#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
Access to Health Care: Does Neighborhood
Residential Instability Matter?*
JAMES B. KIRBY
Agency for Healthcare Research and Quality, U.S. Department of Health and Human Services
TOSHIKO KANEDA
Population Reference Bureau
Journal of Health and Social Behavior 2006, Vol 47 (June): 142–155
Many Americans do not have access to adequate medical care. Previous
research on this problem focuses primarily on individual-level determinants of
access such as income and insurance coverage. The role of community-level factors in helping or hindering individuals in obtaining needed medical care, however, has not received much attention. We address this gap in the literature by
investigating the association between neighborhood residential instability and
access to health care. Using individual-level data from the 2000 Medical
Expenditure Panel Survey and block-group level data from the 2000 decennial
census, we find that individuals who live in neighborhoods with high residential turnover have worse health care access than residents of other neighborhoods. This association persists even when the prevalence of poverty, the supply of health care, and a variety of individual characteristics are held constant.
Delivered by Ingenta to :
We offer explanations for theseUniversity
findings ofand
suggest
North
Dakotadirections for future
research.
Wed, 13 Aug 2008 17:35:46
Many Americans do not have adequate
access to health care services (Institute of
Medicine 2001; U.S. Department of Health
and Human Services 2000). Identifying and
understanding factors that help individuals
obtain needed medical care or that hinder them
from doing so is therefore an important goal
for researchers interested in the U.S. health
care system and, ultimately, in population
health. To date, most research on access to
health care has focused on individual-level
determinants such as insurance coverage,
income, educational attainment, and race
(Brown, Wyn, and Teleki 2000; Center for
Health Economics Research 1993; Collins,
Hall, and Bebus 1999; Institute of Medicine
* The views in this paper are those of the authors,
and no official endorsement by the Agency for
Healthcare Research and Quality and the
Department of Health and Human Services is
intended or should be inferred. Address correspondence to James B. Kirby, Agency for Healthcare
Research and Quality, 540 Gaither Road, 5th floor,
Rockville, MD 20850.
2001). However, there is an increasing amount
of research that suggests community characteristics may also influence individuals’ ability to
obtain needed medical care (Anderson and
Davidson 2001; Anderson, Rice, and Kominski
1996; Berk, Shur, and Cantor 1995; Kirby and
Kaneda 2005). We contribute to this line of
research by investigating neighborhood residential instability—measured by the proportion of individuals living in their current
homes for only a short time—as a possible
community-level determinant of access to
health care.
While substantial geographic variation in
the use of health care has been documented
(Bindman et al. 1995; Carlisle et al. 1995;
Cunningham and Kemper 1998b; Komaromy
et al. 1996; Roderick et al. 1999; Wells and
Horm 1998), it is not always clear whether this
variation is a function of the composition of
individuals in areas or whether it is actually
due to characteristics of the communities
themselves. This distinction between “composition” and “context” is central to this study.
142
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
143
For example, there may be lower average lev- ments was associated with an increased likeliels of access in unstable communities simply hood of having a usual source of care and havbecause such communities are composed dis- ing at least one physician visit.
proportionately of poor individuals who canIn addition to the social and economic facnot afford care regardless of where they live. If tors identified by these studies, factors related
so, the association between residential instabil- to the supply of health care services in a comity and access to care is “compositional.” munity may influence access and use. For
Alternatively, there might be something about example, both Brown et al. (2004) and
living in unstable communities that affects Andersen et al. (2002) found that the number
access regardless of individuals’ personal char- of federally qualified health centers was posiacteristics, in which case the association is tively related to the likelihood of having a
“contextual.” Only recently have studies begun usual source of care. Not all research, howevto investigate the possibility of a contextual er, supports a supply-effect (Grumbach,
relationship between neighborhood character- Vranizan, and Bindman 1997). The organizaistics and access to health care.
tion of the health care market in an area may
Studies reveal substantial variation across also affect individuals’ ability to get care. One
communities in several different indicators of study suggests that the extent to which physiaccess, including subjective reports of access cians in an area derive their income from manproblems (Cunningham and Kemper 1998a; aged care organizations decreases the amount
Cunningham and Kemper 1998b), the number of time they spend providing charity care
of physician visits children have during a given (Cunningham et al. 1999). As a result, poor
year (Long and Marquis 1999), and pre- individuals in such areas may be less able to
ventable hospitalizations (Billings, Anderson, obtain care.
and Newman 1996; Bindman et al. 1995;
All of these studies suggest that factors at
Friedman et al. 1999; Gaskin and Hoffman the community level may affect individuals’
2000). More generally, both hospital-based
access to
net of their personal characterDelivered by Ingenta
to care,
:
care use and ambulatory care use vary
considistics.
Residential
instability, however, has not
University of North Dakota
erably across geographic areas (Carlisle
al. 2008
been17:35:46
examined. Yet sociologists have been
Wed, 13etAug
1995; Komaromy et al. 1996). Some of these interested in residential instability as a predicstudies establish that personal characteristics tor of various community-level and individualsuch as insurance coverage, income, educa- level outcomes for at least five decades (Faris
tion, and race cannot completely account for and Dunham 1939; Shaw and McKay 1942).
this variation. This suggests that associations Research links residential instability to crime
between community-level factors and access to (Sampson, Raudenbush, and Earls 1997), menhealth care may be, at least in part, contextual; tal health (Ross, Reynolds, and Geis 2000),
community characteristics themselves could social relations (Crutchfield, Geerken, and
influence individuals’ ability to obtain needed Gove 1982; Sampson 1985; Sampson and
health care services.
Groves 1989; Sampson et al. 1997), and social
A few studies actually identify specific capital (Coleman 1988, 1990). In addition,
community-level social and economic charac- there is a large amount of work in medical
teristics that are associated with access, net of sociology that links social networks to health
individual-level factors. Looking at a sample care utilization (Earp et al. 2002; Freidson
of low-income individuals in the 100 largest 1970; Horwitz 1977; Kadushin 1966, 1969;
metropolitan areas, Andersen et al. (2002) Levy-Storms and Wallace 2003; McKinlay
found that high unemployment and low per 1973; Pescosolido 1992; Suchman 1964). This
capita income were associated with a reduced research is also relevant because residential
likelihood of having at least one physician visit instability is likely to affect the development
during the year. In a study of 54 urban areas, and functioning of social networks.
Brown et al. (2004) found that the size of the
safety-net population, as proxied by the proportion of individuals who had low incomes THEORY AND HYPOTHESES
and were uninsured, was negatively associated
Residential instability plays a role in several
with the likelihood of having a usual source of
care. This study also found that, among unin- prominent sociological theories. Early formusured individuals, the size of Medicaid pay- lations of both social disorganization theory
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
144
JOURNAL OF HEALTH AND SOCIAL BEHAVIOR
and social control theory posit that low resi- cine, and responses to illness are all affected
dential instability increases social integration by network homogeneity and exclusivity, and
as measured by the likelihood that neighbors that these factors explain racial and ethnic
will know one another, be willing to help one differences in health care utilization.
another, and be able to exert informal social Subsequent studies found that help-seeking
control (Hirschi 1969; Shaw and McKay behavior is affected by the types of social cir1942). In contrast, residents of neighbor- cles in which individuals are involved
hoods characterized by high rates of residen- (Kadushin 1966, 1969). Later research estabtial turnover are more likely to be strangers lished that network structure and content
and less likely to be embedded in social net- affect the timing of seeking care as well as the
works through which they might receive type of care sought (Freidson 1970;
resources and by which social control is exer- McKinlay 1973). More recent work suggests
cised (Crutchfield et al. 1982; Sampson 1985; that the interaction made possible by social
Sampson and Groves 1989; Sampson et al. networks helps individuals handle health1997). Research suggests that residential related difficulties (Pescosolido 1992), that
instability thus leads to high levels of crime community members are more effective at
and victimization (Sampson et al. 1997), low encouraging health care use than health care
levels of exchange (Sampson, Morenoff, and professionals (Earp et al. 2002), and that
Earls 1999), and, in some circumstances, being well-connected to church and civic netreduced psychological well-being (Ross et al. works increases some types of preventive
2000; Sampson et al. 1997).
screenings (Levy-Storms and Wallace 2003).
Residential instability also figures promiThus, over the past four decades, medical
nently in research using the concept of social
sociologists have established that social netcapital. Coleman defines social capital as a
works are important in determining access to
resource that “.|.|. inheres in the structure of
and use of health care services, while
relations between actors and among
actors”by Ingenta to :
Delivered
research in other areas suggests that residen(1988:98). For example, trust, shared
expec-of North Dakota
University
tial instability disrupts social networks. Our
tations, and norms of reciprocityWed,
in commu13 Aug 2008
17:35:46
nities facilitate free exchange of valuable hypotheses are based on these two premises.
information and resources. Coleman’s (1990)
work identified continuity of community
structure as one key to the emergence of
social capital, and it showed that a high level
of population turnover disrupts this continuity. Residential instability not only hinders
the development of new social ties, but the
frequent severing of social ties disrupts entire
social systems within communities (Coleman
1990). Recent research links social capital
and social integration to a variety of healthrelated outcomes (Berkman et al. 2000; Hawe
and Shiell 2000).
A common premise in the theoretical perspectives described above is that residential
instability disrupts the development and functioning of social networks; individuals in residentially unstable communities are less likely to be embedded in social networks through
which information and other resources can be
exchanged. Other research suggests that the
content and structure of social networks are
important determinants of access to and use
of health care services. More than 40 years
ago, Suchman (1964) found that knowledge
about disease, attitudes about modern medi-
Hypotheses
In order to obtain needed health care, individuals need information about the availability of health care resources in their communities. For example, knowledge regarding the
location of facilities that offer affordable
health care services, the safest and most convenient means of getting to such facilities, the
quality of care provided at them, and whether
providers can communicate in a particular
language are extremely helpful, if not essential, in obtaining needed care. Moreover,
much of this information is specific to a given
area and therefore is most readily available
through informal social contacts. As individuals in residentially unstable neighborhoods
are less likely to be embedded in social networks, area-specific information regarding
health care services may be less available to
them. Further, even if individuals are aware of
the health care resources available in their
communities, help in actually obtaining care
may be less available in neighborhoods char-
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
145
acterized by high residential turnover because DATA AND METHODS
support exchanges are less common.
We therefore hypothesize that residential Sources of Data
instability is associated with lower levels of
Data for this study come from three sources.
access to health care, net of other contextuallevel factors such as the prevalence of pover- Individual-level data come from the 2000
ty in a neighborhood and the supply of health Medical Expenditure Panel Survey (MEPS), a
care providers. Further, as the mechanisms we series of longitudinal surveys based on cluspropose are not simply compositional in tered and stratified samples of households that
nature, we expect that any association provide nationally representative estimates of
between instability and access to health care health care use, insurance coverage, and
will persist even after controlling for individ- sociodemographic characteristics for the U.S.
ual-level characteristics. In other words, resi- noninstitutionalized population (Cohen 1996,
dential instability and access to health care 1997). We link individuals in the 2000 MEPS
are associated not simply because unstable to information regarding the supply of health
neighborhoods are composed of individuals care providers from the Health Resources and
who would have poor access regardless of Services Administration. Finally, to obtain
where they live, but because neighborhood neighborhood-level characteristics, we attached
residential instability itself affects the ability longitude and latitude figures to addresses in
the 2000 MEPS sample (often referred to as
of residents to obtain health care services.
A limitation of this study is that we are not “geocoding”), which enabled us to link individable to conduct a direct test of the proposed uals to information from the 2000 decennial
mechanisms that link residential instability to census regarding the block groups in which
health care access; we do not have data on the they live. Block groups are the smallest geographic area for which social statistics are
characteristics of the social networks within
available.
generally contain between 600
Delivered
by
Ingenta
to They
:
neighborhoods or on the volume or type of
and
3,000
people
(U.S. Census Bureau 2000)
University
information that passes through
them.of North Dakota
and 17:35:46
can be considered approximations of neighWed,
13
Aug
2008
Nevertheless, if we are correct in hypothesizborhoods (Auchincloss, Van Nostrand, and
ing that residential instability functions to
Ronsaville 2001).
decrease access by making the transmission
The 2000 MEPS collected data on 25,096
of health care-related information through individuals, 90 percent of whom were successsocial networks less efficient, then other pat- fully linked to a census block group. Though
terns in our data should also be apparent.
differences between individuals with and withOne such pattern has to do with the preva- out block group information were modest,
lence of Spanish speakers at the neighbor- individuals missing block group information
hood level and preference for Spanish at the were more often nonwhite, less educated, and
individual level. If information via word of poor. To minimize sample selection bias, we
mouth is an important determinant of access, imputed missing block-group data and includthen individuals who prefer Spanish over ed a dichotomous variable to identify observaEnglish should have better access in neigh- tions with imputed values. However, the
borhoods where Spanish is widely spoken. dichotomous variable identifying such cases
Conversely, non-Spanish speakers in predom- was never significant in our analysis, nor did
inantly Spanish-speaking neighborhoods our substantive findings change when such
should have worse access than those in other cases were removed. We therefore exclude
neighborhoods. Another pattern that should individuals without block-group information
emerge in our data, if our explanations are from our analyses, yielding a total sample size
correct, pertains to the reasons individuals of 22,656.
have poor access. If our explanations are
valid, neighborhood instability should more
strongly predict poor access due to lack of Access to Health Care
information than poor access due to other factors (e.g., affordability or problems with
Our dependent measure is dichotomous and
insurance coverage). We are able to test both identifies individuals who have poor access to
health care. We use responses from several
these possibilities with our data.
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
146
JOURNAL OF HEALTH AND SOCIAL BEHAVIOR
items in the MEPS questionnaire to construct hood residential instability, measured as the
this variable.1 The questionnaire first asked percentage of residents in a block group who
individuals whether they had a “particular doc- have lived in their current homes for one year
tor’s office, clinic, health center, or other or less. It ranges from 1 to 10, with one unit
place” to which they usually go when they are corresponding to ten percentage points. We
sick or need advice about their health. If a per- also estimated all of our models with alternason reported that they did not have a usual tive cutoffs for duration of residence (2, 3, 4,
source of care, the questionnaire then asked and 5 years). All of these cutoffs produced
why. If an individual reported that they did equivalent results.
have a usual source of care, they were asked
The prevalence of poverty is another imporwhether that source was a hospital emergency tant neighborhood-level characteristic to
room. Finally, household respondents were include in our models. Poverty is correlated
asked whether they or a family member could with residential instability and is also associatnot get care they needed during the previous ed with access to health care (Kirby and
twelve months.2 Using responses from these Kaneda 2005). Our hypotheses, however, prequestions, we coded individuals as having poor dict that neighborhood residential instability
access if any of the following were true: (1) will have an effect on access independent of
they did not have a usual source of care, (2) the prevalence of poverty in a neighborhood.
their usual source of care was a hospital emer- We measure neighborhood poverty as the pergency room, or (3) they reported unmet med- centage of individuals in a neighborhood
ical need. However, if individuals reported not whose income is less than 125 percent of the
having a usual source of care because they “go federal poverty line. Like neighborhood resito different doctors for different needs” or dential instability, the prevalence of poverty in
because they recently moved to an area, they a neighborhood is measured in units of ten perwere not considered to be without a usual centage points and therefore can range from 0
to 10. to :
source of care.
Delivered by Ingenta
Our
final neighborhood-level measure is a
Each of the three criteria described
above
University of North
Dakota
dichotomous
taps an important dimension of access.
Wed, Having
13 Aug 2008
17:35:46 variable that identifies neighbora usual source of care is an important gauge of hoods in which at least 80 percent of the resiaccess because it indicates whether an individ- dents speak Spanish at home. The 2000 census
ual has a specific entry point into the health long form ascertains whether individuals speak
care system if some event necessitates it. Spanish exclusively at home, though many
Previous studies use this measure as a standard may be fluent in English. We interact this
benchmark for whether an individual has neighborhood-level variable with a dichotoaccess to ambulatory care (Zuvekas and mous variable identifying individuals who
Taliaferro 2003). As emergency rooms are not completed the MEPS questionnaire in Spanish.
designed to provide routine care, we consider
the use of emergency rooms for general medical needs to be an indicator of poor access. Health Care Supply Variables
Finally, perceptions of unmet medical need are
An association between neighborhood charimportant because they provide a sense of the
difficulty that people experience in getting acteristics such as residential instability and
care, rather than whether one simply has a access to health care could be in part due to
place they would go to get care. Together, these differences in the supply of health care sermeasures provide an overall sense of the extent vices available to residents. The supply of
to which individuals have difficulty in obtain- health care therefore needs to be included in
ing needed medical care. However, the models our models. Unlike information on residential
yield very similar results when each compo- instability and poverty, information on the supnent of our dependent variable is analyzed sep- ply of health care is not available at the blockgroup level. Moreover, it is probably better to
arately.
measure supply characteristics at higher levels
of aggregation because health service
providers and facilities serve geographic areas
Neighborhood-Level Variables
that are generally larger than block groups. To
Our main independent variable is neighbor- measure the supply of health care in an area,
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
147
cific because Medicare covers nearly everyone
over the age of 64.
Health status is another important predictor
of access that may be related to the likelihood
of living in an unstable neighborhood. Poor
health is likely to be more common in disadvantaged and unstable neighborhoods, and it
is also likely related to at least one of the
three components in our access measure.
Specifically, unhealthy individuals are unlikely to be without a usual source of care. We
measure health status using three variables:
subjective health status, the presence of
chronic conditions, and the presence of functional limitations. Subjective health status is
captured with a battery of dichotomous variables indicating whether individuals rate their
health as excellent, very good, good, fair, or
poor. The variable on chronic conditions indiIndividual-Level Variables
cates how many diagnosed conditions a perIn order to distinguish the compositional son has out of the following conditions: angieffect of neighborhood residential instability na, asthma, coronary heart disease, diabetes,
from its contextual effect, it is important to emphysema, hypertension, heart attack, and
control for individual characteristics that might stroke. Disability is measured with a variable
be associated with both access to health care that indicates whether individuals need help
or supervision
with personal care such as
Delivered
to :
and the likelihood of residing in an
unstableby Ingenta
bathing,
dressing,
or getting around the
neighborhood. Among the mostUniversity
importantof North Dakota
house.
We
originally
operationalized health
Wed,
13
Aug
2008
17:35:46
individual characteristics with respect to
status with an exhaustive set of dichotomous
access to health care are socioeconomic status
variables to capture all conditions and func(SES) and insurance coverage. These factors
tional disabilities. This did not substantially
are also associated with the type of neighborenhance the explanatory power of the models,
hood in which individuals live. We measure
so we elected to use a more parsimonious
individual-level SES using a series of dummy
operationalization.
variables on household income relative to the
In addition to SES, health insurance coverfederal poverty line (less than 125%, age, and health status, we control for other
125%–200%, 200%–400%, or 400% or more) basic demographic characteristics, namely:
and educational attainment (no high school gender, age, marital status, race/ethnicity
degree or GED, a high school diploma only, a (Hispanic, non-Hispanic white, non-Hispanic
college degree, a graduate or professional black, non-Hispanic Asian, or non-Hispanic
degree, or under the age of 25 with a high and another race), and the language of the
school diploma). The “under 25” category is a interview (Spanish or not). All of these characproxy for individuals who may not have com- teristics are associated with access to health
pleted their education. We measure insurance care and may be associated with the type of
status using five dummy variables to indicate neighborhood in which individuals live, makwhether individuals are: age 65 or above and ing these characteristics important to include
insured exclusively by Medicare throughout in our model. Table 1 displays general sample
the year; age 65 or above and insured by characteristics, and Table 2 shows coding and
Medicare plus some private supplemental descriptive statistics for all the variables
insurance plan sometime during the year; included in the analysis.
under age 65 and insured by a private plan any
time during the year; under age 65 and insured
by Medicaid some time during the year (but Analytical Approach
never by a private plan); or under age 65 and
To test our hypotheses, we estimate a series
uninsured all year. Insurance status is age-spewe include the number of general practice or
family practice physicians per 1,000 people in
a “Primary Care Service Area” and the number
of hospitals beds available per 1,000 county
residents. A Primary Care Service Area is a
standardized area developed by the Health
Resources and Services Administration to represent market areas for primary care services,
consisting of one or more zip codes. Figures
for hospital beds per capita are not available at
the service-area level, so the county level is
used instead. Finally, we include a variable that
indicates whether individuals live in a
Metropolitan Statistical Area. This is a general
proxy for the geographic density of health care
providers and facilities.
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
148
JOURNAL OF HEALTH AND SOCIAL BEHAVIOR
TABLE 1. Basic Sample Characteristics
Total number of counties
Total number of Primary Care Service Areas
—(PCSAs)
Total number of block groups
Total number of observations
Mean number of people per county
Mean number of people per PCSA
Mean number of people per block group
Intra-block-group correlation coefficient
(95% confidence interval)
548
998
3,527
22,656
41.34
34.52
6.42
.27
(.25, .28)
ing at the primary sampling unit must also be
considered. If clustering in the sample is
ignored, the standard errors of our estimates
will be biased downward, increasing the
chances of Type I errors. We have chosen to
deal with this problem by calculating standard
errors using a first-order Taylor series linear
approximation method available in Stata,
which adjusts for clustering at the primary
sampling unit level (Levy and Lemeshow
1999; Statacorp 2001). Though this method
was designed to control for clustering due to
sample design rather than nested data, it provides accurate variance estimates because
block groups, service areas, and counties are
all contained entirely within single primary
sampling units (Goldstein 1999). All point
estimates are calculated using sample weights.
of logistic regression models. In the first
model, we include only neighborhood residential instability, thereby investigating the bivariate association between residential instability
and access to care. In the second model, we
add the prevalence of poverty in neighborhoods and the health care supply variables.
This model describes how much of the crude
association between residential instability and RESULTS
access is driven by other community-level
characteristics. In the third model, we control
A simple first step for evaluating the possifor all individual-level variables in addition to bility that block-group characteristics influthose in the previous models. This model indi- ence access to health care is to examine the
cates the extent to which the association intra-block-group correlation coefficient
to :
between neighborhood instability Delivered
and accessby Ingenta
(TableDakota
1). This statistic represents the theoretiUniversity
of
North
to health care is due to the composition of indi- cal upper bound for the proportion of variance
Wed, 13 Aug 2008 17:35:46
viduals within neighborhoods or to neighbor- in the dependent variable attributable to blockhood instability itself. Model 4 tests the possi- group-level characteristics. The intra-blockbility that residing in a neighborhood in which group correlation coefficient is significant (r =
Spanish is widely spoken interacts with .27, p < .05). While far from conclusive, this is
whether the MEPS interview was conducted in at least consistent with the idea that blockSpanish (a proxy for being more comfortable group characteristics explain some of the varicommunicating in Spanish than in English).
ance in access to health care. Our multivariate
To further test our hypotheses, we estimate analysis delves further into this possibility.
an additional model using an alternative
In Table 3, we present odds ratios and t-stadependent variable (model 5). We construct a tistics from four logistic regression models.
dummy variable that takes a value of 1 for indi- The results from the first model provide a
viduals who do not have a usual source of care description of the crude association between
because they do not know where to go to get neighborhood residential instability and the
care, because they believe a usual source of likelihood of having poor access to health care.
care is not available in their area, or because Consistent with our expectations, residents of
they cannot find a provider who speaks their unstable neighborhoods are significantly more
language; the variable is coded 0 for all others. likely to have poor access compared to those in
This variable identifies individuals who do not other neighborhoods. Specifically, the odds
have a usual source of care because they lack ratio from model 1 indicates that a 10 percentinformation, rather than for reasons having to age point increase in the number of residents in
do with affordability, insurance, or other barri- a neighborhood who have lived in their current
ers to care.
homes for one year or less is associated with an
A major methodological challenge in this increase of 23 percent in the odds of having
study is the hierarchical structure of the data; poor access to health care.
individuals are nested within block groups,
In our second model, we include the prevaservice areas, and counties. In addition, MEPS lence of poverty in a neighborhood and the
is a stratified and clustered sample, so cluster- health care supply variables (i.e., doctors and
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
149
TABLE 2. Descriptive Statistics for All Variables
Variable
Weighted Mean
Dependent variable
—Poor access (yes = 1)
.22
—Component variables
——No usual source of care (yes = 1)
.17
——Unmet medical need (yes = 1)
.07
——Usual source of care is an emergency room (yes = 1)
<.01
Neighborhood-level variables
—Percent of persons living 1 year or less at current residence
1.91
—Percent of persons living below 125% of federal poverty line
1.03
—More than 80% of residents speak Spanish (yes = 1)
.01
Health care supply variables
—Specialists per 1,000 PCSA residents
1.33
—General practitioners per 1,000 PCSA residents
.75
—Hospital beds per 1,000 county residents
3.52
—Person resides in an MSA (yes = 1)
.84
Individual-level variables
—Male (yes = 1)
.49
—Married (yes = 1)
.41
—Age
35.65
—Race and ethnicity
——Non-Hispanic white
.71
——Non-Hispanic black
.13
——Non-Hispanic Asian
.03
——Non-Hispanic other
.01
——Hispanic of any race
.12
—Interview conducted in Spanish (yes = 1)
.04
—Highest degree obtained
——No high school diploma/GED
.10
——High school diploma/GED
Delivered by Ingenta to : .37
——College graduate
University of North Dakota .12
——Graduate or professional degree
.06
Wed,
13 Aug 2008 17:35:46 .35
——Younger than 25, inapplicable
—Income relative to federal poverty line
——Under 125%
.15
——125%–200%
.13
——200%–400%
.33
——More than 400%
.39
—Self-rated health
——Excellent
.33
——Very good
.34
——Good
.24
——Fair
.07
——Poor
.02
—Needs help with daily activities (yes = 1)
.01
—Number of serious chronic conditions
.44
—Health insurance
——Private insurance, younger than 65
.67
——Public insurance, younger than 65
.10
——Uninsured, younger than 65
.11
——Private insurance, 65+
.07
——Public insurance, 65+
.05
Notes: PCSA = Primary Care Service Area. MSA = Metropolitan Statistical Area.
hospital beds per 1,000 residents and whether
one resides in a metropolitan statistical area).
The association observed in model 1 remains
statistically significant even after controlling
for these community-level characteristics,
though the odds ratio is smaller. Net of the
prevalence of poverty and the supply of health
care, a 10 percentage point increase in the
number of neighborhood residents with hous-
Minimum
Maximum
0
1.00
0
0
0
1.00
1.00
1.00
0
0
0
10.00
7.74
1.00
0
0
0
0
14.49
3.23
22.64
1.00
0
0
0
1.00
1.00
90.00
0
0
0
0
0
0
1.00
1.00
1.00
1.00
1.00
1.00
0
0
0
0
0
1.00
1.00
1.00
1.00
1.00
0
0
0
0
1.00
1.00
1.00
1.00
0
0
0
0
0
0
0
1.00
1.00
1.00
1.00
1.00
1.00
8.00
0
0
0
0
0
1.00
1.00
1.00
1.00
1.00
ing tenure of one year or less is associated with
a 15 percent increase in the odds of having
poor access to health care. Results from supplementary models (not presented) indicate
that the majority of the decrease in the coefficient for neighborhood instability is due to the
inclusion of neighborhood-level poverty, not
the supply of health care providers or metropolitan residence. Note also that neighbor-
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
150
JOURNAL OF HEALTH AND SOCIAL BEHAVIOR
TABLE 3. Odds Ratios and t-statistics from Four Logistic Regression Models on Having Poor Access
to Health Care
Odds Ratios and t-statistics
Independent Variables
Neighborhood-level variables
—Percent of residents living one year or less in current homes
Model 1
Model 2
Model 3
Model 4
Model 5
1.23
(7.97)**
1.15
(5.52)**
1.31
(7.95)**
1.13
(4.68)**
1.10
(2.76)**
1.13
(4.63)**
1.09
(2.45)*
1.17
(4.40)**
1.00
(.01)
1.06
(1.05)
.66
(2.10)*
.97
(1.99)*
1.26
(2.31)*
1.06
(1.09)
.73
(1.65)
.97
(2.21)*
1.09
(.85)
1.06
(1.12)
.73
(1.67)
.97
(2.15)*
1.08
(.75)
1.26
(.86)
.95
(.69)
.97
(1.83)
.90
(.76)
1.51
(9.38)**
.73
(5.92)**
1.00
(.24)
1.51
(9.37)**
.72
(6.05)**
1.00
(.26)
1.06
(1.03)
.79
(2.37)*
.99
(2.10)*
1.17
(1.76)
1.30
(1.52)
.99
(.04)
1.34
(3.83)**
1.18
(1.86)
1.30
(1.57)
.99
(.02)
1.17
(1.98)*
.66
(2.87)**
1.02
(.07)
.95
(.13)
.89
(.81)
1.18
(2.49)*
1.03
(.37)
.88
(1.25)
.53
(6.32)**
1.14
(1.93)
1.03
(.31)
.87
(1.28)
.52
(6.40)**
1.31
(2.77)**
1.41
(3.02)**
1.43
(2.00)*
.87
(.88)
1.29
(2.49)*
1.76
(6.57)**
1.32
(4.16)**
1.28
(2.37)*
1.74
(6.35)**
1.32
(4.16)**
2.80
(5.66)**
3.55
(7.94)**
2.28
(6.84)**
.88
(1.80)
.89
(1.48)
.91
(.91)
1.36
(2.09)*
.97
(.10)
.88
(1.76)
.89
(1.50)
.91
(.88)
1.37
(2.13)*
.97
(.09)
1.04
(.41)
1.29
(2.20)*
2.03
(5.20)**
3.22
(6.74)**
1.19
(.47)
1.47
(3.49)**
1.46
(3.44)**
2.29
(5.97)**
—Percent of residents living below 125% of federal poverty line
Health care supply variables
—Specialists per 1000 PCSA residents
—Primary care physicians per 1000 PCSA residents
—Hospitals beds per 1000 county residents
—Resides in an MSA (yes = 1)
Individual-level variables
—Gender (female = reference)
—Marital status (unmarried = reference)
—Age in years
—Race and ethnicity (Non-Hispanic white = reference)
——Non-Hispanic black
——Non-Hispanic Asian
——Other non-Hispanic
——Hispanic
Delivered by Ingenta to :
University of North Dakota
Wed, 13 Aug 2008 17:35:46
—Educational attainment (high school/GED = reference)
——No high school diploma/GED
——College graduate
——Professional/graduate degree
——Younger than age 25, inapplicable
—Income relative to federal poverty line (more than 400% = reference)
——Under 125%
——125%–200%
——200%–400%
—Self-reported health (Excellent = reference)
——Very good
——Good
——Fair
——Poor
—Help with ADL (“no” = reference)
—Insurance status (Younger than 65, private insurance = reference)
——Younger than 65, public insurance
(continued on next page)
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
151
TABLE 3. (Continued)
Odds Ratios and t-statistics
Independent Variables
——Younger than 65, uninsured
Model 1
——65 or older, Medicare only
——65 or older, Medicare plus private
—Number of serious conditions
Language and language interaction variables
—80% or more of residents speak Spanish at home (“no” = reference)
—Interview conducted in Spanish (“no” = reference)
—Interaction: Interview language ⫻ neighborhood language
Model 2
Model 3 Model 4 Model 5
4.08
4.04
3.03
(18.14)** (17.85)** (9.83)**
.35
.35
.40
(7.38)** (7.37)** (3.93)**
.59
.59
.76
(3.36)** (3.37)** (1.21)
.76
.76
1.06
(5.71)** (5.67)** (1.40)
1.13
(.51)
1.71
(3.95)**
.54
(2.03)*
1.21
(.40)
.97
(.13)
.65
(.78)
Observations
22,656
22,656
22,656
22,656
22,656
* p < .05; ** p < .01 (two-tailed test)
Notes: Absolute value of t-statistics in parentheses. PCSA = Primary Care Service Area. MSA = Metropolitan
Statistical Area.
hood-level poverty has a large and significant ed information is transmitted through social
networks. To explore this possibility further,
odds ratio of 1.31.
In our third model, we add all individual- model 4 includes a dummy variable identifying
Delivered
to :
neighborhoods
in which at least 80 percent of
level variables to evaluate the extent
to whichby Ingenta
University
Dakota
residents
speak Spanish at home, and it
the association between neighborhood
residen-of North
Wed, care
13 Aug
17:35:46
includes
an interaction between this variable
tial instability and access to health
is 2008
attributable to differences in the composition and whether the MEPS interview was conductof individuals in neighborhoods. Odds ratios ed in Spanish. If information transmission
for the individual-level control variables through social networks is important in obtainshown in model 3 are frequently significant ing needed health care, then Spanish speakers
and are generally in the expected directions. It in predominantly Spanish-speaking neighboris noteworthy that the odds ratio for the preva- hoods should have, on average, better access
lence of poverty in a neighborhood is reduced than Spanish speakers in other neighborhoods.
markedly, from 1.31 to 1.10, suggesting that Conversely, non-Spanish speakers should have
much of the association between neighborhood worse access in neighborhoods in which
poverty and access to health care observed in Spanish is the primary language. The results
the previous models is compositional in nature, from model 4 are consistent with these expecrather than contextual. Supplemental analyses tations, with the odds ratios for both interview
suggest that the inclusion of individual-level language and the interaction term attaining stapoverty accounts for most of the reduction in tistical significance. To gauge the magnitude
odds ratio for neighborhood-level poverty. The of this result, we calculated mean predicted
odds ratio for neighborhood residential insta- probabilities for each group defined by the
bility, unlike that for neighborhood-level cross-level language interaction.3 The mean
poverty, is reduced only nominally, from 1.15 predicted probability of having access probto 1.13. This suggests that the association lems for Spanish speakers in neighborhoods
observed between neighborhood residential with a high prevalence of Spanish speakers
instability and access to health care is mostly was 22 percent, while that for Spanish speakcontextual in nature.
ers in other neighborhoods was 30 percent. For
We theorize that the association between individuals who completed the survey in
residential instability and access to health care English, the pattern is the opposite: 23 percent
exists in part because residential instability of those living in predominantly Spanishreduces the efficiency with which health-relat- speaking areas have problems with access,
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
152
JOURNAL OF HEALTH AND SOCIAL BEHAVIOR
while 21 percent of those living in other neighborhoods have access problems. While these
results are not a direct or thorough test of the
mechanisms that we propose in this paper, they
are consistent with our hypotheses.
Results using an alternative dependent variable are also consistent with our hypotheses. In
model 5, we re-estimate model 4 using a
dummy dependent variable that takes a value
of 1 for individuals who did not have a usual
source of care because they had just moved to
an area, did not know where to find a doctor,
or could not find a doctor who speaks their
language; the variable equals 0 for all others
(see column 5 of Table 3). As expected, residential instability is a stronger predictor in
these models than in our main analysis, though
the differences are modest. In model 4, for
example, neighborhood residential instability
has an odds ratio of 1.13, and the corresponding figure from model 5 is 1.17. While not
conclusive, these results are generally consistent with our network-based explanation for
the association between neighborhood residential instability and access to health care.
The results from this test are consistent with
our explanation. However, future research
should conduct more direct tests of the link. To
do this, data on the size, scope, and density of
social networks within neighborhoods are necessary, together with data on the extent to
which health-related information and support
are transmitted through such networks. Detailed information on neighborhood environments and how residents perceive and respond
to their environments would also be useful to
future research efforts.
Despite its limitations, this study expands
understanding of access to health care by identifying residential instability as a possible
determinant. Given that a major goal of U.S.
health policy is to reduce disparities in access
by race, ethnicity, income, and other characteristics (Institute of Medicine 2001; U.S.
Department of Health and Human Services
2000), and given that the United States is highly segregated by these same factors (Massey
1996), understanding how communities influence access is important. While previous
research links the prevalence of poverty and
the supply
Delivered by Ingenta
to : of health care in communities to
problems
with health care access, this study
University of North Dakota
SUMMARY AND CONCLUSIONS
indicates
that these are not the only important
Wed, 13 Aug 2008
17:35:46
community-level factors. If future research
In this study, we investigated the association confirms that the disruption of social networks
between neighborhood residential instability is one way through which residential instabiliand access to health care. We found that living ty influences access, community health care
in unstable neighborhoods is associated with providers and other community-based organian increased likelihood of having poor access zations should consider developing outreach
to care. This association is reduced but not programs that aid in the dissemination of inforeliminated when the prevalence of poverty in a mation regarding local health care resources.
neighborhood is held constant. Furthermore, Thus, policy makers may wish to develop prothe composition of individuals in a neighbor- grams to help organizations that serve residenhood with respect to various characteristics tially unstable neighborhoods accomplish this
including race, income, health insurance, edu- task.
cation, and health status explains little of the
association between residential instability and
access to health care.
NOTES
We speculate that residential instability is
associated with health care access at least in 1. These are items AC01, AC02, AC03, AC06,
part because residential instability disrupts the
and AC24 from the MEPS panel 4 and
development and flow of health-care-related
panel 5 questionnaires.
information through social networks. A limita- 2. For this variable, everyone in a given famition of our study is that the current data do not
ly takes the same value. This dependence
support direct tests of this proposed mechadoes not pose a problem for our analysis
nism. We conducted an indirect test of our
because we adjust our standard errors for
ideas by examining the interaction between
clustering in the sample. Please refer to the
language ability at the individual level and lan“Analytical Approach” section.
guage prevalence at the neighborhood level. 3. For each group, predicted probabilities were
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
calculated for all observations by fixing the
neighborhood- and individual-level language variables to values corresponding to
that group, leaving all other variables
unchanged. Means of the resulting predicted probabilities are then taken.
153
Center for Health Economics Research. 1993.
“Access to Health Care: Key Indicators for
Policy.” Center for Health Economics
Research, Chestnut Hill, MA.
Cohen, Joel. 1996. “The Medical Expenditure
Panel Survey: A National Health Information
Resource.” Inquiry 33:373–89.
———. 1997. “Design and Methods of the Medical Expenditures Panel Survey Household
REFERENCES
Component.” Agency for Health Care Policy
and Research, Rockville, MD.
Andersen, Ronald M., Hongjian Yu, Roberta Coleman, James C. 1988. “Social Capital in the
Wyn, Pamela L. Davidson, Richard E. Brown,
Creation of Human Capital.” American Jourand Stephanie Teleki. 2002. “Access to Mednal of Sociology 94(Supplement):95–120.
ical Care for Low-Income Persons: How Do ———. 1990. Foundations of Social Theory.
Communities Make a Difference?” Medical
Cambridge, MA: Harvard University Press.
Care Research and Review 59:384–411.
Collins, K. S., A. Hall, and C. Bebus. 1999. “U.S.
Auchincloss, Amy H., Joan F. Van Nostrand, and
Minority Health: A Chartbook.” CommonDonna Ronsaville. 2001. “Access to Health
wealth Fund, New York.
Care for Older Persons in the United States: Crutchfield, R., Michael Geerken, and Walter
Personal, Structural, and Neighborhood CharGove. 1982. “Crime Rates and Social Integraacteristics.” Journal of Aging and Health
tion: The Impact of Metropolitan Mobility.”
13:329–54.
Criminology 20:467–78.
Berk, Marc L., Claudia L. Shur, and Joel C. Can- Cunningham, Peter J., Joy M. Grossman, Robert
tor. 1995. “Ability to Obtain Health Care:
F. Peter, and Cara S. Lesser. 1999. “Managed
Recent Estimates from the Robert Wood JohnCare and Physicians’ Provision of Charity
son Foundation National Access to Care SurCare.” Journal of the American Medical Assovey.” Health Affairs 14:139–46.
ciation 281:1087–92.
Delivered
to : Peter J. and P. Kemper. 1998a. “The
Berkman, Lisa F., Thomas Glass, Ian
Brissette,by Ingenta
Cunningham,
University
Dakota Getting Care: Where You Live Matand Teresa E. Seeman. 2000. “From
Socialof North
Uninsured
13New
Aug 2008ters.”
17:35:46
Integration to Health: DurkheimWed,
in the
Center for Studying Health System
Millennium.” Social Science and Medicine
Change, Washington, DC.
51:843–57.
———. 1998b. “Ability to Obtain Medical Care
Billings, John, Geoffrey M. Anderson, and Laurie
for the Uninsured: How Much Does it Vary
S. Newman. 1996. “Recent Findings on Preacross Communities?” Journal of the Ameriventable Hospitalizations.” Health Affairs 15:
can Medical Association 280:921–27.
239–49.
Earp, Jo Anne, Eugenia Eng, Michael S. O’MalBindman, A. B., K. Grumbach, D. Osmond, M.
ley, Mary Alpeter, Garth Rauscher, Linda
Komaromy, K. Vranizan, N. Lurie, J. Billings,
Mayne, Holly F. Mathews, Kathy S. Lynch,
and A. Stewart. 1995. “Preventable Hospitaland Bahjat Qaqish. 2002. “Increasing Use of
izations and Access to Health Care.” Journal
Mammography among Older, Rural African
of the American Medical Association 274:
American Women: Results from a Community
305–11.
Trial.” American Journal Of Public Health
Brown, E. R., R. Wyn, and S. Teleki. 2000. “Dis92:646–54.
parities in Health Insurance and Access to Faris, Robert E. L. and H. Warren Dunham. 1939.
Care for Residents across U.S. Cities.” ComMental Disorders in Urban Areas. New York:
monwealth Fund and UCLA Center for Health
Hafner.
Policy Research, Los Angeles, CA.
Freidson, Eliot. 1970. Profession of Medicine.
Brown, Richard E., Pamela L. Davidson,
New York: Dodd, Mead.
Hongjian Yu, Roberta Wyn, Ronald M. Ander- Friedman, Bernard, Joanne Jee, Claudia Steiner,
sen, Lida Becerra, and Natasha Razack. 2004.
and Arlene Bierman. 1999. “Tracking the
“Effects of Community Factors on Access to
State Children’s Health Insurance Program
Ambulatory Care for Lower-Income Adults in
with Hospital Data: National Baselines, State
Large Urban Areas.” Inquiry 41:39–56.
Variations, and Some Cautions.” Medical Care
Carlisle, David M., R. Burciaga Valdez, Martin F.
Research and Review 56:440–55.
Shapiro, and Robert H. Brook. 1995. “Geo- Gaskin, Darrell J. and Catherine Hoffman. 2000.
graphic Variation in Rates of Selected Surgical
“Racial and Ethnic Differences in Preventable
Procedures within Los Angeles County.”
Hospitalizations across 10 States.” Medical
Care Research and Review 57:85–107.
Health Services Research 30:27–42.
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
154
JOURNAL OF HEALTH AND SOCIAL BEHAVIOR
Seek Help.” American Journal of Sociology
Goldstein, Harvey. 1999. Multilevel Statistical
97:1096–1138.
Models. London: Arnold Publishers.
Grumbach, Kevin, Karen Vranizan, and Andrew Roderick, P., S. Clements, S. Stone, N. Martin,
and D. Diamond. 1999. “What Determines
B. Bindman. 1997. “Physician Supply and
Geographic Variation in Rates of Acceptance
Access to Care in Urban Communities.”
onto Renal Replacement Therapy in EngHealth Affairs 16:71–89.
land?” Journal of Health Services Research
Hawe, Penelope and Alan Shiell. 2000. “Social
4:139–46.
Capital and Health Promotion: A Review.”
Ross, Catherine E., John R. Reynolds, and Karlyn
Social Science and Medicine 51:871–85.
J. Geis. 2000. “The Contingent Meaning of
Hirschi, T. 1969. Causes of Delinquency. BerkeNeighborhood Stability for Residents’ Psyley: University of California Press.
chological Well-Being.” American SociologiHorwitz, Allan. 1977. “Social Networks and Pathcal Review 65:581–97.
ways to Psychiatric Treatment.” Social Forces
Sampson, Robert J. 1985. “Neighborhood and
56:86–105.
Crime: The Structural Determinants of PerInstitute of Medicine. 2001. Crossing the Chasm:
sonal Victimization.” Journal of Research on
A New Health System for the Twenty-First
Crime and Delinquency 22:7–40.
Century. Washington, DC: National Academy
Sampson, Robert J. and W. Byron Groves. 1989.
Press.
“Community Structure and Crime: Testing
Kadushin, Charles. 1966. “The Friends and SupSocial-Disorganization Theory.” American
porters of Psychotherapy: On Social Circles in
Journal of Sociology 94:774–802.
Urban Life.” American Sociological Review
Sampson, Robert J., Jeffrey D. Morenoff, and Fel31:781–802.
ton Earls. 1999. “Beyond Social Capital: Spa———. 1969. Why People Go to Psychiatrists.
tial Dynamics of Collective Efficacy for ChilNew York: Atherton.
dren.” American Sociological Review
Kirby, James B. and Toshiko Kaneda. 2005.
64:633–60.
“Neighborhood Socioeconomic Disadvantage
and Access to Health Care.” Journal of Health Sampson, Robert J., Stephen W. Raudenbush, and
Felton Earls. 1997. “Neighborhoods and Vioand Social Behavior 46:15–31.
lent Crime:
Komaromy, M., N. Lure, D. Osmond,
DeliveredK.by Ingenta
to : A Multilevel Study of Collective
Efficacy.”
Vranizan, D. Keane, and A. B. Bindman.
1996.of North
University
Dakota Science 227:918–23.
Shaw,
Clifford R. and Henry D. McKay. 1942.
“Physician Practice Style and Rates
of 13
HospiWed,
Aug 2008
17:35:46
Juvenile Delinquency and Urban Areas.
talization for Chronic Medical Conditions.”
Chicago, IL: The University of Chicago Press.
Medical Care 34:594–609.
Levy, Paul S. and Stanley Lemeshow. 1999. Sam- Statacorp. 2001. Stata Statistical Software:
pling of Populations. New York: John Wiley
Release 7.0. College Station, TX: Stata Corpoand Sons.
ration.
Levy-Storms, Lene and Steven P. Wallace. 2003. Suchman, Edward A. 1964. “Sociomedical Varia“Use of Mammography Screening among
tion among Ethnic Groups.” American Journal
Older Samoan Women in Los Angeles County:
of Sociology 70:319–31.
A Diffusion Network Approach.” Social Sci- U.S. Department of Health and Human Services.
ence and Medicine 57:987–1000.
2000. Healthy People 2010: Understanding
Long, Stephen H. and M. Susan Marquis. 1999.
and Improving Health. 2nd ed. Washington,
DC: U.S. Government Printing Office.
“Geographic Variation in Physician Visits for
Uninsured Children.” Journal of the American Wells, B. L. and J. W. Horm. 1998. “Targeting the
Medical Association 281:2035–40.
Underserved for Breast and Cervical Cancer
Massey, Douglas. S. 1996. “The Age of
Screening: The Utility of Ecological Analysis
Extremes: Concentrated Affluence and PoverUsing the National Health Interview Survey.”
ty in the Twenty-First Century.” Demography
American Journal Of Public Health
33:395–412.
88:1484–89.
McKinlay, John B. 1973. “Social Networks, Lay Zuvekas, Samuel H. and Gregg S. Taliaferro.
Consultation, and Help-Seeking Behavior.”
2003. “Pathways to Access: Health Insurance,
Social Forces 51:275–92.
the Health Care Delivery System, and
Pescosolido, Bernice. 1992. “Beyond Rational
Racial/Ethnic Disparities, 1996–1999.” Health
Choice: The Social Dynamics of How People
Affairs 22:139.
James B. Kirby is a service fellow at the Agency for Healthcare Research and Quality in Rockville, MD.
His current research focuses on family and neighborhood influences on health, health behaviors, and access
to health care.
#2788—Jnl of Health and Social Behavior—Vol. 47:2—47204-kirby
RESIDENTIAL INSTABILITY AND ACCESS TO HEALTH CARE
155
Toshiko Kaneda is a policy analyst at the Population Reference Bureau in Washington, D.C. Her research
focuses on socioeconomic inequalities in health, mortality, and access to health care among older populations. She is currently involved in projects examining community-level socioeconomic effects on health and
mortality among older adults in mainland China.
Delivered by Ingenta to :
University of North Dakota
Wed, 13 Aug 2008 17:35:46