American Journal of Community Psychology, Vol. 13, No. 5, 1985
Attachment to Place: Discriminant Validity,
and Impacts of Disorder and Diversity t
Ralph B. Taylor 2 and Stephen D. Gottfredson
Criminal Justice, Temple University
Sidney Brower
School of Social Work and Community Planning,
University of Maryland at Baltimore
We sought to establish the discriminant validity of related concepts: territoriality and attachment to place. Further, we investigated the contextual determinants o f attachment. We hypothesized that persons living in more heterogeneous neighborhoods would be less attached. We also hypothesized that
levels o f perceived or objective disorder would be negatively associated with
attachment levels. The hypotheses were tested using data from a survey
of 687 heads o f household in Baltimore, Maryland. Data on police activity
and neighborhood characteristics were also gathered. A clear territorial
dimension and two attachment dimensions (Rooted and Involved; Acquaintanceship) were identified. Regressions of each dimension supported the distinctness o f territoriality and attachment, as well as the hypothesized impacts
o f diversity and disorder on attachment. Not only are some people more attached to place than others but there are also some places to which people
can become attached more easily. This study is the first to establish the direct
impacts o f diversity on attachment to place.
Clinical psychologists have long recognized that people can and do cathect,
or get attached to, other persons. In everyday language we speak of persons
who are "attached" or "involved" or "invested" in one another. More recently, social scientists have given attention to people's attachment to places. In~Data described here were collected under grant 78-AX-IJ-0134 from the National Institute of
Justice, Ralph B. Taylor, principal investigator. The grant was carried out while the first and
second authors were at the Center for Metropolitan Planning and Research, Johns Hopkins
University. Opinions expressed here are solely those of the authors.
2All correspondence should be sent to Ralph B. Taylor, Department of Criminal Justice, Gladfelter Hall, Temple Universitty, Philadelphia, Pennsylvania 19122.
525
0091-0562/85/1000-0525504.50/0© 1985 Plenum PublishingCorporation
526
Taylor, Gottfredson, and Brower
terest in this topic has been quite broad and has spawned a range of
theoretical models. In addition to the more straightforward notions of attachment to place we have concepts such as place dependence (Stokols &
Shumaker, 1981), place identity (Proshansky, 1980), and residentialareal bonds (Peachey, 1981). Shumaker and Taylor (1983) discussed
attachment to place and these related concepts. They suggested that
there are important differences between the various approaches but
fundamental similarities as well. All of the models of attachment to place
or related concepts are concerned with the psychological meaning of an environment for a person or group, the social and person-place transactions
that precede the development of this meaning, and the positive affective and
social involvements that emerge as part of this bond. "Elements of the system (of attachment) include cognitions of satisfaction and expectations of
stability, feelings of positive affect, greater knowledge of the locale, and behaviors that serve to maintain or enhance the location (investment, improvement, beautification, and so on)" (Shumaker & Taylor, 1983, p. 237).
Research to date on attachment to place has focused for the most part
on what type of people are more attached to place than others. Investigations to date have not yielded consistent results. Gerson, Stueve, & Fisher
(1977) found that stage of the life cycle and social class influenced attachment. Households with young children and of higher social class were more
attached. Kasarda and Janowitz (1974), however, did not find that these two
attributes influenced levels of attachment. Riger and Lavrakas (1981) found
that attributes such as age, income, number of children living at home, and
educational level discriminated persons who showed different patterns of attachment. It is clear from even these few studies that demographic and social class factors are linked to attachment levels. Discrepancies that have
emerged in different studies stem in part from differing operationalizations
of attachment. Studies that had fairly similar measures of attachment, such
as Gerson et al. (1977) and Riger and Lavrakas (1981), have observed similar predictors.
At least as interesting as the question of who is more or less attached
to place is the question of location. In what kinds of settings or contexts
or neighborhoods do we find people who are more or less attached to where
they live? What are the kinds of places to which people are more likely to
become strongly attached? This is one question that is addressed by the present
research.
To date two studies have examined impacts of contextual features on
levels of attachment. Kasarda and Janowitz (1974) examined the impacts of
city size, density, and heterogeneity and found that these factors had no consistent impacts on sentiments of attachment. Gerson et al. (1977) observed
several links between contextual variables and aspects of attachment. In their
all-white sample, persons were more likely to interact frequently with other
Attachment to Place
527
residents the further they lived from blacks, and if they lived in a suburban
and supposedly more homogeneous neighborhood instead of in an urban
one. These linkages suggest that being surrounded by ethnically similar others,
or those of similar class background, is associated with stronger local social
involvement. Interpretation of these findings as conclusive evidence of impacts of diversity on attachment to place is not warranted, however, since
the authors did not have direct measures of diversity or heterogeneity of
respondent's locale. The authors were forced to make the case that their measures probably reflected diversity. In addition, the modest size of the linkages (/3 < .08) also suggests caution.
Why might homogeneity foster, or diversity dampen, attachment to
place? One possible key can be located in the work of Rosenberg (1972, 1975;
see also Gans, 1968; Wilson, 1975) on the impacts of consonant versus dissonant social contexts. Rosenberg (1975) demonstrated that those who live
or go to school in a dissonant context, surrounded by others who are dissimilar on one or more key parameters (e.g., race, religion), have less stable
self-images and lower self-esteem, causing doubts about how they should
behave and obscuring commonly accepted benchmarks by which behavior
can be evaluated. We suggest that in a dissonant context person-place bonds
may be influenced in a similar way. In a more diverse residential setting occupants may be less likely to agree on what is acceptable public behavior.
Norms of how to treat or greet others, or of upkeep and beautification may
be less widely known; or, if known, less widely followed. This lack of clarity
emerging in dissonant contexts may be stress inducing. Through generalization of affect, attachment may consequently be dampened or prohibited from
developing. Or diversity, by reducing consensual validation, may make that
context less reinforcing, resulting in less likelihood of people becoming attached.
Is diversity of residential context associated with lower levels of attachment? This was a major hypothesis we sought to investigate in this study.
In our study we improved upon earlier tests of this hypothesis for two reasons. First, we had available objective data that afforded us direct measures
of diversity. Second, diversity measures from two different levels of analysis were available. We were able to assess diversity at the small group level
of the street block, and at the areal or neighborhood level. As Unger and
Wandersman (1983) and others have shown, the street block can function
as a small group (see also Taylor, in press; Taylor, Gottfredson, & Brower,
1981, 1984). People see one another in passing on a daily basis and are
influenced by the same immediate problems. The street block is like a small
group in that (a) specific people may adopt specific roles (leader, gossiper,
and so on), and (b) norms may exist that are shared widely (Taylor, in press).
We expect that diversity at the street block level would have a negative or
dampening impact on attachment for the above-cited reasons. Further, we
528
Taylor, Gottfredson, and Brower
expect that block diversity would have more of an impact on attachment than
neighborhood diversity, because the block context is more proximal to the
resident.
Of course, diversity can lead to more extreme outcomes than confusion of group norms. It may lead to actual instances of conflict or disorder.
Heterogeneity may result in people arguing more frequently with one another,
more frequent disrespect for the property of others, or residents having to
rely more frequently on agents of formal control, such as police, to quell
disturbances. This disorder may be perceived, as when people report greater
incidences of conflict, minor lawbreaking, and more concern; or, it may be
objectively evident, as when people actually call the police more frequently
to deal with "nuisances" (as contrasted with-serious crime). Therefore, our
second hypothesis was that objective a n d / o r subjective measures of disorder
would be associated with lower levels of attachment.
This second hypothesis could relate to the first one in two ways. Diversity could strongly determine patterns of disorder, in which case the two sets
of variables would compete to explain outcome variance and the independent contribution of each would be much smaller, or perhaps nonsignificant,
compared to the zero-order relationships. Alternatively, disorder and diversity could be related only modestly. If this is the case, each cluster of variables could have sizable, unique impacts on attachment.
Our third hypothesis was that attachment to place can be conceptually
distinguished from human territorial functioning. Territorial functioning is
concerned with access to, control over activities in, and obligations surrounding particular delimited spaces (Sundstrom, 1977; Taylor, 1978). Territorial
functioning represents an interlocked system of attitudes and behaviors that
are goal directed and not instinctual. Territorial cognitions may reflect such
issues as whether strangers or acquaintances are encountered in a particular
place, how much responsibility one has for who enters and what goes on
in a particular location, and how much disorder is experienced in a particular space. Although concerned with smaller scale spaces than attachment,
it is clear that both attachment and territorial functioning reflect local involvement. It is quite possible then that the two may not be distinct.
To date, this possibility has not been assessed. That is, it is not yet
clear if the concept of attachment to place, vis-fi-vis territoriality, has discriminant validity. "For the establishment of construct validity, discriminant
validation is required. Tests can be invalidated by too high correlations with
other tests from which they were intended to differ" (Campbell & Fiske, 1959,
p. 81). To date no assessments of the discriminant validity of attachment
to place have been made. The establishment of attachment's discriminant
validity is important for several reasons. If such validity cannot be established it means that conceptual work on attachment can be integrated with
currrent conceptual work on human territorial functioning. This would greatly
Attachment to Place
529
expand the theoretical framework within which atttachment can be clarified.
Second, if such validity can be established it would mean that person-place
bonds include more than one system of linkages: attachment as well as territoriality. The relations between these different systems of linkages become
of interest.
Related to the discriminant validity question is the question of how
territoriality and attachment are influenced by the predictors that are examined. More specifically, we expect territoriality would be influenced by local levels of disorder, since these make territorial functioning more difficult,
but not by contextual factors reflecting diversity. Different patterns of influence by our predictors would, of course, strengthen even further the discriminant validity of attachment.
In sum, we sought to test three hypotheses: (a) We predict that, controlling for resident social class, persons living in more diverse settings would
be less attached. In this hypothesis we can specifically examine the impacts
of block versus neighborhood diversity. (b) We predict that, in addition to
diversity, higher levels of local disorder would be associated with lower levels of attachment. (c) We expect that measures of attachment would be distinct or isolable from measures of human territorial functioning.
METHOD
Sample and Procedure
A multistage, stratified sample of households in Baltimore City (MD)
was drawn and interviewed. Face-to-face interviews were conducted with
heads from 687 households on 63 blocks nested within 12 neighborhoods.
The 12 neighborhoods were drawn from three types or strata. Neighborhoods
were classified using 1970 census data. The three strata were as follows: Type
1 neighborhoods were neighborhoods with low-income residents and where
the bulk of the housing stock was rental (>60°/o). Type 3 neighborhoods
were neighborhoods with residents from middle- and upper-middle income
levels and predominantly owner-occupied housing units ( > 6 0 % ) . Type 2
neighborhoods were "mixed " i n that they fell between the Type 1 and Type
3 neighborhoods on the income and percentage rental parameters. These mixed
neighborhoods had moderately low income populations and were from 40
to 60% rental. Driving through the Type 2 neighborhoods provided many contrasts. In these neighborhoods it was not unusual to have a block of
predominantly vacant or very run-down housing right around the corner from
a very respectable or well-kept block of houses. Our final sample included
three Type 1 neighborhoods, six Type 2 neighborhoods, and three Type 3
530
Taylor, Gotffredson, and Brower
neighborhoods. We double sampled from the mixed group of neighborhoods
because they far outnumbered the neighborhoods of the other two types.
In the analysis we included two d u m m y variables: one for Type 1 neighborhoods (low income, rental) and one for Type 2 neighborhoods (mixed). The
reference string was therefore Type 3 neighborlloods (medium income, homeowned). Blocks were selected as to include socially cohesive and socially fragmented street blocks. Further details on sampling and interviewing can be
found in Taylor, Gottfredson, Brower, Drain, and Docket (1980), and
Taylor, Gottfredson, and Brower (1981, in press). In the final sample the
owner-renter split was 53 to 47%, average age was 44 years, and average length
dence in the neighborhood was 16 years (median = 12). The sample was
almost 59°70 black. On these and other parameters our sample closely matched
the population of Baltimore City households. The interviews were conducted in the summers of 1979 and 1980.
Police calls for service data were obtained for the years 1978 and 1979
(the latest year available at the time). Police calls were then allocated to specific blocks, using the attached addresses. They were then content analyzed,
based upon type of call. The category of most interest to us here is calls to
deal with social nuisances. These included calls for the following reasons:
disorderly persons, family disturbance, mental case, juvenile disturbance,
exposure case, intoxicated person, person lying in the street, and discharging a firearm. Most of the calls in this class were for family or juvenile disturbances or disorderly persons. Police data for each street block were
subsequently converted to rates using number of occupied housing units per
block as the denominator. Calls to the police to deal with social nuisances
were used as our objective measure of disorder. This category of calls correlated .916 with total service calls (in all categories).
Fronts and backs of each interviewed unit were photographed in the
summer and subsequently rated independently by trained teams of raters.
One feature that was rated was the extensiveness of high-demand gardening. Pictures with the lowest score on this scale would have no grass or shrubs.
Pictures with a moderate score on this dimension would have shrubs a n d / o r
grass, but no high-demand gardening such as flowers or vegetables, both of
which require constant care. Pictures with the highest high score would have
extensive high-demand gardening. Backs of houses offered more yard space
and thus a location where more variation in gardening was present. Using
a 5-point scale, interrater reliability of the two-rater mean on gardening (r
interclass) was .87.
Relevant Interview Items
Attachment. The survey included several items that, based on past
research, clearly tap attachment to place. (a) Owner status (scored 1) versus
Attachment to Place
531
renter status (scored 0) is a crude but significant indicator of capital and probably emotional investment in an area. This was used by Riger and Lavrakas
(1981) as a measure of attachment. (b) Length of address in the neighborhood was used as a measure of the temporal aspect of attachment. Riger
and Lavrakas (1981) also included this as a measure of attachment. We decided to use it as a measure of attachment rather than a predictor of attachment because, as Gerson et al. (1977, p. 148) point out, length of residence
can be a result of local involvement as well as a cause. Several items on the
survey tapped the social aspect of local involvement. All items were couched
in reference to other residents on the street block, not the neighborhood.
(c) Respondents were asked to make an assessment of overall perceived
similarity with neighbors on the block, considering things like values, education, and income. (d) During the course of the interview, the interviewer
produced a block schematic, with all block addresses, and the respondent's
own household, clearly indicated. For each address on the street the respondent was asked to indicate if he or she knew someone who lived there by
face or name. Then, for each respondent we were able to determine the
proportion of addresses on the block where he or she was acquainted with
someone. (e) Respondents were asked if they belonged to any other local
organizations to which other residents on the block also belonged. If they
did, they received a score of 1; others were scored 0. (f) Respondents were
asked if they had asked neighbors on the block to watch their house for them,
take in mail, or water plants when they went away. These three items were
standardized and added up to produce a "reliance on neighbors" scale. Cronbach's alpha for this scale was .70. (g) An index of local involvement used
by Hunter (1974) and others is the ability to provide a neighborhood name.
Hawkes (1951, cited in Relph, 1976) has suggested that this is one way we
become linked to an area. Those who were able to provide a name were scored
1; those who could not were scored 0. (h) Finally, we have our measure of
gardening in back. This measure may reflect attachment since it reflects work
a person has put in on a locale.
Territorial Cognitions. Respondents were presented with 14 statements
of territorial attitudes and asked to answer them for each of four territories:
steps or property in front of the house, yard behind the house, sidewalk directly in front of the house, and alley directly behind the house. These first two
territories were labeled home territories, since they actually were private
property, and the last two labeled near-home territories, given their adjacency
to the home. The territories were presented in randomized order. For each
particular attitude statement, which was answered using a 6-point Likert scale,
we verified that responses to two territories of the same type were more similar
than responses to two territories of different types. The correlations between
the 14 attitude statements were submitted to principal components analysis
using a random half of the sample. Three interpretable factors were produced.
532
Taylor, Gotffredson, and Brower
The same three factors were replicated using the second half of the sample.
(For more details see Taylor et al., 1980.) Thus, the three territorial scales
we produced are quite reliable. One scale reflected problems (e.g., litter,
graffiti) related to a lack of territorial control. Since this problems scale is
conceptually quite close to the perceptions of disorder we are using as predictors, we decided not to use the territorial problems scale as a measure of
territorial functioning. To do so would have been to confuse predictors and
outcomes. The two remaining scales, each operationalized t w i c e - o n c e for
home territories and once for near-home territories-were as follows: (a) One
scale referred to access control. Access control is determined partly by an
ability to discriminate between who belongs and who does not belong in particular locations. Items relevant to this scale include: "I can tell people who
belong there from outsiders"; "If there's a suspicious person hanging around
someone is bound to call the police"; "I see mostly people I know there";
and "I am likely to chat with friends and neighbors." When asked with regard
to home territories these items produced a very acceptable reliability (Cronbach's alpha) of .83. For near-home territories reliability was .78. (b) Three
territorial attitudes combined to form a scale concerned with responsibility.
The items included in this scale were "I feel some responsibility as a member
of the neighborhood for what goes on"; "I feel personally responsible for
what goes on"; and "I have more say than others about what happens." For
home territories scale reliability was .87; for near-home territories it was .82.
Thus we have produced four territorial scales, each of which has acceptable
scale properties. Scale scores were created by taking the mean scores across
the items on a scale.
Perceived Conflict. Respondents rated the severity of 11 diffferent
problems in their neighborhood (see Table II). For each problem they indicated if it was "not a problem," (scored 1), "somewhat of a problem," (scored
2), or "a big problem," (scored 3), in their neighborhood. The problems included direct assessments of conflict between coresidents- people fighting,
for e x a m p l e - a n d less direct measures of conflict such as street hassles,
property damage, and trespassing.
RESULTS
Analysis Overview
We carried out a principal components analysis of the attachment and
territorial variables. One indication of discriminant validity would be if the
territorial variables end up on a component separate from the attachment
variables. Although earlier analyses (Riger & Lavrakas, 1981) used oblique
A t t a c h m e n t to Place
533
rotations to assess dimensionality of attachment, we used a varimax rotation. With the oblique rotation there are serious problems in achieving clear
conceptual clusters (see Gordon, 1968, p. 605).
We make no pretensions to having a complete, exhaustive list of attachment variables at our disposal. Thus, the number of attachment factors
we extract are somewhat arbitrary. Nonetheless, more important than the
number of attachment dimensions extracted is (a) whether or not the dimensions are separate from territoriality and (b) what variables predict the attachment dimensions observed.
Principal components analysis of our 11 problem items, indices of perceived disorder, was also carried out. Then, principal components scores on
the attachment, territoriality, and problems dimensions were created for each
respondent.
The outcomes of interest to us are the principal components scores on
the territoriality and attachment dimensions. Inasmuch as we ended up with
one territoriality component and two attachment components, we had three
outcomes. They are of course independent of one another, since they were
based on independent dimensions. For each outcome we carried out a hierarchical multiple regression. Each regression included variables with significant zero-order correlations (r > .07). Entered first were individual social
class factors, so that this source of variation could be covaried out. On the
second step variables reflecting on-block diversity were allowed to enter. These
are explained below. On the third step variables reflecting neighborhoodlevel diversity were entered. On the last step variables reflecting subjective
and objective measures of disorder were allowed to enter. Variables were
allowed to enter only if they made a significant reduction in unexplained variance at the point they stepped into the model.
Some points about the logic of our hierarchical ordering deserve mention. Entering sociodemographics first allows us to put aside individual level factors which we know, from prior work, may influence attachment. Thus
we can examine contextual factors net of individual sociodemographic differences. Block diversity factors were entered on the second step, because we
feel that the block is a more important context than the larger neighborhood.
It is more immediate than neighborhood, although it is obviously determined
in part by neighborhood composition. By allowing disorder indices to enter
last we are giving them the most stringent test to pass. Since they come after
the diversity measures they merited entry only if they made their own unique
contribution to the outcome, net of heterogeneity.
Measures of block diversity were constructed as follows. For education and income, we simply computed the block-level standard deviation,
based on respondent information gathered in the survey. For race we computed the proportion of block respondents who were black and constructed
534
Taylor, Gottfredson, and Brower
a diversity measure from that. If the block was 50/50 white/black, the measure scored 1.0. As proportion black on the block moved away from 50/50
and toward 0/100 or 100/0, the diversity measure moved toward .50. Thus
the diversity measure increased as the block became racially more integrated. The measure used absolute values and thus was blind to the direction
of the segregation. The formula used was Diversity = 1 - ( a b s o l u t e value
(. 5 0 - proportion black)).
Principal Components Analysis of Territorility and Attachment
The results of our principal components analysis of attachment and
territoriality variables appears in Table I. Three components with eigenvalues
> 1.0 were extracted and rotated to varimax solution. The first component
clearly reflects territorial functioning. All four territorial cognition variables
have loadings of almost .70 or better. No other variables have loadings much
above .20. The responsibility scales load somewhat higher than the access
control scales and the home space scales load somewhat higher than the nearhome space scales. The fact that a clean territorial dimension has emerged,
uncontaminated by attachment to place variables, strongly supports the discriminant validity of the attachment to place concept. If attachment to place
is discriminable from a closely related person-place concept such as territoriality, it is likely also to be distinct from other less similar person-place concepts as well.
The second and third components reflect attachment to place. The second component suggests local investment, reflected in homeownership and
Table 1. Principal Components Analysis of Territoriality and Attachment Variables
Component
Variable
(Q02) Owner status
(Q3C) Proportion households known
(Q9) Overall similarity
(SOCNBR reversed) Trust
(Q14A) Shared organizations
(QI6) Know neighborhood name
(QI8) Length of residence
( T R H O M R S P ) H o m e responsibility
(TRNHMRSP) Near-home responsibility
( T R H O M O U T ) H o m e access control
( T R N H M O U T ) Near-home access control
(BACK4) Gardening in back
Eigenvalue
070 variance explained
I
II
Ill
Mean
SD
.09
.24
.15
.16
.01
.04
-.ll
.81
.76
.74
.69
.08
.68
.20
-.04
.45
.55
.72
.20
.17
.16
.02
-.03
.62
.24
.65
.47
.26
.28
-.18
.70
-.09
.03
.24
.43
-.02
0.53
0.48
1.92
-1.42
1.27
0.68
192
4.98
4.34
4.95
4.52
2.33
0.50
0.30
0.64
0.38
0.44
0.47
183
0.84
1.09
0.64
0.84
0.85
3.17
26.4
1.68
14.0
1.18
9.8
A t t a c h m e n t to Place
535
gardening; local commitment, reflected in shared organizational membership;
and local knowledge reflected in knowing the neighborhood name. The
modest loading (.45) of the reliance scale on this component, and the fact
that the organizational membership variable reflects shared associations, suggests a certain sociality to this dimension of attachment. The high loading
(.62) of gardening in back on this component also reflects investment in locale,
as does homeownership. Gardening as compared to homeownership is a more
behavioral and labile measure of investment. What was o b s e r v e d - levels of
gardening-directly reflects time and effort spent on improvement of the
residence.
Although the structure of Component II is straightforward, the labeling of it is less so. The high loading of homeownership is reminiscent of Riger
and Lavrakas' (1981) Rooted factor on which homeownership had the highest
loading (.66). But at the same time our Component II is more than that.
The high loadings of organizational membership and neighborhood
knowledge suggest that the person who scores high on this component is not
only settled, as evidenced by homeownership, but is also involved in local
collective efforts to improve or maintain the locale.
This covariation of rootedness and involvement is not a complete surprise. Riger and Lavrakas (1981, Table IV) present evidence supportive of
this association. By combining cells from their Table IV, and comparing highversus low-rooted (unweighted) means, we see that rooted persons are more
involved in that they are more likely to regularly read a local paper (62 vs.
4707o) and somewhat more likely to belong to local groups (25 vs. 19070). Thus,
their data provide the suggestion, made clearer by our Component II, of the
association of rootedness and local involvement. We label our Component
lI a Rooted and Involved dimension.
Component Ill picks up two variables reflecting on-block social
climate- proportion of addresses where respondent knows someone, and perceived similarity. The low loading of the near-home access control scale also
reflects social climate. Thus the component appears to be a simple social
dimension, quite similar to the bondedness factor presented by Riger and
Lavrakas (1981).
Nonetheless, there are differences. First, the social variables on Component III reflect low-intimacy social ties, or what Granovetter (1973) has
called "weak" ties. No trust, or cooperation, or shared interest is implied.
These ties are weak in that they do not require high levels of interaction or
trust. Second, the length of residence item has a strong loading on this component, suggesting that a time element is involved in the accretion of these
local ties. Time is involved as a prerequisite for becoming familiar with a
large number of physically proximate others. Perhaps Component III is best
interpreted simply as Local Ties, or a Local Acquaintanceship factor.
Taylor, Gottfredson, and Brower
536
Table 1I. Principal Components Analysis of Problems
Component
Variable
I
II
(Q22) Troublemakers hanging around
(Q23) Neighbors not getting along
(Q24) Excessive drinking in public
(Q25) People who say insulting things
(Q26) Bad elements moving in
(Q27) People who are unpredictable
(Q28) Crime or fear of crime
(Q29) Kids or adults trespassing in yards
(Q30) People fighting
.46
.70
.54
.62
,62
.72
.28
. 12
.67
•08
(Q31) People damaging cars, property
of others
(Q32) Public drug use
Eigenvalue
°70 variance explained
Mean
SD
.52
.02
.44
.29
.23
.18
,64
.68
,22
.79
1.58
1,21
1.53
1.28
1.28
1.28
1.61
1.42
1.28
1.43
0.76
0,50
0,76
0.59
0.56
0.57
0.73
0.67
0.57
0,65
.48
.53
1.47
0.74
4.47
40.6
1.07
9,7
Principal Components Analysis of Problems
In our principal components analysis of problems two components with
eigenvalues > 1.0 were extracted and rotated to a varimax solution. The results
appear in Table II. Component I reflects local conflicts. The items with the
highest loadings mention direct conflicts such as neighbors not getting along
(.70), people fighting (.67), and people who say insulting things or bother
people (.62). These conflicts are associated with a certain perception of chaos,
or at least change as reflected in the high loadings of problems such as bad
elements moving into the neighborhood (.62), and people who are unpredictable and would do just about anything (.72). Thus, in the eyes of residents,
conflict, and a change from the way things used to be are associated. We
simply label this a Conflict dimension.
Component II reflects minor crimes such as trespassing (.68), vandalism (.79), public drug use (.53), and concern about crime (.64). "Troublemakers hanging around" also loads on this component, probably because they
are viewed as the perpetrators of these minor but annoying offenses (.52).
We label this a Minor Crime factor.
Regression Predicting Rooted and Involved Dimension
of Attachment
The results of our hierarchical regression predicting the Rooted and
Involved attachment dimension appear in Table III. The regression explains
almost 40% of the variation in the outcome.
Attachment to Place
537
Table Ill. Regression Predicting Attachment: Rooted and Bonded Dimension ~
Cluster
Variable
Demographics
Race
Income
Education
Racial mix
Income mix
Type 1 (low
income, rent)
Type 2 (mixed)
Conflict
Minor crime
Block diversity
Neighborhood context
Perceived problems
R 2 increment
.192
.094
.024
.009
.006
.011
(p
(p
(p
(p
(p
(p
<
<
<
<
<
<
Final/3
Final t value
.001)
.001)
.001)
.01)
.05)
.01)
-.172
.166
.182
-.082
-.063
-.369
-4.25
4.25
4.81
- 1.98
-1.87
-6.42
(p
(p
(p
(p
(p
(p
<
<
<
<
<
<
.001)
.001)
.001)
.05)
.10)
.0011
.040 (p < .001)
.005 (p < .05)
.005 (p < .05)
-.286
-.072
.069
-6.37 (p < .001)
-2.19 (/9 < .05)
2.12 (p < .05)
Total adjusted R 2 = .376 (F(9, 605) = 40.51; p < .001)
"Listwise deletion used (n = 615).
All three individual-level demographic variables entered. Whites, higher
income persons, and better educated persons were all more attached. Th~ese
results agree with Gerson et al.'s (1977) on social class. Our finding of a race
effect is the first reporting of such. (Riger & Lavrakas, 1981, reported blacks
being higher on a social attachment dimension but did not report race affecting Rootedness.) Further, all three of these class variables remain significant in the final equation, even after context factors have been entered.
Thus, net of type of place, type of person is important.
Two of three possible measures of block diversity merited entry in the
regression. And the impacts of diversity were as expected. Persons living on
blocks that were more racially or economically diverse were less attached.
Thus, net of individual characteristics, block diversity is associated with less
attachment.
Further, in addition to block heterogeneity, neighborhood heterogeneity
was important as well. Both neighborhood dummy variables merited entry
and remained significant in the final equation. In fact, in the final equation
these were the two most powerful predictors. Since these are dummy variables, and since they both entered, in the regression they tell us how respondents differed in the dummy neighborhood compared to the reference string,
which was Type 3 (middle income, predominantly home-owned) neighborhoods. Persons living in lower income, predominantly rental neighborhoods
(Type 1) were less attached than those in Type 3 neighborhoods; and persons in neighborhoods with a mix of owners and renters (Type 2) were less
attached than those in Type 3 neighborhoods. Thus, at the neighborhood
level, lower social class, and diversity, are both associated with lower levels
of attachment.
Entered on the final step, the two perceived problem dimensions contributed significantly to explaining attachment. Persons perceiving more con-
538
Taylor, GoUfredson, and Brower
Table IV. Regression Predicting Attachment: Local Ties Dimension a
Cluster
Variable
Demographics
Education
Income
Racial mix
Educational mix
Block diversity
R2
increment
.042 (p
.009 (p
.050(,o
.017 (p
<
<
<
<
.001)
.05)
.001)
.001)
Final ~
Final t value
-.133
-.038
-.218
.134
-3.02 (t7 < .001)
<1
-5.49 (p < .001)
3.46 (/7 < .001)
Total adjusted R 2 = .113 (F(4, 610) = 19.43; p < .001)
"Listwise deletion used (n = 615).
flict, as expected, were less attached. Contrary to expectation, however, those
perceiving more minor crime problems were more attached.
Thus, the rooted and involved dimension was influenced, as predicted, by block and neighborhood heterogeneity. Further, net of diversity, perceived disorder, in the form of conflict, was associated with lower attachment
levels.
Regression Predicting Acquaintanceship Dimension
of Attachment
The results of the hierarchical regression predicting the Acquaintanceship dimension of attachment appear in Table IV. Two types of measures
entered: individual demographics and block diversity indices. Persons with
less education and less income were stronger on this attachment dimension,
although only the education variable remained significant in the final equation. Our finding of stronger local social ties among lower social class respondents is not new (Gerson et al., 1977; Wellman & Leighton, 1979). Of the
three possible block diversity measures, two merited entry and both remained
significant in the final equation. Residents living on blocks that were more
racially diverse reported lower levels of social attachment. This was as
hypothesized. But those living on more educationally diverse blocks reported higher levels of social attachment. Inasmuch as racial diversity is more
Table V. Regression Predicting Territorial Component"
Cluster
Variable
Demographics
Disorder
Income
Social nuisance
calls
R 2 increment
Final ;3
Final t value
.028 (t7 < .001)
.145
3.61 (p < .001)
.015 (17 < .01)
Total adjusted R 2 = .04 (F(2, 612) = 12.81; p < .001)
"Listwise deletion used (n = 615).
-.125
-3.11 (p < .01)
Attachment to Place
539
immediately apparent than educational diversity, this may explain the former's dampening influence. No measures of neighborhood diversity, or perceived or objective disorder, merited entry in the regression.
Regression Predicting Territoriality Component
The results of our hierarchical regression predicting scores on the territoriality dimension appear in Table V. Although the results are not powerful, two variables merit entry as predictors. The pattern of findings is
interesting when considered in contrast to the regressions predicting the attachment components. Only one individual-level demographic variable merited entry. Persons with higher income levels scored higher on territorial
functioning. Education and race had no impact on this outcome. No diversity measures at either the block or neighborhood level entered, nor did any
measures of perceived problems. Our objective measure of disorder, police
calls for social nuisances, did enter. Territorial functioning was stronger (more
responsibility, better access control) among those individuals living on blocks
with fewer calls to police.
DISCUSSION
All three proposed hypotheses have been supported by the results. Most
clearly supported has been our suggestion that attachment to place was discriminable from territorial functioning. In support of this notion we saw that
the territorial cognitions assessed loaded on their own separate factor. As
even further support of their discriminability we saw that the predictors investigated worked differently for attachment and territoriality. Diversity
measures were relevant to attachment but not to the prediction of territorial
functioning. Further, objective measures of disorder were relevant to territoriality but not to attachment. The fact that we were able to separate attachment from territoriality measures, and the fact that predictor variables
worked differently for the two, strongly supports the notion that the two
concepts, albeit related, are distinct. Although the final word on the relation between the two concepts must await studies with more complete measures of both, at present the best course appears to be to pursue attachment
to place as an independent arena of investigation.
Supported almost as clearly as the above hypothesis was the hypothesis concerning impacts of diversity on attachment. For both attachment
dimensions, racial diversity at the block level was associated with lower levels of attachment. For the Rooted and Involved dimension, areal diversity
540
Taylor, Gottfredson, and Brower
was also associated with lower attachment levels. The impact of racial diversity at the block level confirms Gerson et al.'s (1977) suggestive but indirect
finding. Our study, in contrast to theirs, provides more straightforward measures of diversity. Not only is our measure of heterogeneity less oblique but
we have also observed the impacts of a dissonant context in a much more
localized arena, i.e., the street block. We suggested earlier that racial heterogeneity may interfere with attachment because there is confusion about the
appropriate norms to follow. Our data, of course, do not indicate specifically if this confusion regarding norms flows out of diversity, although earlier
work in other contexts (e.g., schools; Rosenberg, 1975) strongly suggests that
this is the case. Longitudinal and perhaps qualitative investigations of street
block life are needed to clarify further the underlying dynamics that have
been suggested by our results.
Our finding that educational diversity enhanced the Acquaintanceship
attachment score was not predicted and is somewhat puzzling. Our best guess
at this point is that educational diversity, which correlated .342 with the Type
1 (lower income, predominently rental) neighborhoods, was entering as a
contextual proxy for lower social class.
Areal diversity was relevant as well. In the Rooted and Involved regression the dummy variable for mixed (Type 2) neighborhoods that were diverse
on tenure status and income was associated with lower levels of attachment.
The inclusion of this variable along with the block diversity measures confirms that attachment is influenced by factors at multiple levels; that both
neighborhood level attributes and block level qualities can "feed down" to
impact person-place bonds. As Taylor (1981) has discussed, such causal
chains can be complex; there are many pathways via which contextual factors can filter down to influence individuals. An important question is whether
positive block attributes that might enhance attachment can counterbalance
negative neighborhood features that are detrimental to attachment. The final betas in Table III suggest that areal influences predominate. But, our
results here should by no means be taken as final.
Our hypothesis about the impacts of disorder received some but not
unequivocal support. In the Rooted and Involved regression perceived conflicts were, as expected, associated with lower attachment. Thus, net of block
and neighborhood heterogeneity, perceived disorder can interfere with positive people-place involvements. Diversity and perceived disorder did overlap to a degree. Stepping through the regression, when the block diversity
variables entered, the partial correlation of the Conflict dimension with the
outcome was reduced. But the reduction was not sizable (about .02). Thus,
other factors besides diversity contribute to perceived disorder. What these
forces are, of course, is an important question to be pursued. If some persons perceive a degree of local problems greater than seem warranted by the
541
Attachment to Place
diversity of their local setting, this suggests that the dissonant context cannot be blamed as the source for all local disorder.
It has been demonstrated that low levels of attachment are associated
with low levels of health and well-being (Stokols, Shumaker, & Martinez,
1983) and may be associated with low levels of mental health (Shumaker &
Taylor, 1983). Thus, it is important from a community mental health perspective to propose interventions that may help bolster levels of attachment.
Although attachment, to a considerable extent, is a function of time passing
and increasing familiarity with a location, it may be enhanced by small-scale
social groups such as block clubs. It appears that these can enhance levels
of informal neighboring (Unger & Wandersman, 1983). Such involvements
may also enhance the extent to which individuals feel rooted in and involved
with their immediate locale.
The results of the present study are limited. They come from one investigation, carried out in one city, at one point in time, which used a particular type of sampling strategy. Nonetheless, offsetting these limits are several
strengths. First, our investigation is multimethod. In addition to survey data
we used census, police, and physical assessment data. Second, throughout
we have emphasized scale-building techniques to produce consistent indices.
And finally, through the use of orthogonal rotations we have analyzed outcomes that are unconfounded with each other. The extent to which these
strengths "overcome" the limits of our study and allow our results to be replicated elsewhere remains, as it must, a purely empirical question.
In summary, the present research has demonstrated the discriminant
validity of attachment to place vis-~.-vis human territorial functioning. The
two concepts are isolable and have different patterns of predictors. We have
also observed that street block and neighborhood heterogeneity, and perceptions of disorder, are associated with lower levels of attachment. The
results give us some clues about why people can develop positive affective
bonds more easily in some settings. Some places are more attachable to than
others.
REFERENCES
Campbell, D., & Fiske, D. (1959). Convergent and discriminant validation by the multitraitmultimethod matrix. Psychological Bulletin, 56, 81-105.
Gans, H. (1968). People and plans. New York: Basic Books.
Gerson, K., Stueve, S., & Fischer, C. (1977). Attachment to place. In C. Fischer et al. (Eds.), Networks
and places: Social relations in the urban setting (pp. 139-161). New York: Free Press.
Gordon, R. (1968). On the interpretation of oblique factors. American Sociological Review,
33, 601-620.
542
Taylor, Gottfredson, and Brower
Granovetter, M. (1973). The strength of weak ties. American Journal o f Sociology, 78, 1360-1380.
Hawkes, J. (1951). A land. London: Cresset Press.
Hunter, A. (1974). Symbolic communities. Chicago: University of Chicago Press.
Kasarda, J. P., & Janowitz, M. (1974). Community attachment in mass society. American Sociological Review, 39, 328-339.
Peachey, P. (1981). Households and the territorial dispersion o f human activity systems. Paper
presented at the annual meeting of the American Sociological Association, Toronto.
Proshansky, H. M. (1978). The city and self-identity. Environment and Behavior, 10, 142-169.
Riger, S., & Lavrakas, P. J. (1981). Community ties: Patterns of attachment and interaction
in urban neighborhoods. American Journal o f Community Psychology, 9, 55-66.
Relph, E. (1976). Place and placelessness. London: Pion Limited.
Rosenberg, M. (1972). Society and the adolescent self-image. Princeton, N J: Princeton University
Press.
Rosenberg, M. (1975). The dissonant context and the adolescent self-image. In S. E. Dragastin
& G. H. Elder (Eds.), Adolescence in the life cycle. New York: Halsted.
Shumaker, S. A., & Taylor, R. B. (1983). Toward a clarification of people-place relationships:
A model of attachment to place. In N. R. Feimer & E. S. Geller (Eds.), Environmental
psychology: Directions and perspectives (pp. 219-256). New York: Praeger.
Stokols, D., & Shumaker, S. A. (1981). People in places: A transactional view of settings. In
J. Harvey (Ed.), Cognition, social behavior, and the environment (pp. 441-488). Hillsdale, N J: Erlbaum.
Sundstrom, E. (1977). The physical environment and social behavior. In L. Wrightsman (Ed.),
Social psychology (pp. 409-431). Monterey: Brooks/Cole.
Taylor, R. B. (1978). Human territoriality: A review and a moel or future research. Cornell
Journal o f Social Relations, 13, 125-151.
Taylor, R. B. (1981). The neighborhood physical environment and stress. In G. W. Evans (Ed.),
Environmental stress (pp. 286-324). New York: Cambridge University Press.
Taylor, R. B. (in press). Toward an environmental psychology of disorder. In D. Stokols &
I. Airman (Eds.), Handbook o f environmental psychology. New York: Wiley.
Taylor, R. B., Gottfredson, S. D., & Brower, S. (1981). Territorial cognitions and social climate in urban neighborhoods. Basic and Applied Social Psychology, 2, 289-303.
Taylor, R. B., Gottfredson, S. D., & Brower, S. (1984). Understanding block crime and
fear. Journal o f Research in Crime and Delinquency, 21, 303-331.
Taylor, R. B., Gottfredson, S. D., Brower, S., Drain, W., & Dockett, F. (1980). Toward a
resident-based model of community crime prevention: Urban territoriality, social networks, and design. JSAS Catalog o f Selected Documents in Psychology, 10, pp. 29-30.
(Ms. 2044)
Unger, D., & Wandersman, A. (1983). Neighboring and its role in block organizations: An
exploratory report. American Journal o f Community Psychology, ll, 291-300.
Wellman,B., & Leighton,B. (1979). Neighborhoods, networks, and communities: Approaches
to the community question. Urban Affairs Quarterly, 14, 363-390.
Wilson, J. O. (1975). Thinking about crime. New York: Basic.