences do exist, net migration out of poor urban areas was as likely to favor the less well educated as the better-educated. In contrast, net migration out of poor rural and nonpoor urban populations tended to favor the better-educated. These are not the patterns one would have expected if upward mobility accounted for observed health profiles.

Sensitivity Analyses

To check the sensitivity of the results to the exclusion of those cases missing ZIP codes of previous residence, we compared our tabulations of the health profiles of the 2000 populations with tabulations that included the movers with missing ZIP codes. In each case, we know the individual was living in the area in 2000, but we do not know whether that person moved in from another area or changed dwelling units within the same area (i.e., whether they were movers in or stayers). We made two extreme assumptions in separate sensitivity analyses: (1) that all movers with missing ZIP codes had moved from outside the area (were movers in); and (2) that all movers with missing ZIP codes moved within the local area (were stayers).¹⁴ Changing our definition had no measureable effect on the characteristics of the movers, regardless of whether we focused on the fraction with a health-induced disability or the fraction with a high school diploma.

Similarly, we conducted a robustness check using available information on previous county and state. So, for example, when individuals did not report a ZIP code but did report a state of previous residence, we determined whether they moved across or within states. If they reported neither a previous ZIP code nor state, we made two extreme assumptions: (1) that all these individuals moved between states; and (2) that none did. Using this method, the bounds we calculated for the fraction of moves that involved moving from out of state were usually quite similar to the fractions reported.¹⁵ Thus, as a general point, our analyses provide reassurance of data validity to other researchers interested in using the census data on ZIP code of previous residence for analytic purposes related to short-distance moves or moves between local areas (regardless of whether their analytic interests are health-related).

Discussion

Using 2000 census data, we have shed new light on the important question of whether population dynamics of residential areas may be a competing explanation to place or static compositional effects for understanding observed associations between health and place. Analyzing migration streams at a more fine-grained level than was previously possible, we found little evidence that observed migration affected the overall distribution of the healthy or unhealthy populations across our focal study areas. Whereas those who moved out of focal areas tended to be younger, more highly educated, and in better health than those who stayed, those moving in tended to be younger, more highly educated, and in better health. These two residential mobility streams appeared to approximately cancel each other out, leaving cross-sectional estimates of local population health stable.

The notable exception to this regularity is Detroit, where small but statistically significant differences were seen, depending on which year’s snapshot was considered, among white men and women in the high-poverty areas and among black women in the nonpoor area. Although white movers in or out were as likely or less likely to be disabled than stayers, the streams in the two directions were of vastly different magnitude and did not cancel each other out in terms of their effect on the estimated percentage of the population suffering health-induced functional limitations. In addition, for black women in Northwest Detroit, disability rates for movers in were marginally higher than those for stayers, perhaps reflecting the movement of poor blacks from the central cities to inner-ring suburbs—a characteristic of poverty deconcentration efforts (Jargowsky 2003). These exceptions suggest that health and place researchers using cross-sectional estimates should consider whether a recent substantial in- or out-migration has occurred in a specific area that was not countered by a comparably sized migration in the opposite direction. If so, their estimates of the association between health and place may be badly biased.

Investigators have found blacks less likely than whites to translate economic resources into residential movement to better-resourced, safer, or higher-status communities (Alba et al. 1994; Charles 2003; Crowder 2001; Massey and Denton 1987; South and Crowder 1997a,b; Tolnay 2003). Structural barriers are key to this relative lack of upward residential mobility (Charles 2003; Crowder et al. 2012; Massey et al. 1994; Tolnay 2003). Our findings are generally consistent with these earlier studies: we found that about 85 % of whites who moved out of high-poverty areas moved to lower-poverty areas, whereas this was true for only 10 % to 20 % of blacks. Also, consistent with our findings of more muted associations among age, education, health, and residential mobility for women than men, Halliday and Kimmitt (2008) found a relationship between health and mobility among men but not among women, although spouse’s health affected the mobility of married women.

One might wonder whether documented return migration of black populations from the North to the South (Frey 1998, 2001; Long 1987) might influence the size or direction of differences in excess mortality between northern urban versus southern rural high-poverty black areas. However, we found results similar to those of earlier studies indicating that black North-to-South flows have typically been from urban to urban areas and are, to some extent, concentrated among middle-class blacks (Hunt et al. 2008). As such, return migration does not appear relevant to the question of why high-poverty urban populations experience greater excess mortality than high-poverty rural populations.

Although our estimates of the prevalence of health-induced disabilities generally do not suggest that health and place effects are spurious—that they are simply an artifact of selective-health migration—our findings that much movement into and out of the focal study areas is significantly associated with health suggests that investigators interested in understanding the causal effects of place on health face a complicated challenge. The substantial residential mobility across the study populations between 1995 and 2000—ranging from more than one-quarter to almost one-half of each population—mitigates against fully attributing the high rates of mortality found in some areas to the long-term exposure of local residents to specific neighborhood environmental characteristics, especially for mortality from diseases with long latency periods. The primary causes of death that persistently account for important variation in nonelderly adult excess mortality across U.S. population types—such as cardiovascular diseases, cancer, and HIV/AIDS—are such diseases (Geronimus et al. 2011). The possibility of a causal relationship between unique aspects of specific places and health among the residentially mobile might still pertain to health conditions with very short latency periods; to health conditions, even with long latency periods, if they are easily aggravated by short-term environmentally structured changes (e.g., changes in diet or exercise owing to restricted access to healthy foods or safe and inviting exercise spaces; or aggravated impacts of smoking owing to increased environmental air pollution or lead); and to differential access to health services by place, if it affects the rate that injuries or incident cases of disease (acute or chronic) progress into disabilities or become case fatalities. Conceptually, these are secondary or tertiary prevention pathways that are distinct from the pathway in which physical exposures inherent to specific geographic places result in chronic health problems after a long latency period.

To the extent that we found that migration did not affect the health profiles of focal areas, one plausible explanation would be that most moves tended to short-distance migration; or that, especially for blacks, between similar area types in terms of poverty status, urban or rural, and northern or southern location. It is possible that the environmental characteristics of specific types of local areas (high-poverty urban or high-poverty rural) are similar enough that even when moving between specific areas of the same type, individuals are exposed over long periods to the same noxious etiologic agents. Another possibility is that the frequent experiences of loss or disruption that may be implied by the frequency of moves we observed may, in itself, impact health and well-being and may be a common feature of disadvantaged areas (Fullilove 2004; Keene and Geronimus 2010; Keene et al. 2010; Wallace and Wallace 1998).

In addition to threatening material and environmental exposures, the psychosocial stressors inherent in lower structural positions or the contingencies of social identity associated with being black, urban, or poor in the United States may have adverse health effects that track with structural location, regardless of specific ZIP code. Cues to cultural oppression and structurally constrained life chances may trigger repeated or chronic episodes of vigilance, rumination, or physiological stress process activation with concomitant wear and tear on important body systems (Geronimus 2013; Geronimus and Thompson 2004; Geronimus et al. 2006; James 1994; Iznicht and Shmader 2011).

Limitations

Because we used census data to estimate the impact of health-related migration on associations between health and place, we were limited to the use of functional limitation as the health indicator and to a five-year period for residential mobility. Ideally, we would follow individuals’ residential mobility from childhood through adulthood. How might the results differ? One can imagine that residential environmental impacts could have immediate, medium-term, or long-term health implications, and that any long-term health implications may result from cumulative environmental impacts over the life course.¹⁶ The 2000 census data provide two residential snap shots of each individual, in 1995 and 2000. Some of these individuals were youth or young adults in 1995, and others were older. The five-year window allowed us to capture immediate environmental impacts, including ones that interact with chronic disease progression to disability or excess mortality, such as access to health care. Regarding longer-term health implications, we might very well see larger differences if we were able to observe individuals over a 10-, 20-, or 30-year period. Still, if there is important health-selective migration, we would, in general, expect to observe some evidence of this within just the five-year window.

We note an instance in which the five-year window could be misleading. If some poor young adults are upwardly socially mobile and move out of high-poverty neighborhoods as a result, their health trajectory may differ from their childhood neighbors who are not upwardly mobile.¹⁷ Such health differences might not emerge right away. Our consideration of educational characteristics of movers compared with stayers provides some evidence, albeit suggestive, that upward mobility is unlikely to be a serious threat to the validity of the results using the five-year window.

To our knowledge, these are the best currently available data for beginning to address the question of the degree to which cross-sectional associations between health and local residential place may be biased by failure to take into account the dynamic nature of local populations. Additional data are needed—to be collected or made available—that will allow researchers to dig deeper into the associations among health, place, and residential mobility. The ideal data set would have enormous sample sizes that are place-based or can be geocoded, would follow people longitudinally through their lives and would include reliable health measures. Administrative data would be one route. For example, Social Security records linked to death certificates have the potential to tell us where people were working at any point in their working lives, a proxy for place-based exposures as well as a way to track presumed residential mobility. One could imagine various other administrative data sets (such as welfare, Medicaid, or Medicare records) that would also offer needed information. Any one of these data sets would miss groups of people (those who have never worked or those who are not recipients of safety net programs), and matching administrative records to death certificates is imperfect. Yet, if researchers could overcome the confidentiality barriers to accessing these data and perhaps triangulate from different data sources, major strides might be made.

Although subject to noted limitations, this study is the first to attempt to address a critical question in the burgeoning literature on health and place. We found no evidence that excess mortality in U.S. high-poverty urban relative to high-poverty rural areas results from selective migration. Selective migration does not appear to bias cross-sectional estimates in most instances, but the Detroit results suggest that investigators cannot take this as axiomatic. Instead, consideration of whether a specific area studied has been subject to a recent, sizable net one-way migration—in or out—that was racially or socioeconomically structured is indicated for properly interpreting estimates. Our findings provide reassurance that quantitative cross-sectional estimates of the distribution of the healthy and unhealthy population can be used to target medical and social services where they are most needed. Yet, the failure of these estimates to point in a single etiological direction leave important sociological, epidemiological, and public policy questions unanswered regarding the fundamental and proximate causes of geographic disparities in health. Those designing studies to address such questions are advised to consider the implications of residential mobility dynamics in drawing valid inferences from cross-sectional data.

Table 4.

Local area population who moved, 1995

	Urban High-Poverty (%)				Urban Nonpoor (%)					Rural High-Poverty (%)
	Harlem NYC	Eastside Detroit	Southside Chicago	Watts, Los Angeles	Bronx	Queens	NW Detroit	Southwest Chicago	Crenshaw/B aldwin Hills	East North Carolina	Black Belt Alabama	Delta Louisiana
Blacks
All Movers
Stayed in same local area	38	37	40	33	29	24	30	25	37	74	73	68
Movers From Outside Local Area
Moved to same MSA	65	88	79	86	65	62	81	77	81
Moved to other non-southern MSA	14	5	11	8	12	11	8	12	11	NA	16	NA
Moved to southern MSA	16	4	8	5	19	22	9	9	7	70	73	69
Moved to rural non-South	NA	NA	1	NA	2	2	1	1	NA	NA	NA	NA
Moved to rural South	NA	NA	1	NA	2	3	2	1	NA	18	NA	14
Moved to low-poverty area	19	10	17	10	31	41	32	28	18	19	12	21
Moved to moderate-poverty area	26	39	39	37	35	38	23	34	40	43	33	27
Moved to high-poverty area	27	41	21	38	18	13	34	24	26	32	32	28
Moved to very high-poverty area	27	10	23	15	16	9	11	13	17	6	23	25

	Urban High-Poverty (%)		Urban Nonpoor (%)		Rural High-Poverty (%)
	Cleveland	Detroit	West Cleveland	Sterling Heights	Appalachian Kentucky	West NC	Northeast Alabama	South Central Louisiana
Whites
All Movers
Stayed in same local area	42	22	44	22	67	68	77	71
Movers from outside local area
Moved to same MSA	78	81	62	82
Moved to other non-southern MSA	11	8	23	9	10	11	6	5
Moved to southern MSA	4	4	9	4	38	66	71	67
Moved to rural non-South	4	5	4	4	6	4	2	2
Moved to rural South	2	2	1	1	46	18	21	26
Moved to low-poverty area	86	84	84	92	37	74	53	44
Moved to moderate-poverty area	11	11	8	5	41	19	40	49
Moved to high-poverty area	2	3	5	1	15	6	2	4
Moved to very high-poverty area	1	1	3	2	7	1	5	2

Notes: When cell size is too small for the census to permit disclosure, the cell is labeled “NA.” Low poverty = <10 % of adults in poverty; moderate poverty = 10 % to 20 % of adults in poverty; high poverty = 20 % to 30 %; very high poverty = over 30 %. South includes the following states: Virginia, North Carolina, South Carolina, Georgia, Florida, Alabama, Mississippi, Louisiana, Texas, Arkansas, Oklahoma, and Tennessee.