Community Perception: The Ability to Assess the Safety of Unfamiliar Neighborhoods and Respond Adaptively

Abstract

When entering an unfamiliar neighborhood, adaptive social decisions are dependent on an accurate assessment of the local safety. Studies of cities have shown that the maintenance of physical structures is correlated with the strength of ties between neighbors, which in turn is responsible for the crime level. Thus it should be theoretically possible to intuit neighborhood safety through the physical structures alone. Here we test whether people have this capacity for judging urban neighborhoods with 3 studies in which individuals observed photographs of unfamiliar neighborhoods in Binghamton, New York. Each study was facilitated by data collected during previous studies performed by the Binghamton Neighborhood Project studies. In the 1st study, observer ratings on neighborhood social quality agreed highly with reports by those living there. In the 2nd, a separate sample of participants played an economic game with adolescent residents from pictured neighborhoods. Players exhibited a lower level of trust toward adolescents from neighborhoods whose residents report lesser social quality. In the 3rd study, the maintenance of physical structures and the presence of businesses explained nearly all variation between neighborhoods in observer ratings (89%), whereas the specific features influencing play in Study 2 remained inconclusive. These and other results suggest that people use the general upkeep of physical structures when making wholesale judgments of neighborhoods, reflecting a adaptation for group living that has strong implications for the role of upkeep in urban environments.

Full text

1
Community Perception:
The Ability to Assess the Safety of Unfamiliar Neighborhoods and
Respond Adaptively
Daniel Tumminelli O’Brien
Department of Biology, Binghamton University, Binghamton, New York
David Sloan Wilson
Departments of Biology and Anthropology, Binghamton University, Binghamton, New
York
Manuscript+Provisionally+Accepted+at+Journal(of(Personality(and(Social(
Psychology
Please+Do+Not+Cite+or+Distribute.+Comments+Welcome.
Corresponding author:
D. T. O’Brien
Dobrien1@binghamton.edu
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

2
Abstract: When entering an unfamiliar neighborhood, adaptive social decisions are
dependent on an accurate assessment of the local safety. Studies of cities have shown that
the maintenance of physical structures is correlated with the strength of ties between
neighbors, which in turn is responsible for the crime level. Thus, it should be
theoretically possible to intuit neighborhood safety through the physical structures alone.
Here we test if people have this capacity for judging urban neighborhoods with three
studies in which individuals observed photographs of unfamiliar neighborhoods in
Binghamton, NY. Each study was facilitated by data collected during previous studies
performed by the Binghamton Neighborhood Project (BNP) studies. In the first study,
subject ratings on neighborhood social quality agreed highly with reports by those living
there. In the second, a separate sample of subjects played an economic game with
adolescent residents from pictured neighborhoods. Players exhibited a lower level of trust
towards adolescents from neighborhoods whose residents report lesser social quality. In
the final study, the maintenance of physical structures and the presence of businesses
explained nearly all variation between neighborhoods in subject ratings (89%), while the
specific features influencing play in Study 2 remained inconclusive. These and other
results suggest that people use the general upkeep of physical structures when making
wholesale judgments of neighborhoods, reflecting a adaptation for group living that has
strong implications for the role of upkeep in urban environments.
Keywords: Urban social behavior; Prosociality; Evolutionary psychology;
Disorder theory; Environment perception
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

3
During social interaction, people condition their behavior on a variety of signals
provided by the other individual, including facial features, posture, or cultural cues. In
public places, interactions can be unpredictable, but the specific locale could provide
additional information regarding those that might be expected. For example, the
neighborhoods of a city vary in their level of safety, and it would be adaptive for an
individual to use environmental cues to inform her social predisposition, being vigilant in
a potentially dangerous neighborhood, and relaxed otherwise. Prior knowledge would be
helpful in formulating this response, but one entering an unfamiliar neighborhood would
be dependent on the information provided by indirect signals.
Research on people’s perceptions of novel environments—urban or otherwise—
has identified two major factors that make scenes more appealing: the ability to identify
and understand the scene; and the curiosity and exploratory behavior it inspires (Kaplan,
1992). Kaplan takes an evolutionary approach, describing these preferences as
adaptations for an early human lifestyle, helping individuals to take appropriate paths
when hunting and gathering. This research has not considered, however, that group living
may also have exerted an additional selection pressure on environment perception,
leading it to evolve a socially-oriented function similar to person perception. Just as an
individual’s appearance and mannerisms can signal aspects of their personality and
quality as a social partner, the appearance of the streets, buildings and open areas of a
neighborhood reflect the treatment they receive from those who most often utilize them,
and indicate the quality of the social environment. When entering an area inhabited by
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

4
others, individuals able to use such cues to make inferences about a community would be
able to prepare appropriately for the types of interactions that can be expected there.
We refer to this proposed integration of environment and person perceptions as
community perception, and the studies presented in this paper aim to demonstrate its
existence, as well as details of its function. We approach this topic with an evolutionary
perspective, which is to say the focus is organized around two central ideas: the
environmental conditions that place selection pressures on a trait and its function
(ultimate mechanisms); and the specific manner in which the resultant trait operates
(proximate mechanisms). In a culture like our modern cities, distinct neighborhoods sit in
close proximity to each other, and a resident may travel through multiple communities in
a single day. If these communities vary in their safety level, there might be a selection
pressure favoring those who have a capacity for community perception and use it to
condition their social attitudes. To develop a more nuanced set of hypotheses regarding
the trait’s proximate mechanisms and how they interact with community variation, we
turn to the extant literatures on person perception and urban criminology.
Personality Perception: Using the available cues
Humans have the tendency to quickly judge others on a variety of attributes.
Following the “thin slices paradigm” (Ambady, Bernieri, & Richeson, 2000), the
assessment of many traits, including intelligence (Borkenau, Mauer, Riemann, Spinath, &
Angleitner, 2004), sociosexuality (Gangestad, Simpson, DiGeronismo, & Biek, 1992),
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

5
psychological disorders (Oltmanns, Friedman, Fiedler, & Turkheimer, 2004), and
prosociality (Fechtenhauer, Groothuis, & Pradel, 2010; Verplaetse, Vanneste, &
Braeckman, 2007), require only a brief video lasting well under a minute. Depending on
the personality trait, information as limited as a single photo of a person’s face can be
adequate to make an accurate judgment; what is necessary is that attributes correlated
with the behavior in question be made available. For example, Gallup & Wilson (2009)
found that BMI (body-mass index) was a reliable predictor of the level of intrasexual
aggression perpetrated by high school girls. Similarly, independent raters appeared to use
BMI when estimating the aggressiveness of teenage girls in yearbook photos. Agreement
between raters was considerable.
Even if a single photograph does not contain reliable indicators of the behavior in
question, individuals still attempt to make such judgments. This has become particularly
apparent in the literature on cheater (or, conversely, cooperator) detection, in which
subjects are asked to rate the trustworthiness or prosociality (i.e. tendency towards
positive social behavior) of strangers. Individuals are able to discern cheaters from
cooperators after an extended interaction (Brosig, 2002), or a short video (Fechtenhauer,
et al., 2010). Studies providing raters with only a photo, however, find that predictions
are no more accurate than chance. Only one study has violated this rule, and its protocol
is distinct in that the photo was taken at the moment the pictured individual was choosing
whether to cooperate or defect in an anonymous experimental economic game
(Verplaetse, et al., 2007). This suggests that neutral facial features do not signal
prosociality, but that the expressions produced in a social context can. Despite this
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

6
tendency towards inaccuracy, inter-rater agreement is high in all such studies, implying
that raters are relying on a specific set of facial characteristics that, while not correlated
with prosociality, may be informative in some other way. Further research has
demonstrated that these ratings are based on morphometrics that are associated with
anger (Engell, Haxby, & Todorov, 2007; Oosterhof & Todorov, 2008). While not
accurately performing the task at hand, people have responded to cues that may be
independently valuable in a social interaction.
The cheater detection literature is particularly informative in two ways. First,
group living has selected for the ability to quickly judge a stranger’s quality as a social
partner. It would seem reasonable that large-scale society—literally groups of groups—
would promote the evolution of a trait that makes similar judgments about unfamiliar
communities. Second, the methodological breadth of the cheater detection literature
provides a nuanced example of how humans might create prejudgments of individuals,
and we posit that community perception operates similarly. This requires that a
neighborhood’s structures display cues that signal its safety, or trustworthiness. Without
such cues, observers will be unable to accurately assess the safety of a neighborhood,
despite the interest they may have in doing so. In this case, one may make such
judgments inaccurately, relying on features that are informative in other ways, much as
people interpret a seemingly angry facial expression as signaling a tendency to cheat.
Disorganization and Disorder: The cause and symptom of unsafe neighborhoods
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

7
Disorder theory (J. Q. Wilson & Kelling, 1982), also known as the “broken
windows” theory, posits that neighborhoods readily display their safety. Social disorder
(e.g. public alcohol consumption) and physical disorder (e.g. overgrown vegetation)
result when the local residents can not or do not govern and maintain their community.
These signals become crime attractors as they indicate a safe haven for antisocial (i.e.
delinquent) behavior. Residents tend to respond negatively to disorder, as surveys find
that an individual’s opinion of the neighborhood’s social environment correlates highly
with the amount of disorderly behavior she claims to observe there (Markowitz, Bellair,
Liska, & Liu, 2001; Ross & Jang, 2000; Ross, Mirowsky, & Pribesh, 2001; Sampson &
Raudenbush, 1999). Some have suggested that disorder itself might directly trigger fear,
even if it is not accompanied by actually dangerous events, like assault (see Ross & Jang,
2000 for a review). During the 1980’s, disorder theory became very popular among law
enforcement, and New York, NY took a zero-tolerance approach to policing, hoping to
discourage serious crime by stringently enforcing even the mildest of misdemeanors.
There was a decrease in crime during the intervention, and one interpretation of this data
(Corman & Mocan, 2005; Kelling & Sousa, 2001) claimed that the theory had been
validated. An intriguing reanalysis found, however, that it wasn’t the persecution of
misdemeanors that best predicted the decrease in violent crimes, but merely the police
presence (Harcourt & Ludwig, 2006).
This instead provides support for a competing criminological theory, which states
that social regulation of a neighborhood is necessary to prevent its infiltration by criminal
elements (social disorganization theory; Shaw & McKay, 1942/1969). Incidentally, this
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

8
coincides with the evolutionary claim that cooperative systems must include a form of
enforcement to prevent “cheater” strategies from succeeding (D. S. Wilson & Wilson,
2007). A rigorous test of the relationship between disorder, disorganization and crime in
Chicago found that disorder and crime are each symptoms of a neighborhood’s inability
to govern itself (i.e. collective efficacy; Sampson & Raudenbush, 1999), a finding that
has since been replicated in other locales (e.g. Kawachi, Kennedy, & Wilkinson, 1999).
A neighborhood’s safety, then, is a function of the social system constructed by its
residents, a variable that must be assessed through surveys. As a proxy, the level of
disorder, linked to the same deficiency as serious crime, would be a readily available
indicator of local safety. This signal is present in physical structures in the form of poor
maintenance, ranging from loose garbage, to unkempt vegetation, to damaged windows
and doors.
The following studies test the dual hypotheses that individuals can accurately
assess the safety of an unfamiliar neighborhood, and do so by using indicators of
disorder. In each, subjects respond to photos of unfamiliar neighborhoods in Binghamton,
NY. In the first study, subjects report their opinions of a neighborhood’s social
interactions. We examine the accuracy of these inferences using ratings of the
neighborhood’s quality provided by residents. In the second study, an experimental
economics “game” forces subjects to make a social decision that has a monetary effect
for both themselves and a partner living in the pictured neighborhood. This protocol
includes a behavioral measure of trust, permitting us to measure its variation across
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

9
contexts. In the third, we attempt to determine the specific visible features that best
explain the attitudes and behaviors witnessed in the first two studies.
Study 1
Testing the inter-rater consistency of judgments of different neighborhoods would
be as simple as showing a set of subjects a collection of images from a single city. Testing
the accuracy of these impressions, however, would require pre-existing measures of
social quality for those neighborhoods that subjects are observing. This is available as our
study is part of a larger research program called the Binghamton Neighborhood Project
(BNP), a collaboration between Binghamton University and community groups. The
BNP uses an evolutionary perspective to unify multiple disciplines in the study of social
behavior in urban contexts. In a previous BNP study almost 2000 6th-12th grade students
at Binghamton High School responded to a survey that included questions about the
relationships between one’s neighbors (David S. Wilson, O'Brien, & Sesma, 2009). These
act as a measure of neighborhood social cohesion across the city, a reflection of social
organization. Although the strength of a neighborhood’s social organization correlates
positively with income, we hypothesize that observers will assess a neighborhood’s
quality using disorder, not indicators of income, as it is more proximately associated with
safety.
In addition to providing measures of a community’s social quality, the DAP also
contains a variety of questions about personal behavior, including drug use and self-
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

10
esteem. As one might expect, many positive outcomes for youth correlate with a
neighborhood’s social quality, meaning that such characteristics might be accurately
predicted using indicators of a neighborhood’s social quality. There is an important
distinction between individual tendencies and emergent properties of the community. A
neighborhood’s level of disorder is the result of publicly visible behaviors, like littering
or failing to mow one’s lawn, meaning it reflects the ability of the residential community
to collectively enforce social norms. This says nothing of what the individual residents
might do in private, or in other social contexts. The physical structures are thus unlikely
to include signals that are specific to these individual tendencies, leaving naïve observers
with no reliable information to facilitate predictions about them. In Study 1, we ask
subjects to estimate such characteristics of individuals, in addition to the social quality of
the neighborhood. In order to independently test the accuracy of these two different
assessments, we intentionally use a subset of neighborhoods in which community quality
does not correlate with individual tendencies. We hypothesize that subjects will be unable
to assess the behaviors of individual residents, and will base such judgments on disorder,
conflating their impression of individuals with the quality of the community.
Methods
Collecting Neighborhood Images
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

11
On September 23rd, 2007, digital pictures were taken at 20 semi-randomly-
selected addresses from within the city of Binghamton, NY. The initial list contained 25
randomly-selected addresses, and was pared down to 20 that maximized socioeconomic
and geographic variation. At each of these addresses, four photos were taken: one facing
the address, one looking across the street from the address, and one looking each way
down the street. When placed together, these approximate the visual experience of
standing in the street in front of the address (see Figure 1 for examples). No photo
included images of people.
Experiment
On October 15th and 17th, 2007, 143 (45% male) B.U. undergraduates were shown
the collected images. For each of the twenty neighborhoods, the four pictures were
displayed individually for five seconds each. Subsequently, all four images were
displayed together (as seen in Figure 1) for thirty seconds. During this final thirty
seconds, the subjects were asked to rate the pictured neighborhood on the items in Table
1. Images from the next neighborhood were preceded by a slide noting its order number
(i.e. “Location 4”).
Subjects responded to each item on a 5-point Likert scale (1 = Strongly disagree,
5 = Strongly agree). Five items measure a subject’s impression of two aspects of the
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

12
neighborhood’s social environment: the strength of ties between neighbors (social
cohesion) and their ability to govern the neighborhood (social control). Owing to strong
collinearity (r = .88, p < .001), they were combined to form a single measure of social
quality. Field surveys in neighborhoods find the correlation between these two measures
to be of similar strength in vivo (Sampson, Raudenbush, & Earls, 1997). For each
respondent, scale scores for each neighborhood were calculated by summing the
responses to all items and standardizing so that the lowest (all 1’s) and highest (all 4’s)
possible scores were assigned values of 0 and 100 respectively. Finally, three questions
asked the subject to predict the attitudes of adolescents living in the neighborhood
towards their own well-being, healthy habits, and prosociality (see Table 1). Consisting
of only one item, these measures were left on the 1-5 scale. Correlations between all
rating categories at the response level are available in Table 2.
Before the experiment began, subjects reported how well they knew the
geography Binghamton on a 1-5 Likert scale (1 = Not at all, 5 = Extremely well). They
were instructed to leave blank any address they believed they recognized. Fourteen
addresses had at least one subject not respond (M = 4.3, min. = 0, max. = 32), making the
final total 2794 ratings. This methodology was approved by B.U.’s Human Subjects
Resource Review Committee.
Using Previous BNP Studies
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

13
In May 2006, nearly 2000 Binghamton High School students in grades 6-12
responded to the 58-item Developmental Assets Profile (DAP), developed by Search
Institute (http://www.search-institute.org/) to assess the quality of life in adolescents
(David S. Wilson, et al., 2009). Items from this were selected to form scales measuring
well-being, healthy habits, and prosociality (see Table 3 for DAP scales and associated
Cronbach’s alphas). As can be seen by comparing Tables 1 and 3, subjects of the study at
hand attempted to predict the responses of local adolescents on one item from each of
these scales. Three DAP items reference the social quality of one’s neighborhood (see
Table 3), and those used for photo ratings were crafted to closely resemble them both in
word and spirit. The measures of well-being, healthy habits, prosociality, and social
quality are on 0-100 scales as described above. To avoid confusion when discussing these
scales, (DAP) and (Photo) will denote which measure is being referred to.
We used Census Block Groups (CBGs) to approximate neighborhoods, in part
because that permits the use of Census statistics as independent variables. There are 63
CBGs in Binghamton, each intended to contain approximately 1,000 residents (M = 752,
SD = 228, N = 63). Using the mapping software ArcGIS (v. 9.6), responses to the DAP
were mapped across the city and each student was linked to her CBG of residence. We
calculated neighborhood measures for each scale by averaging across the responses of all
residents in a CBG. In addition, the median income of each CBG was accessed from the
2000 Census. Owing to outliers, the variable was log-transformed before analyses.
Descriptive statistics for each of these variables and correlations between them can be
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

14
seen in Tables 4 and 5, respectively. These are reported both for all 63 CBGs and the
sample containing an address that was photographed.
Analysis
We used ArcGIS to link the twenty addresses and their photo ratings to the
appropriate CBG (each was located in a different CBG), creating a design with responses
nested within neighborhoods. In order to partition the variance associated with
descriptors of raters (first-level; e.g. one’s knowledge of Binghamton) from descriptors of
neighborhoods (second-level; e.g. social quality (DAP)), we used the program
Hierarchical Linear Modeling (HLM) 6.06 (Raudenbush, Bryk, Cheong, Congdon, & du
Toit, 2004) to run multilevel regression models. We chose to use the parameters produced
by unit-specific models, which focus on the variation across second-level units, rather
than those that lean towards testing the population-average of the entire sample. We also
used traditional standard errors (as opposed to robust) as we do not expect responses to a
single neighborhood to be influenced by each other. HLM requires that there be no
missing data, so three responses that left out an item were removed before analyses (final
N = 2491 nested in 20 neighborhoods).
Results
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

15
Rating the Social Environment
An initial model, which was essentially a correlation between mean neighborhood
ratings by residents and observers, found that adolescent residents and photo observers
strongly agreed in their ratings of a neighborhood’s social environment (β = 1.35, effect
size = .73, p < .001; see column 1 of Table 6, or Figure 2 for a visual representation). In
fact, the assessments of the two groups shared nearly 50% of their variation. When other
neighborhood descriptors were added as predictors (individual prosociality (DAP) and
median income), social quality (DAP) continued to be positively and significantly related
to the ratings of photo observers (β = 1.08, effect size = .61, p < .01; see column 2 of
Table 6). Individual prosociality was included because of its theoretical relationship to
community social quality, which is essentially a measure of reciprocal prosociality within
the group. The magnitude of the parameter did decrease, but this is to be expected
considering the collinearity between prosociality, median income and social quality
(DAP) (see Table 5).
The proceeding two models first incorporated one’s knowledge of Binghamton
and then interactions between it and the neighborhood’s social quality (DAP). In each, a
greater knowledge of Binghamton was associated with higher neighborhood ratings (β =
1.28, effect size = .07, p < .001, for latter model; see columns 3 and 4 of Table 6).
Additionally, the significant positive parameter for the interaction effect between one’s
knowledge of Binghamton and the neighborhood’s social quality (DAP) shows that those
more acquainted with the city were better able to assess the social quality of a
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

16
neighborhood, rating better neighborhoods higher and lower quality neighborhoods lower
(β = .09, effect size = .04, p < .05). Although each is statistically significant, their actual
influence on ratings is limited. For example, according to the model in column 4, the
maximum predicted difference in the rating of a neighborhood by someone reporting a
“1” and another reporting a “5” for their knowledge of Binghamton is 4.7 points, which is
only half of a standard deviation in the subset of neighborhoods used here (see Table 4).
Rating Other Features of the Neighborhood
Following the above results, we ran three models, one predicting each of the other
three photo ratings: well-being (Photo), healthy habits (Photo), and prosociality (Photo).
Each model included six variables: the corresponding measure from the DAP, one’s
knowledge of Binghamton, and an interaction between the two; the measure of
neighborhood social quality provided by the DAP and an interaction between that and
one’s knowledge of Binghamton; and median income. The findings can be almost
completely generalized across the three measures (see Table 7):
•People were not capable of accurately assessing the level of well-being,
healthy habits, or prosociality in the local youth, as indicated by the lack
of association between their ratings and the measurements taken from the
DAP.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

17
•Instead, these ratings were based almost completely on those cues being
used to assess the social environment. Social quality (DAP) was the most
accurate predictor of the average response to a neighborhood’s photos for
all three measures. The magnitudes of these relationships were nearly
identical to the one with social quality as rated from photos.
•Although unable to accurately assess these three qualities, the third set of
regressions shows that the subjects did not begin using signals of income
to inform their ratings, but relied throughout on indicators of the social
environment.
•Finally, the results regarding the effect of one’s knowledge of Binghamton
on ratings were mixed. In general, it seemed that those more acquainted
with Binghamton viewed the neighborhoods more favorably, although this
was not the case for ratings of prosociality. Also, it seemed that these
people may assess more accurately, following some set of cues in the
photos; although, as above, the effect sizes are small and may be an
artifact of a large sample size.
Discussion
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

18
Using only pictures of a neighborhood’s physical structures, the subjects were
able to accurately judge the social dynamics of a neighborhood. Nearly 50% of this
variation coincided with the ratings provided by neighborhood residents, showing that
these predictions are not just “better than chance,” but quite consistent with reality.
Further, raters were not responding to cues of affluence, but other, unidentified features in
the photos. It seems subjects utilized these same features when asked to judge the
lifestyles of individual residents. Even ratings of individual prosociality were a reaction
to indicators of the neighborhood’s social quality. Being that community ties arise from
reciprocal prosociality, it appears that the pictures were not useful for assessing
individuals in any way, but only for the social environment they have created collectively.
Finding themselves without the necessary information to rate individuals, subjects fell
back on their impressions of the neighborhood as a whole, a misperception that seems
both efficient and adaptive when entering an unfamiliar neighborhood and preparing to
interact with the environment. It is important to note that the lack of correlation between
the features of a community and individual behaviors was part of our experimental
design, and not necessarily the case in actual cities. This simply serves to demonstrate
that the signals visible in a community’s physical structures are only informative about
collective processes, and do not provide direct indicators of individual tendencies. In the
case that the attributes of the community and its individual residents were correlated,
individuals might make accurate assumptions about individual behavior, but only by
virtue of this association.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

19
There is some suggestion that the ability to interpret the physical structures has a
cultural component, as a greater knowledge of Binghamton led to higher and more
accurate ratings. The effect sizes behind these associations were small, however, and even
individuals with little to no experience with the city were able to make accurate
judgments. The limited cultural variation in the subjects—Binghamton University is a
public university with students hailing primarily from New York State—may have
masked a more extensive cultural variation in community perception.
Study II
When interacting with others, the effectiveness of a social strategy is in great part
dependent on the trustworthiness of one’s partner. Fittingly, people are overwhelmingly
more likely to cooperate with strangers they believe to be prosocial (e.g. Brosig, 2002). It
has been oft proposed that the evolved function of attitudes and emotions is to promote
specific action tendencies that produce behaviors adaptive for the immediate
circumstance (Frijda, 1989; Lazarus, 1991; Levenson, 1994; Tooby & Cosmides, 1990).
In this case, the ability to judge the safety of a neighborhood from its physical structures
is only relevant insofar as it influences one’s approach to social interaction. Here we use a
“game” developed by experimental economists called the Sequential Prisoner’s Dilemma
(SPD) to test the hypothesis that people use the information embedded in the physical
structures of a neighborhood to elect appropriate behaviors. In this game, a subject must
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

20
make decisions that will impact monetary payoffs for herself and a single social partner.
Such protocols are especially effective in that they simplify interactions, creating easily
interpreted measurements of social behavior. Further, the real-world implications
associated with monetary payoffs qualify them as performance-based measures of social
behavior that are less vulnerable to the effects of social desirability than standard surveys.
If subjects condition their behavior on the physical appearance of a neighborhood, their
choices when asked to play with a local resident will demonstrate how they would behave
in such an environment. Again, results from previous BNP studies allow us to assess the
ability of subjects to respond appropriately.
Methods
Using an experimental economics game
In the 2-player SPD, each player can either cooperate or defect, with two
cooperators each receiving a greater amount than two defectors (in this version, $30 vs.
$15), but a mixed interaction resulting in a lower payoff for the cooperator than the
defector ($10 vs. $45). A “first-mover” chooses whether to cooperate or defect, enabling
the second player to choose on the basis of the first player’s decision. A player may have
to choose between cooperation or defection in one of three circumstances. As a first
mover, selecting to cooperate (“offers of cooperation”) is an indicator of trust as the
second mover has the opportunity to exploit cooperation. If chosen as a second mover, a
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

21
player will either respond to a cooperative first player or a non-cooperative first player.
The first is a measure of reciprocation, while the latter is an indicator of “self-sacrifice.”
Players did not know which of the three roles they would fill, and were required to make
decisions given each.
Experiment
On January 29th, 2008, 34 (56% male) B.U. undergraduate and graduate students
were shown the images from nine of the 20 neighborhoods used in Study 1. None of
these individuals were participants in Study 1. Initially, we generated a random selection
of nine neighborhoods, then slightly modified it to expand variation in social quality
across neighborhoods. Again, for each neighborhood, the subjects saw each of the four
images alone for five seconds each, and then all four together (as seen in Figure 1) for
thirty seconds. During this longer period, subjects were asked to play the SPD as if they
were playing with an adolescent living in the neighborhood. They were assured that this
would occur and that real monetary payoffs would be given (see below), but that roles
(i.e. first or second player) were to be assigned randomly, so they must provide responses
for each of the three possible situations (being a first mover, second mover with a
cooperator, and second mover with a defector). Again, subjects rated their knowledge of
Binghamton and were asked not to play with a given neighborhood if they believed they
recognized it. For each neighborhood, at least one subject left responses blank (M = 1.3,
min. = 1, max. = 2). The remaining sample size was N = 294 nested in nine
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

22
neighborhoods, each with three measures—offer of cooperation, reciprocation, and self-
sacrifice—coded as dichotomous variables (“1” = cooperate, “0” = defect). We do not
analyze this third measure as it is very rare that a second mover chooses to cooperate with
a non-cooperator, essentially sacrificing $5 and giving the other individual $30 more. The
other two measures allow us to analyze how trust and reciprocation vary with the
appearance of a social partner’s neighborhood.
Using Previous BNP Studies
In May 2007, students at Binghamton High School played the SPD as part of a
BNP study (O’Brien et al., unpublished data). We utilize these data to simulate the social
interactions that subjects of the current study were told they would take part in, satisfying
the requirement of experimental economics to minimize deception regarding payoffs or
purported social partners. The Binghamton City School District provided the home
address of each participant in the high school study. For the nine addresses whose images
were used in Study 2, one student’s data was chosen to represent the neighborhood. This
individual was the CBG resident living closest to the address itself. His or her responses
were only used to facilitate the experiment at hand, and were not intended as a statistical
measure of the neighborhood. We do not use these data in the proceeding analysis.
At the end of the experiment, subjects were notified that only two individuals,
chosen at random, would play the game for money. One would act as a first mover, the
other as a second mover. One of the nine neighborhoods was selected at random, and the
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

23
responses provided by the high school student from that neighborhood filled the role of
first mover in the SPD. The response sheet of one of the university students was
randomly selected to respond as a second mover. Another university student was chosen
at random to be a first mover. A second neighborhood was selected, and the responses
provided by the high school student representing it acted as the second mover’s response.
Money was distributed to the two university subjects as per the rules of the game. This
procedure was intended to ensure realistic responses on the part of the university students
when playing the game. This methodology was approved by B.U.’s Human Subjects
Resource Review Committee.
Analysis
Again, a nested design was created, with responses to a set of photos linked to the
appropriate address, and, in turn, the associated CBG. HLM was used to test logit
models, owing to the dichotomous nature of the two dependent variables, offering
cooperation as a first mover, and reciprocating cooperation as a second mover. In this
study we used unit-specific models but with robust standard errors, owing to a severe
disagreement between the results given by traditional and robust standard errors. In this
case, it is most appropriate to use the robust standard errors because the distribution of
random errors across second-level units is not normally distributed. We hypothesize that
this is because there are obvious cues that can inform behavior in the photos of
neighborhoods with very high and very low social quality, but those neighborhoods rated
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

24
centrally are less likely to elicit consistent decisions when the outcome variable is
dichotomous, as are behaviors in the SPD. This would skew random effects to correlate
with an independent variable, violating the assumptions of the model. To check the
robustness of these findings, we also ran linear regressions using neighborhood
descriptors to predict the proportion of individuals who offered cooperation or reciprocity
when playing with each neighborhood.
Results
The relative success of offering cooperation as a first-mover is influenced by the
likelihood that it will be met with reciprocation, thus we assume that variation in play is
based on the extent to which subjects trusted residents of each neighborhood. For every
point that a neighborhood’s social quality (DAP) increased, a subject was 5% more likely
to offer cooperation, meaning that the images of these neighborhoods elicited greater trust
(β = .05, o.r. = 1.05, p < .01; see column 1 of Table 8 for whole model and Figure 4 for a
visual representation), regardless of their own age, sex or knowledge of Binghamton.
When median income was added to the model, increases in a neighborhood’s social
quality (DAP) remained predictive of trusting behavior, though the effect size was
diminished (see column 2 of Table 8). This was not surprising owing to the strong
correlation between the two variables in this subset of neighborhoods (r = .61, p < .10).
Median income, however, was a non-significant predictor. Because the assumptions of
HLM were violated, we checked the accuracy of the analysis by replicating this last
model in the form of a standard multiple regression. This model corroborated the results
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

25
from HLM, with a neighborhood’s social quality (DAP) positively predicting the
proportion of individuals exhibiting trust when playing with a particular neighborhood (β
= .73, p < .05). Again, median income was a non-significant predictor.
Secondly, we analyzed the tendency of individuals to reciprocate when playing
with a cooperator. The first model found that people were more likely to reciprocate when
playing with a resident from a pictured neighborhood with higher social quality (DAP)
(see column 3 of Table 8). The inclusion of median income in this model, however,
caused the strength of the parameter to shrink to a level of marginal significance (see
column 4 of Table 8). This was also the case for the standard regression (β = .54, p < .10),
although a stepwise version of the regression found a neighborhood’s social quality
(DAP) to be the main predictor of the proportion of people who reciprocated when
playing with an adolescent from a particular neighborhood (β = .72, p < .05; cut-point α
= .05). Interestingly, the most consistent predictor of reciprocation was one’s knowledge
of Binghamton (β = .33, o.r. = 1.38, p < .01), an increase of one point on this scale being
associated with a 38% greater chance of reciprocating. There was no effect of age or sex
on behavior.
Variation in first-mover behavior (i.e. trust) was more closely associated with
differences in the photos than responses to cooperators. This speaks to the different
considerations that go into each decision, as the first involves an assessment of a partner’s
trustworthiness. As in Study 1, it appears that subjects are using information in the photos
to make such judgments. Although this assessment would not be necessary when
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

26
determining whether to reciprocate or not, we still see a relationship between cooperation
and variation in the pictured neighborhoods.
Discussion
Considerable research has found that setting has a strong influence on social
attitudes, as individuals are more likely to exhibit prosocial behavior when in rural
environments than urban ones (see Steblay, 1987). An earlier BNP study (David S.
Wilson, et al., 2009), however, found similar variation to exist across urban environments
using the “lost-letter method” (Milgram, Mann, & Harter, 1965). Stamped envelopes,
addressed to a specific location, were dropped at randomly-selected locations throughout
the city of Binghamton. Envelopes arriving through the mail represented a small act of
prosociality as they had been picked up by a passer-by and put it in a mailbox. Again,
setting was more responsible for prosociality than individuals, as return rate was
positively associated with a neighborhood’s level of social cohesion, rather than the
generalized level of prosociality reported by individual residents.
Similarly, the subjects in this study did not act in a trusting manner if
neighborhoods appeared unwelcoming, reacting adaptively to signals of the quality of the
social environment. Interestingly, they did not respond to cues of median income, despite
the correlation between the two variables. Interestingly, in the experimental economics
study of Binghamton adolescents described above (O’Brien et al., unpublished data), the
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

27
quality of the local social environment did not directly predict trustworthiness of
residents, although population density, which tends to diminish the ability of a
community to govern itself, was associated with a lower rate of reciprocation. As shown
in Study 1 and mentioned above, one’s reaction to a neighborhood is not effective in
judging the residents, but the quality of the environment they have constructed, and the
dangers it may harbor. In this way, a sense of distrust serves to navigate the general
setting as opposed to each individual therein.
In the case of responding to a cooperator (i.e. reciprocation), logic states that there
is no adaptive purpose in varying behavior across neighborhoods as the social partner has
already acted. However, subjects reciprocated more when interacting with certain
neighborhoods than others, and the primary determinant of this variation again seems to
be the local social quality, although the results are less conclusive in this case. This
further supports the theory proposed in Study 1 that subjects observing the images
experience a generalized emotional response to the pictured neighborhood, judging it in a
wholesale manner and reacting accordingly. One caveat to this interpretation, however, is
that subjects may be fabricating variation in response to the experimental setting itself.
If reliable, these results offer support for the theory of strong reciprocity, which
states that humans have evolved a general tendency for prosocial behavior through
generations of group-based survival (Bowles & Gintis, 2004; Gintis, 2000). This
predisposition for cooperation is attuned to the mutual affinity and norms one shares with
a social partner, regardless of whether future interactions are expected. In this case, we
see reciprocation based on the same criteria as offering cooperation; once one determines
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

28
that a neighborhood’s residents are worthwhile social partners, she sees them as both
trustworthy and worth cooperating with, forgoing the temptation to defect. On the other
hand, neighborhoods that appear to harbor untrustworthy individuals do not merit the
social investment associated with reciprocation. Further evidence for strong reciprocity
comes from the relationship between one’s knowledge of Binghamton and the tendency
to reciprocate. While it is generally accepted that individuals tend to cooperate more
often with family members or close friends, here we see people becoming more likely to
exhibit prosociality towards complete strangers based on nothing more than shared
familiarity with a city and, presumably, its culture.
Study III
Studies 1 and 2 have shown that individuals are able to intuit the quality of a
neighborhood’s social environment by viewing its physical landscape, but give no
indication of how they are doing this. As mentioned above, the bonds between neighbors
are an instrumental resource for the informal governance of a neighborhood, and where
there is limited social organization crime is more likely to occur (Sampson, et al., 1997).
Ironically, a direct measurement of these social relationships would require interaction
with residents—examples of the exact experiences for which a passer-by would need
such information. There is a strong correlation, however, between neighborhood
governance, crime and physical indicators of disorder (Markowitz, et al., 2001; Sampson
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

29
& Raudenbush, 1999). Some claim that these signals directly attracts crime as would-be
criminals recognize an area where delinquent behavior entails little risk of punishment
(Corman & Mocan, 2005; Kelling & Sousa, 2001; J. Q. Wilson & Kelling, 1982).
Residents who perceive their neighborhood as being more disorderly also report
their neighborhood as being more dangerous, regardless of whether they have witnessed
or been a victim of a serious crime (Markowitz, et al., 2001; Ross & Jang, 2000; Ross, et
al., 2001). If physical evidence of poor maintenance influences the attitudes and
behaviors of those living there, it would seem logical that those with less a priori
knowledge would base their own attitudes and behaviors on the same cues. In this study
we attempt to establish which forms of disorder—be they deterioration of the houses, the
pavement, the lawns, or other elements of a neighborhood—and other visible items, like
lawn and house decorations, are the best indicators of the social relationships shared by
residents. In turn, we test the hypothesis that the capacity for community perception seen
in Studies 1 and 2 is a response to visible disorder, and attempt to identify those specific
features in an image that are most responsible for these judgments.
Methods
Seven individuals objectively rated each image on the physical features noted in
Table 9. They rated one hundred images—the eighty that were used in Study 1 and
twenty “dummy” images—in a set order. The dummy images comprised the first twenty
to allow individuals to solidify their rating system before rating those images relevant to
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

30
the study. The other eighty images were randomized in such a way that raters did not
know which images came from the same neighborhood. The raters knew that all images
were randomized, but not about the twenty dummy images.
The items rated were meant to reflect the care invested in a neighborhood’s
physical structures (i.e. the level of physical disorder). Ratings were left blank if non-
applicable for a given image (e.g. rating lawn quality in a picture of a street). Ratings
were consistent for those variables that required some level of subjectivity (e.g. “Are the
exteriors well-painted?”; see Table 9), permitting the averaging of all ratings to create an
image-specific score. These scores were then averaged across the four images taken at
each address to create an address-specific score. As for more objective variables, no
images contained broken windows, “junk” vehicles, or graffiti, and only one contained an
abandoned building, leading us to discard these variables before analysis. Fifteen images
from seven different neighborhoods contained businesses, providing enough variation to
include it as a dichotomous variable in analyses (business; “1” = business visible from
address).
Analysis
Those variables rated on a 1-5 scale had considerable shared variation (see Table
10). To reduce these to simpler factors we ran a principle components analysis (PCA). We
elected this form of extraction over a factor analysis because it analyzes all variation, not
just overlapping variance. This seemed appropriate as the variables are naturalistic
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

31
observations. We assumed correlations between any resultant factors and chose a varimax
rotation. The PCA was run at the image level, not the neighborhood level. This is because
20 neighborhoods would be an inadequate sample size to run a PCA. Instead, each image
included in Study 1 (80 in total; four from each neighborhood) was treated as a single
case. Although the sample size is still small, this can be acceptable when factor loadings
are high, as they are here (Guadagnoli & Velicer, 1988; Sapnas & Zeller, 2002). The
results of the PCA informed the creation of variables that, along with the presence of a
business, acted as neighborhood descriptors in multilevel models predicting subject
responses to neighborhoods in Studies 1 and 2. Each model was run in HLM, and used
the same parameters and standard errors as the corresponding model in the previous
studies (see above).
Results
Quantifying Disorder
As mentioned above, the factor analysis was run at the image level, including all
images used in Study 1 (N = 80). For images that did not contain lawns or driveways, we
imputed ratings by using the average rating for the neighborhood’s other images. The two
resulting components reflect the amount of care invested in house (ratings of paint, grass,
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

32
vegetation, garbage, and decorations) and pavement (ratings of driveway, sidewalk, and
streets) (see Table 11 for complete results). Notably, the appearance of the grass had the
greatest shared variance with the quality of other physical features in an image, the same
feature that most highly correlated with a neighborhood’s social environment (r = .73, p
< .001; see Table 10). This suggests that, when present, the front yard would be the most
efficient signal for one to assess, as it is closely predictive of both general disorder and
social disorganization.
We used the PCA results to create two new address descriptors by summing those
variables that loaded on each factor. Two addresses did not contain driveways in any of
the four images, but did contain streets and sidewalks. In order to create scores for
pavement quality, a regression was run for the other 18 addresses predicting the
relationship between these scores with and without the rating of the driveway included.
This linear equation produced the estimated pavement scores for the other two.
Of the three variables derived from images of a neighborhood, house care
correlated positively (r = .59, p < .01) and the presence of a business negatively (r = -.56,
p < .05) with a neighborhood’s social quality. There was no significant relationship
between the social quality and the pavement (r = .28, p = n.s.). When all three variables
derived from images were entered into a regression predicting the social quality, house
care was the primary significant predictor (β = .44, p < .05) and the presence of a
business only approached significance (β = -.39, p < .10). When a stepwise regression
was run with all physical measurements listed in Table 10, the grass quality was the only
significant predictor (β = .92, p < .01; cut-point α = .05).
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

33
Disorder, Disorganization and Attitudes
The two models in Table 12 use features of the images, neighborhood social
quality (DAP), and an individual’s knowledge of Binghamton to predict the responses
seen in Study 1. In order of effect size, the first model found the maintenance of private
houses and lawns (β = 4.80, effect size = .78, p < .001), the absence of businesses (β =
-5.87, effect size = .47, p < .01), and the maintenance of paved surfaces (β = 2.28, effect
size = .39, p < .05) to all give raters a positive impression of an unfamiliar neighborhood,
accounting for nearly all (89%) of the variation between neighborhoods. These factors
fully mediate the relationship between subject and resident ratings of a neighborhood,
implying that these elements are what the subjects used to inform their judgments.
Although we mentioned above that a quick judgment of the front lawn might suffice for
an accurate impression of the neighborhood, this variable was not significantly associated
with ratings when entered into the second model.
Disorder, Disorganization and Behavior
When attempting to determine the physical cues informing behavior in the
sequential Prisoner’s Dilemma, the first model used above was recast as a logit model as
the outcome variables are dichotomous. Because of the extreme collinearity between a
neighborhood’s social quality (DAP) and house care in the nine neighborhoods used in
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

34
Study 2 (r = .92, p < .001), the former was excluded from analyses as we are primarily
interested in those cues in the images that inform behavior. None of those variables—
house care, the presence of a business, or pavement care—significantly predicted one’s
decisions as a first mover, and only the presence of a business was associated with less
reciprocation (β = -.26, p < .05, o.r. = .77; see Table 13). The relationship between one’s
knowledge of Binghamton and reciprocity was unchanged in the new analyses.
Discussion
The first two studies provided evidence for community perception as a cognitive
mechanism that effectively judges the safety of neighborhoods, and adjusts social
behavior accordingly. Here we find some evidence that individuals base these responses
on a thorough observation of the neighborhood’s physical structures, even those whose
upkeep did not actually correlate with social quality (e.g. the maintenance of paved
surfaces). This was clear in the re-analysis of the data from Study 1, and the null results
in the re-analysis of the behavior in the SPD may also support this interpretation, as no
specific feature was able to significantly predict cooperative behavior. While this is a
limitation of the small number of neighborhoods being compared (N = 9), it also suggests
that no specific feature was predominant in influencing behavior.
Despite being the most informative feature in reality (see above), the quality of
the front lawn was not primarily associated with responses. While this seems non-
adaptive, the same correlation might not be consistent across cities or cultures—
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

35
particularly those in which lawns are uncommon—thus community perception would be
unlikely to have evolved with such a narrow orientation. Instead, it appears that subjects
responded to an array of elements, each reflecting the level of effort invested in the
neighborhood’s upkeep. An alternative hypothesis to this tendency for generalized
assessment, though not mutually exclusive, is that individuals differ in the cues they use
to judge a neighborhood. Such differences may be localized in individuals, or correlated
with one’s cultural history. Regarding the negative reaction to the presence of businesses,
it may be that subjects perceive them to be a detriment to the development of a healthy
community, or, more simply, have an aversion to living beside businesses themselves.
Further research will be necessary to answer these sorts of questions.
General Discussion
A previous study in the field of urban planning demonstrated that residents of a
city consistently favor those neighborhoods whose structures appear orderly and well-
maintained (Nasar, 1990). Meanwhile, criminologists working in a variety of cities have
repeatedly shown that these features correlate strongly with a neighborhood’s social
governance and safety (Corman & Mocan, 2005; Harcourt & Ludwig, 2006; Kawachi, et
al., 1999; Kelling & Sousa, 2001; Markowitz, et al., 2001; Ross & Jang, 2000; Ross, et
al., 2001; Sampson & Raudenbush, 1999). Here we replicate both of these findings, and,
at their intersection, provide evidence that our preferences for well-maintained
neighborhoods are not merely aesthetic, but also serve a social function. When observing
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

36
an urban landscape, subjects formulated judgments about the quality of the local social
environment that were highly accurate. These judgments also produced behaviors that
were appropriate for these social expectations. The information encoded in disorder,
however, was only useful in creating impressions of the local community, and not of
individual residents. Judgments of well-being, healthy habits and prosociality were all
inaccurate, meaning that the attention to disorder serves the explicit purpose of assessing
a community’s social environment. Taken together, these results provide ample evidence
for an adaptation specific to group living. At the present moment, there is much work to
be done on the function of community perception, particularly including individual and
cross-cultural differences in assessing neighborhoods.
The results also have strong applications for the field of sociology, where we feel
it will be useful in defining the interplay between two major theories regarding the causes
of crime. One proposes that it is the disorder of a neighborhood that invites crime
(disorder theory; J. Q. Wilson & Kelling, 1982), the other that a lack of social governance
permits crime to occur (social disorganization theory; Shaw & McKay, 1942/1969).
Although there is greater empirical support for the latter, many survey studies have found
that disorder still weighs heavy in the minds of residents, influencing their perception of
crime and the local community (Markowitz, et al., 2001; Ross & Jang, 2000; Ross, et al.,
2001; Sampson & Raudenbush, 1999). Here we find that naïve observers also respond to
the apparent maintenance of an urban landscape, leading to attitudes and behaviors that
are appropriate to the local social environment. Assuming that this same process is
ongoing in residents of a deteriorating neighborhood, it comes as no surprise that
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

37
researchers have found that individuals who live in such an environment tend to have an
exaggerated sense of mistrust and helplessness (Ross & Jang, 2000; Ross, et al., 2001).
Thus, while the relationships between neighbors are most responsible for the actual crime
rate, our attunement to the level of disorder may be influential in the development of
these relationships. In turn, disorder itself becomes a highly relevant factor when
considering how our built environments influence the attitudes and behaviors of residents
and passer-bys alike.
Acknowledgements
We would like to recognize the assistance of a few people who were instrumental
in the fruition of these studies: Tania O’Brien for her photographic expertise; Jeffrey
Carpenter for advice on experimental economics; Andrew Gallup and Omar Eldakar for
assistance in organizing and running Studies 1 and 2; Mara Linek, Rada Maltseva, Alex
Mand, Adam Scheer, Amanda Schneider, and Victoria Zdorodwski for assistance in
running Study 1; and Louis Alerte, Jeremy Cohen, Salvatore Ingrassia, Charles Norton,
Sanjit Parhar, Thomas St. Pierre, and Nicholas White for coding the physical features of
individual photos. We would also like to thank the National Science Foundation for
funding this research through their support of Binghamton University’s Evolutionary
Studies Program and the Binghamton Neighborhood Project.
References
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

38
Ambady, N., Bernieri, F. J., & Richeson, J. A. (2000). Toward a histology of social
behaviour: Judgment accuracy from thin slices of the behaviora stream. Advances
in Experimental Social Psychology, 32, 201-271.
Borkenau, P., Mauer, N., Riemann, R., Spinath, F. M., & Angleitner, A. (2004). Thin
slices of behaviour as cuse of personality and intelligence. Journal of Personality
and Social Psychology, 86, 599-614.
Bowles, S., & Gintis, H. (2004). The evolution of strong reciprocity: Cooperation in
heterogeneous populations. Theoretical Population Biology, 65, 17-28.
Brosig, J. (2002). Identifying cooperative behavior: Some experimental results in a
prisoner's dilemma game. Journal of Economic Behavior and Organization, 47,
275-290.
Corman, H., & Mocan, N. (2005). Carrots, sticks, and broken windows. Journal of Law
and Economics, 48, 235-266.
Engell, A. D., Haxby, J. V., & Todorov, A. (2007). Implicit trustworthiness decisions:
Automatic coding of face properties in the human amygdala. Journal of Cognitive
Neuroscience, 19(9), 1508-1519.
Fechtenhauer, D., Groothuis, T., & Pradel, J. (2010). Not only states but traits--Humans
can dentiy permanent altruistic dispositions in 20 s. Evolution and Human
Behavior, 31, 80-86.
Frijda, N. H. (1989). Relations among emotion, appraisal, and emotional action readiness.
Journal of Personality and Social Psychology, 57(2), 212-228.
Gallup, A. C., & Wilson, D. S. (2009). Body mass index (BMI) and peer aggression in
adolescent females: An evolutionary perspective. Journal of Social, Evolutionary,
and Cultural Psychology, 3(4), 356-371.
Gangestad, S. W., Simpson, J. A., DiGeronismo, K., & Biek, M. (1992). Differential
accuracy in person perceptin across traits: Examination of a functional hypothesis.
Journal of Personality and Social Psychology, 62(688-698), 688.
Gintis, H. (2000). Strong reciprocity and human sociality. Journal of Theoretical Biology,
206(169-176), 169.
Guadagnoli, E., & Velicer, W. F. (1988). Relation to sample size to the stability of
component patterns. Psychological Bulletin, 103(2), 265-275.
Harcourt, B. E., & Ludwig, J. (2006). Broken windows: New evidence from New York
City and a five-city social experiment. The University of Chicago Law Review, 73,
271-320.
Kaplan, S. (1992). Environmental preference in a knowledge-seeking, knowledge-using
organism. In J. H. Barkow, Cosmides, Leda, Tooby, John (Ed.), The adapted
mind: Evolutionary psychology and the generation of culture (pp. 581-598). New
York, NY: Oxford University Press.
Kawachi, I., Kennedy, B. P., & Wilkinson, R. G. (1999). Crime: Social disorganization
and relative deprivation. Social Science and Medicine, 48, 719-731.
Kelling, G., L., & Sousa, W. H. (2001). Do police matter? An analysis of the impact of
New York City's police reforms. New York: Center for Civic Innovation at the
Manhattan Institute.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

39
Lazarus, R. S. (1991). Emotion and Adaptation. New York: Oxford University Press.
Levenson, R. W. (1994). Human emotion: A functional view. In P. Ekman & R. J.
Davidson (Eds.), The Nature of Emotion (pp. 123-126). New York: Oxford
University Press.
Markowitz, F. E., Bellair, P. E., Liska, A. E., & Liu, J. (2001). Extending social
disorganization theory: Modeling the relationships between cohesion, disorder,
and fear. Criminology, 39(2), 293-320.
Milgram, S., Mann, L., & Harter, S. (1965). The lost-letter technique: A tool of social
research. The Public Opinion Quarterly, 29(3), 437-438.
Nasar, J. L. (1990). The evaluative image of the city. Journal of the American Planning
Association, 56(1), 41-55.
Oltmanns, T. F., Friedman, J., Fiedler, E. R., & Turkheimer, E. (2004). Perceptions of
people with personality disorders based on thing slices of behaviour. Journal of
Research in Personality, 38, 216-229.
Oosterhof, N. N., & Todorov, A. (2008). The functional basis of face evaluation.
Proceedings of the National Academy of Sciences of the United States of America,
105(32), 11087-11092.
Raudenbush, S. W., Bryk, A., Cheong, Y. F., Congdon, R., & du Toit, M. (2004). HLM 6:
Hierarchical Linear and Nonlinear Modeling. Lincolnwood, IL: Scientific
Software International.
Ross, C. E., & Jang, S. J. (2000). Neighborhood disorder, fear, and mistrust: The
buffering role of social ties with neighbors. American Journal of Community
Psychology, 28(4), 401-420.
Ross, C. E., Mirowsky, J., & Pribesh, S. (2001). Powerlessness and the amplification of
threat: Neighborhood disadvantage, disorder, and mistrust. American Sociological
Review, 66(4), 568-591.
Sampson, R. J., & Raudenbush, S. W. (1999). Systematic social observation of public
spaces: A new look at disorder in urban neighborhoods. American Journal of
Sociology, 105(3), 603-651.
Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and violent crime:
A multilevel study of collective efficacy. Science, 277, 918-924.
Sapnas, K. G., & Zeller, R. A. (2002). Minimizing sample size when using exploratory
factor analysis for measurement. Journal of Nursing Measurement, 10(2),
135-154.
Shaw, C., & McKay, H. (1942/1969). Juvenile Delinquency and Urban Areas. Chicago:
University of Chicago Press.
Steblay, N. M. (1987). Helping behavior in urban and rural environments: A meta-
analysis. Psychological Bulletin, 102(3), 346-356.
Tooby, J., & Cosmides, L. (1990). The past explains the present: Emotional adaptations
and the structure of ancestral environments. Ethology and Sociobiology, 11,
375-424.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

40
Verplaetse, J., Vanneste, S., & Braeckman, J. (2007). You can judge a book by its cover:
The sequel. A kernel of truth in predictive cheating detection. Evolution and
Human Behavior, 28, 260-271.
Wilson, D. S., O'Brien, D. T., & Sesma, A. (2009). Human prosociality from an
evolutionary perspective: Variation and correlations at a city-wide scale.
Evolution and Human Behavior, 30, 190-200.
Wilson, D. S., & Wilson, E. O. (2007). Rethinking the theoretical foundation of
sociobiology. Quarterly Review of Biology, 82(4), 327-348.
Wilson, J. Q., & Kelling, G., L. (1982). The police and neighborhood safety: Broken
windows. Atlantic Monthly, 127, 29-38.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

41
Table 1.
Descriptors of Neighborhoods and Their Residents Rated by Subjects, Including
Individual Items and Cronbach’s Alpha Scores.
Social Cohesion
“People around here are willing to help their neighbors.”
“There are adults in this neighborhood that children can look up to.”
α = .837
N = 2
Social Control
“This is a safe neighborhood.”
“If there were a fight in this neighborhood, neighbors would interfere.”
“If children in this neighborhood were skipping school and hanging out on a
street corner, neighbors would take action.”
α = .867
N = 3
Well-Beinga
“I feel good about the future.”
α = —
N = 1
Healthy Habitsa
“I avoid things that are dangerous or unhealthy.”
α = —
N = 1
Prosocialitya
“I am sensitive to the needs and feelings of others.”
α = —
N = 1
a-Preceded by the phrase “How do you think an adolescent living in this neighborhood would answer the
following questions?”
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

42
Table 2.
Correlations between Rating Scales.
1
2
3
4
5
1. Social
Cohesion
—
.88***
.80***
.71***
.75***
2. Social Control
—
.79***
.72***
.73***
3. Well-Being
—
.75***
.75***
4. Healthy
Habits
—
.75***
5. Prosociality
—
Note: N = 2791 nested in 20 neighborhoods
*** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

43
Table 3.
Scales from the Developmental Assets Profile (DAP), Including Individual Items and
Cronbach’s Alpha Scores.
Neighborhood Social Quality
“I have a safe neighborhood.”
“I have good neighbors who help me succeed.”
“I have neighbors who help watch out for me.”
α = .724
N = 3
Prosociality
“I think it is important to help other people.”
“I resolve conflicts without anyone getting hurt.”
“I tell the truth even when it is not easy.”
“I am helping to make my community a better place.”
“I am trying to help solve social problems.”
“I am developing respect for other people.”
“I am sensitive to the needs and feelings of others.”
“I am serving others in my community.”
α = .807
N = 8
Well-Being
“I feel in control of my life and future.”
“I feel good about myself.”
“I feel good about the future.”
“I am developing a sense of purpose in my life.”
α = .723
N = 4
Healthy Habits
“I avoid things that are dangerous or unhealthy.”
“I stay away from tobacco, alcohol, and other drugs.”
“I resist bad influences.”
“I am developing good health habits.”
α = .701
N = 4
Note: N = 1942
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

44
Table 4.
Descriptive Statistics for Neighborhood-Level Variables.
Whole-City
Subset
Mean (SD)
Range
Mean (SD)
Range
1. Social Quality
54.66
(11.23)
11.11-80.56
53.29
(8.07)
36.56-
66.41
2. Well-Being
73.83
(5.49)
58.33-
86.54
73.42
(3.70)
62.12-
78.28
3. Healthy
Habits
70.49
(7.09)
41.67-
88.89
70.89
(4.77)
63.89-
81.82
4. Prosociality
61.12
(6.51)
33.00-
77.93
61.59
(4.45)
54.87-
70.00
5. Median
Income
$29,385
($16,465)
$8,430-
$90,143
$28,018
($11,980)
$12,905-
$59,567
Note: All CBGs are included in panel 1 (N = 63), and only the subset of photographed CBGs are in panel 2
(N = 20).
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

45
Table 5.
Correlations between Neighborhood Descriptors.
1
2
3
4
5
1. Social Quality
—
.36**
.58***
.65***
.76***
2. Well-Being
0.14
—
.51***
.52***
.29*
3. Healthy
Habits
-0.12
-.59**
—
.81***
.45***
4. Prosociality
0.05
-0.27
.46*
—
.47***
5. Median
Incomea
.54*
0.29
-0.11
-0.03
—
Note: Whole-city correlations above diagonal (N = 63) and those in the subset of photographed
CBGs beneath (N = 20).
* - p < .05, ** - p < .01, *** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

46
Table 6.
Multilevel Models Using Respondent (First-Level) and Neighborhood (Second-Level)
Descriptors to Predict Ratings of Neighborhood Social Quality.
Parameter
Size (S.E.)
Parameter
Size (S.E.)
Parameter Size
(S.E.)
Parameter Size
(S.E.)
Individual-Level
Predictors
Knowledge of
Binghamton
—
—
1.31***
(.33)
1.28***
(.03)
Neighborhood-Level
Predictors
Social Qualitya
1.35*** (.32)
1.08**
(.35)
1.08**
(.35)
1.08**
(.35)
Median Incomeb
—
18.87
(15.56)
19.81
(15.54)
19.76
(15.56)
Prosociality
—
.93
(.54)
.93
(.54)
.93
(.54)
Cross-Level
Interactions
Social Quality x
Knowledge
—
—
—
.09*
(.04)
Approximate First-
Level R2
—
—
0.01
0.01
Approximate
Second-Level R2
0.48
0.54
0.54
0.54
Note: N = 2791 nested in 20 neighborhoods
a - Accessed from the 2006 application of the DAP.
b- Log-transformed to maintain normality.
* - p < .05, ** - p < .01, *** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

47
Table 7.
Multilevel Models Using Respondent (First-Level) and Neighborhood (Second-Level)
Descriptors to Predict Ratings of Adolescent Well-Being, Healthy Habits, and
Prosociality.
Well-Being
Healthy
Habits
Prosociality
Parameter Size
(S.E.)
Parameter Size
(S.E.)
Parameter Size
(S.E.)
Individual-Level
Predictors
Knowledge of
Binghamton
.03*
(.015)
.03*
(.016)
.002
(.016)
Neighborhood-Level
Predictors
Well-Beinga
-.04
(.028)
—
—
Healthy Habitsa
—
.01
(.02)
—
Prosocialitya
—
—
.03
(.02)
Social Qualitya
.05**
(.015)
.04*
(.015)
.04***
(.01)
Median Incomeb
.92
(.68)
.72
(.65)
.72
(.65)
Cross-Level
Interactions
Well-Being x
Knowledge
.01**
(.004)
—
—
Healthy Habits x
Knowledge
—
-.005
(.004)
—
Prosociality x
Knowledge
—
—
-.01**
(.004)
Social Quality x
Knowledge
.01*
(.002)
.003
(.002)
.008***
(.002)
Approximate First-
Level R2
~.00
~.00
0.01
Approximate
Second-Level R2
0.50
0.44
0.55
Note: N = 2791 nested in 20 neighborhoods
a - Accessed from the 2006 application of the DAP.
b- Log-transformed to maintain normality.
* - p < .05, ** - p < .01, *** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

48
Table 8.
Multilevel Logit Models Predicting the Likelihood of Exhibiting Trust (Offering
Cooperation) and Reciprocation When Playing the SPD with a Resident of a Pictured
Neighborhood.
Offers of Cooperation (Trust)
Reciprocation
Model 1
Model 2
Model 1
Model 2
Parameter
Size (S.E.)
Odds
Ratio
Parameter
Size (S.E.)
Odds
Ratio
Parameter
Size (S.E.)
Odds
Ratio
Parameter
Size (S.E.)
Odds
Ratio
Individual-Level
Predictors
Knowledge of
Binghamton
-.04
(.10)
0.96
-.04
(.10)
0.96
.32**
(.10)
1.38
.33**
(.10)
1.38
Age
.15
(.09)
1.16
.15
(.09)
1.16
.02
(.1)
1.02
.02
(.1)
1.02
Femalea
.03
(.20)
1.03
.03
(.20)
1.03
.08
(.24)
1.09
.09
(.24)
1.09
Neighborhood-Level
Predictors
Social Qualityb
.05**
(.02)
1.05
.03**
(.01)
1.03
.03***
(.007)
1.03
.02+
(.01)
1.02
Median
Incomec
—
—
.90
(.54)
2.46
—
—
.58
(.40)
1.78
Note: N = 294 nested in 9 neighborhoods
a- Dichotomous variable with “1” equal to variable name.
b - Accessed from the 2006 application of the DAP.
c- Log-transformed to maintain normality.
* - p < .05, ** - p < .01, *** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

49
Table 9.
Items Used to Rate Physical Characteristics of Neighborhood Images and Inter-Rater
Reliabilities (Cronbach’s Alphas).
Are the exteriors well painted?
α = .913
Are the driveways well-cared for?
α = .809
Is the grass appropriately trimmed/mowed?
α = .868
Is other vegetation appropriately trimmed?
α = .899
Are there any “junk” vehicles in the street/driveway?a
α = —
Are the lawns/streets kept clean of garbage?
α = .828
Are there any buildings that look abandoned?a
α = —
Are there any broken windows or doors?b
α = —
Is the sidewalk in good repair?
α = .781
Are the streets cracked or unevenly paved?
α = .955
Is there any graffiti present?b
α = —
Are there any lawn/porch decorations?
α = .807
Are there any businesses?a
α = —
Note: N = 80 images rated by 7 individuals. All were rated on a 1-5 scale unless otherwise noted.
Cronbach’s alphas not calculated for variables with no variance across raters.
a- “1” = yes, “0” = no.
b- Three point scale: “None,” “Some,” “Much.”
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

50
Table 10.
Correlations between Physical Features of Images and Neighborhood Descriptors
(Neighborhood-Level Only) at Image- and Neighborhood-Levels.
1
2
3
4
5
6
7
8
9
10
11
1. Paint
(93/20)
.32
(18)
.80***
(20)
.64**
(20)
.13
(20)
.02
(20)
.02
(20)
.65**
(20)
-.64**
(20)
.50*
(20)
.25
(20)
2. Driveway
.41**
(59)
(60/18)
.58*
(18)
.65**
(18)
.24
(18)
.38
(18)
.43+
(18)
.44+
(18)
.02
(18)
.58*
(18)
.41+
(18)
3. Grass
.52***
(89)
.78***
(58)
(95/20)
.84***
(18)
.33
(20)
.23
(20)
.28
(20)
.66**
(20)
-.54*
(20)
.73***
(20)
.45*
(20)
4. Vegetation
.56***
(92)
.69***
(60)
.76***
(95)
(99/20)
.39+
(20)
.40+
(20)
.31
(20)
.73***
(20)
-.16
(20)
.43+
(20)
.36
(20)
5. Garbage
.38***
(93)
.44**
(60)
.45***
(95)
.43***
(99)
(100/20)
.08
(20)
.40
(20)
.42+
(20)
.14
(20)
.21
(20)
.11
(20)
6. Sidewalk
.17
(88)
.44**
(58)
.38***
(88)
.37***
(92)
.15
(93)
(93/20)
.38
(20)
.27
(20)
.30
(20)
.02
(20)
-.13
(20)
7. Streets
.13
(86)
.33*
(57)
.26*
(88)
.24*
(92)
.25*
(93)
.37***
(90)
(93/20)
.05
(20)
.22
(20)
.27
(20)
-.15
(20)
8. Decorations
.17
(63)
.41**
(39)
.31*
(63)
.49***
(64)
.33**
(65)
.21
(59)
.11
(58)
(65/20)
-.27
(20)
.48*
(20)
.41+
(20)
9. Businessa
-.32**
(93)
.06
(60)
.02
(95)
-.01
(99)
.02
(100)
.21*
(93)
.15
(93)
-.09
(65)
(100/20)
-.56*
(20)
-.34
(20)
10. Social
Quality
—
—
—
—
—
—
—
—
—
—
.55*
(20)
11. Median
Income
—
—
—
—
—
—
—
—
—
—
—
Note: Neighborhood-level above diagonal, image-level beneath. The correlations beneath the diagonal were used in
the PCA. Diagonal contains number of images/neighborhoods that include a score for each item. Correlation boxes
note number of cases with scores for both variables in parentheses. Names of ratings of physical features are
simplified. See Table 9 for more information.
+- p < .1, * - p < .05, ** - p < .01, *** - p < .00
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

51
Table 11.
Individual Variable Loadings and Eigenvalues for Resulting Components of Principal
Components Analysis.
I
II
Comm-
unality
Paint
0.72
—
0.54
Driveway
—
0.60
0.59
Grass
0.71
—
0.71
Vegetation
0.77
—
0.74
Garbage
0.68
—
0.46
Sidewalk
—
0.80
0.64
Streets
—
0.76
0.61
Decorations
0.71
—
0.52
Eigenvalue
3.68
1.11
—
Note: Only loadings above .5 are reported (N = 80).
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

52
Table 12.
Multilevel Models Using Ratings of Visible Features of Images to Predict Ratings of
Neighborhood Social Quality.
Parameter
Size (S.E.)
Parameter
Size (S.E.)
Individual-Level
Predictors
Knowledge of
Binghamton
1.27***
(.33)
1.26***
(.33)
Neighborhood-Level
Predictors
Social Qualitya
.10
(.23)
.05
(.25)
House
4.80***
(.87)
3.99*
(1.75)
Pavement
2.28*
(.97)
2.09+
(1.05)
Business
-5.87**
(1.69)
-5.48*
(1.93)
Grass
—
4.28
(8.13)
Cross-Level
Interactions
House x Knowledge
-.29
(.40)
-.27
(.46)
Pavement x
Knowledge
.30
(.23)
-.49
(.27)
Business x Knowledge
-.31
(.24)
.13
(.50)
Grass x
Knowledge
—
2.90
(2.04)
Approximate First-
Level R2
0.01
0.01
Approximate
Second-Level R2
0.89
0.89
Note: N = 2791 nested in 20 neighborhoods
a- Accessed from the 2006 application of the DAP.
* - p < .05, ** - p < .01, *** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

53
Table 13.
Multilevel Logit Models Using Image Features to Predict Trust (Offers of Cooperation)
and Reciprocation When Playing the SPD with a Resident of a Pictured Neighborhood.
Offers of Cooperation
(Trust)
Reciprocation
Parameter
Size (S.E.)
Odds
Ratio
Parameter
Size (S.E.)
Odds
Ratio
Individual-Level
Predictors
Knowledge of
Binghamton
-.04
(.10)
0.96
.32**
(.10)
1.38
Age
.15
(.09)
1.16
.02
(.10)
1.02
Femalea
.03
(.20)
1.03
.09
(.24)
1.09
Neighborhood-Level
Predictors
House
.03
(.08)
1.03
-.02
(.04)
0.98
Pavement
.20+
(.09)
1.22
.10
(.08)
1.11
Business
-.12
(.13)
0.89
-.26*
(.07)
0.77
Note: N = 294 nested in 9 neighborhoods
a- Dichotomous variable with “1” equal to variable name.
+ - p < .10, * - p < .05, ** - p < .01, *** - p < .001
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

54
Figure 1. Example Images From Two Neighborhoods as Seen by Subjects.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

55
Note: Error bars reflect 95% confidence interval for the mean of each neighborhood’s photo-based ratings.
Figure 2. Relationship Between Social Quality as Reported by Resident Adolescents
and as Rated by Subjects Viewing Photos of the Neighborhood.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods

56
Note: Proportion of those offering cooperation (left panel) and of those reciprocating (right panel).
Figure 3. Relationship Between Social Quality as Reported by Resident Adolescents
and Decisions in the SPD When Playing with a Resident of the Neighborhood.
Community Perception: Assessing the Safety of Unfamiliar Neighborhoods