Seeing Disorder: Neighborhood Stigma and the Social Construction of "Broken Windows"
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
Social Psychology Quarterly
2004, Vol. 67, No. 4, 319–342
Seeing Disorder: Neighborhood Stigma
and the Social Construction of “Broken Windows”*
ROBERT J. SAMPSON
Harvard University
STEPHEN W. RAUDENBUSH
University of Michigan
This article reveals the grounds on which individuals form perceptions of disorder.
Integrating ideas about implicit bias and statistical discrimination with a theoretical
framework on neighborhood racial stigma, our empirical test brings together personal
interviews, census data, police records, and systematic social observations situated with-
in some 500 block groups in Chicago. Observed disorder predicts perceived disorder,
but racial and economic context matter more. As the concentration of minority groups
and poverty increases, residents of all races perceive heightened disorder even after we
account for an extensive array of personal characteristics and independently observed
neighborhood conditions. Seeing disorder appears to be imbued with social meanings
that go well beyond what essentialist theories imply, generating self-reinforcing process-
es that may help account for the perpetuation of urban racial inequality.
The concept of neighborhood disorder that residents are indifferent to what goes on
once again has assumed priority in the social in the neighborhood. Few ideas have become
sciences.1 The most visible inquiry has been more influential than “broken windows”
played out in urban sociology and criminolo- (see, e.g., Duneier 1999; Harcourt 2001;
gy. According to the “broken windows” theo- Taylor 2001).
ry of urban decline (Wilson and Kelling Although perhaps less noticed, it is
1982), minor forms of public disorder lead to equally significant that the concept of disor-
serious crime and a downward spiral of urban der has penetrated social psychology. Again
decay (Kelling and Coles 1996). The pre- the notion is that disorder is negative, this
sumed reason is that visual cues such as graf- time with harmful consequences for individ-
fiti, public intoxication, garbage, and ual health and personal well-being. For
abandoned cars are thought to attract crimi- example, a number of recent studies have
nal offenders, who assume from these cues linked perceived disorder to physical decline,
depression, psychological distress, and per-
ceived powerlessness (e.g., Cutrona et al.
* Funding from the American Bar Foundation and
2000; Geis and Ross 1998; Linares et al. 2001;
the John D. and Catherine T. MacArthur Foundation
is gratefully acknowledged, as is the research assis- Mitchell and LaGory 2002; Ross et al. 2000).
tance of Bart Bingenheimer and Nigel Gannon. The On these accounts, residents read signs of dis-
Center for Advanced Study in the Behavioral order as evidence of a deeper neighborhood
Sciences provided terrific sabbatical and intellectual malaise, undermining personal health.
support. We also thank Doug McAdam, Steve Levitt,
Tony Bryk, Felton Earls, Mark Suchman, Per Olof
Yet there remains a puzzling first-order
Wikström, Cecilia Ridgeway, and the reviewers of question about what triggers our perceptions
SPQ for helpful comments on an earlier version. of disorder. Is “seeing” disorder only a matter
Address correspondence to Robert J. Sampson, of the objective level of cues in the environ-
Department of Sociology, Harvard University, ment? Or is disorder filtered through a rea-
Cambridge, MA 02138; rsampson@wjh.harvard.edu.
1 Fascination with disorder has a long history. From soning based on stigmatized groups and
early observers of London in the 1800s such as disreputable areas? Simply put, what makes
Charles Booth (see Pfautz 1967) and Henry Mayhew disorder a problem? We find that most
(1862), to authors of modern classics such as Jane research on disorder is based on individual-
Jacobs (1961), visual signs of disorder in public spaces
level perceptions decoupled from a systemat-
have been viewed as central to understanding the
civic fabric of cities. ic concern with the disorder-generating
319#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
320 SOCIAL PSYCHOLOGY QUARTERLY
environment. This is unfortunate for our whether disorder is a problem, combining
understanding of how perceptions line up uncertain evidence with prior beliefs and cul-
with independently observed ecological tural stereotypes.2 Evidence from cognitive
assessments, and how such perceptions are psychology suggests that categories are espe-
socially structured. As Harcourt (2001) cially important for the organization of infor-
argues, theoretical reflection and empirical mation in everyday life (Fiske 1998;
evidence bearing on the meaning of disorder Huttenlocher, Hedges and Vevea 2000). If
are remarkably thin. cultural stereotypes are pervasive and if resi-
Thus we set aside the usual questions: dents have uncertain information about dis-
whether disorder does or does not cause order, then they may, in a Bayesian way,
crime, or whether disorder is linked causally augment that information with contextual
to poor health. Instead we examine what pre- cues about categories of people who can be
dicts individuals’ perceptions that disorder, seen on the streets.
defined in the manner of “broken windows,” What categories matter? Research on
is a problem. Drawing on independent sets of implicit bias and cultural stereotyping sug-
linked data, we examine how the racial, eth- gests that Americans hold persistent beliefs
nic, and socioeconomic structure of neigh- linking blacks and disadvantaged minority
borhood contexts shapes perceptions above groups to many social images, including but
and beyond observable conditions of disor- not limited to crime, violence, disorder, wel-
der. fare, and undesirability as neighbors (e.g.,
Bobo 2001; Bobo and Kluegel 1997; Quillian
Implicit Bias and the Social Meaning of and Pager 2001). Beliefs about disorder are
Disorder reinforced by the historical association of
nonvoluntary racial segregation with concen-
In the first instance, “race” is a mode of per- trated poverty, which in turn is linked to insti-
ceptual categorization people use to navigate tutional disinvestments and neighborhood
their way through a murky, uncertain world decline (Massey and Denton 1993; Skogan
(Loury 2002:17). 1990; Wilson 1987). As Loury (2002) argues
in his theory of racial inequality, dark skin is
We begin by noting that the “broken win- an easily observable trait that has become a
dows” theory assumes an essentialist notion statistical marker in American society, one
both of disorder and its connection to per- imbued with meanings about crime and dis-
ception: visual cues are unambiguous and order that stigmatize not only people but also
natural in meaning (also see Harcourt 2001). the places in which they are concentrated.
A similar train of thought is seen in the The use of racial context to encode disor-
health and social psychological literature, der does not necessarily mean that people
where perceived disorder is thought to reflect are racially prejudiced in the sense of person-
external problems that influence local resi- al hostility. The power of cultural stereotypes
dents’ mental and physical health. This is all is that they operate beneath the radar screen,
quite reasonable, especially if residents’ per- forming what has been termed implicit bias
ceptions map neatly onto objectively observ- (Banaji 2002; Bobo 2001; Fiske 1998).
able aspects of disorder such as garbage, Suppose, for example, that someone without
broken bottles, litter, graffiti, abandoned cars, racial animus has nonetheless been exposed
and drug paraphernalia. to the historically and structurally induced
Objective cues of “disorder” are certain- inequality that is urban America: on average,
ly salient. So also, we argue, are cultural levels of observable disorder are higher in
stereotypes about disorder in American soci- black neighborhoods than in white. Implicit
ety. Stereotypes loom large when residents bias arises when this person automatically
are not trained as systematic or neutral concludes from such a statistical generaliza-
observers and are likely to have uncertain or tion that a specific all-black neighborhood
ambiguous information about the neighbor-
hood as a whole. It follows that individuals 2
For an overview of Bayesian reasoning see
may act somewhat like Bayesians in judging Sedlmeier and Gigerenzer (2001).#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
NEIGHBORHOOD STIGMA 321
has a disorder problem, even when empirical Neighborhood Racial Stigma
inquiry would lead to contrary evidence.
We believe that the literature on implicit
Research in social psychology has shown that
bias can be linked productively with Loury’s
automatic racial stereotypes can persist
(2002) emphasis on racial stigma, as well as
regardless of conscious or personal rejection with previous but often neglected writings on
of prejudice toward blacks (Devine 1989). the “ecological contamination” of places.
This situation leads to what Bobo (2001:292) Although Goffman’s (1963) concept of stig-
calls “laissez-faire racism.” Negative cultural ma originally was advanced at the individual
stereotyping, rather than normative preju- level, its contextual or group forms are equal-
dice, also has been shown to account for ly compelling. A contextual stance was taken
white Americans’ widespread unwillingness some time ago by Werthman and Piliavin
to share residential space with blacks and (1967), who argued that the police divide up
other minority groups (Bobo and Massagli the territories they patrol into readily under-
2001). We argue that such cognitive bias, via standable, and racially shaped, categories.
the stereotypical association of blacks with The result is a process of what they called
disorder, works in a similar, implicit way. ecological contamination, whereby all per-
The power of implicit bias is seen in a sons encountered in “bad” neighborhoods
recent and fascinating experimental study are viewed as possessing the moral liability of
(Correll et al. 2002). Researchers used a the neighborhood itself. This process has gen-
videotaping strategy to examine the effect of eral implications insofar as citizens them-
selves impute the character of disreputability
race on shoot/don’t shoot decisions in scenar-
(Hagan 1994:150) to neighborhoods contain-
ios where subjects were told to shoot armed
ing stigmatized minorities and the “rabble
targets and not to shoot unarmed targets.
class” (Irwin 1985; Wacquant 1993). Such
Participants made the correct decision to stigmatization appears to be an enduring
shoot an armed target more quickly if the tar- mechanism. In the 1800s, for example,
get was African American than if he was Charles Booth produced color-coded maps
white. Interestingly, the magnitude of this of London’s poor, with the lowest grade
racial bias in shooting decisions varied with “inhabited principally by occasional labour-
perception of cultural stereotypes but not ers, loafers, and semi-criminals—the ele-
with personal racial prejudice. In fact, the ments of disorder” (quoted in Pfautz
study revealed equivalent levels of shooting 1967:191).
bias in African American and in white partici- That the social structure of public places
pants. This finding underscores the potential- is tied to race and class sharply reinforces the
ly far-reaching consequences of statistical production of disrepute (Duneier 1999). As
discrimination and cultural stereotypes that Stinchcombe (1963) argued, access to private
reside below the level of conscious racial space is structured such that disorder by the
prejudice (also see Devine 1998). As the disadvantaged consists of doing many things
authors argue, ethnicity can influence the in public that would be (and are) legitimate
decision to shoot because cultural traits asso- in private (e.g., drinking, hanging out). That
is, privileged status enhances private access,
ciated with African Americans, namely “vio-
reducing exposure to public disorder. The
lent” or “dangerous,” act as a schema to
resulting social structure of public spaces
influence perceptions of an ambiguously
reinforces the stereotype that disorder is a
threatening target (Correll et al. 2002:1325). problem mainly in poor, African-American
African Americans are unlikely to be racially communities (Fagan and Davies 2000); this
prejudiced against their own ethnic group, stereotype feeds racial stigma and the cre-
but they are exposed, as is everybody, to ation of a durable spoiled identity for the
dominant cultural stereotypes.3 modern American ghetto (Wacquant 1993).
The insidious quality of implicit bias is
3 On implicit bias, also see https://implicit. realized further when predictions become
harvard.edu/implicit/ and Banaji (2002). self-confirming, leading to actions that#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
322 SOCIAL PSYCHOLOGY QUARTERLY
increase the statistical association between status of persons rather than on effects of
race and the observable behavior (Loury neighborhood racial contexts.
2002:23). For example, if more affluent resi- The evidence on disorder relevant to our
dents use a neighborhood’s racial composi- theoretical approach thus dwindles to a
tion as a gauge for the level and seriousness handful of studies, but we find support to
of disorder, unconsciously or not, they may motivate further inquiry. One effort is pre-
disinvest in predominately minority areas or sent in Taylor’s (2001) long-term study of the
move out; such actions would tend to city of Baltimore. In 1981–1982 he and his
increase physical disorder in those neighbor- colleagues rated street blocks and conducted
hoods. In this way implicit bias leads to rein- a survey of residents in each of 66 neighbor-
forcing mechanisms that perpetuate the hoods. In 1994, Taylor returned to 30 of the
connection of race to disorder, therefore original neighborhoods to conduct on-site
helping to explain the dynamics reinforcing ratings and another round of interviews. A
racial segregation (Charles 2002, 2003). key finding was that the perception of disor-
der varied widely between individuals even
within the same neighborhood. Perkins et al.
Prior Evidence and Hypotheses
(1993), using similar procedures in New York
Surprisingly little research bears on our City, found that residents’ perceptions and an
theoretical framing of the social structures independent rating of physical disorder were
that shape perceptions of disorder as a prob- not significantly correlated.
lem. One stream of research has been stimu- A second and perhaps more relevant
lated by the theory of broken windows and study was conducted by Quillian and Pager
crime (Wilson and Kelling 1982), but tests of (2001), who capitalized on residential surveys
this theory (e.g., Sampson and Raudenbush and police data from Seattle, Chicago, and
1999; Taylor 2001) do not adjudicate our con- Baltimore to assess the association of racial
cerns. composition with perceptions of neighbor-
A second stream of studies explores the hood crime. Using individual- and neighbor-
connection of neighborhood conditions to hood-level measures, they found a positive
individual-level health and psychological association between perceived neighborhood
functioning. An important finding has been crime and the percentage of young black
men. This finding was robust when the
the persistent correlation of perceived disor-
authors controlled for police-recorded crime
der with mental health outcomes such as
rates, survey-reported victimization, and
depression, psychological distress, and per-
measures of neighborhood deterioration.
ceived powerlessness (e.g., Aneshensel and
Quillian and Pager (2001) argue that the rela-
Sucoff 1996; Cutrona et al. 2000; Geis and
tionship between race and perceived crime
Ross 1998; Mitchell and LaGory 2002; Ross
may identify a key mechanism in the process
2000; Ross et al. 2000; for a recent review, also of white flight and residential segregation
see Sampson, Morenoff, and Gannon- (Massey and Denton 1993; Schelling 1971).
Rowley 2002). Similarly, measures of hostili- In addition, they underscore the importance
ty, mistrust, and conduct disorder have been of conceptualizing perceived crime and dis-
linked to perceived disorder (e.g., order as distinct from actual crime rates.
Aneshensel and Sucof, 1996; Ewart and A third, ethnographic study also suggests
Suchday 2002). Nearly all of the studies in the the symbolic importance attached to the
health field measure disorder at the individ- intersection of race and disorder. In a study
ual rather than the contextual level, however; of a white working-class Chicago neighbor-
thus it is difficult to isolate the neighborhood- hood, Kefalas (2003) sought to understand
level mechanisms at work in shaping per- the fastidiousness with which residents kept
ceived disorder. up their property and why they seemed to be
A third stream of studies, as noted above, obsessed with physical signs of order. She
connects race to implicit bias in stereotypes found that homeowners fretted about “the
about crime and violence. The literature on last Garden” and the threats that disorder
implicit bias, however, focuses on the racial were thought to bring on the neighborhood#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
NEIGHBORHOOD STIGMA 323
(2003:11, 14, 62, 74). No act of vandalism was Denton 1993). Thus our main hypothesis is
too minor; no unkempt yard was too trivial to that neighborhood racial/ethnic and class
escape notice. On the basis of numerous composition predicts perceptions of neigh-
interviews, Kefalas argues that residents did borhood disorder, even with adjustments for
not care so much about disorder per se, but disorder as measured systematically in those
were threatened subjectively by the urban neighborhoods by video cameras and trained
underclass, blacks in particular. In Chicago, observers. This does not imply a simple “rela-
residents perceive the history of the tivist” stance: we expect residents’ percep-
Southwest Side as a long westward march of tions of disorder to be based in part on their
decline, where the first signs were the visual observations of objective indicators such as
cues of disorder. Thus in many ways the resi- trash in the streets, broken bottles, graffiti,
dents of Kefalas’s Beltway had a “broken abandoned cars, drug paraphernalia in gut-
windows” theory in mind, but one with a ters, and the presence of loitering, drunken,
decidedly black face (2003:43). The impor- or hostile adults. Our larger point, however, is
tance attached to disorder went well beyond that the social and especially the racial com-
the level of its occurrence in the neighbor- position of a local area, which is associated
hood, so much so that the deeper meaning of statistically with disorder, is highly salient in
disorder for residents motivated Kefalas’s contemporary culture and deeply imbued
study originally. Pattillo-McCoy’s (1999) with stereotypes. If a Bayesian view of
study of a black middle-class neighborhood human reasoning is integrated with the
reveals a similar concern with the perceived research on implicit bias, perceptions of dis-
threat of disorder, although in this case the order then would be constituted as a combi-
threat came from lower-class residents within nation of inferences based on observable
the black community. evidence of disorder and correlated informa-
A small but intriguing body of research tion, especially skin color. Because racial
thus motivates our general proposition that composition is observed easily and carries
perceptions of disorder are socially con- powerful stereotypes (even if subconscious-
structed and are shaped by much more than ly), by this logic it will generate a compara-
actual levels of disorder. We assess this theo- tively large weight; social context may even
retical framework by combining census and “trump” actual disorder.
police data on key neighborhood structural If racial composition independently pre-
characteristics with original data collection in dicts the subjective perception of disorder, a
the form of personal interviews and the sys- question then arises: whether this perception
tematic social observation of city streets. If reflects pure racial prejudice rather than
essentialist theories of disorder are correct, what social psychologists call implicit or
we should find considerable agreement on automatic bias (e.g., Bobo and Massagli
perceived disorder within neighborhoods, 2001:93), or what Loury (2002) would call
few if any systematic variations by social statistical discrimination.Antiblack prejudice
position within the same neighborhood, and, would be likely to affect the perceptions of
most important, few if any between-neigh- whites (or Latinos or Asians) more strongly
borhood variations in perceived disorder than blacks. That is, insofar as nonblacks are
linked to social structure when interpersonal- particularly infected with negative stereo-
ly observed disorder is taken into account. types of blacks as a group, they might be
By contrast, insofar as disorder is socially expected to report more disorder in predom-
encoded, we expect that perceptions of disor- inately black neighborhoods while overlook-
der are shaped by individual social position ing similar levels of disorder that might exist
and by neighborhood stratification—espe- in a predominately white (or nonblack) area.
cially by race and class. Disorder, in our Prejudice, in other words, would be the dri-
framework, is part of a larger cultural narra- ving force. By contrast, drawing on the idea
tive or generalized stereotype that is tightly of implicit bias coupled with the closely allied
bound up in American cities with racially and notion of statistical discrimination, we argue
spatially understood meanings (Bobo and that the association between racial composi-
Massagli 2001; Loury 2002; Massey and tion and perceptions of disorder ought also to#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
324 SOCIAL PSYCHOLOGY QUARTERLY
be independent of the perceiver’s ethnicity. block-group boundaries typically are drawn
As an analogy, consider the black citizen who on the basis of street use, such that pedestrian
crosses the street when walking late at night streets connect with one another without
to avoid a group of approaching young black crossing major thoroughfares or highways.
males. Therefore we predict no interaction of Perceptions of disorder are measured
individual-level race with racial composition from six questions put to respondents. In
in perceiving disorder. keeping with the disorder literature (see
especially Skogan 1990:51–53; Taylor
DATA SOURCES 2001:56), residents were asked “how much of
a problem” (“a big problem,” “somewhat of a
Our theoretical framework calls for a problem,” “not a problem”) they considered
focus on how individual perceptions of disor- litter/trash, graffiti, and vacant housing/store-
der vary within and between neighborhoods, fronts (defined as physical disorder), as well
linked to independent assessments of disor- as drinking in public, selling or using drugs,
der that are reliable and ecologically valid. and teenagers causing a disturbance (social
We achieve this by integrating four sources of disorder). 4 We constructed scales of per-
data. ceived physical and social disorder with relia-
bilities at the block-group level of .65 and .67
Neighborhood Survey respectively; the combined scale is .70.5 These
The first source is a neighborhood survey reliabilities are relatively high, given that the
conducted in 1995 as part of a larger study. A average block-group sample is under 10, and
stratified probability sample of approximate- reflect the fact that 34 percent of the varia-
ly 500 block groups nested within clusters of
196 Chicago census tracts was selected to 4 Specifically, interviewers told respondents: “I’m
maximize variation by race/ethnicity and going to read a list of things that are problems in some
SES. Within areas, households then were neighborhoods. For each, please tell me how much of
a problem it is in your neighborhood. (a) How much
selected according to a multistage probability
of a problem is litter, broken glass, or trash on the
sample. At stage 1, city blocks were sampled sidewalks and streets? Would you say it is a big prob-
randomly within tracts; at stage 2, dwelling lem, somewhat of a problem, or not a problem in your
units were sampled randomly within blocks; neighborhood?” This question was repeated for (b)
at stage 3, one adult resident (age 18 or older) “graffiti on buildings and walls,” (c) “vacant or desert-
ed houses or storefronts,” (d) “drinking in public,” (e)
was chosen randomly within each selected “people selling or using drugs,” and (f) “groups of
dwelling unit and was interviewed personally. teenagers or adults hanging out in the neighborhood
The final sample size for our purposes was and causing trouble.” Hence all items refer to physical
3,585 persons living in 478 block groups, or social aspects that are observed in public spaces, in
reflecting a response rate of 78 percent. keeping with the coding scheme of our observational
approach. In a recent pilot test conducted in service of
We chose the census block group as the a planned replication of the Chicago community sur-
smallest and most ecologically meaningful vey, interviewers asked the questions about perceived
context to assess our main hypotheses. disorder in two ways. They asked first about the per-
Census block groups average about 1,300 res- ceived volume of disorder (e.g., a lot, some, a little),
idents in the sample to be described, com- and second, as in the earlier survey, about the extent
to which each aspect of disorder “is a problem.” At
pared with more than 4,000 for the more the individual level, measures of perceived disorder
commonly used unit of census tracts. As derived from the two ways of wording the questions
Grannis (1998) has shown, block groups also were correlated at r = .95. On the basis of this
reflect surprisingly well the layout of pedes- extremely high correlation, we conclude that resi-
dents perceive aspects of disorder “to be a problem”
trian streets and patterns of social interac-
primarily to the extent that they “see” those aspects of
tion. In fact, neighborhood networks tend to disorder.
concatenate within what Grannis (1998) calls 5
Reliability is defined as Σ[τ2/(τ2 + σ2/ nk)] / K, the
tertiary communities, defined as areas within average of block-group specific reliabilities across the
which houses can be connected without set of areas (K = 478). This reliability is a function of
(1) the sample size (nk) in each of the block groups
crossing a nonresidential street or thorough- and (2) the proportion of the total variance between
fare. Block groups are similar to or nest eco- neighborhoods (τ2) relative to the amount within
logically within tertiary communities because neighborhoods (σ2).#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
NEIGHBORHOOD STIGMA 325
tion in perceived disorder lies between block Police Records
groups.
Third, we collected the police-recorded
From the neighborhood survey we
number of violent offenses (robbery, homi-
selected a core set of 12 demographic or
cide, rape, aggravated assault) geocoded to
background characteristics at the person the block-group level. We then constructed
level that we believe influence perceptions of the log rate of violent crimes.
disorder. A key concern is race/ethnicity,
which we measure with indicators for Latino Systematic Social Observation (SSO)
American, non-Latino African American, and
other (primarily Asian); non-Latino whites The fourth data source is systematic
serve as the reference category. This strategy observation conducted within each of the
allows us to directly compare how blacks and block groups. Building on Reiss (1971) and
Latinos assess disorder in relation to whites. Raudenbush and Sampson (1999), by system-
atic we mean that observation and recording
Other controls include a composite measure
are conducted according to explicit rules that
of socioeconomic status (first principal com-
permit replication, and that the means of
ponent of the factor analysis of education,
observation are independent of that which is
income, and occupational prestige), sex (1 = observed. During the time of the community
female, 0 = male), current marital status survey, observers trained at the National
(composed of separate indicators for mar- Opinion Research Center (NORC) drove a
ried, separated or divorced, and single), sport utility vehicle (SUV) at three to five
homeownership, residential mobility (number miles per hour down every street within the
of moves in the past five years), residential sample of almost 500 block groups. The origi-
tenure (years in the neighborhood), and age. nal geographic unit of recorded observation
was the face block: the block segment on one
Census Data side of a street. A unique geographic identifi-
cation code was created for each face block
We collected a second set of data from to permit linking to the block group. As the
the census at the block-group level for 1990, SUV was driven down the street, a pair of
which includes the following theoretically video recorders, one located on each side of
relevant measures for assessing disorder: pro- the SUV, captured social activities and physi-
portion of families in poverty, population cal features of both face blocks simultaneous-
size, 6 density of settlement (persons per ly. At the same time, two trained observers,
square kilometer), and proportion black and one on each side of the SUV, recorded their
Latino. Hence we define racial/ethnic com- observations onto an observer log for each
position in the same fashion at both the per- face block. Face blocks were observed ran-
son and census structural level. Unlike larger domly and videotaped from 7 a.m. to 7 p.m.
aggregations, block groups provide the addi- In all, 23,816 face blocks were observed
and videorecorded for an average of almost
tional information necessary to address the
50 per block group. Because videotapes
well-known multicollinearity among ecologi-
required the expensive and time-consuming
cal variables. For example, poverty and per-
task of first viewing and then coding, we
cent black are correlated at only .37 at the selected a random subsample of 15,141 face
block-group level; this allows us the statistical blocks for videotape coding, an average of
efficiency to disentangle independent effects. about 30 per block group. Observer logs were
Throughout all analyses we performed coded for all face blocks. Coders were trained
checks for multicollinearity and influential in multiple sessions, including an intercoder
observations. reliability training session where 90 face
blocks were double coded independently, dif-
6 We estimated all models with a logged version of
ferences were resolved, and coding proce-
population size. The results were largely the same,
dures were revised. As a second check on
although in most cases raw size was a slightly stronger agreement, new observers recoded a random
predictor. 10 percent of all coded face blocks, and the#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
326 SOCIAL PSYCHOLOGY QUARTERLY
results were compared. This test produced specific aspects of disorder than does the sur-
over 98 percent interpersonal agreement, an vey, including gangs, prostitution, and multi-
essential condition for the present study ple types of graffiti.9
(Carter, Dougherty, and Grigorian 1996; Our observational scale of physical dis-
NORC 1995). order is limited to behavioral manifestations
We constructed six theoretically motivat- (e.g., graffiti, garbage in the streets) that can
ed scales based on the SSO face-block obser- be conceptually decoupled from structural
vations. These scales purposefully measure resources. Therefore we examine the sepa-
not only traditionally defined disorder but rate contribution, to perceived disorder, of
also physical decay, land use, commercial physical decay that can arise from institution-
building security, alcohol/tobacco advertis- al disinvestments. We coded whether on each
ing, and bars/liquor stores. This strategy face block there was evidence of any of the
allows us to assess whether residents are five following conditions: vacant houses;
influenced more strongly by easily observed burned-out, boarded-up, or abandoned com-
physical conditions (e.g., security fences on mercial/industrial buildings; burned-out,
commercial buildings, abandoned houses) boarded-up or abandoned houses; badly
than by comparatively rare social conditions deteriorated residential units; and badly
deteriorated recreation facilities. The decay
(e.g., prostitution, gangs in public) when
scale was derived from videotapes; again,
making inferences about disorder. Physical
double-blind coding produced interrater reli-
disorder is defined by 10 items from the
ability in the high 90s. Although physical
observer logs and videotaped coding that
decay and physical disorder are related posi-
capture the presence or absence of cigarettes
tively (r = .40), the data permit us to assess
or cigars in the street or gutter, garbage or lit-
how they predict perceived disorder inde-
ter on street or sidewalk, empty beer bottles
pendently. When the physical decay and
visible in the street, tagging graffiti, graffiti physical disorder items are combined, no sur-
painted over, gang graffiti, abandoned cars, vey disorder item is missing in the SSO mea-
condoms on the sidewalk, needles/syringes surement scheme.
on the sidewalk, and political message graffi- Commercial building security is mea-
ti.7 sured by videotapes of each face block, with
Social disorder was coded from the indicators for whether iron security gates or
videotapes. The scale items tap the presence “pull-downs” were present on the building
or absence of adults loitering or congregat- fronts and whether the windows were cov-
ing,8 drinking alcohol in public, peer group ered with security gates. Alcohol/tobacco
with gang indicators, public intoxication, advertising is based on videotapes: simple
adults fighting or arguing in a hostile manner, yes/no codes specify whether there were
selling drugs, and street prostitution. Taken signs advertising either substance.
together, the SSO items cover many more Bars/liquor stores is a two-item scale based
on videotaped assessments of the presence or
7 Graffiti not painted over were classified by type absence of bars and establishments with visi-
on the basis of guidelines informed by gang research ble signs of alcohol sales. Finally, we used one
in Chicago and internal Chicago police memos on item from the SSO, the presence of commer-
gang identification (NORC 1995). Tag graffiti was
identified by stylized forms such as block-letter art or
by attempts to create some form of visual expression.
Gang graffiti ordinarily was distinguished by the 9 Again in line with prior research, the survey asked
absence of tag art and typically involved a combina- about groups hanging out and “causing a distur-
tion of stars, crowns, emblems, and specific colors that bance.” In the SSO we matched the survey as closely
distinguish among gangs. Political graffiti was defined as we could, coding separate items not only for loiter-
as political messages and slogans. ing but also for peer groups with gang indicators,
8
We limited the definition of loitering to groups of fighting and arguing on the street, intoxicated people,
three or more adults not waiting for scheduled activi- and visible evidence of prostitution. Although per-
ties or business. For example, groups of adults waiting haps we did not achieve a perfect match, any omitted
for public transportation or standing in line to enter a SSO item that is an obvious manifestation of “causing
store would not be included, nor would children play- a disturbance” would have to be uncorrelated with
ing in public. the items we included to overturn the basic results.#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
NEIGHBORHOOD STIGMA 327
cial buildings in blocks with residential units, As in structural equation models, our mea-
to calculate the percentage of face blocks in a surement strategy for observed disorder thus
block group with mixed land use (see Table corrects for unreliability in the scale scores in
1). assessing the estimated effects of other inde-
Descriptive statistics on constructed pendent predictors such as racial composi-
measures from the four data sources are tion (Whittemore 1989).
shown in Table 1. The full list of SSO items
for the six multi-item scales is also shown in PREDICTING PERCEIVED DISORDER
the appendix and in Appendix Table A1,
along with parameter estimates for a multi- To address the nested structure of per-
level measurement model. We correct for ceived disorder in our research design, we
three major components of measurement adapt models that account for the depen-
error in constructing scales of observed dis- dence of residents’ responses within block
order to be used as independent predictors: groups. Each model can be conceived in
item inconsistency within a block group, ran- terms of a within-neighborhood and a
domly missing data, and temporal variation. between-neighborhood equation, though the
Table 1. Means and Standard Deviations for Linked Data
Mean SD
Survey Data, Person-Level Demographics
—Female .58 .49
—Married .38 .49
—Separated/divorced .17 .38
—Single .31 .46
—Homeowner .43 .49
—Black .34 .47
—Latino .33 .47
—Other .07 .26
—Mobility 1.04 1.48
—Age 41.79 16.34
—Unemployed .40 .49
—SES scale –.09 1.36
Survey Data, Person-Level Perceived Disordera
—Litter/trash .28 .45
—Graffiti .20 .40
—Vacant houses .13 .33
—Public drinking .25 .43
—Selling drugs .30 .46
—Group loitering .27 .44
Census Data, Block Group
—Density 7,451.93 4,732.33
—Poverty .21 .17
—Black .36 .42
—Latino .26 .30
—Population size 1,335.63 893.51
Police Data, Block Group
—Ln (violence rate) 8.61 .69
Systematic Social Observation Data, Block Group
—% mixed land use 24.11 11.31
—Bars/liquor storesb .01 .98
—Alcohol/tobacco advertisementsb .00 .62
—Commercial building securityb .00 .81
—Physical decayb .00 .87
—Physical disorderb .04 .77
—Social disorderb .00 .78
a Proportion of respondents perceiving disorder item to be a “big problem.”
b
See Appendix Table A1 for scale items.#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
328 SOCIAL PSYCHOLOGY QUARTERLY
two equations are estimated simultaneously where 00 is the adjusted mean perceived dis-
by means of maximum likelihood order score, 01 through 05 are the regression
(Raudenbush and Bryk 2002). The within- coefficients for the census-based structural
neighborhood (person-level) model specifies characteristics, and 06 through 12 are the
perceptions of disorder as a function of a regression coefficients of the effects of SSO-
core set of individual characteristics that based assessments. The random effects u0k, k
have been hypothesized or shown in prior = 1,...K, are assumed to be identically and
research to be influential. Our person-level independently distributed with mean zero
model is and variance τ2. Thus our strategy permits
12
(Disorder)jk = 0k + Σ
q=1
qk
Xqjk + rjk, (1)
variations at both the person and the block-
group levels, with appropriate measurements
for each. Although it is possible to allow one
where (Disorder)jk is disorder in neighbor- or more level 1 regression coefficients to vary
hood k as perceived by respondent j in that over neighborhoods, in the interest of sim-
neighborhood, 0k is a neighborhood-specific plicity we set qk = q0 for q = 1,...,12 in our
intercept, Xqjk is the value of covariate q asso- initial models. All coefficients are estimated
ciated with respondent j in neighborhood k, simultaneously.
qk is the neighborhood-specific partial effect
of that covariate on disorder, and rjk is a per- Predictors
son-specific random error assumed to be dis-
tributed independently and normally with Person-level. Although we focus mainly
constant variance σ2. Covariates include indi- on neighborhood variations, it is important to
cators for gender, ethnicity, socioeconomic understand how individuals within the same
status, employment status, age, marital status, environment perceive disorder. Thus in Table
homeownership, and mobility. To estimate 2 we begin to examine how perceptions of
how these covariates are associated with per- disorder vary as a function of the perceiver’s
ceptions within block groups (in Table 2 personal characteristics, with all neighbor-
below), we center each Xqjk around its neigh- hood conditions controlled.
borhood mean. This ensures that between- The results show that older residents per-
block-group variation in each Xqjk plays no ceive less disorder than do younger residents,
role in the estimation. For this purpose, we those who are separated or divorced perceive
allow the block-group intercept, 0k, to vary more disorder than do widowed persons, and
randomly around its overall mean and fix
other coefficients qk, q = 1,...,12 to be con- Table 2. Within-Neighborhood Predictors of
stant. Perceived Disorder
When we estimate associations between
Coeff. SE t-Ratio
block-group covariates and perceptions,
however, we center each Xqjk around its over- Covariate
—Intercept 1.764 .019 92.69**
all mean, ensuring that contributions of
—Female .041 .020 2.06*
block-group covariates to perceptions will be —Married .011 .033 .34
adjusted for person-level covariates (as in —Separated/divorced .081 .035 2.32*
Table 3 below). For this purpose our —Single .046 .036 1.29
between-neighborhood model is —Homeowner .037 .022 1.61
—Black –.157 .035 –4.45**
0k = 00 + 01(poverty)k + 02(black)k —Latino –.059 .035 –1.88
—Other –.116 .039 –2.89**
+ 03(Latino)k + 04(size)k + 05(density)k
—Mobility .010 .006 1.58
+ 06(SSO physical disorder)k —Age –.002 .001 –3.13**
+ 07(SSO social disorder)k —Unemployed –.042 .022 –1.91
+ 08(SSO physical decay)k (2) —SES scale –.014 .009 –1.56
Level 1 Variance .231
+ 09(SSO building security)k Level 2 Variance .123
+ 10(SSO bars/liquor stores)k
Notes: N = 3,116 persons and 478 block groups.
+ 11(SSO alcohol advertising)k
Covariates are centered within block groups.
+ 12(SSO mixed land use)k + u0k , *p < .05; **p < .01#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
NEIGHBORHOOD STIGMA 329
females tend to perceive more disorder than order and whites more. The basic psychologi-
do males. Employment status, SES, mobility, cal mechanism involves the perception of dis-
and homeownership, however, are unrelated crepancies based on expectation,
to perceptions of disorder. Most relevant and underscoring the fact that perceived disorder
most evident are the effects of ethnicity: reflects more than meets the eye.
blacks perceive significantly less disorder Neighborhood-level. We now turn to the
than do whites living in the same block group main question. How are perceptions of disor-
(coefficient = –.157, t = –4.45). This is true as der influenced by neighborhood-level char-
well for “other” races and for Latinos (p < acteristics? To answer this question, we
.10). introduce our independent measures of sys-
This pattern makes sense if blacks and tematically observed disorder to assess
other minorities have been exposed to more between-neighborhood variations in per-
disorder in the past; such exposure in turn ceived disorder (Table 3, Model 1). We adjust
influences the threshold for considering it a for compositional differences in all person-
problem. In the segregated and racialized city level covariates through grand mean center-
that is Chicago, for example, a white person ing, but because of space limitations, the
living in an all-white area would expect to individual-level coefficients are not repeated
see, on average, relatively small amounts of in Table 3 (available on request). There is
disorder. In this case even minor amounts of clear evidence that observed disorder strong-
disorder might be perceived as a problem. A ly predicts perceptions of disorder, under-
black person living in an all-black neighbor- scoring the validity of the measurement
hood, however, would see more disorder, on scheme. For example, we see large and inde-
average; thus the level must rise to a higher pendent contributions of observed physical
level to be perceived as a problem. In an inte- and social disorder (t = 10.06 and 6.06 respec-
grated area, which (by definition) Table 2 tively), with 73 percent of the neighborhood
models, this argument implies that the two level variance explained (unconditional vari-
groups are judging disorder by the norms ance = .235). Furthermore, when we add the
that have been generated in past, segregated SSO indicators of the physical aspects of pub-
environments: hence blacks perceive less dis- lic space in Model 2, we see that a neighbor-
Table 3.Main-Effect Estimates for Neighborhood-Level Predictors of Total Perceived Disorder, With Separate
Indicators for Observed Physical and Social Characteristics
Model 1 Model 2 Model 3
(SSO Disorder) (Adding SSO Physical) (Adding Social Context)
Coeff. t-Ratio Coeff. t-Ratio Coeff. t-Ratio
Neighborhood Level
—Intercept 1.747 115.12** 1.693 40.72** 1.261 25.46**
—Physical disorder .225 10.06** .146 5.89** .038 1.55
—Social disorder .123 6.06** .096 4.74** .044 2.24*
—Bars/liquor stores .016 .43 .017 .56
—Alcohol/tobacco advertisements .045 1.36 .052 1.50
—Commercial building security .010 .43 .004 .18
—Physical decay .129 6.98** .068 3.77**
—% mixed land use .002 1.34 .001 .73
—Poverty .756 7.43**
—Black .414 8.03**
—Latino .442 6.63**
—Population size .015ⴒ10–3 1.10
—Density .001ⴒ10–3 .44
Level 1 Variance .233 .232 .230
Level 2 Variance .063 .051 .031
Notes: N = 3,316 persons and 478 block groups. Neighborhood estimates are adjusted for compositional differ-
ences in the 12 person-level covariates shown in Table 2.
*p < .05; **p < .01#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
330 SOCIAL PSYCHOLOGY QUARTERLY
hood’s observable physical decay is related variables. The first component appears to tap
positively and significantly to perceived dis- commercial areas that sell and/or advertise
order (t = 6.98) net of observed disorder. alcoholic beverages or tobacco, and where
In Model 3 we add neighborhood ethnic building security is extensive. The second
and social composition to test our main the- component combines physical and social dis-
sis. With controls for observed disorder and order with physical decay.
physical structure in addition to the person- To achieve greater parsimony we thus
level predictors, neighborhood social and substitute these two principal components
ethnic composition are linked powerfully to for the six SSO predictors in predicting per-
perceptions of disorder. In particular, con- ceived disorder. In addition, we disaggregate
centrated poverty, proportion black, and pro- perceived disorder by type (social and physi-
portion Latino are related positively and cal): presumably physical disorder is less sen-
significantly to perceived disorder. sitive to cultural stereotypes or statistical
Moreover, adjusting for racial context greatly discrimination based on racial context than is
reduces the association between SSO and the more nebulous concept of social disorder.
perceived disorder: the coefficient for social This strategy simultaneously addresses biases
disorder is cut in half, and the association that might emerge from item discrepancies
between physical observed disorder and per- between the survey and the SSO method by
ceptions is eliminated entirely. (Compare type of disorder. In particular, the overlap
coefficients for these variables moving between survey and SSO items is not exact
between Model 2 and Model 3.) Thus much within subscales (in the SSO scale, for exam-
of the explained variance of SSO disorder in ple, vacant houses are considered signs of
Models 1 and 2 is spurious through its associ- physical decay and disinvestment rather than
ation with racial and class composition. of disorder), but in combination the coverage
is full.
Principal Components of Disorder We also control for two new variables to
test the robustness of our main findings. One
The six scales derived from systematic is the log violence rate at the neighborhood
social observation are correlated, to some level.10 Violence, especially violent crimes
degree; they also vary in reliability (see that make the news, may influence subjective
appendix). Therefore it is appropriate to perceptions of threat. In particular, the “bro-
investigate the contributions of a reduced set ken windows” thesis might be read to imply
of constructs that capture common sources of that residents will perceive disorder as a
variation among these six measurements. problem primarily when it is connected to
Table 4 displays the results of a principal- public (or official) accounts of personal vio-
components analysis. We found two compo-
nents that are reasonable on conceptual 10 Substituting a survey measure of violence pro-
grounds, accounting for approximately 65 duces quite similar but weaker results. We retain offi-
percent of the variation among the six SSO cial violence because it provides the stronger test.
Table 4. Higher-Order Principal-Components Analysis of Systematic Social Observation (SSO) Scales
Rotated Loadings
Factor 1 Factor 2
SSO Scale
—Bars/liquor stores .88 .06
—Alcohol/tobacco advertisements .90 .09
—Commercial building security .62 .46
—Physical decay .03 .74
—Physical disorder .31 .75
—Social disorder .07 .69
Eigenvalues 2.07 1.82
% Variance Explained 34.59 30.32
Note: N = 478 block groups.#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
NEIGHBORHOOD STIGMA 331
lence. Because race and poverty are related borhood are invited?” “When a neighbor is
to official crime, a spurious connection may not at home, how often do you and other
exist between perceiving disorder as a prob- neighbors watch over their property?” “How
lem and large concentrations of minority and often do you and other people in this neigh-
poor residents (Quillian and Pager 2001). borhood visit in each other’s homes or on the
The other control stems from a long his- street?” “How often do you and other people
tory of research in urban sociology on the in the neighborhood ask each other advice
importance of social exchange and ties about personal things such as childrearing or
among neighbors (e.g., Fischer 1982). We job openings?” (scale mean = 2.47, sd = .73).
specifically introduce a scale measuring reci- Overall this measure taps the flow of infor-
procated exchange among neighbors. The mation exchange and network ties among
items comprising the scale are: “About how neighbors.
often do you and people in your neighbor- Table 5 assesses the new model specifica-
hood do favors for each other? By favors we tion at both the person and the neighborhood
mean such things as watching each other’s level. The outcomes are total perceived disor-
children, helping with shopping, lending gar- der (Model 1), perceived physical disorder
den or house tools, and other small acts of (Model 2) and perceived social disorder
kindness” (never, rarely, sometimes, or (Model 3). In every case, both SSO principal
often). “How often do you and people in this components are positive and statistically sig-
neighborhood have parties or other get- nificant predictors. Yet for all three out-
togethers where other people in the neigh- comes, the neighborhood’s social and ethnic
Table 5. Main-Effect Estimates of Neighborhood- and Person-Level Predictors of Perceived Disorder, by
Type, With Principal Components for SSO Disorder/Decay and Alcohol Density, Plus Additional Controls
Model 1 Model 2 Model 3
(Total Disorder) (Physical Disorder) (Social Disorder)
Coeff. t–Ratio Coeff. t–Ratio Coeff. t–Ratio
Neighborhood Level
—Intercept .327 1.26 .635 2.45* .060 .19
—Poverty .661 6.28** .690 5.71** .662 5.74**
—Black .334 5.77** .250 4.12** .426 6.09**
—Latino .386 5.90** .329 4.66** .441 5.74**
—Density .003ⴒ10–3 .92 .000ⴒ10–3 .10 .005ⴒ10–3 1.15
—Mixed land use .001 .52 .002 1.42 –.001 –.61
—Population size .016ⴒ10–3 1.31 –.000ⴒ10–3 –.01 .034ⴒ10–3 2.14*
—SSO alcohol density .040 2.91** .040 2.57** .045 2.89**
—SSO disorder/decay .080 3.84** .078 3.45** .084 3.54**
—Ln (violence rate) .116 3.71** .077 2.51* .150 3.97**
Person Level
—Female .026 1.35 .019 .93 .034 1.40
—Married .021 .67 –.026 –.78 .076 2.02*
—Separated/divorced .076 2.27* .045 1.19 .118 3.02**
—Single .035 1.03 –.012 –.33 .096 2.3*
—Homeowner .039 1.84 .057 2.47* .023 .86
—Black –.164 –4.95** –.215 –5.97** –.119 –2.89**
—Latino –.077 –2.60** –.121 –3.84** –.026 –.68
—Other –.128 –3.54** –.166 –4.03** –.079 –1.70
—Mobility .007 1.03 –.001 –.17 .017 1.99*
—Age –.003 –3.67** –.003 –3.83** –.002 –1.93
—Unemployed –.022 –1.06 .010 .44 –.055 –2.21*
—SES scale –.022 –2.64** –.014 –1.68 –.033 –3.14**
—Exchange/ties –.061 –4.24** –.056 –3.81** –.068 –3.81**
Level 1 Variance .229 .273 .352
Level 2 Variance .029 .034 .034
Notes: N = 3,116 persons and 478 block groups.
*p < .05; **p < .01#1821—Social Psychology Quarterly—VOL. 67 NO. 4—67401-sampson
332 SOCIAL PSYCHOLOGY QUARTERLY
composition remains positively and highly isolated. It may be that those enjoying more
significantly related to perceived disorder. social support are less likely to report more
Even for perceived physical disorder, which problems of a wide variety. In any case, the
is the strictest test of our thesis, the estimated racial pattern remains intact.
effect of racial composition is substantively
large, with controls for observed disorder. Cross-Level Interactions With Race/Ethnicity
For example, if we use the model presented
in Table 5 to evaluate how perceived physical The results so far support the hypothesis
disorder changes as a function of increasing that neighborhood racial context helps shape
observed physical disorder/decay and racial perceptions of disorder, with controls for
composition from the 25th to the 75th per- carefully observed disorder. Does this ten-
centile, the effect of racial composition is dency depend on the resident’s race? Do the
approximately three times larger than that of effects of an individual’s race on perceptions
observed disorder/decay, with controls for all of disorder vary randomly across neighbor-
personal characteristics and neighborhood hoods? We tested this specification; once
ecology.11 The magnitude of the differential again, black residents reported less disorder
by racial composition is so substantial that than whites (coeff. = –.13, t = –3.11), an effect
measurement error is not a credible explana- that varied randomly across neighborhoods.
tion, especially in this case, where observed More interesting, however, the contextual
physical disorder is measured with reliability effect of racial composition is largely inde-
greater than .90. Also, the simple correlation pendent of the observer’s ethnicity. 13
of percent black with observed disorder is Specifically, blacks were not significantly
positive, as expected: r = .15 for SSO social more or less likely than whites to view pre-
dominately black neighborhoods as high in
disorder and r = .11 for SSO physical disorder
disorder, with controls for observable disor-
(p < .05). The magnitude of the racial compo-
der and other covariates.
sition effect on perceptions is all the more
A notable exception was the interaction
revealing in this light.12
between Latino ethnicity and block-group
Two other findings presented in Table 5
percent black. Perceptions of disorder
are noteworthy. First, the officially reported
increase as a function of percent black for
violence rate in a block group contributes sig-
members of each ethnic group, but this ten-
nificantly to residents’ perceptions of disor-
dency is significantly pronounced for Latinos.
der, in line with Quillian and Pager (2001).
This relationship is graphed in Figure 1; pre-
Survey-measured reports of personal victim-
dictors not displayed are held constant at
ization also are linked positively to perceived
their means. In neighborhoods less than 25
disorder. Yet in neither case is the effect of
percent black, whites and Latinos essentially
racial composition eliminated. Second, resi-
do not differ in their perceptions of disorder.
dents embedded in networks of greater reci-
At roughly 25 percent black, however, a
procal exchange, and thus presumably with
threshold suggested by prior research as par-
access to better information, perceive less
ticularly salient (e.g., Schelling 1971), Latinos
disorder than those who are more socially
begin to diverge sharply from whites: when
neighborhoods reach 75 percent black or
11 Moving from the 25th to 75th percentile of more, Latinos perceive substantially more
observed physical disorder is associated with about a disorder than do whites (also see Charles
.08 gain in the perceived physical disorder scale. 2000). Overall, as percent black in the block
Moving from the 25th to 75th percentile of percent
black, by contrast, shifts the perceived disorder scale
upward by about .25.
12 13 In testing cross-level interactions, we estimated a
We also disaggregated at the item level to exam-
ine exact matches of physical disorder; again we reduced model that retained significant predictors of
obtained similar results. For example, the t-ratio total perceived disorder at the neighborhood level
reflecting the effect of racial composition on per- shown in Table 5. We estimated a series of alternative
ceived graffiti, with controls for observed disorder models with the same results. We also freed sequen-
(including three types of graffiti) and the other factors tially the slope variance for each racial/ethnic group;
reported in Table 5, was 5.08 (p < .01). For trash it was it was significant only for blacks. (Full table results are
3.41 (p < .01). available on request.)You can also read
