CRIMINOLOGY ARTICLE – [less tables, references and appendix]
Previous research frequently has observed a positive cross-sectional
relationship between
racial/ethnic minorities and crime and generally
has posited that this relationship is entirely because of the effect of
minorities on neighborhood crime rates. This study posits that at least
some of this relationship might be
a result of the opposite effect—
neighborhood crime increases the number of racial/ethnic minorities.
This study employs a
unique sample (the American Housing Survey
neighborhood sample) focusing on housing units nested in
microneighborhoods across three waves from 1985 to 1993. This format
allows one to test and find that such racial/ethnic transformation
occurs because of the following effects: First, White households that
perceive more crime in the neighborhood or that live in
microneighborhoods with more commonly perceived crime are more
likely to move out of such neighborhoods. Second, Whites are significantly
less likely to move into a housing unit in a microneighborhood
with more commonly perceived crime. And third, African American
and Latino households are more likely to move into such units.
One consistent finding of prior research is that neighborhoods and cities
with a greater proportion of racial/ethnic minorities tend to have higher
levels of crime (Krivo and Peterson, 1996; McNulty, 2001; Ouimet, 2000;
Roncek, 1981; Roncek and Maier, 1991).
Despite the almost exclusive
focus on cross-sectional data in these studies, researchers usually conclude
from this evidence that the presence of
racial/ethnic minorities leads to
more crime.
The reasons given for such a relationship are numerous—
from a culture of violence theory in which African Americans are posited
to be inherently more violent (Wolfgang and Ferracuti, 1967), to a structural
cultural explanation in which neighborhoods with high levels of poor
racial/ethnic minorities lack the economic resources and social institutions
to provide the social control that
otherwise would reduce the level of
crime (Sampson and Wilson, 1995),
to a structural explanation that economic
dislocation and unemployment in minority-dominant neighborhoods
leads to an increase in the number of broken households and to a
subsequent decrease in the ability to provide social control that otherwise
would reduce the amount of crime (Sampson, 1987). Nonetheless, a commonality
in such theories is
their assumption that the causal direction runs
from the presence of minority residents to more crime.
Given that these
are almost exclusively
cross-sectional studies, this assumption generally
neither is questioned nor can it easily be tested.
Fewer studies have asked whether in fact this process at least in part
might work in the opposite causal direction. That is, could higher levels of
crime in neighborhoods
cause an increase in the proportion of minority
residents residing there?
If in fact the causal direction, at least in part, is
reversed, then prior research employing cross-sectional data has overestimated
the size of the effect if it assumes that the relationship is entirely
because of the effect of racial/ethnic minorities on crime rates. The voluminous
segregation literature (Farley and Frey, 1994; Fischer et al., 2004;
Massey and Denton, 1987, 1993; Van Valey, Roof, and Wilcox, 1977)
as
well as the literature showing differential access to neighborhoods based
on race/ethnicity as a result of steering and discriminatory behavior (South
and Crowder, 1997a, 1997b; Turner et al., 2000)
has suggested a potential
mechanism through which crime in neighborhoods might change the
racial/ethnic composition of the neighborhood.
As will be elaborated in
more detail in this article, to the extent that households wish to avoid
neighborhoods with higher levels of crime,
and to the extent that racial/
ethnic minorities have constrained choices when selecting a neighborhood
in which to move,
neighborhoods with more crime might experience an
increase in racial/ethnic minorities throughout time.
It is not entirely novel to suggest that crime might affect the racial/ethnic
composition of a neighborhood,
as scholars occasionally have raised
this possibility (Bursik, 1986; Schuerman and Kobrin, 1986; Skogan, 1990;
Taylor, 1995). Nevertheless, few studies have addressed this question rigorously.
For instance, although two studies viewing the relationship
between crime rates and racial/ethnic composition of cities throughout
time are suggestive of such a relationship (Liska and Bellair, 1995; Liska,
Logan, and Bellair, 1998),
measuring this process at the level of the city is
too crude a measure to test precisely whether this neighborhood-level process
is present. Studies using neighborhood-level aggregated data are more
appropriate, although they cannot determine whether this process represents
disproportionate out-mobility, in-mobility, or both (Bursik, 1986;
Morenoff and Sampson, 1997).
Instead, I suggest that this question is multilevel
in which crime in a neighborhood might affect the household mobility
decisions of people living in the neighborhood as well as those
considering moving into the neighborhood.
This latter point raises an
important question; if crime indeed changes the racial/ethnic composition
of an area because of differential access to neighborhoods by race/ethnicity,
then does this neighborhood change occur because of differential ability
to leave the neighborhood
or because of differential likelihood of
entering the neighborhood?
The current study exploits a unique sample design to explore the following
key questions: 1) What effect does the common perception of crime in
a microneighborhood have on the relative likelihood of Whites and racial/
ethnic minorities
for moving out of a housing unit, and 2) what is the relative
likelihood of Whites and racial/ethnic minorities moving into a housing
unit in a
microneighborhood with a greater common perception of
crime? Thus, although this study is limited to measuring
microneighborhood crime based on the perceptions of residents, it has the
advantage of being able to drill down to the housing unit in viewing mobility
in and out of a unit. Furthermore, these perceptions of the crime context
are measured at the level of the local microneighborhood, which
provides advantages when studying these processes. Prior work aggregating
crime measures to larger units of analysis has not obtained such geographic
precision.
RESIDENTIAL TRANSITION
DOES RACIAL/ETHNIC COMPOSITION AFFECT THE CRIME
RATE?
Considerable prior scholarship has explored the cross-sectional relationship
between the
presence of racial/ethnic minorities and the amount of
crime in a city (Baumer et al., 1998; Chamlin and Cochran, 1997; Miethe,
Hughes, and McDowall, 1991)
as well as this relationship measured at the
geographic level of neighborhoods (Hipp, 2007a; Krivo and Peterson,
1996; McNulty, 2001; Ouimet, 2000; Roncek, 1981; Roncek and Maier,
1991).
These studies commonly assume that the presence of racial/ethnic
minorities in the particular geographic unit results in the higher levels of
crime and disorder.
One reason given for such a relationship in an early
body of research in the social disorganization tradition was that the presence
of minorities increased the level of racial/ethnic heterogeneity in a
neighborhood, which reduced the number of social contacts between
residents, leading to a subsequent reduced ability to provide the sort of
informal social control that otherwise would address problems of disorder
and crime (Shaw and McKay, 1942). More recently, researchers have
pointed out that the presence of minorities themselves are not necessarily
an indicator of heterogeneity, as neighborhoods that become populated
almost entirely by one racial/ethnic minority group are actually homogeneous
(Hipp, 2007a; Roncek and Maier, 1991; Sampson and Groves, 1989;
Warner and Rountree, 1997).
Other theories, however, have posited an even more direct effect from
the presence of racial/ethnic minorities to the occurrence of crime and disorder.
For instance, cultural theories such as the culture of violence perspective
posit that an African American culture exists that does not
sanction negatively violent behavior as strongly as does mainstream culture,
which results in more violent behavior on the part of residents (Wolfgang
and Ferracuti, 1967). Another perspective adopts a structural/cultural
approach in arguing that the culture in minority-dominated neighborhoods
is shaped by the larger structural system that brings about economic dislocation
in these neighborhoods
(Sampson and Wilson, 1995). That is, the
exodus of middle-class minority residents from these neighborhoods eliminates
positive role models espousing more conventional norms, and the
remaining low-income minority residents develop a nonnormative culture
because of this structural imposition. Such neighborhoods, therefore, are
populated with the truly disadvantaged, who lack the economic resources
and time to support the neighborhood’s institutions and who do not provide
the type of role models that would increase neighborhood youths’
desire to embrace
middle-class values (Wilson, 1987). This isolation causes
these neighborhoods to develop norms that accept the use of violence and
crime.
EXITING HIGH-CRIME NEIGHBORHOODS: DIFFERENTIAL ABILITY?
Despite this dominant paradigm assuming that it is the presence of
racial/ethnic minorities that brings about more crime, a small but growing
literature has suggested that at least some of this relationship might occur
because
minority residents are pushed into such neighborhoods. To understand
this perspective, first consider the possibility that crime in a neighborhood
might induce residential mobility.
Of course, this argument
contradicts social disorganization theory’s postulate that residential instability
leads to more crime.
The notion that neighborhood crime might induce residential mobility is
not new, as scholars occasionally have suggested this possibility in recent
years (Bursik, 1986; Schuerman and Kobrin, 1986; South and Messner,
2000; Xie and McDowall, 2008).
Although this suggestion is hardly controversial—
clearly, most households wish to avoid neighborhoods with
higher levels of
crime—fewer studies have tested this possibility empirically.
Nonetheless, the few studies exploring this question have provided
some supportive evidence. For instance, a study using the National Crime
Victimization Survey found that experiencing a crime event increased the
likelihood of exiting a neighborhood (Dugan, 1999). A recent study with
the same data
found that victimization experienced by one’s nearest four
neighbors also increased residential mobility (Xie and McDowall, 2008).
Skogan (1990) found for 40 neighborhoods that crime rates caused dissatisfaction
and a desire to move.
Likewise, a study of census tracts in Los
Angeles found that tracts with higher levels of violent or property crime in
1 year experienced a larger volume of home sales the following year
(Hipp, Tita, and Greenbaum, 2009).
Studies using data aggregated to cities
have found that
higher levels of crime led to a greater population loss
throughout time (Cullen and Levitt, 1996; Sampson and Wooldredge,
1986) and that higher rates of violent crime in central cities relative to
suburbs
spurred city-to-suburb mobility and inhibited suburb-to-city
moves (South and Crowder, 1997b).
Although it is clear why households might want to leave a neighborhood
with
high levels of crime, it is less clear why another household would be
willing to enter such a neighborhood.
One plausible explanation is that no
households in fact wish to enter such neighborhoods but rather that such
movement is driven by an economic process.1 That is, to the extent that
low-crime neighborhoods are more desirable, they will have higher rents
and higher home values.
As a consequence, only households with the
greatest economic resources
will be able to reside in such neighborhoods.
However, an increase in crime will decrease the desirability of the neighborhood,
pushing down rents and home values. Indeed, cross-sectional
studies have found that neighborhoods with higher rates of crime have
lower home values (e.g., Buck and Hakim, 1989; Schwartz, Susin, and
1. Another possible explanation is information asymmetry. That is, current
residents of the neighborhood likely have a better sense of how the rate of crime
in a neighborhood is changing compared with potential new residents. In such an
instance, a household might be willing to move into the neighborhood simply
because its inhabitants are unaware that the level of crime is higher than they
expected. Some evidence supports this theory; households who have lived longer
in the neighborhood perceive more crime (Hipp, 2010; Sampson, Raudenbush,
and Earls, 1997) and more risk of crime (Taylor, Gottfredson, and Brower, 1984).
Another study found that residents who lived longer in the neighborhood
expressed more fear of walking in their local block at night and more fear of
walking in the broader neighborhood during both the day and the night (Taylor,
2001). This possibility has little effect on the posited model here, except that it
introduces a stochastic element into the in-mobility decisions of households. As a
consequence, it might slow the racial/ethnic transition of a neighborhood but not
reverse or stop it. Most likely, this trend would be manifested in a greater likelihood
of White residents abandoning a neighborhood but no difference in their
likelihood of entering the neighborhood (assuming complete asymmetry in information
regarding neighborhood crime). The models presented later can assess
the extent to which this situation is actually the case.
Voicu, 2003; Thaler, 1978). Evidence also suggests that increasing crime in
the neighborhood decreases home values (Tita, Petras, and Greenbaum,
2006) and that tracts with higher rates of crime in 1 year experience a
relative decrease in home values the following year (Hipp, Tita, and
Greenbaum, 2009).
This economic argument has implications for racial/ethnic minorities in
neighborhoods. To the extent that such minorities have fewer economic
resources in general,
they will be disproportionately unlikely to leave such
neighborhoods (Massey and Denton, 1985).
This process implies that
throughout time, a high-crime neighborhood not only will increase the
number of low-income residents, but it also will increase the number of
low-income racial/ethnic minorities. As a consequence, high-income White
residents disproportionately will abandon the neighborhood. This trend
suggests that a change in the neighborhood’s racial/ethnic composition will
result from the amount of crime and disorder. Of course, this argument
implies that no difference should be observed in the race/ethnicity of those
who leave in response to higher crime
if the economic resources of neighborhood
residents
are taken into account.
Nonetheless, the legacy of segregation in the United States implies a
possible explanation for why racial/ethnic minorities might have a constrained
ability to leave an undesirable neighborhood beyond their limited
economic resources and underlies place stratification theory. The highly
segregated nature of racial/ethnic minority communities is well documented
(Frey and Farley, 1996; Massey, Gross, and Shibuya, 1994; Massey
and Hajnal, 1995),
and the greater tendency for racial/ethnic minorities to
enter neighborhoods dominated by members of their same race/ethnicity
also is established firmly (Logan, Alba, and Leung, 1996; Massey and Mullan,
1984; Rosenbaum, 1994; Rosenbaum and Argeros, 2005; South and
Crowder, 1997b).
Thus, place stratification theory posits that such mobility
constraints limit the neighborhoods that racial/ethnic minorities can enter.
Given that the number of neighborhoods dominated by minorities is far
smaller than those dominated by Whites, the implication is that racial/ethnic
minorities face considerable constraints when choosing where to move.
Indeed, evidence suggests that although racial/ethnic minorities express an
equal desire to leave neighborhoods as Whites (Lee, Oropesa, and Kanan,
1994),
in fact, they are less likely to do so (Boehm, Herzog, and
Schlottmann, 1991; Deane, 1990).
What might be the mechanisms explaining such a constrained choice for
racial/ethnic minorities? No shortage of possible explanations are available
as shown in the voluminous literature describing the role of gatekeepers,
steering, and discriminatory behavior. Studies have shown that
gatekeepers (such as real estate agents)
are an important source of segregation
as they often
present racial/ethnic minority homebuyers with a
more limited number of neighborhoods (La Gory and Pipkin, 1981; Turner
et al., 2000).
In these instances, this steering pushes racial/ethnic minorities
toward neighborhoods already highly populated with fellow group members
and away from White-dominated neighborhoods. Additionally, a
large literature shows discriminatory behavior on the part of potential
landlords and property management companies.
For instance, audit studies
have shown consistently that racial/ethnic minorities often are turned
down from housing options despite identical credentials to White candidates
(Turner et al., 2000).
One study even found such evidence when conducting
over-the-phone audits; in this instance, speaking in a Black
vernacular yielded fewer offered residences (Fischer and Massey, 2004).
As a consequence, racial/ethnic minorities likely have fewer options when
it comes to choosing a neighborhood in which to reside.
Because of these limited mobility options, a racial/ethnic minority
household living in an area with increasing levels of crime might be unable
to leave the neighborhood. That is, a White household might respond to
increasing crime rates by abandoning the neighborhood. However, a
racial/ethnic minority household also might wish to leave and proceed to
engage in a search for an alternative neighborhood. If this search is constrained
by the aforementioned mechanisms, then they will be less likely
to find a suitable alternative.
As a consequence, the household might be
less likely to leave given that the alternatives seem no better and not
because the household members are not concerned about the changes in
the neighborhood.
Despite the plausibility of these hypotheses, limited empirical evidence
is available that tests whether Whites have a differential ability to leave
high-crime neighborhoods compared with racial/ethnic minorities. Two
studies employed city-level longitudinal data to test and found that higher
levels of crime resulted in a greater concentration of
non-White populations
in cities (Liska and Bellair, 1995; Liska, Logan, and Bellair, 1998).
Another study measuring crime rates based on city-level crime rates actually
found that African Americans in the central city were more likely
than Whites to move to a different tract in response to a higher ratio of
city-to-suburb violent crime (South and Crowder, 1997b). Despite the
importance of these studies, these results only can be suggestive given that
the high level of aggregation precludes testing these processes at the more
appropriate geographic level of housing units nested within local neighborhoods.
One longitudinal study of Chicago census tracts found that the
rate of homicide in the census tract led to a general population loss of both
Whites and African Americans (Morenoff and Sampson, 1997),
which
does not support the notion of disproportionate mobility, at least when
measured aggregated to census tracts.
Another study of neighborhoods in
Chicago did find that the delinquency rate in 1960 increased the number of
non-Whites in 1970 (Bursik, 1986). This evidence is suggestive, although
the focus on a single city 40 years ago, along with the small number of
control variables, leaves open the question of whether such a process
might be observed with more recent data focusing on transition in the
actual housing unit.
WHO IS MOVING IN?
The previous discussion implies a second consideration; not only might
racial/ethnic minorities be less likely to leave a high-crime neighborhood,
but they also might be more likely to enter a high-crime neighborhood.
Similar to the process described earlier regarding residential mobility out
of the neighborhood,
if racial/ethnic minorities have a constrained choice
on where to move, then they might be more likely to move into a neighborhood
with higher levels of crime. That is, the household will have fewer
neighborhoods to choose from when deciding on a new residence and, as a
consequence, might be forced to consider neighborhoods with more crime.
Crime would not necessarily be desirable, of course, but such households
simply might have few other options in neighborhoods to consider. This
trend implies that racial/ethnic minorities will be more likely than Whites
to move into neighborhoods with more crime.
It is an empirical question
whether either, or both, of these processes
regarding out-mobility and in-movement are at work. Both can affect the
neighborhood’s racial/ethnic composition. For instance, a neighborhood
would experience a racial/ethnic transition if racial/ethnic minorities are
less likely to leave a
high-crime neighborhood even though the racial/ethnic
composition of the entering
households does not change. It can be
shown easily that throughout time, this pattern will lead to a change in the
racial/ethnic composition of
the neighborhood. Alternatively, a neighborhood
also would experience a racial/ethnic transition if racial/ethnic
minorities are no more likely to leave a high-crime neighborhood than
Whites but racial/ethnic minorities are more likely to enter the neighborhood.
A third possibility is that both of these processes are at work—
Whites are more likely to leave a high-crime neighborhood and racial/ethnic
minorities are more likely to enter it—which would lead to the most
rapid transformation of the neighborhood’s racial/ethnic composition.
It is instructive to note that approximately 30 years ago, Aldrich and
colleagues (Aldrich and Reiss, 1976; Aldrich, Zimmer, and McEvoy, 1989)
studied an analogous question when asking whether a transformation of
the racial/ethnic composition of a neighborhood’s residents leads to a
change in the racial/ethnic composition of the business owners. Studies
from both the United States (Aldrich, Zimmer, and McEvoy, 1989) and
the United Kingdom (Aldrich and Reiss, 1976) found that although White
business owners were no more likely to abandon such a neighborhood,
White business owners were far less likely to enter such neighborhoods.
Thus, the racial/ethnic composition of the business owners in such neighborhoods
changed across time because of this differential likelihood of
entering the neighborhood.
Such a finding is important for understanding
the process through which such change occurs and allows more appropriate
policy interventions.
Likewise, in the present study, although an important first question is
whether a neighborhood with higher levels of crime is more likely to transition
to a greater proportion of racial/ethnic minorities, an equally crucial
second question centers on how this transition takes place. If it occurs
because Whites are more likely to abandon such neighborhoods, then that
will imply a need to focus on the differential ability of racial/ethnic minorities
to leave such neighborhoods. If it occurs because racial/ethnic minorities
are more likely to enter such neighborhoods, then that will imply a
different set of policy choices in response.
Nevertheless, little information is available on whether racial/ethnic
minorities are disproportionately likely to move into or out of higher
crime neighborhoods. Studies finding that higher crime neighborhoods
have more racial/ethnic minorities throughout time (Bursik, 1986) cannot
determine whether this occurs because of an increased likelihood of
Whites abandoning such neighborhoods or because of an increased likelihood
of racial/ethnic minorities entering such neighborhoods. Testing for
in-mobility requires a sample that follows housing units across time—as
employed here—rather than households.
SUMMARY
The present study asks whether higher levels of
commonly perceived
crime lead to a transformation in the racial/ethnic composition of the
neighborhood. This study provides
the following important contributions
to the literature: 1) By focusing on housing units, it
avoids the challenges
inherent in aggregating residential mobility measures to larger geographic
units such as census tracts, given that such change must occur at the level
of the housing unit; 2)
by using a study design that follows housing units
throughout time, it allows for the assessment of the degree to which any
change occurs
as a result of the disproportionate likelihood of Whites
leaving such neighborhoods or the disproportionate likelihood of racial/
ethnic minorities entering such neighborhoods;
and 3) by measuring crime
based on
the perceptions of residents in the immediate area (as measured
by the 11 closest housing units), it provides a more geographically precise
estimate of the crime environment experienced by the housing unit rather
than crime measures aggregated to the level of a census tract or even
larger geographic unit. Furthermore, I can assess the extent to which a
household’s own perception of neighborhood crime translates into
mobility.
DATA AND METHODOLOGY
DATA
The subsample of the American Housing Survey (AHS) employed here
is suited uniquely to address these research questions. This special neighborhood
subsample of the AHS initially randomly selected 680 housing
units in 1985 from the full AHS that were located in either urban or suburban
locations and then interviewed the ten closest neighbors of the initial
respondent.2 In what follows, I refer to these 11 households as a
“microneighborhood.” This unique data set thus has housing units nested
within microneighborhoods as the units of analysis, allowing testing of
whether the structural characteristics of the local microneighborhood
affect residential housing transition. Only microneighborhoods with at
least five respondents were included; thus, 663 microneighborhoods were
present in 1985. Three waves of data follow these housing units throughout
time. At each wave of data collection, interviewers returned to the exact
same housing units; thus, no attrition is in this sample design but only possible
nonresponse (as discussed subsequently). Furthermore, to account
for new housing developments, the samples in 1989 and 1993 were augmented
with new microneighborhoods. Thus, information from 1985 is
used to predict the likelihood of the household leaving by 1989 and the
new household’s characteristics in 1989, and information from 1989 is used
to predict the likelihood of
the household leaving by 1993 and the new
household’s characteristics in 1993. As a consequence, a sample of 6,865
units is included for the in-mobility models (units that changed residents
during one of these two periods), and samples of 1,172 Latino residents,
2. In the AHS, sample units were selected from the 1980 Census Sample Housing
Unit Record File. A housing-unit coverage study was performed to locate units
missed by the 1980 U.S. Census, and an additional sample was selected from the
units located by this study (such as nonresidential to residential units, new mobile
home parks, etc.). Building permits also were sampled to represent newly constructed
housing since the 1980 Census and were the source of the new
microneighborhoods added in the 1989 and 1993 waves. To construct the frame
for building permits, clusters of four new construction units were formed using
information from sample building permit offices; one construction unit was subsampled
from each cluster. They used the 1980 characteristics of the units in
these enumeration districts as stratifiers based on 1) geographic location (central
city, urbanized area outside of central city, urban outside of urbanized area, and
rural), 2) tenure, 3) number of rooms, and 4) value of unit or gross rent. (For a
more complete description of the AHS sampling design, see Aldrich and Reiss,
1976; Hadden and Leger, 1995.)
2,032 African American residents, and 11,590 White residents are included
in the two waves in the residential mobility models.3
OUTCOME MEASURES
One outcome measure is an indicator of whether a new household lived
in the unit 4 years later (showing outward residential mobility). A new
household is defined as one with a new household head; it is not considered
mobility if only submembers of the household leave between the two
waves.4 The other three outcome measures are the race/ethnicity of the
household head of the new household residing in a residence
4 years later.
These measures indicate whether the household is White, African American,
or Latino.
In the relatively rare cases in which a household did not
participate in a survey, these missing data are treated using a multiple
imputation strategy, which will be described subsequently.
CRIME IN THE ENVIRONMENT
I measure crime in the local environment by combining the reported
perceptions of the 11 households in the microneighborhood. This strategy,
therefore, is measuring the common perception of crime of these residents.
The AHS asks respondents a series of three questions about crime in the
neighborhood (as defined by the respondent); is crime a problem, is it so
much of a problem that it is a bother, and is it such a bother that the
respondent wishes to move. These questions are nested, as a respondent
only was asked the latter questions if they answered the previous questions
affirmatively. These responses were combined into a 4-point
response in which the respondent either replies “no” to all questions,
replies “yes” to one, “yes” to two, or “yes” to all three. I accounted for
likely measurement error in these responses by estimating this as a
microneighborhood-level latent variable of common perception of crime,
as described in the Methods section. Given the theorizing and evidence
from prior research that residents’ perceptions of crime or disorder might
be just as important for affecting behavior as more objective measures
(Deane, 1990; Sampson, 2009; Sampson and Raudenbush, 2004), this
3. The number of households in these microneighborhoods can vary. Across these
three waves, just .7 percent of the microneighborhoods have between 5 and 7
households, 5.1 percent have between 8 and 10, 78.5 percent have 11, 8.2 percent
have 12, 4.2 percent have 13, .8 percent have 14, 2.1 percent have between 15 and
19, and .4 percent have 20 or 21.
4. I can assess the degree to which mobility varies across residents within the same
microneighborhood compared with across microneighborhoods. I find that about
25 percent of the variance occurs across microneighborhoods, whereas the other
75 percent occurs within. This finding suggests a fair amount of clustering of
movers within microneighborhoods.
household-level perception of crime measure also was included in the residential
mobility models to test whether it affects mobility beyond this
microneighborhood-level effect.5
Of course, measuring crime is always a challenging task, and no perfect
measure exists of the construct. Although studies often measure crime
based on official crime reports, the na¨ıve assumption that official crime
rates are infallible has given way to the acknowledgment that a considerable
amount of underreporting occurs for such official statistics. For
instance, the 2005 National Crime Victimization Survey reported that only
62 percent of aggravated assaults, 60 percent of robberies, and 56 percent
of burglaries were reported to the police (Klaus and Maston, 2006). A
second approach, victimization studies, has desirable properties; however,
these studies require large samples to get reasonably useful estimates of
crime in small geographic areas. Furthermore, they do not capture crimes
experienced by victims not living in the neighborhood. A third approach—
measuring crime based on the reported perceptions of residents—is potentially
fallible as well given the possible systematic biases of certain types of
households when assessing this issue.
Despite the imperfections of different measures of crime, some evidence
suggests that these measures might not be as bad as some have assumed.
For instance, one study found that perceptions of crime can map on reasonably
to official crime rates (Perkins, Meeks, and Taylor, 1992). Another
exhaustive study used seven waves of data throughout a 25-year period
and found a relatively high correlation (approximately .70) between the
official violent crime rate in a census tract and residents’ combined perceptions
of crime (Hipp, 2007b). It should be noted that even this might be
an underestimate given the aggregation to census tracts, which might
experience heterogeneity in the amount of crime across the blocks within a
tract. Another study demonstrated criterion validity for three measures of
crime, finding that the structural measures in their model had very similar
effects regardless of whether they measured the outcome of crime by the
official crime rate, victimization reports, or the common perception of
crime (Sampson, Raudenbush, and Earls, 1997). Furthermore, given that I
am combining the reports of 11 households living adjacent to one another
5. No evidence was found that including this individual-level measure along with
the microneighborhood-level measure of common perception of crime caused
estimation difficulties. The correlation between the two measures was .55, which
is not excessively high. Furthermore, estimating a model without this individuallevel
measure showed a similar effect for the common perception of crime measure
(with a slightly larger coefficient) and a relatively similar standard error.
Thus, little evidence was found that including this individual-level measure introduced
multicollinearity problems.
rather than combining the reports of residents scattered throughout a census
tract, I argue that this estimate of the amount of crime is more geographically
precise. Studies using measures of crime aggregated to census
tracts implicitly make the strong assumption that the rate of crime is constant
across all blocks in a census tract.
I also point out that for the analyses viewing the characteristics of
residents moving into the neighborhood, that it is residents who lived in
the neighborhood 4 years previously who are assessing the level of crime.
The new residents’ perceptions of crime are immaterial to this measure
and thus cannot bias the results. Although the prior household’s inaccurate
perception of more crime in the neighborhood might increase the
likelihood that they will leave (and indeed this possibility explicitly is
accounted for in the models predicting the likelihood of exiting such housing
units), little reason exists to suspect that such a household will be
replaced systematically by a household of any given race/ethnicity.
It is possible that households might use the racial/ethnic composition of
the microneighborhood as a proxy for the amount of crime. For instance,
Krysan (2002) suggested that Whites often find mixed-race neighborhoods
undesirable because of an inflated assessment of the amount of crime in
such neighborhoods. However, although evidence suggests that White
residents who perceive a more racially mixed environment are more likely
to fear crime (Chiricos, Hogan, and Gertz, 1997; Rountree, 1998), studies
often have found no relationship between the actual racial/ethnic composition
of the environment and Whites’ perceptions of crime (Chiricos,
Hogan, and Gertz, 1997; Rountree, 1998). For instance, one study of
residents in three cities found that White residents perceived more crime
than African Americans in census tracts with more young African American
males in Seattle, but no such effect was detected in the samples of
Chicago and Baltimore (Quillian and Pager, 2001). However, the fact that
this same study found that all residents (both Black and White) perceived
more crime in census tracts with more young Black males suggests that
perceptions in fact might be impacted by the racial/ethnic composition. Of
course, given that studies have shown consistently that racial/ethnic heterogeneity
is associated with higher levels of official rates of crime (Hipp,
2007a; Roncek and Maier, 1991; Sampson and Groves, 1989; Warner and
Rountree, 1997), it is unclear whether residents’ reports of more crime
when living in racially/ethnically mixed neighborhoods are misperceptions
or are simply reflecting actual higher levels of crime. Given this possible
bias, it is important to account for the racial/ethnic composition in the
models.
HOUSEHOLD AND MICRONEIGHBORHOOD-LEVEL PREDICTORS
In the models predicting the race/ethnicity of the new residents, the
race/ethnicity of the household at the previous time point was taken into
account by creating indicators of whether the household was White, African
American, Latino, or other race.
Given that the racial/ethnic composition
of the
microneighborhood affects its desirability and hence mobility
behavior (Charles, 2000), measures of the percent White, African American,
Latino,
and other race were created.
For the residential mobility model, I also included several householdlevel
measures and individual-level measures (based on the characteristics
of the household head) that are likely important predictors of
mobility. A
measure of the age of the respondent, measures of the number of children
aged
0–5 years, aged 6–12 years, or aged 13–18 years in the home, and
dichotomous indicators for marital status (married, divorced, widowed,
with single as the reference category) account for stage of the life course. I
measured community investment with an indicator of whether the respondent
owned their residence and a measure of the length of time in the
residence (log transformed). To account for mismatch with the housing
unit,
a measure of the persons per room (log transformed) was created to
capture overcrowding. Socioeconomic status effects were captured with
measures of household income (logged) and
years of education of the
respondent.
In ancillary models, I tested whether the household-level measures in
the residential mobility model explained who moved into the unit at the
next time point. As expected, nearly all these measures were insignificant.
The one exception was the measure of length of residence of the previous
household. Given that it had a significant effect in certain models, and the
fact that it is plausible to suppose that long-term residents indeed might be
different in whom they transition the unit to, I left this measure in the inmovement
models.
I also took into account several characteristics of the microneighborhood.
The average household income in the microneighborhood accounts
for economic resources. Residential stability is measured by the percentage
of new households in the last 5 years, the percentage of vacant units,
and the percentage of homeowners in the microneighborhood. Given that
crowding might affect who enters a neighborhood, a measure of the percent
of households living in crowded conditions (more than one person
per room) was constructed. Finally, a measure of the percent households
with children and an indicator of the wave of the survey were constructed.
The summary statistics for the variables used in the analyses are shown in
table 1.
I tested for and found no evidence of collinearity problems in the
estimated models (all variance inflation values were below 4).
METHODOLOGY
Simply summing the perception of crime of residents in the
microneighborhood would ignore the certain measurement error in these
responses. I instead adopted an approach that explicitly accounts for this
measurement error by estimating this as a latent variable of commonly
perceived crime (for a detailed description of this approach, see Ludtke et
al., 2008). This equation is expressed as follows:
xik = L1xk + eik (1)
where xik is the combined 4-point response in the AHS regarding the level
of crime reported by the ith respondent of I respondents in the kth
microneighborhood, xk is the latent variable of common perception of
crime in the microneighborhood, L1 measures the impact of perceived
crime on the respondent’s report of crime (because the ordering of
respondents in neighborhoods is random, these ls are constrained equal),
and eik is a disturbance term (the variances of the es are constrained
equal).6 Such an approach was adopted by Bollen and colleagues in different
substantive contexts (Bollen and Paxton, 1998; Speizer and Bollen,
2000), as well as by Sampson, Raudenbush, and Earls (1997), who used an
item response theory (IRT) approach (which is identical to the approach
here).7
To account for possible different mobility processes by race/ethnicity, I
estimated the household residential mobility model as a multiple groups
analysis in which the following equation is estimated separately for
Whites, Latinos, and African Americans:
yik(t+1) = ßxkxk(t) + GXikXik(t) + GXkXk(t) + GYRYR (2)
6. The average R-square for the household-level perceived crime measures is .23,
suggesting that approximately 23 percent of these responses are accounted for by
this neighborhood-level measure of commonly perceived crime, whereas the
other 77 percent is measurement error. This result emphasizes the importance of
explicitly taking into account this measurement error. The latent measure has a
reliability of .75, which is explicitly taken into account in the estimation strategy.
7. A large literature shows that hierarchical linear models (HLMs) and structural
equation models (SEMs) will provide identical results for several types of models.
Numerous studies have shown that HLM and SEM will yield identical estimates
for trajectory models (Chou, Bentler, and Pentz, 1998; Guo and Hipp,
2004; MacCallum et al., 1997; Mehta and West, 2000; Raudenbush, 2001). Recent
work has shown the two techniques will yield identical results even when the data
are unbalanced across level 2 units (Bauer, 2003; Lee and Tsang, 1999). Two nice
didactic articles showing how HLM and SEM can be used to obtain identical
results are Bauer (2003) and Mehta and Neale (2005). It is also important to
point out that although some scholars might assume that using an IRT approach
to create scales would be preferable to the SEM approach, in fact an article
shows the exact mathematical relationship between SEM and IRT model parameters
(Kamata and Bauer, 2008).
where yik(t+1) is the probability that the household will move of the ith
respondent of I respondents in the kth microneighborhood; xk(t) is the
latent variable of neighborhood crime, which has ßxk effect on the outcome
(the k-subscript makes explicit that this value is measured at the
microneighborhood level); Xik(t) is a matrix of independent variables with
values for each household i in microneighborhood k at time t (the i- and ksubscripts
make explicit that these values are measured at the individual
or household level); GXik shows the effect of these predictors on the
probability of moving;
Xk(t) is a matrix of microneighborhood-level independent
variables for microneighborhood k at time t; GXk shows the effect
of these predictors on the probability of moving;
and YR indicates which
year of the sample the observation comes from and has a GYR effect on the
outcome. To account for the ordinal nature of the crime assessments by
households (xik) and the dichotomous outcome of moving (yik), I created a
polychoric correlation matrix and estimated the model using a diagonally
weighted least-squares estimator in the Mplus 4.1 software (Muth´en &
Muth´en, Los Angeles, CA). This approach assumes that these ordinal
measures have unobserved continuous measures underlying them that are
distributed normally. This strategy is analogous to estimating simultaneously
each of these outcomes as probit or ordered probit equations. The
clustering in the data was accounted for by correcting the standard errors
using robust standard errors corrected for microneighborhood-level
clustering.
For the models testing the characteristics of the household moving into
the unit, the equation predicting that the new household is, for instance,
White is
expressed as follows:
yik(t+1) = ßyik(t) + ßxkxk(t) + GXikXik(t) + GXkXk(t) + GYRYR + eik(t) (3)
where yik(t+1) is an indicator of whether the new household in the unit is
White of the ith respondent of I respondents who are new in the kth
microneighborhood; yik(t) is a vector showing the race/ethnicity of the prior
residents
in the unit and has a vector of ß effects on the outcome; xk(t) is
the latent variable of neighborhood crime that has a ßxk effect on the outcome;
Xik(t) is a matrix of independent variables with values for each individual
i in microneighborhood k; GXik shows the effect of these predictors
on the outcome; Xk(t) is a matrix of microneighborhood-level independent
variables for microneighborhood k; GXk shows the effect of these
predictors on the outcome; and YR indicates which year of the sample the
observation comes from and has a GYR effect on the outcome. For maximum
efficiency, these three outcomes were estimated simultaneously in
Mplus 4.1 using a diagonally weighted least-squares estimator. To allow
inferences to the entire sample,
I accounted for nonmovers by estimating a
selection model in which the outcome is whether
the household moved
during the 4-year period. This probit selection model was estimated, and
the inverse Mills ratio from this selection model was included in the final
models.8 Missing data were addressed with a multiple imputation strategy
(Rubin,
1987).9 This approach requires the less stringent assumption of
missing at random
rather than the missing completely at random assumption
of listwise deletion. Five data sets were imputed, and the results were
combined with appropriate standard errors using the standard formulas
that account for the variability both within imputed data sets and across
data sets (Rubin, 1987; Schafer, 1997).10
RESULTS
RESIDENTIAL MOBILITY
I begin by focusing on the models predicting movement out of the
neighborhood.11 The model was estimated separately for White, African
8. This selection model included the measures for the residential mobility model
shown in table 2. Given that these household-level characteristics predict residential
mobility (the selection process) but have little effect on the type of households
that move into the neighborhood, the level of multicollinearity is reduced
in the final model relative to other implementations of such a selection model. I
nonetheless performed sensitivity analyses by estimating the models without the
inverse Mills ratio. The results were extremely similar to those in the models
presented. Thus, the results are essentially the same whether or not selection
effects are taken into account.
9. I used the Proc MI procedure in SAS (SAS Institute, Inc., Cary, NC) to perform
the imputations. Only information from the current wave was included when
imputing values (given that the household in other waves could be a different
one). The imputation model included all variables contained in the substantive
models as well as several other possibly important measures to get more precise
estimates of the missing values.
All imputed values were constrained to fall
within the range of values in the original measure, and
values were not rounded
to integers given Monte Carlo simulation evidence that such an approach has
poor properties (Allison, 2005).
10. Only modest amounts of data were missing. For example, among housing units
that responded to the survey, less than 1 percent of data were missing for any
variable. For the neighborhood common perception of crime measure, 6.2 percent
of data were missing for the indicators of this latent variable (as missing data
for this measure occurs in instances in which a household does not respond to the
survey, or the housing unit was empty). See appendix A for an explanation of the
data structure.
11. Given that these models are estimated within the SEM framework rather than
the HLM, the overall fit of the model can be assessed. However, given that I am
estimating parallel logit models, overidentification in the model only occurs
because of the latent variable for the microneighborhood common perception of
crime. I therefore assessed the fit of the confirmatory factor analysis (CFA) of
the common perception of crime and found an excellent fit for the overall sample:
a root-mean-square error of approximation (RMSEA) of .018 and values of
.99 for the Tucker Lewis Index (TLI) and the Comparative Fit Index (CFI) (values
less than .05 for the RMSEA, and greater than .90 for the other measures
American, and Latino households. There is a differential likelihood by
race/ethnicity of leaving a housing unit in an area with a higher common
perception of crime. Model 1 in table 2 shows strong evidence that White
households are more likely to
leave a neighborhood when they perceive
more crime, and a contextual effect also is noted in which they leave when
there is a greater common perception of crime. A White household that
perceives that crime is a problem is .045 units more likely to leave the
neighborhood than
one that does not perceive a problem, and a 1-unit
increase in commonly perceived crime in the local microneighborhood further
increases the probability of moving .111 units. Thus, perceiving 1 unit
more crime increases the predicted probability of moving by 4 percent for
Whites (for a household at the mean on all other characteristics), and a 1
standard deviation increase in the microneighborhood common perception
of crime increases the predicted probability of moving by 5 percent.12
However, no evidence in models 2 and 3 in table 2 is found that African
Americans or Latinos are more likely to move out of a
microneighborhood if they perceive more crime, or if a higher level of
commonly perceived crime persists. I simply find no evidence that members
of these two racial/ethnic minority groups are more likely to abandon
a housing unit if they perceive more crime or if the microneighborhood
has higher reported levels of commonly perceived crime. Although evidence
indicates that Whites are more likely to leave a microneighborhood
in response to higher levels of commonly perceived crime and a lack of
evidence to indicate that Latinos and African Americans do so, the models
lack the statistical power to distinguish between these different group
effects. That is, although the model has the statistical power to determine
that the effect of crime on White mobility is significantly different than
zero, it lacks the power to conclude that the difference in the effect of
crime on White mobility significantly differs from that on African American
or Latino mobility.13 Note that statistical power alone is not the reason
for this different conclusion for Whites, as the estimated coefficients
generally are considered an indicator of a very good fit). These values remained
very good when estimating the CFA separately by race/ethnicity: for instance, the
RMSEA values were .051 for Latinos, .041 for African Americans, and .021 for
Whites. The model fits were similar for the residential out-mobility models; furthermore,
the in-mobility model with the combined sample also showed an excellent
fit with an RMSEA of .017 and CFI and TLI values of .97.
12. This figure is computed by setting the other variables in the model to their mean
value and then by setting the variable of interest to specific values and computing
the predicted probability of moving for these values (using the standard normal
cumulative distribution because these outcomes are estimated as a probit model).
I then computed the percentage change between these two values.
13. This figure is assessed by estimating additional models constraining the parameter
for the effect of the common perception of crime to be equal across these
for the effect of crime on mobility for Latinos and African Americans are
much smaller than that estimated for Whites. However, it is the case that
the effect of the household’s own perception of crime on residential mobility
is significantly different between Whites and these two minority
groups.
I briefly note that the control variables in these models showed results
consistent with prior literature. Residents
who have lived longer in the
residence are much less likely to move,
and an additional contextual effect
is noted in which microneighborhoods with more new residents at one
time point (residential instability) increase the likelihood that the household
will move by the next time point. Another robust finding consistent
with prior research is that homeowners are much less likely to move
regardless of their race/ethnicity. The presence of young children (younger
than 6 years of age) increases the likelihood of residential mobility for
Whites and African Americans, possibly in response to the need for quality
schools.
NEW HOUSEHOLD IS WHITE
I next turn to the in-mobility model. Given that this model simultaneously
estimated three separate equations, I refer to each of the outcomes
as “equations” in what follows. Focusing first on the equation predicting
whether the new household in the unit 4 years later is White, the expected
strong stasis effects are observed for the race/ethnicity of the previous
household. Housing units that were occupied by non-White residents at
time 1 are far less likely to have White residents move in 4 years later, as
shown in equation 1 in table 3. The coefficients for the different race/
ethnicities suggest differing effects by race; the presence of an African
American
household at the previous time point has the strongest negative
effect
on the likelihood that the new household will be White, whereas the
presence of a Latino
household at the prior time point has a strong,
although slightly weaker, negative effect, and the presence of other race
households does not have a significantly negative effect. The presence of
an African American household at the prior time point reduces the predicted
probability that the new household will be White by 70 percent for
a household at the mean on all of these measures.
Aggregated effects also suggest that the race/ethnicity of the
microneighborhood has strong effects on the likelihood that the new
household is White beyond the effect of race/ethnicity of the prior household.
Based on the standardized coefficients (Bs), a prior African American
household and the percent African American in the
groups and by assessing the change in the overall fit of the model as a chi-square
test.
microneighborhood strongly reduce the likelihood that the new household
will be White. Likewise, the standardized effects of both a Latino household
and the percent Latino in the microneighborhood of about –.24 are
quite strong. Thus, a White household is unlikely to move into a unit occupied
by an African American or a Latino at the previous time point and is
even less likely as the percent African American or Latino in the
microneighborhood increases.
Although these race/ethnicity effects are strong, evidence exists here
that the common perception of crime at the previous time point has a
significantly negative effect on the likelihood that the new household will
be White. A 1 standard deviation increase in commonly perceived crime
decreases the likelihood that the new household will be White .068 standard
deviations and reduces the predicted probability that the new household
will be White 4 percent for a household at the mean on all other
characteristics.
Recall that these perceptions are of the residents in the
microneighborhood at the prior time point, and therefore, the new household’s
perception of crime has no effect on this measure. It is notable that
this finding is one of the strongest nonrace effects in this model; indeed,
the only other microneighborhood characteristic that shows a significant
effect is the measure of the percent living in crowded conditions, which
shows a negative effect. None of the other measures (e.g., residential stability,
economic resources, or vacant units) have an effect. These findings
imply that more commonly perceived crime in the microneighborhood
reduces the percent White in the microneighborhood by reducing the likelihood
that the new household will be White.
NEW HOUSEHOLD IS AFRICAN AMERICAN
I next ask whether similar factors determine the likelihood that the new
residents will be African American.
Again strong race/ethnicity effects are
found; the strong positive effect for the presence of an African American
household at the previous time point indicates that the presence of a
White or Latino household at the previous time point reduces the likelihood
that the new residents will be African American based on equation 2
in table 3. Thus, African Americans are far more likely to move into a unit
that previously was occupied by an African American household. Once
again, strong contextual effects are found; the standardized coefficient for
percent African American is nearly as large as that for the indicator that
the previous household was African American. This finding implies that
African Americans are more likely to
enter microneighborhoods that are
mostly African American.
Despite these strong race/ethnicity effects, again a significant effect is
noted from the common perception of crime. The common perception of
crime in the microneighborhood at the previous time point significantly
increases the likelihood that the new residents will be African American,
even controlling for the very strong race/ethnicity stasis effects. A 1 standard
deviation increase in crime increases the likelihood that the new
residents will be
African American .074 standard deviations and increases
the predicted probability 20 percent. Again, it is notable that virtually no
other contextual measures have a significant effect on this racial/ethnic
transition. It thus seems that African Americans are more likely to enter
microneighborhoods with higher levels of commonly perceived crime.
NEW HOUSEHOLD IS LATINO
Finally, these same processes explain the movement of a Latino household
into a housing unit in equation 3 in table 3. The same stasis effects are
found for race/ethnicity whether measured at the level of the previous
household, or the level of the context of the microneighborhood. Despite
these very strong tendencies for Latinos to move into Latino-dominated
neighborhoods, evidence again indicates that crime is an important factor
in explaining such moves. A 1 standard deviation increase in commonly
perceived crime at the previous time point increases the likelihood that
the new household will be
Latino .055 standard deviations and increases
this predicted probability 15 percent. Once again, this result is the only
significant contextual effect outside of the race/ethnicity effects.
SENSITIVITY ANALYSES
Although I attempted to take into account the measurement error that
is contained in the measure of commonly perceived crime, the possibility
of systematic bias also exists in these reports.14 That is, certain types of
14. The possibility of measurement error also exists in these individual assessments
of perceived crime. Indeed, as noted in footnote 5, approximately 77 percent of
these reports are measurement error. I tested for and found that the degree of
measurement error differed by race/ethnicity, as it tended to be somewhat lower
for Whites compared with Latinos and African Americans. I estimated additional
models, taking into account this measurement error (by creating a latent variable
of individual perceived crime with this single indicator and an error variance constrained
to a specific value that achieves appropriate reliability), and unsurprisingly,
the perception of crime for a household did not affect mobility for any of
these groups. The effect of the common perception of crime on the mobility of
Whites remained robust in these models. If it is assumed that the perception is of
theoretical interest, regardless of the degree of “error” it contains, then this
study’s modeling approach (and that of virtually all prior studies) of ignoring this
measurement error is appropriate.
respondents and households systematically might report more or less perceived
crime than would their neighbors living in the same
microneighborhood (Hipp, 2010). I assessed this issue by adopting an
approach used by others (see Morenoff, 2003) in which a fixed-effects
model conditioning on the microneighborhood of residence is estimated
first.15 From this model, these coefficient estimates of the systematic bias
for such individuals were obtained, and then a new measure of perceived
crime was constructed purged of this bias defined as follows:
yp = y – XB
where yp is the perception of crime purged of these biases, y is the respondent’s
reported perception of crime, B contains the coefficient estimates
from the fixed-effects models estimated, and X is a vector containing the
values for the respondent on these various characteristics. These crime
perceptions purged of biases then were substituted for the original individual
crime perception variable in the models.
The initial findings were robust to this new specification, as the results
accounting for this systematic bias were actually somewhat stronger. In the
residential mobility models, the effect of the perception of crime on White
mobility was 22 percent stronger (.055 vs. .045 in the original models) for
the individual household’s perception of crime and 30 percent stronger
(.144 vs. .111 in the original model) for the microneighborhood common
perception of crime (results available upon request). The results for Latinos
and African Americans remained unchanged even when accounting
for this systematic bias, as they showed no greater likelihood to move out
of a residence in which they perceived more crime or in which the
microneighborhood collectively perceived more crime. Once again, the
models lacked the statistical power to detect a difference in the effect of
commonly perceived crime on residential mobility between Whites and
these two minority groups.
The results were also consistent, and were somewhat stronger, for the
in-mobility models. Again, microneighborhoods with more commonly perceived
crime reduced the likelihood that the new household would be
White and increased the likelihood that the new household would be
Latino or African American (results available upon request). The size of
this effect based on the unstandardized coefficients increased 23 percent
for Latinos, 24 percent for African Americans, and 32 percent for Whites.
15. I included the following individual- and household-level measures in the equation:
female, age, Black, Latino, other race, years of education, household
income, length of residence
(logged), an indicator of whether it is the first year in
the residence, ownership status, currently married, currently divorced, currently
widowed, the presence of children younger than 5 years of age, the presence of
children 6–12 years of age, and the presence of children 13–18 years of age.
The other variables in these ancillary residential mobility and in-mobility
models remained essentially unchanged.
CONCLUSION
Prior research frequently has found a relationship between the presence
of racial/ethnic minorities in a neighborhood and the rate of crime at one
point in time.
Although they sometimes posit different mechanisms, these
studies almost always conclude that the causal direction runs from the
presence of such minorities to higher rates of crime.
The present study has
proposed that at least some of this relationship might be because crime
actually increases the percentage of minorities in a neighborhood. This
postulate was based on the voluminous segregation literature and on studies
showing discriminatory behavior limiting racial/ethnic minorities’
access to some neighborhoods. I used a unique data set that allowed focus
to be placed on the housing unit to assess the extent to which disproportionate
mobility occurs in and out of high-crime neighborhoods based on
the race/ethnicity of residents.
The findings showed that although White
residents who perceive more nearby crime and those who live in
microneighborhoods with more commonly perceived crime are more
likely to move out of the housing unit,
no such effect was detected for
analogous Latino and African American residents. Furthermore, although
White residents are less likely to move into housing units in
microneighborhoods in which more commonly perceived crime occurs,
Latino and African American households are more likely to move into
such housing units.
This study’s findings suggest that the amount of crime in the
microneighborhood might play an important role in how the racial/ethnic
composition changes
across time in geographic areas. A crucial implication
of this finding is that prior work testing for a cross-sectional relationship
between the racial/ethnic composition of a neighborhood and the crime
rate and assuming that the presence of minorities increases the crime rate
might have the causal explanation, at least in part, reversed. The evidence
here suggests that Latinos and African Americans are moving into
microneighborhoods with more crime, explaining some of these cross-sectional
relationships, at least based on the measure of commonly perceived
crime. Furthermore, Whites are more likely to leave housing units in
microneighborhoods with higher levels of commonly perceived crime than
are African Americans or Latinos.
Thus, evidence is found that the limited residential mobility options of
African Americans
and Latinos affect them both coming and going.
Although modest evidence suggested that Whites might be more likely
than racial/ethnic minorities to leave a housing unit in a
microneighborhood with a greater common perception of crime, much
stronger differences in the likelihood were found by race/ethnicity of
entering a housing unit in microneighborhoods with more perceived crime.
These strong in-mobility effects highlight an important avenue through
which racial/ethnic transformation might occur in neighborhoods as a
result of the presence of crime. I emphasize that these results were found
for both African Americans and Latinos, mirroring recent work demonstrating
that the housing options of Latinos are not much greater than
those for African Americans (Iceland and Nelson, 2008).
It also should be emphasized that, other than the racial composition,
commonly perceived crime was the only consistently predictive contextual
factor in these models. No evidence was found that minorities are more
likely to enter neighborhoods with
lower economic resources or more residential
instability. This result highlights the important role crime seems to
play in the process of racial/ethnic transformation.
The findings here complement and extend recent research suggesting
that
victimization might lead to residential mobility. Although Dugan
(1999) found that being victimized increased mobility, and Xie and
McDowall (2008) found that even having nearby neighbors experience a
victimization led to greater out-mobility,
the present study’s results suggest
that the common perception of crime has important effects, at least
for White residents. Furthermore, the findings suggested that these common
perceptions of crime were most important for the racial/ethnic transformation
of neighborhoods through their effect on disproportionate inmobility
into such microneighborhoods by racial/ethnic minorities.
Although this study has provided important new insights for understanding
the relationship between the presence of racial/ethnic
minorities
and crime, certain limitations should be acknowledged. First, the discussion
of the difficulty of measuring crime suggests that it would be useful
for future studies to replicate these findings using alternative measures of
crime rather than the common perceptions of residents. Of course, for this
study’s approach to be flawed implies that racial/ethnic minorities are less
likely to leave
neighborhoods in which residents incorrectly perceive more
crime than actually exists (whereas Whites are more likely to exit such
neighborhoods), and they are more likely to enter neighborhoods in which
residents incorrectly perceive more crime than actually exists. It is not
clear what theory would predict such a process. Nonetheless, replication
using other measures of neighborhood crime would enhance confidence in
the findings. Second, although this study measured the contextual effects
at the microneighborhood level, it still might be the case that a different
level of geographic aggregation is appropriate.
Future studies will need to
test
this possibility using other levels of aggregation, such as block groups
or various spatial smoothing approaches. Third, although
the findings are
informative, they still were limited to the period of 1985–1993. Future
studies will need to test the extent to which racial/ethnic minorities remain
constrained in their access to other neighborhoods and the extent to which
that affects their ability to either exit
high-crime neighborhoods or avoid
entering them. Fourth,
I acknowledge that more than one household
change might have occurred during the 4-year period between waves.
Thus, it is possible that in the in-mobility models, a different type of
household resided in the unit between the households observed at the two
time
points.16
An important takeaway point from this study is that racial/ethnic minorities
are more likely to
enter housing units located in microneighborhoods
with more commonly perceived crime. The consequences are important
for understanding how neighborhoods evolve
across time. The assumption
that the positive relationship between the
presence of racial/ethnic minorities
and crime is caused entirely by such racial/ethnic minorities clearly
needs reconsideration. At least some of this relationship seems to be a
result of the role crime plays in changing the racial/ethnic composition of
such neighborhoods.
This outcome suggests the need for future research to
model explicitly the change in neighborhood crime and the change in
racial/ethnic composition in a dynamic framework to tease out these
effects. It seems that these constrained housing choices lead African
Americans and Latinos into more dangerous neighborhoods.
16. Of course, if the intervening household were of the same race/ethnicity as the
new household observed,
then the results would be unchanged. It is only if the
intervening household(s) were of a different race/ethnicity that results could
change; of course, the fact that the intervening household did not stay long in the
housing unit suggests that missing them might not be a serious limitation to the
design. The present study also
would miss a household that moved out and then
moved back in.
Such temporary moves are of less theoretical interest, as they do
not change the composition of the neighborhood.