Salienko Evgenii/Shutterstock

Broken Windows Theory

Reviewed by Psychology Today Staff

The broken windows theory states that visible signs of disorder and misbehavior in an environment encourage further disorder and misbehavior, leading to serious crimes. The principle was developed to explain the decay of neighborhoods, but it is often applied to work and educational environments.

  • What Is the Broken Windows Theory?
  • Do Broken Windows Policies Work?

Syda Productions/Shutterstock

The broken windows theory, defined in 1982 by social scientists James Wilson and George Kelling, drawing on earlier research by Stanford University psychologist Philip Zimbardo, argues that no matter how rich or poor a neighborhood, one broken window would soon lead to many more windows being broken: “One unrepaired broken window is a signal that no one cares, and so breaking more windows costs nothing.” Disorder increases levels of fear among citizens, which leads them to withdraw from the community and decrease participation in informal social control.

The broken windows are a metaphor for any visible sign of disorder in an environment that goes untended. This may include small crimes, acts of vandalism, drunken or disorderly conduct, etc. Being forced to confront minor problems can heavily influence how people feel about their environment, particularly their sense of safety.  

With the help of small civic organizations, lower-income Chicago residents have created over 800 community gardens and urban farms out of burnt buildings and vacant lots. Now, instead of having trouble finding fresh produce, these neighborhoods have become go-to food destinations. This example of the broken windows theory benefits the people by lowering temperatures in overheated cities, increasing socialization, reducing stress , and teaching children about nature.

George L. Kelling and James Q. Wilson popularized the broken windows theory in an article published in the March 1982 issue of The Atlantic . They asserted that vandalism and smaller crimes would normalize larger crimes (although this hypothesis has not been fully supported by subsequent research). They also remarked on how signs of disorder (e.g., a broken window) stirred up feelings of fear in residents and harmed the safety of the neighborhood as a whole.

The broken windows theory was put forth at a time when crime rates were soaring, and it often spurred politicians to advocate policies for increasing policing of petty crimes—fare evasion, public drinking, or graffiti—as a way to prevent, and decrease, major crimes including violence. The theory was notably implemented and popularized by New York City mayor Rudolf Giuliani and his police commissioner, William Bratton. In research reported in 2000, Kelling claimed that broken-windows policing had prevented over 60,000 violent crimes between 1989 and 1998 in New York City, though critics of the theory disagreed.


Although the “Broken Windows” article is one of the most cited in the history of criminology , Kelling contends that it has often been misapplied. The implementation soon escalated to “zero tolerance” policing policies, especially in minority communities. It also led to controversial practices such as “stop and frisk” and an increase in police misconduct complaints.

Most important, research indicates that criminal activity was declining on its own, for a number of demographic and socio-economic reasons, and so credit for the shift could not be firmly attributed to broken-windows policing policies. Experts point out that there is “no support for a simple first-order disorder-crime relationship,” contends Columbia law professor Bernard E. Harcourt. The causes of misbehavior are varied and complex.

The effectiveness of this approach depends on how it is implemented. In 2016, Dr. Charles Branas led an initiative to repair abandoned properties and transform vacant lots into community parks in high-crime neighborhoods in Philadelphia, which subsequently saw a 39% reduction in gun violence. By building “palaces for the people” with these safe and sustainable solutions, neighborhoods can be lifted up, and crime can be reduced.  

When a neighborhood, even a poor one, is well-tended and welcoming, its residents have a greater sense of safety. Building and maintaining social infrastructure—such as public libraries, parks and other green spaces, and active retail corridors—can be a more sustainable option and improve the daily lives of the people who live there.

According to the broken windows theory, disorder (symbolized by a broken window) leads to fear and the potential for increased and more severe crime. Unfortunately, this concept has been misapplied, leading to aggressive and zero-tolerance policing. These policing strategies tend to focus on an increased police presence in troubled communities (especially those with minorities and lower-income residents) and stricter punishments for minor infractions (e.g., marijuana use).  

Zero-tolerance policing metes out predetermined consequences regardless of the severity or context of a crime. Zero-tolerance policies can be harmful in an academic setting, as vulnerable youth (particularly those from minority ethnic/racial backgrounds) find themselves trapped in the School-to-Prison Pipeline for committing minor infractions. 

Aggressive policing practices can sour relationships between police and the community. However, problem-oriented policing—which identifies the specific problems or “broken windows” in a neighborhood and then comes up with proactive responses—can help reduce crime. This evidence-based policing strategy  has been shown to be effective. 

broken window theory health

In reports of police use of force, officers and minority citizens are often portrayed as natural enemies. Here is what we know and some suggestions for a helpful approach.

broken window theory health

Once we've mastered self-rationalization, our "inner weasels" can silence our consciences.

broken window theory health

Part 2: A high murder rate. A cry for help. And I was walking down the street…

broken window theory health

Pace yourself. It all comes down to this month, and it doesn't. It's ongoing, uphill work.

broken window theory health

Lying is always bad except when it's good. Recursion suggests how to find exceptions to moral rules.

broken window theory health

Defund the police is the rallying cry you hear as people feel more resources need to help residents deal with mental health issues. In one city, therapists respond with the police.

broken window theory health

We live in a time of political chaos, terrorism, civil unrest, and economic unpredictability. We must build resilient communities if we are to survive and thrive. Here's how.

  • Find a Therapist
  • Find a Treatment Center
  • Find a Psychiatrist
  • Find a Support Group
  • Find Teletherapy
  • United States
  • Brooklyn, NY
  • Chicago, IL
  • Houston, TX
  • Los Angeles, CA
  • New York, NY
  • Portland, OR
  • San Diego, CA
  • San Francisco, CA
  • Seattle, WA
  • Washington, DC
  • Asperger's
  • Bipolar Disorder
  • Chronic Pain
  • Eating Disorders
  • Passive Aggression
  • Personality
  • Goal Setting
  • Positive Psychology
  • Stopping Smoking
  • Low Sexual Desire
  • Relationships
  • Child Development
  • Therapy Center NEW
  • Diagnosis Dictionary
  • Types of Therapy

September 2023 magazine cover

As the lines between real and fake blur, Americans increasingly chase the idea of authenticity. The first step may be to consider self-knowledge, truthfulness, and other building blocks on the road to personal growth.

  • Coronavirus Disease 2019
  • Affective Forecasting
  • Neuroscience
  • Social Anxiety Disorder
  • Bipolar Disorder
  • Kids Mental Health
  • Therapy Center
  • When To See a Therapist
  • Types of Therapy
  • Best Online Therapy
  • Best Couples Therapy
  • Best Family Therapy
  • Managing Stress
  • Sleep and Dreaming
  • Understanding Emotions
  • Self-Improvement
  • Healthy Relationships
  • Relationships in 2023
  • Student Resources
  • Personality Types
  • Verywell Mind Insights
  • 2023 Verywell Mind 25
  • Mental Health in the Classroom
  • Editorial Process
  • Meet Our Review Board
  • Crisis Support

Broken Windows Theory

How Environment Impacts Behavior

Rachael is a New York-based writer and freelance writer for Verywell Mind, where she leverages her decades of personal experience with and research on mental illness—particularly ADHD and depression—to help readers better understand how their mind works and how to manage their mental health.

broken window theory health

Akeem Marsh, MD, is a board-certified child, adolescent, and adult psychiatrist who has dedicated his career to working with medically underserved communities.

broken window theory health

Verywell / Dennis Madamba

Origins and Explanation

  • Application
  • Impact on Behavior
  • Positive Environments

The broken windows theory was proposed by James Q. Wilson and George Kelling in 1982, arguing that there was a connection between a person’s physical environment and their likelihood of committing a crime.

The theory has been a major influence on modern policing strategies and guided later research in urban sociology and behavioral psychology . But it’s also come under increasing scrutiny and some critics have argued that its application in policing and other contexts has done more harm than good.

The theory is named after an analogy used to explain it. If a window in a building is broken and remains unrepaired for too long, the rest of the windows in that building will eventually be broken, too. According to Wilson and Kelling, that’s because the unrepaired window acts as a signal to people in that neighborhood that they can break windows without fear of consequence because nobody cares enough to stop it or fix it. Eventually, Wilson and Kelling argued, more serious crimes like robbery and violence will flourish.

The idea is that physical signs of neglect and deterioration encourage criminal behavior because they act as a signal that this is a place where disorder is allowed to persist. If no one cares enough to pick up the litter on the sidewalk or repair and reuse abandoned buildings, maybe they won’t care enough to call the police when they see a drug deal or a burglary either.

How Is the Broken Windows Theory Applied?

The theory sparked a wave of “broken windows” or “zero tolerance” policing where law enforcement began cracking down on nonviolent behaviors like loitering, graffiti, or panhandling. By ramping up arrests and citations for perceived disorderly behavior and removing physical signs of disorder from the neighborhood, police hope to create a more orderly environment that discourages more serious crime.

The broken windows theory has been used outside of policing, as well, including in the workplace and in schools. Using a similar zero tolerance approach that disciplines students or employees for minor violations is thought to create more orderly environments that foster learning and productivity .

“By discouraging small acts of misconduct, such as tardiness, minor rule violations, or unprofessional conduct, employers seek to promote a culture of accountability, professionalism, and high performance,” said David Tzall Psy.D., a licensed forensic psychologist and Deputy Director for the Health and Wellness Unit of the NYPD.

Criticism of the Broken Window Theory

While the idea that one broken window leads to many sounds plausible, later research on the topic failed to find a connection. “The theory oversimplifies the causes of crime by focusing primarily on visible signs of disorder,” Tzall said. “It neglects underlying social and economic factors, such as poverty, unemployment, and lack of education, which are known to be important contributors to criminal behavior.”

When researchers account for those underlying factors, the connection between disordered environments and crime rates disappears.

In a report published in 2016, the NYPD itself found that its “quality-of-life” policing—another term for broken windows policing—had no impact on the city’s crime rate. Between 2010 and 2015, the number of “quality-of-life” summons issued by the NYPD for things like open containers, public urination, and riding bicycles on the sidewalk dropped by about 33%.

While the broken windows theory would theorize that serious crimes would spike when the police stopped cracking down on those minor offenses, violent crimes and property crimes actually decreased during that same time period.

“Policing based on broken windows theory has never been shown to work,” said Kimberly Vered Shashoua, LCSW , a therapist who works with marginalized teens and young adults. “Criminalizing unhoused people, low socioeconomic status households, and others who create this type of ‘crime’ doesn't get to the root of the problem,”

Not only have policing efforts that focus on things like graffiti or panhandling failed to have any impact on violent crime, they have often been used to target marginalized communities. “The theory's implementation can lead to biased policing practices as law enforcement officers can concentrate their efforts on low-income neighborhoods or communities predominantly populated by minority groups,” Tzall said.

That biased policing happens, in part, because there’s no objective measure of disordered environments so there’s a lot of room for implicit bias and discrimination to influence decision-making about which neighborhoods to target in crackdowns.

Studies show that neighborhoods where residents are predominantly Black or Latino are perceived as more disorderly and prone to crime than neighborhoods where residents are mostly white, even when police-recorded crime rates and physical signs of physical deterioration in the environment were the same.

Moreover, many of the behaviors that are used by police and researchers as signs of disorder are influenced by racial and class bias . Drinking and hanging out are both legal activities that are viewed as orderly when they happen in private spaces like a home or bar, for example. But those who socialize and drink in parks or on stoops outside their building are viewed as disorderly and charged with loitering and public drunkenness.

The Impact of Physical Environment on Behavior

While the broken windows theory and its application are flawed, the underlying idea that our physical environment can influence our behavior does hold some water. On one hand, “the physical environment conveys social norms that influence our behavior,” Tzall explained. “When we observe others adhering to certain norms in a particular space, we tend to adjust our own behavior to align with them.”

If a person sees litter on the street, they might be more likely to litter themselves, for example. But that doesn’t necessarily mean they’ll make the leap from littering to robbery or violent assault. Moreover, litter can often be a sign that there aren’t enough public trashcans available on the streets for people to throw away food wrappers and other waste while they’re out. In that scenario, installing more trashcans would do far more to reduce litter than increasing the number of citations for littering.

“The design and layout of spaces can also signal specific expectations and guide our actions,” Tzall explained. In the litter example, then, the addition of more trashcans could also act as an environmental cue to encourage throwing trash away rather than littering.

How to Create Positive Environments to Foster Safety, Health, and Well-Being

Ultimately, reducing crime requires addressing the root causes of poverty and social inequality that lead to crime. But taking care of public spaces and neighborhoods to keep them clean and enjoyable can still have a positive impact on the communities who live in and use them.

“Positive environments provide opportunities for meaningful interactions and collaboration among community members,” Tzall said. “Access to green spaces, recreational facilities, mental health resources, and community services contribute to physical, mental, and emotional health,” said Tzall.

By creating more positive environments, we can encourage healthier lifestyle choices—like adding protected bike lanes to encourage people to ride bikes—and prosocial behavior —like adding basketball courts in parks to encourage people to meet and play a game with their neighbors.

At the individual level, Tzall suggests people “can initiate or participate in community projects, volunteer for local organizations, support inclusive initiatives, engage in dialogue with neighbors, and collaborate with local authorities or community leaders.” Create positive environments by taking the initiative to pick up litter when you see it, participate in tree planting initiatives, collaborate with your neighbors to establish a community garden, or volunteer with a local organization to advocate for better public spaces and resources. 

Wilson JQ and Kelling GL. Broken Windows: The Police and Neighborhood Safety . The Atlantic Monthly. 1982.

Harcourt B, Ludwig J. Broken windows: new evidence from new york city and a five-city social experiment . University of Chicago Law Review. 2006;73(1).

Peters M, Eure P. An Analysis of Quality-of-Life Summonses, Quality-of-Life Misdemeanor Arrests, and Felony Crime in New York City, 2010-2015 . New York City Department of Investigation Office of the Inspector General for the NYPD; 2016.

Sampson RJ. Disparity and diversity in the contemporary city: social (Dis)order revisited . The British Journal of Sociology. 2009;60(1):1-31. Doi:10.1111/j.1468-4446.2009.01211.x

By Rachael Green Rachael is a New York-based writer and freelance writer for Verywell Mind, where she leverages her decades of personal experience with and research on mental illness—particularly ADHD and depression—to help readers better understand how their mind works and how to manage their mental health.

By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts.

  • Research article
  • Open access
  • Published: 04 December 2020

An empirical application of “broken windows” and related theories in healthcare: examining disorder, patient safety, staff outcomes, and collective efficacy in hospitals

  • Louise A. Ellis   ORCID: 1 ,
  • Kate Churruca 1 ,
  • Yvonne Tran 1 ,
  • Janet C. Long 1 ,
  • Chiara Pomare 1 &
  • Jeffrey Braithwaite 1  

BMC Health Services Research volume  20 , Article number:  1123 ( 2020 ) Cite this article

8500 Accesses

5 Citations

7 Altmetric

Metrics details

Broken windows theory (BWT) proposes that visible signs of crime, disorder and anti-social behaviour – however minor – lead to further levels of crime, disorder and anti-social behaviour. While we acknowledge divisive and controversial policy developments that were based on BWT, theories of neighbourhood disorder have recently been proposed to have utility in healthcare, emphasising the potential negative effects of disorder on staff and patients, as well as the potential role of collective efficacy in mediating its effects. The aim of this study was to empirically examine the relationship between disorder, collective efficacy and outcome measures in hospital settings. We additionally sought to develop and validate a survey instrument for assessing BWT in hospital settings.

Cross-sectional survey of clinical and non-clinical staff from four major hospitals in Australia. The survey included the Disorder and Collective Efficacy Survey (DaCEs) (developed for the present study) and outcome measures: job satisfaction, burnout, and patient safety. Construct validity was evaluated by confirmatory factor analysis (CFA) and reliability was assessed by internal consistency. Structural equation modelling (SEM) was used to test a hypothesised model between disorder and patient safety and staff outcomes.

The present study found that both social and physical disorder were positively related to burnout, and negatively related to job satisfaction and patient safety. Further, we found support for the hypothesis that the relationship from social disorder to outcomes (burnout, job satisfaction, patient safety) was mediated by collective efficacy (social cohesion, willingness to intervene).


As one of the first studies to empirically test theories of neighbourhood disorder in healthcare, we found that a positive, orderly, productive culture is likely to lead to wellbeing for staff and the delivery of safer care for patients.

Peer Review reports

A long tradition exists in criminology and social-psychology research on the concept of neighbourhood disorder and in what ways disorder relates to anti-social behaviour and poor outcomes [ 1 ] . Interest in neighbourhood disorder is readily apparent in Broken Window Theory (BWT) [ 2 ], as well as in alternative perspectives of disorder involving shared expectation and cohesion—more broadly known as collective efficacy [ 3 , 4 , 5 ]—that are consistent with social disorganisation theory. The current study draws from these various theories and insights into neighbourhood disorder and applies them to hospital settings. At this point, we must make clear our intentions in applying neighbourhood disorder theories to healthcare. It is perilous to expect theories of neighbourhood disorder can be perfectly replicable in an organisational setting, nor do we consider that all elements of the theories are applicable to hospital settings (such as the concept of fear) [ 6 ] . We particularly reject the flawed ramifications of these theories that saw victimisation and blame attributed to individual neighbourhood members. However, here, we consider that concepts from neighbourhood studies may have considerable promise to shed new light on the relationships between the physical and social environments of hospitals on the one hand, and the health, wellbeing and behaviour of staff and patients, on the other [ 7 ] . We begin by reviewing the history and evolution of these theories before considering their application to healthcare.

Broken windows: a theory of disorder in neighbourhoods

Broken windows theory (BWT), as a social-psychological theory of urban decline, was originally developed almost 40 years ago by Wilson and Kelling [ 2 ]. Proponents of this theory argue that both physical disorder (e.g., broken windows, graffiti, litter) and social disorder (e.g., vandalism, antisocial activities) provide important environmental cues to the kinds of negative actions that are normalised and tolerated in an area, fuelling further incivility and more serious crime. For example, signs of disorder can signal potential safety issues to residents of a neighbourhood, leading to their withdrawal from public spaces, and thereby a reduction in informal social control, further perpetuating the effects of disorder [ 2 ].

Defining disorder

Although debates have occurred in the literature as to what counts as disorder, it has usually been defined as representing “minor violations of social norms” ([ 8 ] p4923). Some researchers have made a distinction between physical and social disorder, with physical disorder relating to the overall appearance of an area and social disorder directly involving people [ 9 ]. Thinking about disorder in this way, neighbourhoods with high levels of physical disorder were defined as: noisy, dirty, and run-down; buildings are in disrepair or abandoned; and vandalism and graffiti are common [ 10 ]. On the other hand, signs of social disorder in neighbourhoods may include the presence of people hanging out on the streets, drinking, or taking drugs [ 10 ]. Researchers highlight the importance of measuring perceptions of physical and social disorder as separate factors [ 9 , 11 ] with recent studies finding differential impacts of the two types of disorder [ 12 ].

Rethinking disorder: the role of collective efficacy

The BWT originally proposed by Wilson and Kelling [ 2 ] suggested a causal relationship with disorder leading to crime, which had a significant bearing upon subsequent controversial policy developments, such as ‘zero-tolerance policing’ [ 13 ] and ‘stop-and-frisk’ programs [ 14 ]. Under this approach, police pay attention to every facet of the law, including minor offences, such as public drinking and vandalism, with the aim of preventing more serious crimes from occurring [ 13 ]. The level of support these policing strategies have received has been surprising, given that BWT has not received a commensurate amount of study to date, and the research on crime that does exist is equivocal [ 12 ]. In particular, there has been an ongoing debate in the academic literature over whether BWT posits a direct or indirect relationship between disorder and crime. Most prominently, Sampson and Raudenbush [ 4 ] reconsidered the claims of BWT and argued instead that physical and social disorder were not generally causal antecedents to more serious crimes. Consistent with social disorganisation theory [ 3 ], Sampson and Raudenbush [ 4 ] suggested that collective efficacy has a significant influence on criminality in neighbourhoods. They defined collective efficacy as “social cohesion among neighbours combined with their willingness to intervene on behalf of the common good” ([ 5 ] p918). Empirical results supported their conceptual ideas in that the positive relationship between disorder and crime was mediated by collective efficacy [ 4 ].

Other lines of research have found a direct association between disorder and crime even when controlling for collective efficacy (e.g., [ 15 ]). For example, Plank et al. [ 16 ] studied disorder and collective efficacy in a school setting. They found a robust association between both disorder and violence (i.e., crime) while controlling for collective efficacy. They concluded that “fixing broken windows and attending to the physical appearance of the school cannot alone guarantee productive teaching and learning, but ignoring them greatly increases the chances of a troubling downward spiral” ([ 16 ] p244). In summary, the results are mixed as to the extent that there is direct effect of disorder on crime or other poor outcomes, but the evidence clearly suggests that there is at least an indirect effect. The key problem is what people do with this information. There is no justification for blaming individuals or demonising groups or neighbourhoods for their behaviour. We do not in any way condone seriously erroneous and consequential victimisation of people or groups as a result of the application of BWT. But we do think this is an area worthy of study.

Applying broken windows theory to healthcare

Following recent interest in applying BWT to smaller, more circumscribed environments, such as workplaces [ 17 , 18 ], researchers have started to consider the application of BWT to healthcare settings [ 7 , 19 , 20 ]. There are several well-studied trends in health services research that support this application. Theories and studies of increasing popularity include: the normalisation of deviance [ 21 ], behavioural modelling in hand hygiene [ 22 ], hospital workplace violence [ 23 ], and the association between staff’s safe work practices and their perceiving their work area as cluttered and disorderly [ 24 ].

Disorder in hospitals may include negative deviations, trade-offs or workarounds that manifest continuously in complex, dynamic and time-pressured environments, which can contribute to poor staff outcomes [ 25 , 26 , 27 ]. While trade-offs and workarounds occur in every setting, and they may have many benefits including signalling productive flexibility and staff capacity for manoeuvring, they can also represent risk in healthcare. For example, some researchers have shown that small deviations such as violating recommended processes for use of local anaesthesia can be detrimental, potentially even leading to death [ 28 ]. In line with BWT logic, there is evidence to suggest that the physical hospital environment influences the health and wellbeing of staff and patients [ 29 ]. Similarly, evidence shows that social disorder (e.g., bullying, violence) can influence staff in healthcare organisations [ 23 , 30 ]. All of these examples highlight the potential negative perpetuating effects of disorder in healthcare organisations and how disorder may detrimentally affect patients, such as through poor patient safety outcomes (see Fig.  1 [ 7 ]). Despite the elevated interest in BWT, we could find no empirical study of disorder in hospitals, nor any examination of the role of collective efficacy on staff outcomes or patient safety.

figure 1

Proposed model of disorder in hospitals Source: Churruca, Ellis et al., 2018 [ 7 ]

Aims of the present study

The primary purpose of the present study is to empirically examine the relationship between hospital disorder and three key outcomes: staff burnout, staff job satisfaction, and patient safety. We also sought to address the contention in the literature regarding the role of collective efficacy (defined here as social cohesion among hospital staff and their willingness to intervene to address problems) between hospital disorder and outcomes. The first aim was to develop a short but valid and reliable survey instrument for measuring physical disorder, social disorder, social cohesion and willingness to intervene in hospital settings. Based on previous research, physical and social disorder were kept as separate constructs. We then sought to test the following three research questions:

Is there a significant association between hospital disorder (physical disorder, social disorder) and staff outcomes (burnout, job satisfaction)?

Is there a significant association between hospital disorder (physical disorder, social disorder) and patient safety?

What is the function of “collective efficacy” (social cohesion, willingness to intervene) in hospitals? Specifically, does staff collective efficacy mediate the relationship between disorder and outcomes? Figure  2 demonstrates the simplified hypothesised mediation model.

figure 2

Hypothesised mediation model

Participants and setting

The study employed a cross-sectional survey of staff from four major hospitals in Australia. All hospital sites were public hospitals in metropolitan areas with over 200 beds. The sites were selected based on the similarity in the types of services offered (e.g., emergency department, intensive care, surgical, medical, geriatric care) and that they were located within areas of varying relative socio-economic disadvantage [ 31 ]. All hospital staff were invited to participate in the study through an invitation sent to their work email address. The email included a link to an online version of the survey via Qualtrics [ 32 ].

Survey development

The Disorder and Collective Efficacy survey (DaCEs) for hospital staff was developed for the present study based on an extensive review of the BWT literature. An initial pool of items was formed to assess the hypothesised constructs of the DaCEs: Physical disorder (19 items), social disorder (13 items), and collective efficacy, represented by social cohesion (12 items) and willingness to intervene (10 items). Some of the items were adapted from existing scales [ 16 , 24 , 33 , 34 , 35 ], and others were purpose-developed by the research team (see Supplementary File  1 ). Items were modified to make them relevant to a hospital context. All items were answered on a five-point Likert scale (1 = strongly disagree to 5 = strongly agree). A panel of experts in healthcare ( n  = 10; hospital staff and researchers) reviewed and provided feedback on the wording of items mapping onto each of the hypothesised constructs and checked for possible misinterpretations of questions, instructions and response format. Minor adjustments were made to the initial item pool (see Supplementary File  1 ). The aim was then to refine the item pool to produce a survey that would be short enough to be completed by busy hospital workers, but which has satisfactory psychometric properties.

Staff outcomes

The survey included existing validated scales to measure staff burnout and job satisfaction. Burnout was measured through a 10-item version of the Maslach Burnout Inventory (MBI) [ 36 , 37 , 38 ]. Two subscales of burnout—emotional exhaustion and depersonalisation—were used for the current survey as the third subscale, personal accomplishment, was deemed less relevant to nonclinical staff. Burnout items were answered on a seven-point Likert scale (1 = strongly disagree to 7 = strongly agree). The job satisfaction section of the Job Diagnostic Survey (5 items) was selected to capture individual’s feelings about their job [ 39 ]. Job satisfaction items were answered on a five-point Likert scale (1 = strongly disagree to 5 = strongly agree).

Patient safety

An item taken from the Hospital Survey of Patient Safety Culture (HSOPSC) was used as an indicator of patient safety [ 40 ]. This item is an outcome measure for patient safety that asks staff to provide an overall patient safety grade for their hospital (1 = excellent to 5 = failing).

Data analysis

Participants missing more than 10% of survey data were excluded. Remaining missing values were imputed using the Expectation Maximisation (EM) Algorithm within SPSS, version 25 [ 41 ]. Some items were then reversed coded so that higher item-response scores indicated a greater extent of job satisfaction, burnout, disorder, willingness to intervene, and patient safety (See Supplementary File  1 for individual recoded items). Frequency distributions were calculated to test whether items violated the assumption of univariate normality (i.e., skewness index ≥3, kurtosis index ≥10). As a number of the items were skewed (i.e., skewness index ≥3), the chi-square significance value was corrected for bias using the Bollen-Stine bootstrapping method [ 42 ] based on 1000 bootstrapped samples.

Items were evaluated psychometrically via confirmatory factor analysis (CFA), using a two-stage process. First, to refine the initial item pool, four one-factor congeneric models (of physical disorder, social disorder, social cohesion and willingness to intervene items) were run using AMOS, version 25 [ 43 ]. Here, our analytic plan involved removing one item at a time from each model using the following strategy: (i) removing items with the lowest factor loadings while maintaining the theoretical content and meaning of the proposed construct; (ii) removing items as long as each construct contained at least four observed variables; and (iii) items were removed as long as the resulting model demonstrated an improved model fit [ 44 , 45 ]. Differences in model fit were assessed using the chi-square difference test [ 46 ]. Second, two two-factor models were used to assess the factor structure of items related to disorder (i.e., physical disorder, social disorder) and collective efficacy (i.e., social cohesion, willingness to intervene) using the reduced item sets. Each item was loaded on the one factor it purported to represent. Further item refinement was undertaken as required through inspection of factor loadings, standardised residuals and modification indices to reduce each scale to three or four items. Goodness-of-fit was assessed using the Tucker Lewis Index (TLI), Comparative Fit Index (CFI) and Root Mean Square Error of Approximation (RMSEAs), and chi-square, with significance value supplemented by the Bollen-Stine bootstrap test. The TLI and CFI yield values ranging from zero to 1.00, with values greater than .90 and .95 being indicative of acceptable and excellent fit to the data [ 47 ]. For RMSEAs, values less than .05 indicate good fit, and values as high as .08 represent reasonable errors of approximation in the population [ 48 ]. For the Bollen-Stine test, non-significant values indicate that the proposed model is correct. Reliability of each of the subscales was assessed through Cronbach’s alpha (using SPSS, version 25) and composite reliability (using AMOS, version 25).

The hypothesised mediation model (Fig.  2 ) was assessed using structural equation modelling (SEM) in AMOS, version 25 [ 43 ]. First, we tested the direct effects from disorder (physical and social) to each outcome (burnout, job satisfaction, patient safety), followed by the indirect effect from disorder to outcomes, through collective efficacy (social cohesion, willingness to intervene). A parametric bootstrapping approach was used to test mediation. Under the bootstrapping approach, indirect effects are of interest and based on bootstrapped standard errors (with 1000 draws) [ 49 , 50 ]. Model fit was evaluated using CFI, TLI, RMSEA, and chi-square.

Descriptive statistics, distribution, reliability and confirmatory factor analysis

Participants were 415 staff from four hospitals in Australia. Once participants with more than 10% of survey data missing were excluded, the remaining sample was reduced to 340. Of the 340 participants, most were female (77.5%), worked as a nurse (34.2%), and had been working in the same hospital for three or more years (76.1%). The characteristics of the survey respondents are presented in Table  1 .

Descriptive statistics and data pertaining to assumptions of normality for all items are presented in Supplementary File  1 . The vast majority of the social disorder, social cohesion and willingness to intervene items demonstrated a skewness index greater than three, while only three items demonstrated a kurtosis index greater than 10 (SD7, SD10, SC6). As a result, Bollen-Stine bootstrapping was conducted in order to improve accuracy when assessing parameter estimates and fit indices.

To refine the initial item pool, first four one-factor congeneric models were run for items designed to measure physical disorder, social disorder, social cohesion and willingness to intervene. Based on an examination of modification indices and standardised factor loadings, items were removed one at a time, until the four strongest items remained. As shown in Table  2 , the reduced four-item constructs demonstrated much improved model fit statistics relative to the full models with all items. Chi-squared difference tests for all four constructs were significant, indicating that the reduced item constructs were significantly better models. The results of the chi-squared difference tests were: Physical disorder, (χ 2 difference = 139, df = 18, p  < .001), social disorder (χ 2 difference = 680, df = 63, p  < .001), social cohesion (χ 2 difference = 302, df = 52, p  < .001), and willingness to intervene (χ 2 difference = 243, df = 33, p  < .001).

Two two-factor models of disorder (physical disorder, social disorder) and collective efficacy (social cohesion, willingness to intervene) were then tested through CFA each using eight of their respective items. Each item was loaded on the one factor it purported to represent. Where required, further item refinement was undertaken through inspection of factor loadings, standardised residuals and modification indices. The two-factor model of disorder, including four physical disorder items and four social disorder items produced an adequate fit to the data, χ 2 (19) = 54.06, TLI = .96, CFI = .97, RMSEA = .08, though the Bollen-Stine bootstrap was significant ( p  = .005). Inspection of the standardised factor loadings for items PD3 and SD3 suggested that their removal may improve model fit. The removal of these two items resulted in an improved model fit, χ2 (8) = 18.28, TLI = .979, CFI = .989, RMSEA = .062, and the Bollen-Stine bootstrap ( p  = .057). The standardised factor loadings for the six items remaining ranged from .71 to .90. The correlation between physical disorder and social disorder was low, but significant ( r  = .17, p  = .007). Next, a two-factor model of collective efficacy consisting of four social cohesion items and four willingness to intervene items were tested. This model produced an excellent fit to the data, χ2 (19) = 25.36, TLI = .99, CFI = 1.00, RMSEA = .06, and the Bollen-Stine bootstrap was not significant ( p  = .458). The standardised factor loadings for the six items ranged from .68 to .90, and the correlation between social cohesion and willingness to intervene was strong, r  = .69, p  < .001. The retained items from the two-factor models are presented in Table  3 , along with their factor loadings. Cronbach’s alpha and composite reliability for the final items is also shown in Table  3 , demonstrating that all four scales demonstrated acceptable levels of reliability.

Research question 1: is there a significant association between hospital disorder and staff outcomes?

In order to examine the relationship between hospital disorder and staff outcomes, four separate models were run (i.e., models were run separately for physical disorder and social disorder, each with burnout and job satisfaction as dependent variables). Findings are presented in Supplementary File  2 . The results showed that physical disorder was significantly associated with higher burnout (β = .26, p  < .001) and lower job satisfaction (β = −.40, p  < .001). Similarly, social disorder was significantly associated with higher burnout (β = .23, p  < .001) and lower job satisfaction (β = −.54, p  < .001).

Research question 2: is there a significant association between hospital disorder and patient safety?

Two separate models were run for physical disorder and social disorder (Supplementary File  2 ). Physical disorder was significantly associated with lower patient safety scores (β = −.15, p  = .008). Likewise, a greater extent of social disorder was significantly associated with lower levels of patient safety (β = −.26, p  < .001).

Research question 3: does staff collective efficacy mediate the relationship between disorder and outcomes?

We then tested three separate mediation models for each outcome measure where the relationship between disorder and outcomes was mediated by collective efficacy via bootstrapping. For burnout, the model fit the data well, χ2 (81) = 142.75, TLI = .97, CFI = .98, RMSEA = .05. The findings presented in Fig.  3 show that there were significant negative paths from: social disorder to social cohesion (β = −.45, p  = .003); social disorder to willingness to intervene (β = −.49, p  = .002); social cohesion to burnout (β = −.23, p  = .022); and willingness to intervene to burnout (β = −.33, p  = .004). However, the paths from physical disorder to social cohesion (β = −.11, p  = .077) and from physical disorder to willingness to intervene (β = −.04, p  = .466) were not significant. Alongside these parameters, there was a significant direct effect from physical disorder to burnout (β = .18, p  = .001), but not from social disorder to burnout (β = −.07, p  = .351). Importantly, bootstrapped analyses for indirect effects indicated a significant indirect path from social disorder to burnout via social cohesion and willingness to intervene (β = .26, p  = .001). However, the indirect path from physical disorder to burnout was not significant (β = .04, p  = .205).

figure 3

Model of disorder and burnout, mediated by collective efficacy

For job satisfaction, the model provided an adequate fit to the data, χ2 (125) = 274.69, TLI = .95, CFI = .96, RMSEA = .06 (Fig.  4 ). The findings show that there was a significant path from social cohesion to job satisfaction (β = .34, p  = .002) and from willingness to intervene to job satisfaction (β = .38, p  = .001). The direct effects from physical disorder to job satisfaction (β = −.06, p  = .233) and from social disorder to job satisfaction (β = −.04, p  = .575) were not significant. Bootstrapped analyses for indirect effects indicated a significant indirect path from social disorder to job satisfaction via social cohesion and willingness to intervene (β = −.34, p  = .001). However, the indirect path from physical disorder to burnout was not significant (β = −.05, p  = .171).

figure 4

Model of disorder and job satisfaction, mediated by collective efficacy

For patient safety, the model fit provided a satisfactory fit to the data, χ2 (81) = 171.26, TLI = .96, CFI = .97, RMSEA = .06. The findings are presented in Fig.  5 and show that there was a significant path from willingness to intervene to patient safety (β = .23, p  = .041). The path from social cohesion to patient safety just failed to reach significance (β = .20, p  = .057). The direct effects from physical disorder to patient safety (β = −.08, p  = .155) and from social disorder to patient safety (β = −.04, p  = .612) were not significant. The indirect effects indicated a significant indirect path from social disorder to patient safety via social cohesion and willingness to intervene (β = −.20, p  = .001). However, the indirect path from physical disorder to burnout was not significant (β = −.03, p  = .174).

figure 5

Model of disorder and patient safety, mediated by collective efficacy

BWT and related theories of neighbourhood disorder were used here as a novel way of studying the influence of hospital environment on staff outcomes and patient safety. In this study, we developed and validated a survey instrument of disorder and collective efficacy for hospital staff—the DaCEs. In response to our research questions, we found that both social and physical disorder were positively related to burnout and negatively related to job satisfaction and patient safety. This indicated that the greater the perceived disorder in hospitals the higher the burnout and lower job satisfaction in hospital staff, and lower ratings of patient safety. Although neighbourhood disorder theories are not perfectly applicable to a hospital setting, our findings are broadly analogous with previous neighbourhood research and suggest that while attending to the physical appearance of the hospital cannot alone guarantee better staff and patient outcomes, ignoring them can significantly increase the chances of poorer outcomes. The present study also found support for the contention that collective efficacy mediated the relationship between social disorder and outcomes (burnout, job satisfaction, patient safety), but not for physical disorder.

This study is one of the first to empirically evaluate neighbourhood disorder theories in healthcare. Consistent with the original BWT, we found that perceptions of social and physical disorder were associated with potential safety issues [ 2 ], in this case, low patient safety ratings in hospitals. Past research on neighbourhood disorder supports the association between perceived neighbourhood disorder and poor mental health [ 51 ], corresponding with the present study’s findings that hospital disorder was associated with low job satisfaction and high burnout. These findings shed light on the potential relationship between culture and disorder in hospitals. We recognise that BWT has received considerable criticism over the years [ 1 ], particularly in response to controversial policy developments that were based on the BWT perspective. At this point, we must make clear that we do not advocate such policies, and find them abhorrent. However, we do contend that it seems likely that disorder is a marker for a poorer workplace culture compared to a workplace that is perceived as more orderly by hospital staff. This represents further converging evidence that having a productive, functional, more orderly culture is good for both staff and patients and not having a collective, efficacious, productive, collaborative culture is not [ 52 ].

Consistent with previous research, our study findings demonstrate the differential effects of physical and social disorder on outcome measures [ 11 , 53 ]. While both types of disorder were found to be directly related to all outcomes, once collective efficacy was added to the model, the relationship between social disorder and each of the outcomes became non-significant. In summary, consistent with the assertions of Sampson and Raudenbush [ 4 ] and in concordance with social disorganisation theory, we found that the relationship between social disorder and all outcome measures was significantly mediated by collective efficacy; however, this was not the case for physical disorder. As for the potential reasons for these findings, from a research standpoint, social disorder and physical disorder are qualitatively different: neighbourhood social disorder has been described as “episodic behaviour” involving individuals “which only lasts for a limited amount of time”, whereas neighbourhood physical disorder instead refers to “the deterioration of urban landscapes” and “does not necessarily involve actors” ([ 53 ] p5). Similarly, in a hospital setting, physical disorder may be perceived by staff as a more stable and constant presence in the hospital environment. In other words, hospital staff may be “inoculated” ([ 12 ] p411) to the presence of physical disorder in the hospital environment, with collective efficacy being less likely to alter or affect the relationship between physical disorder and outcomes.

A further explanation as to why the relationship between social disorder and all three outcome measures were mediated by collective efficacy, but not for physical disorder, is because when social disorder manifests in hospitals (e.g., non-compliance, wasting time), healthcare staff must work together to ‘pick up the slack’ to avoid serious threats to the safety and quality of care delivered. For example, if certain staff are absent or late in a particular hospital ward, the rest of the staff in that ward must work together to negate the likelihood of patient safety issues. Working as a team to make up for the social disorder may prevent any one individual staff member experiencing burnout and low job satisfaction. Indeed, this is consistent with past research showing that collaboration in hospitals has a positive effect on staff and patient outcomes, including patient safety, burnout, and job satisfaction [ 54 ]. This differs to physical disorder (e.g., run-down hospital, vandalism) where it is not necessarily seen as the responsibility of hospital staff to work collaboratively and address this form of disorder. That is, while staff must work together to address issues of social disorder such as someone being absent or late, physical disorder is more likely to be seen to be needing to be dealt with on the organisational level. For example, a hospital being in need of repair needs intervention from the government, NHS Trust, Board of Governors or local health district which can provide the necessary resources to redevelop the infrastructure.

This study thereby contributes to the broader BWT and related neighbourhood disorder field as it highlights the importance of keeping social and physical disorder as separate constructs when assessing disorder. Further, this study highlights the importance of encouraging collective efficacy among hospital staff as it can act as a barrier between social disorder and poor staff outcomes and patient safety issues.

Strengths and limitations

A strength of this study was the development of an initial psychometric profile for the measure of disorder and collective efficacy for hospitals, with its psychometric properties being assessed across four hospital sites in Australia. As to limitations, the study was based on self-reports of staff and, as with all research of this kind, is reflective of the perceptions of the agents involved. We did not include patients’ self-reports or observational research. The data was collected at one time point and therefore cannot identify any causal influence of physical and social disorder on outcomes which would require longitudinal studies involving repeated sampling on the same set of study participants. The findings concerning patient safety would need to be replicated in view of the fact that only one item was used to assess patient safety and therefore the measure has unestablished reliability. The DaCEs also warrants further cross-validation of its factor structure, as the final items were selected on the basis of results from our four included hospitals, and may not be generalisable to all hospital systems. Optimally, CFA should be randomly divided into subgroups (calibration and validation samples) to validate and verify the factor structure of the tool [ 55 ]. However, the current study was limited by the relatively modest sample size, and further work would be needed to verify the validity of the tool.

As one of the first studies to empirically test theories of neighbourhood disorder in healthcare, we found that a positive, orderly, productive culture is likely to lead to wellbeing for staff and better safety for patients, and vice versa. This is a modified study of BWT and related theories in hospitals, and one of the few studies to assess associations between different forms of disorder, collective efficacy, and staff and patient outcomes. Our hypothesised mediation model was supported, showing that the relationship between social disorder and outcomes (job satisfaction, burnout, patient safety) was mediated by collective efficacy. Having established and tested the robustness of the model, we offer it for new applications and future studies on this topic and highlight the importance of studying physical and social disorder as separate constructs. This study demonstrates the potential benefits of encouraging collective efficacy among hospital staff as it can act as a barrier to poor staff wellbeing and patient safety issues when there is social disorder.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.


  • Broken windows theory

Disorder and Collective Efficacy Survey

Confirmatory factor analysis

Structural equation modelling

Maslach Burnout Inventory

Hospital Survey of Patient Safety Culture

Expectation Maximisation

Tucker Lewis Index

Comparative Fit Index

Root Mean Square Error of Approximation

Kubrin CE. Making order of disorder: a call for conceptual clarity. Criminol Pub Pol’y. 2008;7:203.

Article   Google Scholar  

Wilson JQ, Kelling GL. Broken windows: the police and neighborhood safety. Atl Mon. 1982;211:29–38.

Google Scholar  

Sampson RJ, Groves WB. Community structure and crime: testing social-disorganization theory. Am J Sociol. 1989;94:774–802.

Sampson RJ, Raudenbush SW. Systematic social observation of public spaces: a new look at disorder in urban neighborhoods. Am J Sociol. 1999;105:603–51.

Sampson RJ, Raudenbush SW, Earls F. Neighborhoods and violent crime: a multilevel study of collective efficacy. Science. 1997;277:918–24.

Article   CAS   PubMed   Google Scholar  

Ranasinghe P. Jane Jacobs’ framing of public disorder and its relation to the ‘broken windows’ theory. Theor Criminol. 2012;16:63–84.

Churruca K, Ellis LA, Braithwaite J. ‘Broken hospital windows’: debating the theory of spreading disorder and its application to healthcare organizations. BMC Health Serv Res. 2018;18:201.

Article   PubMed   PubMed Central   Google Scholar  

Yang S-M. Social disorder and physical disorder at places. Encyclopedia of Criminology and Criminal Justice; 2014. p. 4922–32.

Book   Google Scholar  

Skogan WG. Disorder and decline: crime and the spiral of decay in American neighborhoods. Berkeley: UC Press; 1992.

Ross CE, Mirowsky J. Neighborhood disorder, subjective alienation, and distress. J Health Soc Sci. 2009;50:49–64.

Yang S-M. Assessing the spatial–temporal relationship between disorder and violence. J Quant Criminol. 2010;26:139–63.

Article   CAS   Google Scholar  

Hinkle JC. The relationship between disorder, perceived risk, and collective efficacy: a look into the indirect pathways of the broken windows thesis. Crim Justice Stud. 2013;26:408–32.

Doran BJ, Burgess MB. Putting fear of crime on the map: investigating perceptions of crime using geographic information systems. London: Springer; 2011.

Fagan J, Davies G. Street stops and broken windows: Terry, race, and disorder in New York City. Fordham Urb LJ. 2000;28:457.

Xu Y, Fiedler ML, Flaming KH. Discovering the impact of community policing: the broken windows thesis, collective efficacy, and citizens’ judgment. JRCD. 2005;42:147–86.

Plank Stephen B, Bradshaw Catherine P, Young H. An application of “broken-windows” and related theories to the study of disorder, fear, and collective efficacy in schools. Am J Educ. 2009;115:227–47.

Ramos J, Torgler B. Are academics messy? Testing the broken windows theory with a field experiment in the work environment. RLE. 2012;8:563–77.

Lim MSC, Hellard ME, Aitken CK. The case of the disappearing teaspoons: longitudinal cohort study of the displacement of teaspoons in an Australian research institute. BMJ. 2005;331:1498–500.

Kayral İH. Can the theory of broken windows be used for patients safety in city hospitals management model? Hacettepe Sağlık İdaresi Dergisi. 2019;22:677–94.

Demirel ET, Emul E. Evaluation of the Broken Windows Theory in Terms of Patient Safety. Multidimensional Perspectives and Global Analysis of Universal Health Coverage. United States: IGI Global; 2020. p. 266–84.

McNamara SA. The normalization of deviance: what are the perioperative risks? AORN J. 2011;93:796–801.

Article   PubMed   Google Scholar  

Erasmus V, Brouwer W, van Beeck EF, Oenema A, Daha TJ, Richardus JH, et al. A qualitative exploration of reasons for poor hand hygiene among hospital workers: lack of positive role models and of convincing evidence that hand hygiene prevents cross-infection. Infect Control Hosp Epidemiol. 2009;30:415–9.

Zhou C, Mou H, Xu W, Li Z, Liu X, Shi L, et al. Study on factors inducing workplace violence in Chinese hospitals based on the broken window theory: a cross-sectional study. BMJ Open. 2017;7:e016290.

Gershon RRM, Karkashian CD, Grosch JW, Murphy LR, Escamilla-Cejudo A, Flanagan PA, et al. Hospital safety climate and its relationship with safe work practices and workplace exposure incidents. Am J Infect Control. 2000;28:211–21.

Debono DS, Greenfield D, Travaglia JF, Long JC, Black D, Johnson J, et al. Nurses’ workarounds in acute healthcare settings: a scoping review. BMC Health Serv Res. 2013;13:175.

Ekstedt M, Cook R. In: Wears RL, Hollnagel E, Braithwaite J, editors. The Stockholm Blizzard of 2012. Surrey, England, UK: Ashgate Publishing Limited; 2015.

Wears RL, Hollnagel E, Braithwaite J. Resilient health care: the resilience of everyday clinical work. Farnham, Surrey: Ashgate Publishing, Ltd.; 2015.

Amalberti R, Vincent C, Auroy Y, de Saint Maurice G. Violations and migrations in health care: a framework for understanding and management. Qual Saf Health Care. 2006;15:i66–71.

Eijkelenboom A, Bluyssen PM. Comfort and health of patients and staff, related to the physical environment of different departments in hospitals: a literature review. Intell Build Int. 2019;0:1–19.

Laschinger HKS, Wong CA, Grau AL. The influence of authentic leadership on newly graduated nurses’ experiences of workplace bullying, burnout and retention outcomes: a cross-sectional study. Int J Nurs Stud. 2012;49:1266–76.

Australian Bureau of Statistics. IRSD INTERACTIVE MAP. 2033055001 - Census of Population and Housing: Socio-Economic Indexes for Areas (SEIFA), Australia, 2016. Canberra: ABS; 2016.

Qualtrics. Qualtrics 2014 [Available from: .

Coyne I, Bartram D. Personnel managers’ perceptions of dishonesty in the workplace. Hum Resour Manag J. 2000;10:38.

NSW Health. 2015 YourSay workplace Cultue survey: overall. Sydney, NSW, AUS: ORC International; 2015.

Sexton JB, Helmreich RL, Neilands TB, Rowan K, Vella K, Boyden J, et al. The safety attitudes questionnaire: psychometric properties, benchmarking data, and emerging research. BMC Health Serv Res. 2006;6:44.

Schaufeli W, Enzmann D, Girault N. Measurement of burnout: a review. In: Schaufeli WB, Maslach C, Marek T, editors. Professional burnout: recent developments in theory and research. Washington: Taylor & Francis; 1993. p. 199–215.

Maslach C, Jackson S. The measurement of experienced burnout. J Occup Behav. 1981;2:99–113.

Maslach C, Schaufeli W, Leiter MP. Job burnout. Annu Rev Psychol. 2001;52:397–422.

Bowling N, Hammond G. A meta-analytic examination of the construct validity of the Michigan organizational assessment questionnaire job satisfaction subscale. J Vocat Behav. 2008;73:63–77.

Agency for Healthcare Research and Quality. Surveys on patient safety culture. Rockville, MD, US: United States Department of Health & Human Services; 2017.

Corp IBM. IBM SPSS statistics for windows, version 25.0. IBM Corp: Armonk, NY; 2017.

Bollen KA, Stine RA. Bootstrapping goodness-of-fit measures in structural equation models. Sociol Methods Res. 1992;21:205–29.

Arbuckle J. Amos user's guide: Version 25. Chicago: IBM SPSS; 2017.

Bollen KA. Structural equations with latent variables. New York: Wiley; 1989.

Schumacker RE, Lomax RG. A beginner's guide to structural equation modeling. L. Erlbaum Associates: Mahwah, NJ; 1996.

Bagozzi RP, Yi Y, Phillips LW. Assessing construct validity in organizational research. Adm Sci Q. 1991;36:421–58.

McDonald RP, Marsh HW. Choosing a multivariate model: noncentrality and goodness of fit. Psychol Bull. 1990;107:247–55.

Browne MW, Cudeck R. Alternative ways of assessing model fit. Sociol Methods Res. 1993;154:230–58.

MacKinnon DP, Lockwood CM, Hoffman JM, West SG, Sheets V. A comparison of methods to test the significance of the mediated effect. Psychol Methods. 2002;7:83–104.

Shrout PE, Bolger N. Mediation in experimental and nonexperimental studies: new procedures and recommendations. Psychol Methods. 2002;7:422.

Polling C, Khondoker M, Hatch S, Hotopf M, Team SS. Influence of perceived and actual neighbourhood disorder on common mental illness. Soc Psychiatry Psychiatr Epidemiol. 2014;49:889–901.

Article   CAS   PubMed   PubMed Central   Google Scholar  

Braithwaite J, Herkes J, Ludlow K, Testa L, Lamprell G. Association between organisational and workplace cultures, and patient outcomes: systematic review. BMJ Open. 2017;7:e017708.

Yang S-M. Causal or merely co-existing: a longitudinal study of violence and disorder at places. Unpublished Doctoral Dissertation; 2007.

Pomare C, Long JC, Churruca K, Ellis LA, Braithwaite J. Interprofessional collaboration in hospitals: a critical, broad-based review of the literature. J Interprof Care. 2020;34:509–19.

Byrne BM. Structural equation modeling with AMOS, EQS, and LISREL: comparative approaches to testing for the factorial validity of a measuring instrument. Int J Test. 2001;1:55–86.

Download references


The authors thank all hospital staff that participated in the survey.

This work is supported in part by National Health and Medical Research Council grants held by JB (APP9100002, APP1176620 and APP1135048). The funding body had no role in the design of the study and collection, analysis, and interpretation of data.

Author information

Authors and affiliations.

Centre for Healthcare Resilience and Implementation Science, Australian Institute of Health Innovation, Macquarie University, Sydney, Australia

Louise A. Ellis, Kate Churruca, Yvonne Tran, Janet C. Long, Chiara Pomare & Jeffrey Braithwaite

You can also search for this author in PubMed   Google Scholar


LAE, KC, JCL and JB conceived the study. LAE, KC, JCL and CP designed the DaCEs and drafted the paper. LAE, YT and CP performed the analysis. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Louise A. Ellis .

Ethics declarations

Ethics approval and consent to participate.

The ethical conduct of this study was approved by South Eastern Sydney Local Health District (HREC ref. no: 16/363). Governance approvals to conduct the research were obtained for each site. Participation was voluntary and anonymous. Participants provided written consent.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1., additional file 2., rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit . The Creative Commons Public Domain Dedication waiver ( ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Cite this article.

Ellis, L.A., Churruca, K., Tran, Y. et al. An empirical application of “broken windows” and related theories in healthcare: examining disorder, patient safety, staff outcomes, and collective efficacy in hospitals. BMC Health Serv Res 20 , 1123 (2020).

Download citation

Received : 14 April 2020

Accepted : 25 November 2020

Published : 04 December 2020


Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

  • Collective efficacy

BMC Health Services Research

ISSN: 1472-6963

broken window theory health

share this!

May 16, 2019

Researchers find little evidence for 'broken windows theory,' say neighborhood disorder doesn't cause crime

by Greg St. Martin, Northeastern University

Researchers find little evidence for ‘broken windows theory,’ say neighborhood disorder doesn’t cause crime

The methodology behind the findings

"there are other ways to think about disorder'.

Journal information: Social Science & Medicine

Provided by Northeastern University

Explore further

Feedback to editors

broken window theory health

Seeing the unseen: How butterflies can help scientists detect cancer

13 hours ago

broken window theory health

Physicists ask: Can we make a particle collider more energy efficient?

15 hours ago

broken window theory health

How salt from the Caribbean affects our climate

broken window theory health

Large herbivores such as elephants, bison and moose shown to contribute to tree diversity

16 hours ago

broken window theory health

Researchers can now visualize osmotic pressure in living tissue

17 hours ago

broken window theory health

Moroccan archaeologists unearth new ruins at Chellah, a tourism-friendly ancient port near Rabat

broken window theory health

Scorching, seven-planet system revealed by new Kepler exoplanet list

18 hours ago

broken window theory health

Vacuum in optical cavity can change material's magnetic state without laser excitation

broken window theory health

El Niño may be drying out the southern hemisphere: Here's how that affects the whole planet

19 hours ago

broken window theory health

New approach to water electrolysis for green hydrogen

Relevant physicsforums posts, cover songs versus the original track, which ones are better.

7 hours ago

Music to Lift Your Soul: 4 Genres & Honorable Mention

Nov 1, 2023

What are your favorite Disco "Classics"?

Oct 31, 2023

The Origin of Me and Bobbi McGee

Oct 30, 2023

Today's Fusion Music: T Square, Cassiopeia, Rei & Kanade Sato

Oct 26, 2023

Origin of this phrase: "We shall see what we shall see..."

Oct 24, 2023

More from Art, Music, History, and Linguistics

Related Stories

broken window theory health

New research challenges 'broken windows theory' of crime prediction

Sep 3, 2015

How 'broken windows' impact political activism

Feb 12, 2014

Interpersonal conflict is the strongest predictor of community crime and misconduct

Jun 2, 2015

Epileptics in high-crime neighborhoods have three times as many seizures

Dec 3, 2018

Looks do matter, particularly when it comes to neighborhoods

Apr 26, 2011

broken window theory health

The presence of sexual violence in neighborhoods erodes feelings of safety for women—but not men

Jan 10, 2018

Recommended for you

broken window theory health

Climate-induced loss is impeding human rights in the Pacific, new study finds

Nov 2, 2023

broken window theory health

Study warns restrictions to application programming interfaces by social media platforms threaten research

broken window theory health

In measuring how online news influence political views, do mouse clicks speak louder than words?

broken window theory health

New study finds that being alone does not closely correlate with feelings of loneliness

broken window theory health

Workplace discrimination saps everyone's motivation, even if it works in your favor

broken window theory health

Why all languages have words for 'this' and 'that'

Let us know if there is a problem with our content.

Use this form if you have come across a typo, inaccuracy or would like to send an edit request for the content on this page. For general inquiries, please use our contact form . For general feedback, use the public comments section below (please adhere to guidelines ).

Please select the most appropriate category to facilitate processing of your request

Thank you for taking time to provide your feedback to the editors.

Your feedback is important to us. However, we do not guarantee individual replies due to the high volume of messages.

E-mail the story

Your email address is used only to let the recipient know who sent the email. Neither your address nor the recipient's address will be used for any other purpose. The information you enter will appear in your e-mail message and is not retained by in any form.

Newsletter sign up

Get weekly and/or daily updates delivered to your inbox. You can unsubscribe at any time and we'll never share your details to third parties.

More information Privacy policy

Donate and enjoy an ad-free experience

We keep our content available to everyone. Consider supporting Science X's mission by getting a premium account.

E-mail newsletter

How Broken Windows Do —and Do Not— Matter

  • Public Health
  • Criminal Justice

Connect with the author


“Broken windows theory” has been one of the most influential – and controversial – perspectives generated by the social sciences in the last thirty years. According to this theory, signs of urban disorder such as graffiti, panhandling, and dilapidation can directly hurt affected neighborhoods – either by encouraging serious crime or by harming the health of residents. This perspective informs “zero-tolerance” policing, which operates on the assumption that cracking down on low-level misdemeanors can head off serious crimes. Critics of zero-tolerance policing say it can be counterproductive, by encouraging racial profiling and stoking contention between police and people in vulnerable communities.  

A more fundamental issue remains unresolved, however: Do the basic tenets of broken windows theory hold true? We have tackled this question by assembling and reviewing all studies that tested the effects of disorder on residents, including hundreds of studies that explore a variety of outcomes and use different research methodologies. Our review specifically probed the impact of neighborhood disorder on crime and public health, and we find little support for key claims in broken windows theory. This brief summarizes our findings and suggests alternative ways for policymakers to think about disorder in urban areas.

Little Evidence for Broken Windows Theory

Does disorder lead to crime in neighborhoods? Proponents of this claim believe that disorder signals that crime will go unpunished. People inclined toward criminal activity and aggression, the argument goes, will be emboldened by signs of neighborhood disorder to perpetrate serious crimes; and other members of the community will perceive the environment as threatening and retreat from public spaces. However, our review of available studies found no consistent evidence that disorder induces higher levels of aggression or makes residents feel more negative toward the neighborhood.

We also assessed evidence for three pathways through which proponents of this theory posit that disorder worsens public health. We considered whether disorder discourages physical activity, whether it encourages risky behavior, and whether disorder may create signals that worsen people’s mental health. We do find that perceived disorder consistently predicts lower mental health and self-reported health problems, as well as associated problems such as substance abuse. But we find no evidence that neighborhood disorder leads to other risky behaviors, like unprotected sex, or to diminished physical activity.

In both of our reviews, claims in support of broken windows theory disproportionately came from studies that relied on weak research designs. Two deficiencies were widespread. Many studies failed to consider key variables, such as socioeconomic status, that can account for positive correlations between disorder and criminal behavior. What is more, the strongest evidence in favor of broken windows’ claims comes from studies that measure disorder via surveys completed by the same people whose experiences were in question. People’s pessimism about the maintenance of their neighborhoods could be confounded with their worries about crime or associated with the mental health problems many report. Such associations are not evidence that neighborhood disorder directly causes misbehavior or undermines wellbeing.

Alternative Perspectives on the Effects of Disorder

Our results question the premises of broken windows theory – especially as applied to policing. However, we are not saying that disorder is irrelevant, just that we need to reconsider how and why disorder matters in the urban landscape. Here we articulate four perspectives that are consistent with current research, each providing its own distinct logic for why disorder matters. These perspectives can prove useful for policymakers and practitioners.

The psychosocial model of disadvantage. Sociologists have argued that many of the negative outcomes experienced by disadvantaged people are attributable to high levels of stress in their daily lives and environments. Disorder contributes to such stress and is ever-present in poor neighborhoods. We find support for this perspective, because our reviews show that residents of neighborhoods with high levels of disorder consistently have diminished mental health.

Ecological advantages for criminals. Disorder may not elicit crime, but it can provide advantages for people already inclined toward such activity. Abandoned buildings, for example, offer hiding spots for illicit drugs, guns, or other contraband. Municipalities might be wise to fix disorderly places that facilitate crime, rather than seeking to eliminate all forms of disorder.

Social escalation. Some forms of social disorder may be precursors of serious violence. For instance, studies have shown that domestic disputes, landlord-tenant troubles, and other types of interpersonal conflict in neighborhoods do tend to predict increases in violent crime. If left unchecked, such interpersonal conflicts can escalate or spill into public spaces. This finding suggests the value of policing that works with community members to short-circuit social disputes, an approach in many ways at odds with zero-tolerance efforts that target minor infractions such as panhandling.

Custodianship . Long before broken windows theory was first formulated in 1982, urbanists emphasized neighborhood disorder, but as a symptom of social health rather than a cause of crime. C ustodial steps that community members take to prevent or eliminate disorder can be a key indicator of how well communities accomplish shared goals. Hundreds of cities have implemented 311 call systems to engage residents in maintaining public spaces and infrastructure – a theme further explored by O’Brien in a recent book on “the urban commons.”

Our systematic research reviews found little support for the broken windows theory and zero-tolerance policing. Nevertheless, we do find evidence that neighborhood disorder harms mental health; and we conclude that policy and practice seeking to improve urban life would be better served by other perspectives on neighborhood disorder.

Read more in Daniel T. O’Brien, Chelsea Farrell, Brandon C. Welsh, “ Looking Through Broken Windows: The Impact of Neighborhood Disorder on Aggression and Fear of Crime is an Artifact of Research Design,” Annual Review of Criminology and Social Science & Medicine, 2, (2019): 53-71; and Daniel T. O’Brien, The Urban Commons: How Data and Technology Can Rebuild Our Communities, (Harvard University Press, 2018).

Related Content

Broken Windows Theory of Criminology

Charlotte Ruhl

Research Assistant & Psychology Graduate

BA (Hons) Psychology, Harvard University

Charlotte Ruhl, a psychology graduate from Harvard College, boasts over six years of research experience in clinical and social psychology. During her tenure at Harvard, she contributed to the Decision Science Lab, administering numerous studies in behavioral economics and social psychology.

Learn about our Editorial Process

Saul Mcleod, PhD

Educator, Researcher

BSc (Hons) Psychology, MRes, PhD, University of Manchester

Saul Mcleod, Ph.D., is a qualified psychology teacher with over 18 years experience of working in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.

On This Page:

The Broken Windows Theory of Criminology suggests that visible signs of disorder and neglect, such as broken windows or graffiti, can encourage further crime and anti-social behavior in an area, as they signal a lack of order and law enforcement.

Key Takeaways

  • The Broken Windows theory, first studied by Philip Zimbardo and introduced by George Kelling and James Wilson, holds that visible indicators of disorder, such as vandalism, loitering, and broken windows, invite criminal activity and should be prosecuted.
  • This form of policing has been tested in several real-world settings. It was heavily enforced in the mid-1990s under New York City mayor Rudy Giuliani, and Albuquerque, New Mexico, Lowell, Massachusetts, and the Netherlands later experimented with this theory.
  • Although initial research proved to be promising, this theory has been met with several criticisms. Specifically, many scholars point to the fact that there is no clear causal relationship between lack of order and crime. Rather, crime going down when order goes up is merely a coincidental correlation.
  • Additionally, this theory has opened the doors for racial and class bias, especially in the form of stop and frisk.

The United States has the largest prison population in the world and the highest per-capita incarceration rate. In 2016, 2.3 million people were incarcerated, despite a massive decline in both violent and property crimes (Morgan & Kena, 2019).

These statistics provide some insight into why crime regulation and mass incarceration are such hot topics today, and many scholars, lawyers, and politicians have devised theories and strategies to try to promote safety within society.

Broken Windows Theory

One such model is broken windows policing, which was first brought to light by American psychologist Philip Zimbardo (famous for his Stanford Prison Experiment) and further publicized by James Wilson and George Kelling. Since its inception, this theory has been both widely used and widely criticized.

What Is the Broken Windows Theory?

The broken windows theory states that any visible signs of crime and civil disorder, such as broken windows (hence, the name of the theory), vandalism, loitering, public drinking, jaywalking, and transportation fare evasion, create an urban environment that promotes even more crime and disorder (Wilson & Kelling, 1982).

As such, policing these misdemeanors will help create an ordered and lawful society in which all citizens feel safe and crime rates, including violent crime rates, are low.

Broken windows policing tries to regulate low-level crime to prevent widespread disorder from occurring. If these small crimes are greatly reduced, then neighborhoods will appear to be more cared for.

The hope is that if these visible displays of disorder and neglect are reduced, violent crimes might go down too, leading to an overall reduction in crime and an increase in public safety.

Broken Windows Theory

Source: Hinkle, J. C., & Weisburd, D. (2008). The irony of broken windows policing: A micro-place study of the relationship between disorder, focused police crackdowns and fear of crime. Journal of Criminal Justice, 36(6), 503-512.

Academics justify broken windows policing from a theoretical standpoint because of three specific factors that help explain why the state of the urban environment might affect crime levels:

  • social norms and conformity;
  • the presence or lack of routine monitoring;
  • social signaling and signal crime.

In a typical urban environment, social norms and monitoring are not clearly known. As a result, individuals will look for certain signs and signals that provide both insight into the social norms of the area as well as the risk of getting caught violating those norms.

Those who support the broken windows theory argue that one of those signals is the area’s general appearance. In other words, an ordered environment, one that is safe and has very little lawlessness, sends the message that this neighborhood is routinely monitored and criminal acts are not tolerated.

On the other hand, a disordered environment, one that is not as safe and contains visible acts of lawlessness (such as broken windows, graffiti, and litter), sends the message that this neighborhood is not routinely monitored and individuals would be much more likely to get away with committing a crime.

With a decreased likelihood of detection, individuals would be much more inclined to engage in criminal behavior, both violent and nonviolent, in this type of area.

As you might be able to tell, a major assumption that this theory makes is that an environment’s landscape communicates to its residents in some way.

For example, proponents of this theory would argue that a broken window signals to potential criminals that a community is unable to defend itself against an uptick in criminal activity. It is not the literal broken window that is a direct cause for concern, but more so the figurative meaning that is ascribed to this situation.

It symbolizes a vulnerable and disjointed community that cannot handle crime – opening the doors to all kinds of unwanted activity to occur.

In neighborhoods that do have a strong sense of social cohesion among their residents, these broken windows are fixed (both literally and figuratively), giving these areas a sense of control over their communities.

By fixing these windows, undesired individuals and behaviors are removed, allowing civilians to feel safer (Herbert & Brown, 2006).

However, in environments in which these broken windows are left unfixed, residents no longer see their communities as tight-knit, safe spaces and will avoid spending time in communal spaces (in parks, at local stores, on the street blocks) so as to avoid violent attacks from strangers.

Additionally, when these broken windows are not fixed, it also symbolizes a lack of informal social control. Informal social control refers to the actions that regulate behavior, such as conforming to social norms and intervening as a bystander when a crime is committed, that are independent of the law.

Informal social control is important to help reduce unruly behavior. Scholars argue that, under certain circumstances, informal social control is more effective than laws.

And some will even go so far as to say that nonresidential spaces, such as corner stores and businesses, have a responsibility to actually maintain this informal social control by way of constant surveillance and supervision.

One such scholar is Jane Jacobs, a Canadian-American author and journalist who believed sidewalks were a crucial vehicle for promoting public safety.

Jacobs can be considered one of the original pioneers of the broken windows theory. One of her most famous books, The Death and Life of Great American Cities, describes how local businesses and stores provide a necessary sense of having “eyes on the street,” which promotes safety and helps to regulate crime (Jacobs, 1961).

Although the idea that community involvement, from both residents and non-residents, can make a big difference in how safe a neighborhood is perceived to be, Wilson and Keeling argue that the police are the key to maintaining order.

As major proponents of broken windows policing, they hold that formal social control, in addition to informal social control, is crucial for actually regulating crime.

Although different people have different approaches to the implementation of broken windows (i.e., cleaning up the environment and informal social control vs. an increase in policing misdemeanor crimes), the end goal is the same: crime reduction.

This idea, which largely serves as the backbone of the broken windows theory, was first introduced by Philip Zimbardo.

Examples of Broken Windows Policing

1969: philip zimbardo’s introduction of broken windows in nyc and la.

In 1969, Stanford psychologist Philip Zimbardo ran a social experiment in which he abandoned two cars that had no license plates and the hoods up in very different locations.

The first was a predominantly poor, high-crime neighborhood in the Bronx, and the second was a fairly affluent area of Palo Alto, California. He then observed two very different outcomes.


After just ten minutes, the car in the Bronx was attacked and vandalized. A family first approached the vehicle and removed the radiator and battery. Within the first twenty-four hours after Zimbardo left the car, everything valuable had been stripped and removed from the car.

Afterward, random acts of destruction began – the windows were smashed, seats were ripped up, and the car began to serve as a playground for children in the community.

On the contrary, the car that was left in Palo Alto remained untouched for more than a week before Zimbardo eventually went up to it and smashed the vehicle with a sledgehammer.

Only after he had done this did other people join the destruction of the car (Zimbardo, 1969). Zimbardo concluded that something that is clearly abandoned and neglected can become a target for vandalism.

But Kelling and Wilson extended this finding when they introduced the concept of broken windows policing in the early 1980s.

This initial study cascaded into a body of research and policy that demonstrated how in areas such as the Bronx, where theft, destruction, and abandonment are more common, vandalism would occur much faster because there are no opposing forces to this type of behavior.

As a result, such forces, primarily the police, are needed to intervene and reduce these types of behavior and remove such indicators of disorder.

1982: Kelling and Wilson’s Follow-Up Article

Thirteen years after Zimbardo’s study was published, criminologists George Kelling and James Wilson published an article in The Atlantic that applied Zimbardo’s findings to entire communities.

Kelling argues that Zimbardo’s findings were not unique to the Bronx and Palo Alto areas. Rather, he claims that, regardless of the neighborhood, a ripple effect can occur once disorder begins as things get extremely out of hand and control becomes increasingly hard to maintain.

The article introduces the broader idea that now lies at the heart of the broken windows theory: a broken window, or other signs of disorder, such as loitering, graffiti, litter, or drug use, can send the message that a neighborhood is uncared for, sending an open invitation for crime to continue to occur, even violent crimes.

The solution, according to Kelling and Wilson and many other proponents of this theory, is to target these very low-level crimes, restore order to the neighborhood, and prevent more violent crimes from happening.

A strengthened and ordered community is equipped to fight and deter crime (because a sense of order creates the perception that crimes go easily detected). As such, it is necessary for police departments to focus on cleaning up the streets as opposed to putting all of their energy into fighting high-level crimes.

In addition to Zimbardo’s 1969 study, Kelling and Wilson’s article was also largely inspired by New Jersey’s “Safe and Clean Neighborhoods Program” that was implemented in the mid-1970s.

As part of the program, police officers were taken out of their patrol cars and were asked to patrol on foot. The aim of this approach was to make citizens feel more secure in their neighborhoods.

Although crime was not reduced as a result, residents took fewer steps to protect themselves from crime (such as locking their doors). Reducing fear is a huge goal of broken-windows policing.

As Kelling and Wilson state in their article, the fear of being bothered by disorderly people (such as drunks, rowdy teens, or loiterers) is enough to motivate them to withdraw from the community.

But if we can find a way to make people feel less fear (namely by reducing low-level crimes), then they will be more involved in their communities, creating a higher degree of informal social control and deterring all forms of criminal activity.

Although Kelling and Wilson’s article was largely theoretical, the practice of broken windows policing was implemented in the early 1990s under New York City Mayor Rudy Giuliani. And Kelling himself was there to play a crucial role.

Early 1990s: Bratton and Giuliani’s implementation in NYC

In 1985, the New York City Transit Authority hired George Kelling as a consultant, and he was also later hired by both the Boston and Los Angeles police departments to provide advice on the most effective method for policing (Fagan & Davies, 2000).

  Giulian Broken Window Theory NYC

Five years later, in 1990, William J. Bratton became the head of the New York City Transit Police. In his role, Bratton cracked down on fare evasion and implemented faster methods to process those who were arrested.

He attributed a lot of his decisions as head of the transit police to Kelling’s work. Bratton was just the first to begin to implement such measures, but once Rudy Giuliani was elected as mayor in 1993, tactics to reduce crime began to really take off (Vedantam et al., 2016).

Together, Giuliani and Bratton first focused on cleaning up the subway system, where Bratton’s area of expertise lay. They sent hundreds of police officers into subway stations throughout the city to catch anyone who was jumping the turnstiles and evading the fair.

And this was just the beginning.

All throughout the 90s, Giuliani increased misdemeanor arrests in all pockets of the city. They arrested numerous people for smoking marijuana in public, spraying graffiti on walls, selling cigarettes, and they shut down many of the city’s night spots for illegal dancing.

Conveniently, during this time, crime was also falling in the city and the murder rate was rapidly decreasing, earning Giuliani re-election in 1997 (Vedantam et al., 2016).

To further support the outpouring success of this new approach to regulating crime, George Kelling ran a follow-up study on the efficacy of broken windows policing and found that in neighborhoods where there was a stark increase in misdemeanor arrests (evidence of broken windows policing), there was also a sharp decline in crime (Kelling & Sousa, 2001).

Because this seemed like an incredibly successful mode, cities around the world began to adopt this approach.

Late 1990s: Albuquerque’s Safe Streets Program

In Albuquerque, New Mexico, a Safe Streets Program was implemented to deter and reduce unsafe driving and crime rates by increasing surveillance in these areas.

Specifically, the traffic enforcement program influenced saturation patrols (that operated over a large geographic area), sobriety checkpoints, follow-up patrols, and freeway speed enforcement.

Albuquerque’s Safe Streets Program

The effectiveness of this program was analyzed in a study done by the U.S. National Highway Traffic Safety Administration (Stuser, 2001).

Results demonstrated that both Part I crimes, including homicide, forcible rape, robbery, and theft, and Part II crimes, such as sex offenses, kidnapping, stolen property, and fraud, experienced a total decline of 5% during the 1996-1997 calendar year in which this program was implemented.

Additionally, this program resulted in a 9% decline in both robbery and burglary, a 10% decline in assault, a 17% decline in kidnapping, a 29% decline in homicide, and a 36% decline in arson.

With these promising statistics came a 14% increase in arrests. Thus, the researchers concluded that traffic enforcement programs can deter criminal activity. This approach was initially inspired by both Zimbardo’s and Kelling and Wilson’s work on broken windows and provides evidence that when policing and surveillance increase, crime rates go down.

2005: Lowell, Massachusetts

Back on the east coast, Harvard University and Suffolk University researchers worked with local police officers to pinpoint 34 different crime hotspots in Lowell, Massachusetts. In half of these areas, local police officers and authorities cleaned up trash from the streets, fixed streetlights, expanded aid for the homeless, and made more misdemeanor arrests.

There was no change made in the other half of the areas (Johnson, 2009).

The researchers found that in areas in which police service was changed, there was a 20% reduction in calls to the police. And because the researchers implemented different ways of changing the city’s landscape, from cleaning the physical environment to increasing arrests, they were able to compare the effectiveness of these various approaches.

Although many proponents of the broken windows theory argue that increasing policing and arrests is the solution to reducing crime, as the previous study in Albuquerque illustrates. Others insist that more arrests do not solve the problem but rather changing the physical landscape should be the desired means to an end.

And this is exactly what Brenda Bond of Suffolk University and Anthony Braga of Harvard Kennedy’s School of Government found. Cleaning up the physical environment was revealed to be very effective, misdemeanor arrests were less so, and increasing social services had no impact.

This study provided strong evidence for the effectiveness of the broken windows theory in reducing crime by decreasing disorder, specifically in the context of cleaning up the physical and visible neighborhood (Braga & Bond, 2008).

2007: Netherlands

The United States is not the only country that sought to implement the broken windows ideology. Beginning in 2007, researchers from the University of Groningen ran several studies that looked at whether existing visible disorder increased crimes such as theft and littering.

Similar to the Lowell experiment, where half of the areas were ordered and the other half disorders, Keizer and colleagues arranged several urban areas in two different ways at two different times. In one condition, the area was ordered, with an absence of graffiti and littering, but in the other condition, there was visible evidence for disorder.

The team found that in disorderly environments, people were much more likely to litter, take shortcuts through a fenced-off area, and take an envelope out of an open mailbox that was clearly labeled to contain five Euros (Keizer et al., 2008).

This study provides additional support for the effect perceived order can have on the likelihood of criminal activity. But this broken windows theory is not restricted to the criminal legal setting.

Other Domains Relevant to Broken Windows

There are several other fields in which the broken windows theory is implicated. The first is real estate. Broken windows (and other similar signs of disorder) can indicate low real estate value, thus deterring investors (Hunt, 2015).

As such, some recommend that the real estate industry adopt the broken windows theory to increase value in an apartment, house, or even an entire neighborhood. They might increase in value by fixing windows and cleaning up the area (Harcourt & Ludwig, 2006).

Consequently, this might lead to gentrification – the process by which poorer urban landscapes are changed as wealthier individuals move in.

Although many would argue that this might help the economy and provide a safe area for people to live, this often displaces low-income families and prevents them from moving into areas they previously could not afford.

This is a very salient topic in the United States as many areas are becoming gentrified, and regardless of whether you support this process, it is important to understand how the real estate industry is directly connected to the broken windows theory.

Another area that broken windows are related to is education. Here, the broken windows theory is used to promote order in the classroom. In this setting, the students replace those who engage in criminal activity.

The idea is that students are signaled by disorder or others breaking classroom rules and take this as an open invitation to further contribute to the disorder.

As such, many schools rely on strict regulations such as punishing curse words and speaking out of turn, forcing strict dress and behavioral codes, and enforcing specific classroom etiquette.

Similar to the previous studies, from 2004 to 2006, Stephen Plank and colleagues conducted a study that measured the relationship between the physical appearance of mid-Atlantic schools and student behavior.

They determined that variables such as fear, social order, and informal social control were statistically significantly associated with the physical conditions of the school setting.

Thus, the researchers urged educators to tend to the school’s physical appearance to help promote a productive classroom environment in which students are less likely to propagate disordered behavior (Plank et al., 2009).

Despite there being a large body of research that seems to support the broken windows theory, this theory does not come without its stark criticisms, especially in the past few years.

Major Criticisms

At the turn of the 21st century, the rhetoric surrounding broken windows drastically shifted from praise to criticism. Scholars scrutinized conclusions that were drawn, questioned empirical methodologies, and feared that this theory was morphing into a vehicle for discrimination.

Misinterpreting the Relationship Between Disorder and Crime

A major criticism of this theory argues that it misinterprets the relationship between disorder and crime by drawing a causal chain between the two.

Instead, some researchers argue that a third factor, collective efficacy, or the cohesion among residents combined with shared expectations for the social control of public space, is the causal agent explaining crime rates (Sampson & Raudenbush, 1999).

A 2019 meta-analysis that looked at 300 studies revealed that disorder in a neighborhood does not directly cause its residents to commit more crimes (O’Brien et al., 2019).

The researchers examined studies that tested to what extent disorder led people to commit crimes, made them feel more fearful of crime in their neighborhoods, and affected their perceptions of their neighborhoods.

In addition to drawing out several methodological flaws in the hundreds of studies that were included in the analysis, O’Brien and colleagues found no evidence that the disorder and crime are causally linked.

Similarly, in 2003, David Thatcher published a paper in the Journal of Criminal Law and Criminology arguing that broken windows policing was not as effective as it appeared to be on the surface.

Crime rates dropping in areas such as New York City were not a direct result of this new law enforcement tactic. Those who believed this were simply conflating correlation and causality.

Rather, Thatcher claims, lower crime rates were the result of various other factors, none of which fell into the category of ramping up misdemeanor arrests (Thatcher, 2003).

In terms of the specific factors that were actually playing a role in the decrease in crime, some scholars point to the waning of the cocaine epidemic and strict enforcement of the Rockefeller drug laws that contributed to lower crime rates (Metcalf, 2006).

Other explanations include trends such as New York City’s economic boom in the late 1990s that helped directly contribute to the decrease of crime much more so than enacting the broken windows policy (Sridhar, 2006).

Additionally, cities that did not implement broken windows also saw a decrease in crime (Harcourt, 2009), and similarly, crime rates weren’t decreasing in other cities that adopted the broken windows policy (Sridhar, 2006).

Specifically, Bernard Harcourt and Jens Ludwig examined the Department of Housing and Urban Development program that placed inner-city project residents into housing in more orderly neighborhoods.

Contrary to the broken windows theory, which would predict that these tenants would now commit fewer crimes once relocated into more ordered neighborhoods, they found that these individuals continued to commit crimes at the same rate.

This study provides clear evidence why broken windows may not be the causal agent in crime reduction (Harcourt & Ludwig, 2006).

Falsely Assuming Why Crimes Are Committed

The broken windows theory also assumes that in more orderly neighborhoods, there is more informal social control. As a result, people understand that there is a greater likelihood of being caught committing a crime, so they shy away from engaging in such activity.

However, people don’t only commit crimes because of the perceived likelihood of detection. Rather, many individuals who commit crimes do so because of factors unrelated to or without considering the repercussions.

Poverty, social pressure, mental illness, and more are often driving factors that help explain why a person might commit a crime, especially a misdemeanor such as theft or loitering.

Resulting in Racial and Class Bias

One of the leading criticisms of the broken windows theory is that it leads to both racial and class bias. By giving the police broad discretion to define disorder and determine who engages in disorderly acts allows them to freely criminalize communities of color and groups that are socioeconomically disadvantaged (Roberts, 1998).

For example, Sampson and Raudenbush found that in two neighborhoods with equal amounts of graffiti and litter, people saw more disorder in neighborhoods with more African Americans.

The researchers found that individuals associate African Americans and other minority groups with concepts of crime and disorder more so than their white counterparts (Sampson & Raudenbush, 2004).

This can lead to unfair policing in areas that are predominantly people of color. In addition, those who suffer from financial instability and may be of minority status are more likely to commit crimes in the first place.

Thus, they are simply being punished for being poor as opposed to being given resources to assist them. Further, many acts that are actually legal but are deemed disorderly by police officers are targeted in public settings but aren’t targeted when the same acts are conducted in private settings.

As a result, those who don’t have access to private spaces, such as homeless people, are unnecessarily criminalized.

It follows then that by policing these small misdemeanors, or oftentimes actions that aren’t even crimes at all, police departments are fighting poverty crimes as opposed to fighting to provide individuals with the resources that will make crime no longer a necessity.

Morphing into Stop and Frisk

Stop and frisk, a brief non-intrusive police stop of a suspect is an extremely controversial approach to policing. But critics of the broken windows theory argue that it has morphed into this program.

With broken-windows policing, officers have too much discretion when determining who is engaging in criminal activity and will search people for drugs and weapons without probable cause.

However, this method is highly unsuccessful. In 2008, the police made nearly 250,000 stops in New York, but only one-fifteenth of one percent of those stops resulted in finding a gun (Vedantam et al., 2016).

And three years later, in 2011, more than 685,000 people were stopped in New York. Of those, nine out of ten were found to be completely innocent (Dunn & Shames, 2020).

Thus, not only does this give officers free reins to stop and frisk minority populations at disproportionately high levels, but it also is not effective in drawing out crime.

Although broken windows policing might seem effective from a theoretical perspective, major valid criticisms put the practical application of this theory into question.

Given its controversial nature, broken windows policing is not explicitly used today to regulate crime in most major cities. However, there are still traces of this theory that remain.

Cities such as Ferguson, Missouri, are heavily policed and the city issues thousands of warrants a year on broken window types of crimes – from parking infractions to traffic violations.

And the racial and class biases that result from such an approach to law enforcement have definitely not disappeared.

Crime regulation is not easy, but the broken windows theory provides an approach to reducing offenses and maintaining order in society.

What is the broken glass principle?

The broken glass principle, also known as the Broken Windows Theory, posits that visible signs of disorder, like broken glass, can foster further crime and anti-social behavior by signaling a lack of regulation and community care in an area.

How does social context affect crime according to the broken windows theory?

The Broken Windows Theory proposes that the social context, specifically visible signs of disorder like vandalism or littering, can encourage further crime.

It suggests that these signs indicate a lack of community control and care, which can foster a climate of disregard for laws and social norms, leading to more severe crimes over time.

How did broken windows theory change policing?

The Broken Windows Theory influenced policing by promoting proactive attention to minor crimes and maintaining urban environments.

It led to strategies like “zero-tolerance” or “quality-of-life” policing, focusing on reducing visible signs of disorder to prevent more serious crime.

Braga, A. A., & Bond, B. J. (2008). Policing crime and disorder hot spots: A randomized controlled trial. Criminology, 46(3), 577-607.

Dunn, C., & Shames, M. (2020). Stop-and-Frisk data . Retrieved from

Fagan, J., & Davies, G. (2000). Street stops and broken windows: Terry, race, and disorder in New York City. Fordham Urb. LJ , 28, 457.

Harcourt, B. E. (2009). Illusion of order: The false promise of broken windows policing . Harvard University Press.

Harcourt, B. E., & Ludwig, J. (2006). Broken windows: New evidence from New York City and a five-city social experiment. U. Chi. L. Rev., 73 , 271.

Herbert, S., & Brown, E. (2006). Conceptions of space and crime in the punitive neoliberal city. Antipode, 38 (4), 755-777.

Hunt, B. (2015). “Broken Windows” theory can be applied to real estate regulation- Realty Times. Retrieved from

Jacobs, J. (1961). The Death and Life of Great American Cities . Vintage.

Johnson, C. Y. (2009). Breakthrough on “broken windows.” Boston Globe.

Keizer, K., Lindenberg, S., & Steg, L. (2008). The spreading of disorder. Science, 322 (5908), 1681-1685.

Kelling, G. L., & Sousa, W. H. (2001). Do police matter?: An analysis of the impact of new york city’s police reforms . CCI Center for Civic Innovation at the Manhattan Institute.

Metcalf, S. (2006). Rudy Giuliani, American president? Retrieved from

Morgan, R. E., & Kena, G. (2019). Criminal victimization, 2018. Bureau of Justice Statistics , 253043.

O”Brien, D. T., Farrell, C., & Welsh, B. C. (2019). Looking through broken windows: The impact of neighborhood disorder on aggression and fear of crime is an artifact of research design. Annual Review of Criminology, 2 , 53-71.

Plank, S. B., Bradshaw, C. P., & Young, H. (2009). An application of “broken-windows” and related theories to the study of disorder, fear, and collective efficacy in schools. American Journal of Education, 115 (2), 227-247.

Roberts, D. E. (1998). Race, vagueness, and the social meaning of order-maintenance policing. J. Crim. L. & Criminology, 89 , 775.

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. (2004). Seeing disorder: Neighborhood stigma and the social construction of “broken windows”. Social psychology quarterly, 67 (4), 319-342.

Sridhar, C. R. (2006). Broken windows and zero tolerance: Policing urban crimes. Economic and Political Weekly , 1841-1843.

Stuster, J. (2001). Albuquerque police department’s Safe Streets program (No. DOT-HS-809-278). Anacapa Sciences, inc.

Thacher, D. (2003). Order maintenance reconsidered: Moving beyond strong causal reasoning. J. Crim. L. & Criminology, 94 , 381.

Vedantam, S., Benderev, C., Boyle, T., Klahr, R., Penman, M., & Schmidt, J. (2016). How a theory of crime and policing was born, and went terribly wrong . Retrieved from

Wilson, J. Q., & Kelling, G. L. (1982). Broken windows. Atlantic monthly, 249 (3), 29-38.

Zimbardo, P. G. (1969). The human choice: Individuation, reason, and order versus deindividuation, impulse, and chaos. In Nebraska symposium on motivation. University of Nebraska press.

Further Information

  • Wilson, J. Q., & Kelling, G. L. (1982). Broken windows. Atlantic monthly, 249(3), 29-38.
  • Fagan, J., & Davies, G. (2000). Street stops and broken windows: Terry, race, and disorder in New York City. Fordham Urb. LJ, 28, 457.
  • Fagan, J. A., Geller, A., Davies, G., & West, V. (2010). Street stops and broken windows revisited. In Race, ethnicity, and policing (pp. 309-348). New York University Press.

broken window theory health

U.S. flag

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

  • Publications
  • Account settings
  • Advanced Search
  • Journal List
  • HHS Author Manuscripts

Logo of nihpa

Broken Windows, Informal Social Control, and Crime: Assessing Causality in Empirical Studies

An important criminological controversy concerns the proper causal relationships between disorder, informal social control, and crime. The broken windows thesis posits that neighborhood disorder increases crime directly and indirectly by undermining neighborhood informal social control. Theories of collective efficacy argue that the association between neighborhood disorder and crime is spurious because of the confounding variable informal social control. We review the recent empirical research on this question, which uses disparate methods, including field experiments and different models for observational data. To evaluate the causal claims made in these studies, we use a potential outcomes framework of causality. We conclude that, although there is some evidence for both broken windows and informal control theories, there is little consensus in the present research literature. Furthermore, at present, most studies do not establish causality in a strong way.


A contemporary criminological controversy concerns the interrelationships among neighborhood disorder, informal social control, and crime. This controversy derives from a rich set of theoretical ideas explaining these relationships. According to Wilson & Kelling’s (1982) broken windows thesis, physical and social disorders exert a causal effect on criminal behavior. Disorder does so directly, as it signals to criminals community indifference to crime, and indirectly, as disorder undermines informal social control. By contrast, theories of informal social control argue that the association between disorder and crime is not causal but is instead spurious because of the confounding variable neighborhood informal control ( Sampson & Raudenbush 1999 ). This theoretical divergence has important implications for criminological theory and public policy. Therefore, the conclusions of empirical research on this controversy are of paramount importance. This review discusses the controversy between broken windows and informal social control by reviewing the current state of empirical research. Perhaps the most important question in evaluating the empirical literature is the degree to which studies approximate causal relations. We use a potential outcomes or counterfactual definition of causality, which has gained prominence in statistics and social science ( Morgan & Winship 2015 , Rubin 2006 ), to assess recent research. We try, whenever possible, to give our own assessments of the relative strengths and weaknesses of the studies—our evaluation of the plausibility of the assumptions made in different research designs. This assessment is open to debate and criticism, but we feel that stating our opinion provides a point of departure for subsequent debate. We conclude our discussion with what we think are important avenues for future research.

Rather than exhaustively covering all studies, we focus on those that are well executed both theoretically and methodologically. Because we are principally concerned with how well studies approximate causality, we organize our discussion by methodological design. We acknowledge that causality is not the only important issue for evaluating empirical studies. Extensive literature exists on the important issues of proper measurement of disorder and informal control ( Hipp 2007 , 2010 , 2016 ; Kubrin 2008 ; Sampson & Raudenbush 2004 ; Skogan 2015 ; Taylor 2001 , 2015 ), implications for public policy—particularly order maintenance policing ( Braga et al. 2015 , Fagan & Davies 2000 , Harcourt 1998 , Kelling & Coles 1997 , Weisburd et al. 2015 )—and micro–macro relationships ( Matsueda 2013 , 2017 ; Taylor 2015 ). We set these aside, referring the reader to the extant literature. We also set aside detailed examination of observational studies of individual-level mechanisms of broken windows evaluated recently by O’Brien et al. (2019) , including fear of crime ( Hinkle 2015 ).


Social disorganization theory.

From their exhaustive mapping of delinquency across Chicago neighborhoods, Shaw & McKay (1931 , 1969 ) identified a strong statistical association between disorder and delinquency in which delinquency clustered in zones of transition, characterized by rapid population turnover, impoverished immigrant groups, and few homeowners. Also present were signs of physical and social disorder: dilapidated buildings, vacant lots, homeless and unsupervised youth, panhandling, and other incivilities. Delinquency rates followed a gradient—highest in the central city and progressively lower in the periphery—and remained that way over decades despite drastic changes in neighborhood ethnic composition. To explain these patterns, Shaw & McKay (1969) developed their theory of social disorganization and cultural transmission, in which rapid in- and out-migration and lack of homeownership, as well as high rates of poverty, ethnic diversity, and immigrants, undermined local social organization. Social disorganization—weak and unlinked local institutions—led to unsupervised street youth, who forged a delinquent tradition transmitted from older gangs to unsupervised youth. Shaw & McKay treated physical and social disorder as a manifestation—and, consequently, an indicator—of social disorganization. Disorder does not cause crime but instead indexes disorganization, which causes crime via weak informal control, the prevalence of unsupervised youth, and the creation and transmission of a delinquent tradition across age-graded youth groups.

Broken Windows Theory

Wilson & Kelling’s (1982) broken windows thesis posits that disorder and crime are causally linked in a developmental sequence in which unchecked disorder spreads and promotes crime. Both physical disorder (e.g., abandoned buildings, graffiti, and litter) and social disorder (e.g., panhandlers, homeless, unsupervised youths) exert causal effects on crime directly and indirectly. Directly, disorder signals to potential criminals that residents are indifferent to crime, emboldening criminals to commit crimes with impunity. This individual-level causal mechanism implies a rational actor: Motivated offenders perceive disorder to mean the absence of capable guardians ( Cohen & Felson 1979 ). Indirectly, disorder induces residents to fear crime, which causes them to avoid unfamiliar people, restrict travel to safe spaces, and withdraw from public life. Disengaged from the neighborhood, fearful residents increasingly feel that combatting disorder and crime is the duty of others. Ironically, signs of local disorder create fear of crime in residents because they assume a causal effect of disorder on crime. Eventually, as disorder and crime increase, residents with sufficient resources begin to leave the neighborhood, taking their capital with them, which undermines both community resources and the capacity for informal social control ( Wilson & Kelling 1982 ). This indirect effect is a neighborhood-level causal mechanism: Rampant disorder causes residents to withdraw, eroding neighborhood control, which fosters crime.

These two pathways form feedback loops, creating a cascading effect of crime and disorder spreading across physical spaces. As Wilson & Kelling (1982) note, one broken window (signaling indifference) is often followed by another and so on, until all windows are broken. This is an informational cascade, as the observation of disorder and crime provides information signaling the absence of social control. Disorder causes residents to withdraw from the community, weakening objective informal social control, and fostering additional crime, disorder, and incivilities, which, in turn, further undermine informal control, leading to more crime. This is a social interactional cascade in which the key causal mechanism is local residents disengaging from community attempts to control disorder and crime. Left unabated, these feedback loops can produce a crime epidemic spreading across time and space.

Figure 1 diagrams the causal relationships among disorder, informal social control, and crime specified by broken windows. Due to reciprocal pathways, this is a nonrecursive model that is underidentified for cross-sectional data without additional information such as instrumental variables (IVs) or a panel design with repeated observations.

An external file that holds a picture, illustration, etc.
Object name is nihms-1686796-f0001.jpg

Conceptual model of broken windows theory: disorder, social control, and crime. Two paths link disorder to crime: a direct path, in which ( a ) disorder signals community indifference, which increases crime; and an indirect path, in which ( b ) disorder elicits actual community indifference, which weakens social control, which in turn ( c ) increases crime. These effects are reinforced as ( d ) weakened social control stimulates more disorder and ( e ) crime weakens social control. Two feedback pathways ( d and e ) mean this is a nonrecursive model.

Collective Efficacy Theory of Informal Social Control

Sampson (2012) and others (e.g., Morenoff et al. 2001 , Sampson et al. 1997 ) extend Shaw & McKay’s theory of social disorganization by refining the causal mechanism of informal control, which translates neighborhood social organization into safe neighborhoods. They argued that social cohesion, including social capital, is a crucial resource for neighborhoods to solve problems collectively. Drawing from Coleman (1990) , they asserted that neighborhoods rich in social capital—intergenerational closure (parents know the parents of their children’s friends), reciprocated exchange (neighbors exchange favors and obligations), and generalized trust—have greater resources to prevent neighborhood disorder, incivilities, and crime. Such resources are translated into action via child-centered control. Borrowing from Bandura (1986) , they called this entire causal sequence collective efficacy ( Sampson et al. 1997 ). Sampson et al. (1999) specified potential spillover effects for collective efficacy, in which collective efficacy in one neighborhood affects contiguous neighborhoods, producing a social interactional cascade.

Sampson & Raudenbush (1999) used collective efficacy theory to specify causal relationships among disorder, informal control, and crime and in the process offered a critique of broken windows theory. They maintained that collective efficacy not only keeps neighborhoods safe but also keeps them clean. Because social disorder, physical disorder, and crime pose similar problems, neighborhoods high in collective efficacy are able to combat all three problems. Sampson & Raudenbush (1999) argue that, in contrast to broken windows, the correlation between disorder and crime is spurious due to the confounding variable, collective efficacy. Figure 2 depicts the collective efficacy model of disorder, informal control, and crime. This model is a restrictive recursive model nested within the broken windows model. If these restrictions are valid—crime and disorder are related solely because of confounding by exogenous collective efficacy—this model is fully recursive and identified.

An external file that holds a picture, illustration, etc.
Object name is nihms-1686796-f0002.jpg

Conceptual model of collective efficacy theory: disorder, social control, and crime. The direct path between disorder and crime is spurious (A = 0), and collective efficacy is an exogenous cause of both crime and disorder (B,E = 0). This is a recursive model.


Broken windows and collective efficacy specify competing causal relationships among disorder, informal control, and crime. To adjudicate empirically between the two requires research methods that closely approximate causal relations. To evaluate the disparate research designs used in the empirical literature, we need a framework for establishing causality. The potential outcomes framework, or Rubin causal model, is an approach to causal inference based on counterfactual reasoning using the ideal of a controlled experiment. Rather than considering only the factual statement “a given treatment happened and we observed a particular outcome,” one also considers the counterfactual statement “if a given treatment had not happened, we would have observed a particular (potential) outcome.” These two statements correspond to treatment and control groups in an ideal controlled experiment. Treatment here refers to a variable of primary interest believed to have a causal effect on the outcome under examination. In the classic experimental design, values of the treatment are assigned (manipulated) by the investigator (e.g., in randomized controlled trials, treatments are randomly assigned to subjects). For a variable to be a cause, it must have been manipulated—or, short of that, at least be manipulable in principle ( Holland 1986 ). Thus, this framework is sometimes termed an interventionist definition of causality ( Woodward 2003 ). Although potential outcome(s) is not the only causal framework ( Morgan & Winship 2015 ), it has increasingly become the dominant approach to causality in statistics and the social sciences.

If Y i 1 is the potential outcome of individual i in the treatment state, and Y i 0 is the potential outcome of individual i in the control group, the individual treatment effect is

The fundamental problem of causal inference is that for those in the treatment group, we cannot observe their outcome in the control group; conversely, for those in the control group, we cannot observe their outcome in the treatment group ( Holland 1986 ). Therefore, we cannot compute individual causal treatment effects (Δ i ). Under additional assumptions, however, we can estimate (causal) average treatment effects (ATEs). For example, we can assume, in a randomized experiment with a treatment group and a control group, treatment assignment is ignorable:

where T = 0 , 1 denotes treatment assignment, and ⊥ denotes statistical independence. Here, the difference in the sample means for assignments T = 1 and T = 0 estimates E ( Y 0 − Y 1 ), the ATE of T on Y .

In an observational study, Equation 2 is unlikely to hold due to selectivity or confounding, but treatment assignment may be ignorable after conditioning on covariates Z :

Equation 3 includes the additional identification condition that at each level of the covariates, there is a positive probability of receiving either treatment. The set of conditions described in Equation 3 is known as strong ignorability given covariates ( Rosenbaum & Rubin 1983 ). Here, the conditional ATE (CATE) E ( Y 1 − Y 0 | Z = z ) can be used to estimate ATEs using a properly specified regression or propensity score match that includes all relevant covariates Z . The major difficulty of establishing causality in observational (nonexperimental) studies is the problem of controlling for all relevant Z to achieve conditional ignorability.

Observational studies of disorder, informal control, and crime have used different methods to approximate CATEs. Cross-sectional studies of neighborhoods use observed covariates to control for confounding, under the assumption of no reciprocal causation. Cross-lagged panel models relax this assumption and examine lagged endogenous predictors over time, under the assumptions that there is sufficient temporal variation to obtain stable estimates and that observed covariates achieve conditional ignorability. Fixed-effects panel models relax the assumption that all relevant time-stable confounders are included in the model. By estimating within-individual (neighborhood) variation over time, fixed-effects models control for both observed and unobserved time-stable covariates. Fixed-effects models, however, still require that all relevant time-varying confounders are included. In reviewing the empirical literature on disorder and informal control, we will assess the degree to which studies achieve conditional ignorability.

The definition of unit causal effects makes the stable unit treatment value assumption (SUTVA), a term coined by Rubin (1986 , p. 961): “the value of Y for unit u when exposed to treatment t will be the same no matter what mechanism is used to assign treatment t to unit u and no matter what treatments the other units receive.” SUTVA implies two distinct assumptions: consistency and absence of interference. Consistency means that the mechanism used to assign treatment can be ignored because the outcomes of the treated observations will be invariant to different assignment mechanisms. Thus, an individual assigned a treatment in an experimental setting exhibits the same outcome as if they naturally received the treatment in the real world. Consistency is less likely in experimental settings because the treatment assignments are carried out by the researcher instead of assigned through natural processes in the real world. Results of the experiment may not generalize to real-world settings because of differences in the treatment-assignment process. By contrast, consistency is more likely to hold for observational data given that treatments are assigned naturally in the real world and not through an artificial assignment process.

No interference means that the treatment assignment of one subject does not affect the outcomes of other subjects. This form of contamination can bias treatment estimates in both experimental and nonexperimental designs. Interference often arises via social processes such as spillover effects, displacement, and cascades ( Matsueda 2017 , Nagin & Sampson 2019 ). Interference violates the assumption typically made in observational studies of identically and independently distributed observations (conditional on covariates). When the form of dependence is known, it can be addressed by specific models, such as autoregressive spatial models for spillover effects between contiguous observations or social network models of contagion across individuals.

Beginning with experimental studies, we review the quantitative empirical research on disorder, informal control, and crime, with an eye toward adjudicating between competing theories of broken windows and collective efficacy and evaluating causal claims.


Controlled experiments begin with treatment and control groups and manipulate the treatment by intervening in the experimental group. The key to achieving ignorability is ensuring that treatment and control groups are equivalent before the intervention. Randomized experiments ensure groups are probabilistically equivalent by randomly assigning subjects to groups. Because the treatment is manipulated by the researcher, it is strongly exogenous to the outcome, ruling out reciprocal causation. Thus, a well-executed randomized experiment can achieve ignorability and, therefore, strong internal validity. The external validity of experiments is often compromised in three ways. First, observations are rarely obtained through representative sampling, limiting inferences to larger populations of interest. Second, treatment assignments of experiments often differ from the natural way that treatments are assigned in the real world, compromising consistency, and therefore causality, and again limiting generalizability to relevant populations. Third, interference can occur. In individual-level studies, subjects may influence each other on the basis of treatment assignment; in aggregate spatial studies, treatment effects may spill over and affect contiguous aggregate units.

Although a number of studies have attempted to test aspects of broken windows and informal social control using laboratory experiments (e.g., Diekmann et al. 2015 , Engel et al. 2014 ), it is our opinion that no studies we found exhibited sufficient external validity to provide evidence for or against the causal pathways in Figure 1 . Consequently, we examine only field experiments. Compared with laboratory experiments, field experiments trade off some internal validity for gains in external validity. They are conducted in the natural social contexts in which disorder, informal control, and crime are likely to occur and use local subjects who are the typical actors involved. Field experiments use interventions that closely approximate the real-world treatment of interest. This increase in external validity comes at a cost. Because they are conducted in natural settings, field experiments are unable to control the environment of the experiment and, thus, less able to rule out potential confounding factors. Interference, in which subjects of a treatment group affect the outcomes of others, is more likely to occur. Field experiments are typically conducted in a single or small number of geographical locations and rarely use representative sampling of locations or subjects.

A number of field experiments examine the individual-level hypothesis, derived from broken windows, that disorder exerts a direct causal effect on crime (norm violation). The most significant and highly cited broken windows field experiment, Keizer et al.’s (2008) study published in Science , has led to a resurgence of interest in the use of field experiments to study broken windows. Therefore, we discuss this study in some detail. Following Cialdini (2003) , Keizer et al. (2008) conceptualize broken windows as a cross-norm inhibition effect. Descriptive norms reflect common behaviors in a setting, while injunctive norms reflect what is commonly held to be proper in society. Observing a descriptive norm (e.g., seeing litter on the ground) that conflicts with an injunctive norm (e.g., it is wrong to litter) inhibits other injunctive norms as well (e.g., it is wrong to steal). Disorder thus causes crime by reducing inhibitions against criminal behavior. Keizer et al. (2008) conducted six related field experiments in which they manipulated a signal that a contextual norm had been violated (treatment) and then observed whether a target (injunctive) norm was more likely to be violated (outcome).

In the first experiment, the contextual norm was antigraffiti and the target norm was antilittering. Flyers were placed on the handlebars of bicycles parked in an alley of a shopping district. On the wall was a “no graffiti” sign. For the treatment condition, the wall was covered with graffiti; for the control condition, no graffiti was visible. The dependent variable was whether the owner of the bicycle littered the flyer upon returning. Of 77 subjects in each condition, 33% littered in the control condition compared with 69% in the treatment condition, a significant difference. Another experiment placed flyers on bicycles parked in a bicycle shed near a train station, with a treatment condition of the sound of fireworks set off illegally. Again, differences in the incidence of littering the flyers were significant: 26 (52%) littered in the control condition compared to 37 (80%) in the fireworks condition.

A third experiment used a private setting of a supermarket parking garage. The contextual norm was indicated by a “please return your shopping carts” sign and the outcome was littering flyers attached to the windshields of parked cars. In the treatment condition, with unreturned shopping carts strewn about, 35 of 60 (58%) shoppers littered the flyer compared with 13 of 60 (30%) shoppers in the control condition. The fourth experiment used a public setting of a car parking lot, in which the contextual norm was designated by a police ordinance “locking bicycles to the fence is prohibited” sign on a fence outside the lot. The target norm was indicated by a second police ordinance “do not enter” sign at an opening of the fence. In the treatment condition in which four bicycles were locked to the fence, 40 of 49 (82%) subjects violated the “do not enter” sign, whereas only 12 of 44 (27%) violated the norm in the control condition.

The experiment most relevant to broken windows examined theft. Keizer et al. (2008) left a mailing envelope—which was addressed and stamped and had a 5-euro note visible in the envelope’s window—hanging out of a mailbox’s mailing slot. The dependent variable was whether passersby stole the envelope. Two treatment conditions consisted of litter on the ground near the mailbox ( N = 72) and graffiti spray-painted on the mailbox ( N = 60). In the control condition of no graffiti and litter ( N = 71), nine passersby (13%) stole the envelope, compared to 18 (25%) in the litter condition and 16 (27%) in the graffiti condition.

Keizer et al.’s (2008) article is a citation classic, having received nearly 1,000 Google Scholar citations in approximately ten years. It has also spawned a number of studies of broken windows using similar research designs. Nevertheless, the paper has been subject to sharp criticism. Wicherts & Bakker (2014) , in particular, argue that the study is fraught with methodological weaknesses, such as failing to address potential confounding, observer bias, and measurement error; using inflated Type I error rates due to dependencies among subjects; using sequential testing; and failing to control for multiple testing. A major limitation of the study is that each experiment was carried out in a single geographic neighborhood in Groningen, Netherlands, compromising external validity. This criticism has been partially addressed by attempts to replicate Keizer et al.’s (2008) results in different settings. Volker (2017) attempted to replicate Keizer et al.’s (2008) mailbox letter theft experiment in the identical neighborhood as the original study and failed to find significant effects. In their follow-up, Keizer et al. (2011) found the effect of disorder on norm violation stronger in the presence of a sign prohibiting the form of disorder present; however, Wicherts & Bakker (2014) offered similar criticisms to those leveled at the first study. A third study found negative effects of norm violation on prosocial behavior ( Keizer et al. 2013 ).

Keuschnigg & Wolbring (2015) replicated Keizer et al.’s (2008) experiments in two German neighborhoods differing in social capital measured with administrative data. With abandoned and damaged bicycles as a disorder manipulation, they dropped envelopes with five-, ten-, or one-hundred-euro notes and used theft of the envelopes as the outcome. They found treatment heterogeneity: The probability of envelope theft was higher in the disorder condition but only in the low-social-capital neighborhood and for smaller monetary values. This study is significant because it attempts to address the role of informal social control in the disorder–crime relationship. A drawback is that using only two neighborhoods to control for social capital ignores myriad other differences between neighborhoods that may affect theft.

Berger & Hevenstone (2016) conducted field experiments testing the relationship between litter and sanctioning of litterers in Bern and Zurich, Switzerland, and New York. A confederate dropped a bottle near a trash receptacle in view of pedestrians while another recorded whether participants verbally sanctioned the confederate, subtly sanctioned (e.g., an angry glance) the confederate, or picked up the dropped bottle. The researchers manipulated the treatment conditions by introducing bags of garbage and stray litter or conducting the drop farther from the trash can. The manipulation moderately reduced both forms of sanctioning and strongly reduced picking up the bottle. In contrast, dropping the bottle farther from the trash receptacle reduced picking up the bottle but had no effect on sanctioning. Berger & Hevenstone (2016) interpreted their findings as a local effect of litter on both informal social control and cleanup of additional litter, which could produce a cascade effect of littering. They note the presence of interference: In 6.4% of trials, after a participant reacted to the littering, a second individual subsequently joined in sanctioning the confederate. Thus, sanctioning may be a contagious behavior.

These individual-level experiments approximate ignorability by manipulating treatments of graffiti and litter and using naturally occurring passersby, making treatment and control subjects different by the timing of their appearance. Thus, unless treatment and control conditions differ by some confounding event occurring for one but not the other, equivalence seems assured. Furthermore, because the treatment conditions are one-shot transitory events, subjects are unlikely to interfere with each other. The transitory nature of treatment, however, means that these experiments cannot test the hypothesis that repeated exposure to disorder is necessary for norm violations.

A second set of field experiments intervene at the neighborhood level and examine aggregate neighborhood outcomes (see Kondo et al.’s 2018 review). Branas et al. (2018) conducted a randomized experiment of neighborhood disorder in which the main intervention cleaned up physical disorder in vacant lots, created a park-like atmosphere, and maintained the lots on a regular schedule. A second intervention only cleaned up physical disorder. They found that the main intervention, but not cleanup alone, was significantly negatively associated with survey-recorded perceived crime, vandalism, and staying inside due to safety concerns, and positively associated with socializing outside. The main intervention was also positively associated with people watching out for each other but only in neighborhoods below the poverty line. By contrast, both the main intervention and cleanup alone were negatively associated with an index of crimes.

Branas et al. (2018) estimated intent-to-treat models, which estimate treatment effects regardless of whether experimental subjects complied with treatment. If a policy implementing the treatment would result in similar noncompliance, intent-to-treat estimates will give the policy effect expected in the real world. By contrast, if interest is in the effect of actual neighborhood disorder, noncompliance is a problem, and intent-to-treat estimates may be biased. To overcome this, a model controlling for noncompliance uses the randomized intent-to-treat variable as an IV for compliance, yielding an unbiased estimate of the complier average causal effects (CACE) (see Imbens & Rubin 2015 ). Branas et al. (2018) found that CACE and intent-to-treat estimates were similar, suggesting that noncompliance was not a major problem.

This study suggests that neighborhood disorder may undermine social cohesion as well as increase neighborhood crime. The use of randomization ensures ignorability. The manipulated treatment—cleaning up vacant lots—suggests a treatment amenable to public policy intervention, where noncompliance is likely to occur. Nevertheless, we cannot rule out the ecological fallacy because we cannot know for certain if the individuals perceiving high disorder are the same ones withdrawing from the community or committing more crimes. In a related study, Branas et al. (2016) found a stronger reduction in firearm violence near vacant buildings that were boarded up and had their exteriors cleaned. The relevance of these studies to broken windows hinges on an untested assumption of symmetric causality ( Lieberson 1985 ): Removing disorder reduces crime; therefore, introducing disorder increases crime. Researchers clearly cannot introduce disorder at the neighborhood level due to the potential for harm but it is conceivable that natural experiments—sudden exogenous increases in disorder—could be exploited to confirm this symmetry.

A third set of experiments come from the moving to opportunity (MTO) studies, which used a randomized quasi-experimental design ( Harcourt & Ludwig 2006 ). Beginning in 1994 in five major cities, MTO randomly assigned 4,600 low-income families—who were living in public housing or Section 8 project-based housing in high-poverty neighborhoods—to one of three groups. A treatment group was offered housing vouchers for moving to neighborhoods having poverty rates of ten percent or less. A Section 8 group was offered housing vouchers to move to any neighborhood. A control group was not offered housing vouchers. Random assignment rules out the potential biasing effects of self-selection into neighborhoods. Approximately half of the families complied with the treatment by relocating through MTO; therefore, intent-to-treat (ITT) models of the opportunity to move were augmented with ATEs on the treated (ATT) using treatment assignment as an IV. Compared with controls, members of the treatment and Section 8 groups moved to neighborhoods lower in poverty and higher in reported informal social control. Nevertheless, in neither ITT nor ATT models did the treatment groups show lower arrest rates, delinquency, or behavior problems by 2001 ( Kling et al. 2005 ). Harcourt & Ludwig (2006) conclude that broken windows is unsupported: Either declines in community disorder do not reduce criminality or their effects are offset by increases in neighborhood socioeconomic status.

Although MTO studies rule out self-selection, for our purposes, they have three weaknesses. First, the treatment is a compound treatment, consisting of movement to a neighborhood with different poverty, disorder, collective efficacy, and other unmeasured characteristics. The studies cannot distinguish between causal effects of these different treatments. Second, analyses do not consider spillover effects. Sobel (2006) has argued persuasively that the no-interference assumption may have been violated. Families given vouchers may be reluctant to move unless their neighborhood friends also move and may be unable to find suitable housing in a tight housing market when many others are given vouchers. Such interference could bias estimated treatment effects. Third, Sampson (2008) suggested that studies using MTO data must be interpreted carefully: Any treatment effect also includes disruptive effects of moving; voucher users moved to destinations lower in poverty but similar on other indicators of disadvantage and embedded in larger disadvantaged areas; and the treatment resulted in only modest changes in conditions. The sample also constitutes a small, highly disadvantaged group subject to years of cumulative deprivation, limiting external validity.

Our review of field experiments of disorder, social control, and crime suggests mixed results. Keizer et al. (2008) consistently conclude that their experiments support the broken windows thesis (direct effect of disorder on norm violation), but those experiments have been criticized on methodological grounds and have failed to replicate in one instance and were replicated only in a neighborhood poor in social capital in another. Some experimental evidence finds support for the effects of disorder on crime at the neighborhood level, although the mechanism is unclear. Disorder may also impede social control behavior at the individual level.


Observational studies typically combine survey data on individuals within neighborhoods with administrative data from police and the census. In principle, such nested data allow estimation of combined micro–macro models, but in practice, this research typically models macro relationships among variables aggregated to neighborhoods. A major advantage of observational studies of neighborhoods is they examine natural variation across a representative sample of neighborhoods, making external validity strong. Because the research environment lacks the controls of experiments, however, there are greater threats to internal validity, as ignorability is difficult to approximate. Observational designs differ in how they address the problem of ignorability. Research on disorder and control can be categorized into four nonexperimental designs and models: ( a ) recursive cross-sectional models; ( b ) nonrecursive simultaneous equation models; ( c ) fixed-effects panel models; and ( d ) cross-lagged panel models.

Cross-Sectional Recursive Models

Cross-sectional recursive models are commonly used to examine community theories of crime. The key independent variables, disorder and social control, are not manipulated by the researcher but rather are endogenous. To rule out what econometricians term endogeneity bias, these models rely on two strong assumptions. First, treatment assignment—the process by which neighborhoods attain a level of social control or disorder—is ignorable conditional on exogenous control variables. Thus, all relevant covariates are included in the model. Second, reverse causality—crime affecting either disorder or social control—is absent.

In an important cross-sectional study of neighborhood disorder and crime, Skogan (1990) pooled surveys covering 40 areas in six major US cities to examine a model of disorder and decay. Under the assumption that his path models were well specified, he found a strong effect of perceived disorder on crime, in which disorder mediated the effects of poverty, residential instability, and racial heterogeneity on crime, supporting broken windows. Disorder was measured with an index of social and physical disorder indicators that displayed convergent and discriminant validity. Skogan noted that informal social control is negatively correlated with disorder; however, he did not include it in his structural models. Harcourt (1998) reanalyzed Skogan’s data, excluded a small number of neighborhoods with unusually high disorder and crime, and found results were not robust, although, as Xu et al. (2005) pointed out, Harcourt’s reanalysis lacks statistical power.

Cross-sectional studies have also supported theories of informal control (e.g., Sampson & Groves 1989 ). Using police data, census data, and survey data on 8,782 residents from 343 Chicago neighborhoods from the Project on Human Development in Chicago Neighborhoods (PHDCN), Sampson et al. (1997) examined collective efficacy, sociodemographic structure, and crime. Collective efficacy was measured by residents’ reports of informal control (e.g., would neighbors intervene if children were committing deviance?) and social cohesion (e.g., neighbors help each other) adjusted for differential composition of informants across neighborhoods. Using three-level hierarchical linear modeling (HLM) models, they found that collective efficacy was strongly negatively related to survey-measured victimization and perceived crime as well as police-reported homicides. Collective efficacy also mediated much of the effects of neighborhood disadvantage, residential instability, and immigration. Although this study was carefully conducted—particularly in addressing measurement issues—the cross-sectional design could not rule out reciprocal effects.

Subsequent cross-sectional studies replicated Sampson et al.’s (1997) results in other settings or under varying specifications (e.g., Mazerolle et al. 2010 , Sampson & Wikström 2008 , but not Bruinsma et al. 2013 ). Matsueda & Drakulich (2016) augmented Sampson et al.’s (1997) models to adjust individual perceptions of collective efficacy for perceived deviance. Using the Seattle Neighborhoods and Crime Survey, they found that respondents who observe deviance in their neighborhood report a lower likelihood of neighbors intervening. Nevertheless, controlling for observed deviance resulted in a stronger association between collective efficacy and crime.

Neighborhoods in closer proximity tend to be more similar than those far apart, resulting in spatial autocorrelated data. This can be due to substantive processes, such as spillover or cascade effects, or methodological artifacts, such as spatial mismatch in which the true unit of analysis in causal processes differs from the unit used in the study (see Sampson et al. 1999 , Taylor 2015 ). In either case, the result can be interference as, for example, social capital in one neighborhood (treatment) spills over into a low-social-capital neighborhood (control), lowering its crime rate. Sampson et al. (1999) reanalyzed PHDCN data using first-order spatial autoregressive lag models and found evidence of spillover in neighborhood collective efficacy.

Morenoff et al. (2001) reanalyzed PHDCN data to explore spillover effects in models of social ties, collective efficacy, and future homicide rates. By controlling for homicide rates for three years preceding the survey, they partially address ignorability as prior homicide partly absorbs unobserved (omitted) stable covariates. They find organizations and social ties important for predicting collective efficacy (but not crime). They also find that collective efficacy predicts lower homicide rates but does not mediate socioeconomic disadvantage as strongly as found in Sampson et al. (1997) . Building on this model, Browning et al. (2004) found that collective efficacy predicts lower homicide rates, but the effect is attenuated in the presence of dense social ties (see also Bellair & Browning 2010 ). This suggests that measures of network density moderate collective efficacy and cannot serve as a proxy for informal control (e.g., Markowitz et al. 2001 ).

In sum, cross-sectional studies find effects of neighborhood disorder on crime and effects of informal control on crime that persist in the face of spatial autoregression. These studies do not manipulate disorder or informal control and thus make strong assumptions about causal order (no reciprocal causation) and ignorability (controls for social disorganization achieve conditional ignorability).

Nonrecursive (Simultaneous Equation) Models

In principle, simultaneous equation models with IVs resolve the problem of reverse causality for nonexperimental studies in which treatments are not manipulated but rather are assigned through an endogenous process (e.g., Greene 2003 ). Figure 3 depicts a nonrecursive model in which social control and crime are simultaneously determined. The problem here is that, in the crime (social control) equation, the endogenous predictor, social control (crime), is correlated with the disturbance ε 1 ( ε 2 ), which violates a key assumption of the general linear model, causing estimates to be biased. To resolve this, at least one IV that strongly predicts social control but has no direct effect on crime—holding social control constant—is needed to identify the effect of social control on crime. Similarly, at least one IV that strongly predicts crime but not social control—holding crime constant—is needed to identify the effect of crime on social control. These exclusionary restrictions, in which γ 1 = γ 2 = 0, are indicated in Figure 3 .

An external file that holds a picture, illustration, etc.
Object name is nihms-1686796-f0003.jpg

Nonrecursive model of social control and crime with lagged instrumental variables (IVs). Social control and crime are reciprocally related ( β 1 , β 2 ) with correlated errors ( ε 1 , ε 2 ). The assumption that the IVs for social control and crime have no cross-lagged effects—signified by the restrictions γ 1 = γ 2 = 0 ( dotted lines )—permits identification.

Sampson & Raudenbush (1999) used nonrecursive models to estimate causal relationships among collective efficacy, disorder, and crime while controlling for crime’s effect on collective efficacy. They used systematic social observation (SSO), an innovative method of measuring disorder across neighborhoods: Videos of street blocks taken by SUVs driving through streets of Chicago during daylight were coded for signs of social and physical disorder. Following Sampson et al. (1997) , they measured collective efficacy using a multilevel measurement model of PHDCN data; crime is captured by police-reported homicide, robbery, and burglary. To identify the reciprocal effects between collective efficacy and crime, the authors assumed that reciprocated exchange among neighbors and attachment to neighborhood (IVs for collective efficacy) affect crime only indirectly through collective efficacy. They used geocoded victim-based homicides from death records as an IV for police-reported crimes under the assumption that resident-based homicide affects collective efficacy solely through police-reported crimes. Sampson & Raudenbush (1999) found, contrary to broken windows, no direct relationship between disorder and crime net of collective efficacy for homicide or burglary. They did, however, find support for a “feedback loop, whereby disorder entices robbery, which in turn undermines collective efficacy, leading over time to yet more robbery” ( Sampson & Raudenbush 1999 , p. 637). Researchers have critiqued Sampson & Raudenbush for assuming that crime does not feedback on disorder and disorder does not feedback on collective efficacy, and therefore, there could be an indirect pathway of disorder on crime through collective efficacy ( Gault & Silver 2008 , Xu et al. 2005 ). O’Brien & Kauffman (2013) replicated Sampson & Raudenbush’s (1999) main results using a survey of rural youth with respondent prosociality—rather than crime—as an outcome. They found rater-assessed and respondent-perceived disorder were unrelated to prosociality, but collective efficacy predicted both disorder and low adolescent prosociality. Although Sampson & Raudenbush specify collective efficacy as a composite of cohesion and expectations for social control, Taylor (1996) found disorder negatively related to social control and positively related to cohesion—effects that canceled out in reduced forms.

Sampson & Wikström (2008) used data from 3,992 individuals in 200 Stockholm neighborhoods and Chicago’s PHDCN to make a cross-national comparison of relationships among collective efficacy, perceived disorder, and crime, controlling for indicators of social disorganization. They found collective efficacy negatively associated with neighborhood crime and victimization in both cities. Sampson & Wikström (2008) found that, controlling for collective efficacy, disorder had a strong positive association with reported violent crimes in Stockholm but not in Chicago. Although these results provide evidence for cross-national consistency in collective efficacy, they also reveal the disorder–crime relationship in Stockholm survives controlling for confounding collective efficacy, a finding that supports broken windows.

Using data from the British Crime Survey, Markowitz et al. (2001) applied nonrecursive models to relationships among burglary, social cohesion, fear of crime, and perceived disorder. To identify the simultaneous parameters, they used lagged versions of endogenous variables as IVs. Thus, the exclusion restriction is no lagged effects in the presence of contemporaneous effects. The authors used survey measures of disorder (e.g., neighborhood litter, vandalism, and loitering teenagers), social cohesion (organizational participation, helping behavior, and neighborhood satisfaction), and fear of crime (fear walking after dark and worried about burglary or robbery). They summed and aggregated the indicators to create neighborhood-level indices. They controlled for disorganization (ethnic heterogeneity, family disruption, and urbanization) to achieve ignorability. In cross-sectional models, Markowitz et al. (2001) found a nonsignificant effect of disorder on burglary, holding constant cohesion and previous burglary. Both nonrecursive and cross-lagged panel models reveal cohesion and disorder are reciprocally related (each at −0.18 standardized), as are burglary and disorder. Furthermore, they found a feedback loop in which social cohesion reduced burglary and disorder, each of which increased fear of crime, which in turn, fed back to reduce social cohesion. These findings generally support broken windows.

Previous studies using nonrecursive models also found evidence that crime—particularly robbery—is associated with less informal social control. Liska & Warner (1991) modeled the reciprocal effects between crime and constrained social behavior (a combination of fear of crime, going out at night, and limiting activities because of crime). Using the US National Crime Survey, they analyzed crime victimization (robbery and a general index of felony crimes) for 26 cities. To identify their simultaneous equations, they used population density as an instrument for crime (assuming no direct effect on constrained social behavior) and media coverage of homicide for constrained social behavior (assuming no direct effect on crime). These are very strong assumptions. They find a reciprocal relationship between robbery and constrained social behavior: Constrained social behavior reduces robbery and other victimization, but robbery also increases constrained social behavior.

Using data on 100 Seattle neighborhoods, Bellair (2000) estimated nonrecursive models of informal surveillance and crime, comparing burglary with a combined measure of robbery and stranger assault. Following Sampson & Raudenbush (1999) , Bellair used reciprocated exchange among neighbors as an instrument for informal control. He used unsupervised teenage groups as an instrument for crime. Because the presence of unsupervised teens is likely to elicit more surveillance even when crime is held constant, Bellair (2000) tried other instruments for crime, including percentage of bars and clubs in the neighborhood and percentage of 16–19-year-olds in the neighborhood (note that each instrument could also affect surveillance directly). Nevertheless, Bellair found that robbery and stranger assault reduce informal surveillance by increasing perceived risk of victimization, which leads to withdrawal from public spaces. Conversely, burglaries lead to greater surveillance. Furthermore, controlling for perceived risk of attack, surveillance is negatively associated with robbery and assault, but not burglary.

In sum, nonrecursive models appear to find a reciprocal causal relationship between informal control (social cohesion) and disorder. Sampson & Raudenbush (1999) found collective efficacy reduces crime but not vice versa, and the effect of disorder on crime is largely spurious due to the confounder, collective efficacy. Other studies, however, find reciprocal effects between informal control and crime, and one finds support for the broken windows indirect pathway in which social cohesion undermines disorder, disorder fosters fear of crime, and fear of crime feeds back to reduce cohesion ( Markowitz et al. 2001 ).

In principle, simultaneous equation models allow researchers to estimate feedback effects; however, in practice, such models require researchers to make strong assumptions to identify key parameters. Consequently, recent applications of simultaneous equations have searched for naturally occurring strong instruments, such as lotteries that randomize treatments to subjects, as in the military draft (e.g., Angrist & Krueger 2001 ), or naturally occurring exogenous shocks that create strong instruments. For example, Kirk (2015) used Hurricane Katrina as an exogenous intervention that dispersed parolees geographically to estimate the effects of returning parolees to their local neighborhoods on recidivism rates. Unfortunately, such strong instruments have not been available to identify nonrecursive relationships among disorder, control, and crime, leaving results open to question.

Panel Models: Cross-Lagged Effects and Fixed Effects

Panel designs collect data on samples of observations repeatedly over short time spans. Researchers examining disorder, informal control, and crime have typically used one of two panel models. First are cross-lagged panel models, in which the interrelationships among time-varying endogenous variables are modeled as first-order lagged variables. Thus, the models estimate (residualized) change in endogenous variables. In Figure 4 , these lagged effects include stability effects (e.g., social control on itself) and cross-lagged effects (e.g., social control on crime and vice versa). Cross-lagged panel models address causality in several ways. The cross-lagged effects specify a causal order among variables consistent with their temporal order. To obtain ATEs in dynamic panel models, one assumes sequential ignorability (conditional ignorability at each time point) ( Rosenbaum & Rubin 1983 ). Sequential ignorability is addressed by including potential time-invariant confounders (exogenous controls in Figure 4 ) as well as stability effects, which help absorb unobserved heterogeneity. Thus, selection into endogenous variables is assumed to be captured by a combination of observed confounders, stabilities, and cross-lags. To obtain stable estimates, cross-lagged panel models make substantial demands of the data: Endogenous variables must have changed sufficiently to model change, and contemporaneous correlations among endogenous predictor variables must be low enough to provide sufficient statistical power, given modest samples of neighborhoods.

An external file that holds a picture, illustration, etc.
Object name is nihms-1686796-f0004.jpg

Cross-lagged panel model of social control and crime. T1, T2, and T3 represent time periods. Social control (crime) impacts both social control and crime in the next period. Double-headed arrows between ε indicate correlated errors.

Second are fixed-effects models. These models pool the time-series cross-sectional data, yielding NT observations, where N is the number of neighborhoods and T is the number of time periods (waves). Fixed-effects models control for unobserved heterogeneity (time-stable omitted confounders) by estimating within-neighborhood variation by, for example, including N – 1 dummy variables. Thus, fixed-effects models capitalize on panel data to attain conditional ignorability by controlling for all unobserved stable confounders (see Sobel 2012 for assumptions needed to obtain ATEs in fixed-effects models). All relevant time-varying covariates are assumed to be included in the model. Fixed-effects models can include lagged variables—including crosslags—which make greater demands of the data ( Allison et al. 2017 , Wooldridge 2010 ).

Taylor (2001) used two waves of data spaced 12 years apart in 66 tracts in Baltimore to examine the effect of disorder on crime, fear of crime, avoiding dangerous locations, and intentions to move. Disorder was measured by raters observing neighborhoods and separate respondent-perceived physical and social disorder. Taylor found that each indicator of disorder was linked to a different form of crime. This implies different types and measures of disorder may operate as different treatments and combining them into a single measure may mask their separate effects. Using multilevel models, he found both assessed and perceived disorder were associated with fear of crime at the individual level, although only perceived social disorder was associated with avoiding dangerous places and intentions to move. Taylor argued that rater assessments of disorder failed to capture social disorder most troubling to residents—because it is transient and relatively rare—which draws into question objective measures like SSO in capturing a key form of disorder. Taylor does not examine spatial or reciprocal effects.

Using administrative data and ecometric measurement models of survey responses from the Boston Neighborhood Survey, O’Brien & Sampson (2015) examined disorder and crime with two-period cross-lagged panel models. Applying exploratory factor models to police dispatch data, they obtain four measures of conflict and disorder: public social disorder (e.g., public intoxication), public violence (e.g., assault), private violence (e.g., domestic violence), and gun prevalence (e.g., shootings). Physical disorder is measured with counts of reports for private neglect (e.g., housing issues) or public denigration (e.g., detritus) (see O’Brien et al. 2015 ). All measures were aggregated to census tracts. They also included time-invariant controls: percent Hispanic, percent black, income, and baseline collective efficacy.

O’Brien & Sampson (2015) found that no form of disorder or violence was predicted by (or predictive of) public physical disorder. Public social disorder did, however, moderately predict public violence (standardized β = 0.11), although less strongly than did private conflict (0.17). Public social disorder was strongly predicted by private conflict (0.33) and public violence (0.22). This indicates a reciprocal relationship between public violence and public social disorder (but not physical disorder), which supports broken windows theory. The authors could not examine whether social disorder operated through informal social control—the indirect pathway of broken windows—because collective efficacy was measured only at the baseline. The authors also report a second feedback loop of personal conflict in which private conflicts escalate to gun violence: Private conflict predicts gun violence directly (0.19) and via public violence (0.17), which also predicts gun violence (0.20), and gun violence in turn feeds back on future private conflict (0.33). They interpret this result as supporting a social escalation model, in which private conflicts spill over into public spaces. These analyses are limited to residential neighborhoods only, which may inhibit generalizability, as crime and disorder are often concentrated in nonresidential areas ( Yang 2010 ).

Using quasi-experimental methods and panel models, Wheeler et al. (2018) examined the effect of vacant property demolitions on crime. Vacant properties are a form of disorder that may increase crime by signaling low social control (broken windows) or providing opportunities to commit crimes out of sight (situational opportunity). Increases in crime may, however, be spurious if crime and vacant lots are each the result of social disorganization. To examine this, Wheeler and colleagues capitalize on 2,000 demolitions occurring in Buffalo, New York, between 2010 and 2015 to model changes in crime at neighborhood and microplace levels. They estimate a difference-indifference model that matches demolished properties to nondemolished controls using propensity scores based on pretreatment crime and local demographic composition. Comparing crimes before and after demolition at exact addresses and varying distances, they found reductions in crime averaged 90% at demolished parcels. Significant reductions were seen out to more than 1,000 feet. To minimize interference between treatment and controls, the authors estimated a neighborhood-level spatial panel model relating counts of demolitions in census tracts to changes in crime. At the tract level, they found mixed results, as demolitions exerted no significant effects on violent crimes or total police calls, and only a significant spatial effect on nonviolent crimes. That is, demolitions in adjacent neighborhoods are associated with crime reductions, but demolitions within the neighborhood are not. Together these results suggest that removal of vacant properties may reduce crime at the microlevel but may not suppress overall crime in a neighborhood.

Using a two-period cross-lagged, spatial autoregressive model, Boggess & Maskaly (2014) examined effects of disorder on robbery, assault, and disorder in Reno, Nevada. They measured disorder using police calls about intoxication, unwanted persons, graffiti, abandoned vehicles, litter, and dumping and used police-reported robbery and assault as outcomes. They found that, controlling for neighborhood sociodemographic composition, disorder predicts robbery and assault. They also found weak spatial relationships and a modest feedback effect of crime on disorder. However, with no measures of informal control or fear of crime, they cannot rule out spuriousness, and cannot estimate indirect paths between disorder and crime. Their use of police reports for both disorder and crime may introduce a response set, as invoking police is a form of social control. Although Boggess & Maskaly did not provide interwave correlations for disorder, the magnitude of the lagged disorder coefficients (greater than 0.96), large standard errors of other predictors, and a sample of only N = 117 suggest low statistical power.

Wheeler (2017) examined the effects of 311 calls (complaints to the city for physical disorder) on crimes at microplaces (street segments and intersections) in Washington, DC. He divided 311 calls into two categories—detritus (e.g., garbage, abandoned vehicles, illegal dumping) and infrastructure (e.g., potholes, damaged sidewalks, graffiti)—and created an index of serious police-reported crime. The models were carefully done, addressing a number of threats to internal validity. To address ignorability, Wheeler used fixed-effects models to eliminate stable unobserved neighborhood effects, with neighborhoods defined as 500-m squares. To address reciprocal causation, he controlled for prior crime in models of future crime. To model cascading effects of broken windows, he used a first-order, spatial autoregressive lagged crime variable. Wheeler (2017) found that both forms of physical disorder were modestly associated with future crime: An increase of 50 disorder calls for service was significantly associated with one fewer crime. This study is limited exclusively to physical disorder, which may be less consequential for crime than social disorder ( St. Jean 2007 , Yang 2010 ). Furthermore, Wheeler acknowledges that 311 calls are likely to be correlated with informal control against crime, and therefore, disorder could be capturing effects of unobserved informal control.

Although most studies of disorder and informal control use data from the United States, a few apply panel models to other nations. Steenbeek & Hipp (2011) used ten-year panel data from 74 neighborhoods in Utrecht, Netherlands, and distinguished potential informal control (social cohesion and shared expectations for control) from behavioral informal control in cross-lagged models of disorder, social cohesion, and informal control. They did not model crime. Their cross-sectional models replicated previous findings in which social control reduces disorder. Their panel models, however, showed no effect of social control or cohesion on future disorder. By contrast, they found disorder to be negatively associated with future control potential (consistent with broken windows) and residential stability; stability, in turn, is positively associated with future disorder. Disorder is positively associated with future control behavior, which has no significant effect on later disorder. Using first-order spatial- (and temporal-) lagged dependent variables, they find substantial spillover effects between neighborhoods. The authors’ intertemporal correlation tables suggest very high correlations for demographic variables, disorder (0.92–0.96), and cohesion (0.95–0.96), suggesting little change to be explained and potentially weak power of the tests ( Steenbeek & Hipp 2011 ).

Several related papers have examined collective efficacy and crime using panel data from the Australian Community Capacity Study (ACCS), which collected survey data on 4,334 residents across 148 neighborhoods in Brisbane. Hipp & Wickes (2017) followed Sampson et al. (1997) in measuring collective efficacy as a composite of willingness to intervene, cohesion, and trust and controlling for differential distributions of informant characteristics across neighborhoods that may bias reports of collective efficacy. They estimated cross-lagged models for collective efficacy, violence, and neighborhood characteristics (e.g., disadvantage, residential stability, age, population density) and included spatially lagged neighborhood characteristics to control for spillover. They found that, contrary to collective efficacy theory, controlling for prior violence, collective efficacy is significantly associated with future violence—but in the wrong direction. This result held in models using five-year lags and two-year lags and in simultaneous equations using lagged dependent variables as instruments. They also found a small negative indirect effect of collective efficacy on violence through concentrated disadvantage. Although the authors did not present intertemporal correlations among variables, they reported stability coefficients of 0.82 for violence, suggesting modest change, which, combined with the sign flip for collective efficacy, may suggest weak power of statistical tests. It is noteworthy that Sampson (2012) reported evidence for a reciprocal relationship between violent crime and collective efficacy in Chicago using hierarchical cross-lagged panel models. This suggests that the divergent findings may be the result of contextual, rather than methodological, differences between the Brisbane and Chicago studies: Chicago is a larger city with more variation in violent crime.

Wickes & Hipp (2018) used similar models on the same ACCS data set but included three measures of collective efficacy—child-centered social control, reciprocated exchange, and exercise of informal control (attend a meeting, sign a petition, solve a problem with neighbors)—which they hypothesize should have independent effects on crime. They found reciprocal relationships between disadvantage, nearby disadvantage, and all three measures of informal control. Moreover, Wickes & Hipp (2018) found that, contrary to collective efficacy theory, no measure of collective efficacy consistently predicted future crime in the expected direction. Social ties were significantly associated with property crime and drug crime in the wrong direction, control expectations were negatively associated with drug crime only, and exercise of social control was negatively associated with violence only. Interestingly, at the bivariate level, child-centered control is significantly correlated (0.3–0.4) with all crime measures in the direction hypothesized by collective efficacy theory, while the other measures of social control are uncorrelated with all crimes (see appendix 2 in Wickes & Hipp 2018 ). This, with fairly high stabilities for violent crime and property crime, raises the issue of the power of tests of informal control as well as whether the models are controlling for different aspects of the same concept (collective efficacy).

Most studies of disorder, social control, and crime use data from large urban areas, which is consistent with the model of urban growth underlying social disorganization theory. Do results generalize to less-urban settings, where the dynamics of neighborhood residential patterns may be different? Hipp (2016) used block-group-level data from rural North Carolina to examine disorder, informal social control, and perceptions of neighborhood crime in three-wave cross-lagged panel models. Using conventional measures of social cohesion and collective efficacy, Hipp controlled for potential bias in neighborhood reports due to compositional differences in residents across neighborhoods. To measure crime, he asked respondents whether they may have seen or heard acts of violence, arrests, and drug dealing around their neighborhoods and then aggregated responses to the block group. The measures of disorder asked respondents their general impressions of the neighborhood (Do respondents believe neighbors take care of homes and respect property? Is there too much drug use in the neighborhood?).

Using a cross-sectional model, Hipp (2016) replicates Sampson & Raudenbush’s (1999) finding of collective efficacy negatively associated with crime and of both collective efficacy and cohesion negatively associated with disorder. In cross-lagged panel models, he found perceived disorder and crime negatively associated with future collective efficacy, which he interprets as evidence of updating: Respondents’ perceptions of crime and disorder signal weak social control, causing them to update their perceptions of collective efficacy downward (see also Matsueda & Drakulich 2016 ). Furthermore, in a main-effects model, Hipp found that, contrary to collective efficacy theory, neither collective efficacy, social cohesion, nor a composite of the two significantly predicted future perceived crime or disorder. He does, however, find evidence of an interaction effect between social cohesion and collective efficacy. By contrast, consistent with broken windows, disorder predicts future crime. This study provides perhaps the most direct support for broken windows over collective efficacy and contrasts with Sampson & Raudenbush’s (1999) findings. This divergence of findings may be due to differences in measures of crime and disorder (Hipp’s are notably weaker), in simultaneous equation models versus panel models, and in urban versus rural settings.

In sum, panel models find mixed results. In models of disorder and crime, research finds a modest effect of disorder on violence and robbery, even controlling for collective efficacy. Demolitions were modestly associated with future crime at addresses and microplaces but not at the tract level. Cross-lagged panel models find collective efficacy either unrelated to crime, positively related to crime in Brisbane, or moderated by cohesion in rural North Carolina.


Our review of recent causal claims about disorder, informal control, and crime finds a lack of consensus across studies. Turning first to causal links in models of informal social control ( Figure 2 ), some evidence suggests that crime undermines informal control, but this may be limited to robbery. With the exception of one panel study, most research using different designs finds that informal social control is negatively associated with future disorder. Research on the key proposition of collective efficacy and informal social control is mixed. Cross-sectional studies find strong inverse effects of informal control on criminal behavior in different cities in the United States and several other countries. Such studies are unable to address potential reciprocal relationships between informal control and crime, which could result in upward bias. Nonrecursive models address this issue, but those that identify collective efficacy with reciprocated exchange as an instrument find informal control generally affects crime, whereas those that use lagged informal control do not. Cross-lagged panel models, however, show little effect of collective efficacy on future crime in Brisbane or on disorder in Utrecht, and only an interaction effect with cohesion in rural North Carolina, drawing into question theories of informal control. Given that panel models explain change in crime, it could be that collective efficacy can explain variation in crime across neighborhoods but not over time. Alternatively, the panel data sets used may lack sufficient statistical power to detect effects of informal control on temporal variation in crime. Thus, to date, an important counterfactual remains unanswered: In a data set with sufficient change in key endogenous variables and sufficient statistical power of tests, would we find effects of informal social control on changes in crime and disorder?

Evidence on the causal links implied by broken windows theory ( Figure 1 ) is also mixed. The experimental studies of Keizer et al. (2008 , 2011 ) find disorder associated with minor norm violations in Groningen, but another study failed to replicate this result in Groningen while another found treatment heterogeneity by social capital in Germany. With one exception, cross-sectional studies find modest effects of combined social and physical disorder on neighborhood crime. Of the four studies testing the broken windows hypothesis that disorder fosters crime when controlling for informal control, one nonrecursive cross-sectional model found little effect in Chicago, whereas three cross-lagged panel models found modest but significant effects in Boston, rural North Carolina, and Brisbane. Experiments in Bern, Zurich, and New York, the panel study in North Carolina, and one nonrecursive model in Britain found disorder undermines future social control, whereas a second nonrecursive model failed to find a significant effect in Chicago. The positive results would suggest support for disorder affecting crime indirectly through informal control, except that, unlike cross-sectional studies, cross-lagged panel models find little effect of collective efficacy on crime.

In evaluating this research literature, we have come to five tentative conclusions about the relative merits of different research designs as implemented to date. First, individual-level field experiments of disorder on norm violations are promising for testing the specific behavioral principles underlying broken windows. Such experiments can approximate ignorability when conducted with care. The results of recent experiments, however, are questionable because of methodological weaknesses ( Wicherts & Bakker 2014 ). Furthermore, when applied to broken windows and informal control theories, these experiments lack consistency and external validity, and, therefore, to be relevant to criminological debates they must be augmented with studies of naturally occurring crime. Second, we have greater enthusiasm for field experiments that intervene in neighborhoods by manipulating urban blight ( Kondo et al. 2018 ). These experiments manipulate, in a policy-relevant way, the key concept of disorder and examine serious crime. Unfortunately, we could not find parallel interventions seeking to manipulate informal social control. Third, MTO studies, which found few effects of individual moves on crime while eliminating selectivity, are less useful for our task because they cannot disentangle disorder, informal control, and other neighborhood characteristics.

Fourth, we began assuming that well-specified nonrecursive models are stronger than cross-sectional recursive models but weaker than panel models. In this applied literature, however, nonrecursive models are only as valid as the identifying restrictions on IVs. Panel models, while superior in principle, require sufficient change in dependent variables and sufficient statistical power of tests, which may be lacking in applications. Statistical power may also be an issue in simultaneous equation models, given that the power of simultaneous parameters is dependent on, among other things, the strength of IVs ( Bielby & Matsueda 1991 ). Rather than treating such models as panaceas for the possibility of feedback effects, each study needs to be carefully examined for whether the data are up to the assumptions of the models. The model could be correct, but the data are insufficient to estimate the model’s parameters. Fifth, interference is often present in neighborhood models as revealed by spatial analyses; such studies, however, typically do not discuss the degree of bias resulting when interference is ignored.

Although we find mixed results on most key hypotheses, we can make a tentative assessment of where the preponderance of the evidence currently lies. First, informal social control appears to be negatively associated with crime and disorder in urban areas. The negative evidence from panel studies may be the result of inadequate power of tests. Second, disorder—particularly social disorder—appears to be positively associated with future crime and disorder. The causal mechanism could be the broken windows hypothesis that disorder signals weak neighborhood control, or that such disorder generates crime opportunities or conflict, as suggested by Branas et al. (2018) , O’Brien & Sampson (2015) , St. Jean (2007) , and Wheeler et al. (2018) . This association drops substantially when holding constant informal social control but appears not to drop to zero. Third, disorder appears to be negatively associated with future informal social control, implying the possibility of a small indirect effect of disorder on crime operating through informal social control, as suggested by broken windows.

These conclusions are tentative because they will likely change as more research is accumulated. We have deliberately stated these conclusions as empirical associations rather than causal effects because, taken as a whole, these studies of disorder, informal control, and crime remain far from approximating causality as defined by a potential outcomes approach. Consistency is a problem in most experiments, interference is an issue in MTO studies, and ignorability is questionable in most observational studies.

Because these research designs have distinct strengths and weaknesses, more studies within each design are called for, with the hope that consistent results emerge across disparate designs. Future research is needed to address shortcomings in the literature. Individual-level field experiments need to conduct power analyses to ensure sufficient power of tests and conduct experiments in multiple neighborhoods to increase external validity and examine treatment heterogeneity by neighborhood. Given the issues of insufficient change and weak statistical power of tests in neighborhood panel studies, researchers may want to consider modifying research designs. Larger samples, perhaps on smaller neighborhood units, are needed. If the focus is on relatively short-term change, as in most panel studies, studies might examine multiple cities undergoing dynamic change rather than studying older stable metropolitan areas. Within cities, neighborhoods exhibiting change in local organization, disorder, and crime might be oversampled and followed for longer periods, maximizing the likelihood of change. Incorporating neighborhood interventions into panel studies would further leverage change.

Innovative interventions at the neighborhood level, such as randomly assigned demolitions, cleanup campaigns, and greening programs, are needed to examine whether exogenous changes in neighborhood disorder affect crime and informal control. More pressing is the need for studies of informal social control that use experimental interventions to manipulate social capital and collective efficacy across neighborhoods. Finally, while we have focused on causality as the key issue in evaluating the empirical literature on disorder, informal control, and crime, we hope our evaluation will stimulate not only future empirical research but also the further development of theories of disorder, crime, and informal control.


The research underlying this article was supported by grants from the National Science Foundation (SES-1625273) and the Royalty Research Foundation, University of Washington. Additional support came from a Eunice Kennedy Shriver National Institute of Child Health and Human Development research infrastructure grant (P2C HD042828) to the Center for Studies in Demography and Ecology at the University of Washington. The opinions expressed in this article are those of the authors and do not necessarily reflect the positions of the funding agencies. We thank Jerald R. Herting and Robert J. Sampson for helpful comments on an earlier draft of this article.


The authors are not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.


  • Allison PD, Williams R, Moral-Benito E. 2017. Maximum likelihood for cross-lagged panel models with fixed effects . Socius 3 :1–17 [ Google Scholar ]
  • Angrist JD, Krueger AB. 2001. Instrumental variables and the search for identification: from supply and demand to natural experiments . J. Econ. Perspect 15 ( 4 ):69–85 [ Google Scholar ]
  • Bandura A 1986. Social Foundations of Thought and Action: A Social Cognitive Theory . Englewood Cliffs, NJ: Prentice-Hall [ Google Scholar ]
  • Bellair PE. 2000. Informal surveillance and street crime: a complex relationship . Criminology 38 ( 1 ):137–70 [ Google Scholar ]
  • Bellair PE, Browning CR. 2010. Contemporary disorganization research: an assessment and further test of the systemic model of neighborhood crime . J. Res. Crime Delinquency 47 ( 4 ):496–521 [ Google Scholar ]
  • Berger J, Hevenstone D. 2016. Norm enforcement in the city revisited: an international field experiment of altruistic punishment, norm maintenance, and broken windows . Ration. Soc 28 ( 3 ):299–319 [ Google Scholar ]
  • Bielby WT, Matsueda RL. 1991. Statistical power in nonrecursive linear models . Sociol. Methodol 21 :167–97 [ Google Scholar ]
  • Boggess LN, Maskaly J. 2014. The spatial context of the disorder-crime relationship in a study of Reno neighborhoods . Soc. Sci. Res 43 :168–83 [ PubMed ] [ Google Scholar ]
  • Braga AA, Welsh BC, Schnell C. 2015. Can policing disorder reduce crime? A systematic review and meta-analysis . J. Res. Crime Delinquency 52 ( 4 ):567–88 [ Google Scholar ]
  • Branas CC, Kondo MC, Murphy SM, South EC, Polsky D, MacDonald JM. 2016. Urban blight remediation as a cost-beneficial solution to firearm violence . Am. J. Public Health 106 ( 12 ):2158–64 [ PMC free article ] [ PubMed ] [ Google Scholar ]
  • Branas CC, South E, Kondo MC, Hohl BC, Bourgois P, et al. 2018. Citywide cluster randomized trial to restore blighted vacant land and its effects on violence, crime, and fear . PNAS 115 ( 12 ):2946–51 [ PMC free article ] [ PubMed ] [ Google Scholar ]
  • Browning CR, Dietz RD, Feinberg SL. 2004. The paradox of social organization: networks, collective efficacy, and violent crime in urban neighborhoods . Soc. Forces 83 ( 2 ):503–34 [ Google Scholar ]
  • Bruinsma GJN, Pauwels LJR, Weerman FM, Bernasco W. 2013. Social disorganization, social capital, collective efficacy and the spatial distribution of crime and offenders: an empirical test of six neighbourhood models for a Dutch city . Br. J. Criminol 53 ( 5 ):942–63 [ Google Scholar ]
  • Cialdini RB. 2003. Crafting normative messages to protect the environment . Curr. Dir. Psychol. Sci 12 ( 4 ):105–9 [ Google Scholar ]
  • Cohen LE, Felson M. 1979. Social change and crime rate trends: a routine activity approach . Am. Sociol. Rev 44 ( 4 ):588–608 [ Google Scholar ]
  • Coleman JS. 1990. Foundations of Social Theory . Cambridge, MA: Belknap Press [ Google Scholar ]
  • Diekmann A, Przepiorka W, Rauhut H. 2015. Lifting the veil of ignorance: an experiment on the contagiousness of norm violations . Ration. Soc 27 ( 3 ):309–33 [ Google Scholar ]
  • Engel C, Beckenkamp M, Glöckner A, Irlenbusch B, Hennig-Schmidt H, et al. 2014. First impressions are more important than early intervention: qualifying broken windows theory in the lab . Int. Rev. Law Econ 37 :126–36 [ Google Scholar ]
  • Fagan J, Davies G. 2000. Street stops and broken windows: terry, race, and disorder in New York City . Fordham Urban Law J . 28 :457–504 [ Google Scholar ]
  • Gault M, Silver E. 2008. Spuriousness or mediation? Broken windows according to Sampson and Raudenbush (1999) . J. Crim. Justice 36 ( 3 ):240–43 [ Google Scholar ]
  • Greene WH. 2003. Econometric Analysis . Upper Saddle River, NJ: Prentice Hall. 5th ed. [ Google Scholar ]
  • Harcourt BE. 1998. Reflecting on the subject: a critique of the social influence conception of deterrence, the broken windows theory, and order-maintenance policing New York style . Mich. Law Rev 97 ( 2 ):291–389 [ Google Scholar ]
  • Harcourt BE, Ludwig J. 2006. Broken windows: new evidence from New York City and a five-city social experiment . Univ. Chic. Law Rev 73 :271–320 [ Google Scholar ]
  • Hinkle JC. 2015. Emotional fear of crime versus perceived safety and risk: implications for measuring “fear” and testing the broken windows thesis . Am. J. Crim. Justice 40 ( 1 ):147–68 [ Google Scholar ]
  • Hipp JR. 2007. Block, tract, and levels of aggregation: neighborhood structure and crime and disorder as a case in point . Am. Sociol. Rev 72 ( 5 ):659–80 [ Google Scholar ]
  • Hipp JR. 2010. Resident perceptions of crime and disorder: How much is “bias”, and how much is social environment differences? Criminology 48 ( 2 ):475–508 [ Google Scholar ]
  • Hipp JR. 2016. Collective efficacy: How is it conceptualized, how is it measured, and does it really matter for understanding perceived neighborhood crime and disorder? J. Crim. Justice 46 :32–44 [ PMC free article ] [ PubMed ] [ Google Scholar ]
  • Hipp JR, Wickes R. 2017. Violence in urban neighborhoods: a longitudinal study of collective efficacy and violent crime . J. Quant. Criminol 33 ( 4 ):783–808 [ Google Scholar ]
  • Holland PW. 1986. Statistics and causal inference . J. Am. Stat. Assoc 81 ( 396 ):945–60 [ Google Scholar ]
  • Imbens GW, Rubin DB. 2015. Causal Inference for Statistics, Social, and Behavioral Sciences: An Introduction . New York: Cambridge Univ. Press [ Google Scholar ]
  • Keizer K, Lindenberg S, Steg L. 2008. The spreading of disorder . Science 322 ( 5908 ):1681–85 [ PubMed ] [ Google Scholar ]
  • Keizer K, Lindenberg S, Steg L. 2011. The reversal effect of prohibition signs . Group Process. Intergroup Relat 14 ( 5 ):681–88 [ Google Scholar ]
  • Keizer K, Lindenberg S, Steg L. 2013. The importance of demonstratively restoring order . PLOS ONE 8 ( 6 ):e65137. [ PMC free article ] [ PubMed ] [ Google Scholar ]
  • Kelling GL, Coles CM. 1997. Fixing Broken Windows: Restoring Order and Reducing Crime in Our Communities . New York: Touchstone [ Google Scholar ]
  • Keuschnigg M, Wolbring T. 2015. Disorder, social capital, and norm violation: three field experiments on the broken windows thesis . Ration. Soc 27 ( 1 ):96–126 [ Google Scholar ]
  • Kirk DS. 2015. A natural experiment of the consequences of concentrating former prisoners in the same neighborhoods . PNAS 112 ( 22 ):6943–48 [ PMC free article ] [ PubMed ] [ Google Scholar ]
  • Kling JR, Ludwig J, Katz LF. 2005. Neighborhood effects on crime for female and male youth: evidence from a randomized housing voucher experiment . Q. J. Econ 120 ( 1 ):87–130 [ Google Scholar ]
  • Kondo MC, Andreyeva E, South EC, MacDonald JM, Branas CC. 2018. Neighborhood interventions to reduce violence . Annu. Rev. Public Health 39 :253–71 [ PubMed ] [ Google Scholar ]
  • Kubrin CE. 2008. Making order of disorder: a call for conceptual clarity . Criminol. Public Policy 7 ( 2 ):203–13 [ Google Scholar ]
  • Lieberson S 1985. Making It Count: The Improvement of Social Research and Theory . Berkeley: Univ. Calif. Press [ Google Scholar ]
  • Liska AE, Warner BD. 1991. Functions of crime: a paradoxical process . Am. J. Sociol 96 ( 6 ):1441–63 [ Google Scholar ]
  • Markowitz FE, Bellair PE, Liska AE, Liu J. 2001. Extending social disorganization theory: modeling the relationships between cohesion, disorder, and fear . Criminology 39 ( 2 ):293–319 [ Google Scholar ]
  • Matsueda RL. 2013. Rational choice research in criminology: a multi-level framework. In Handbook of Rational Choice Social Research , ed. Wittek R, Snijders TAB, Nee V, pp. 283–321. Stanford, CA: Stanf. Univ. Press [ Google Scholar ]
  • Matsueda RL. 2017. Toward an analytical criminology: the micro-macro problem, causal mechanisms, and public policy . Criminology 55 ( 3 ):493–519 [ Google Scholar ]
  • Matsueda RL, Drakulich KM. 2016. Measuring collective efficacy: a multilevel measurement model for nested data . Sociol. Methods Res 45 ( 2 ):191–230 [ Google Scholar ]
  • Mazerolle L, Wickes R, McBroom J. 2010. Community variations in violence: the role of social ties and collective efficacy in comparative context . J. Res. Crime Delinquency 47 ( 1 ):3–30 [ Google Scholar ]
  • Morenoff JD, Sampson RJ, Raudenbush SW. 2001. Neighborhood inequality, collective efficacy, and the spatial dynamics of urban violence . Criminology 39 ( 3 ):517–58 [ Google Scholar ]
  • Morgan SL, Winship C. 2015. Counterfactuals and Causal Inference: Methods and Principles for Social Research . New York: Cambridge Univ. Press. 2nd ed. [ Google Scholar ]
  • Nagin DS, Sampson RJ. 2019. The real gold standard: measuring counterfactual worlds that matter most to social science and policy . Annu. Rev. Criminol 2 :123–45 [ Google Scholar ]
  • O’Brien DT, Farrell C, Welsh BC. 2019. Looking through broken windows: The impact of neighborhood disorder on aggression and fear of crime is an artifact of research design . Annu. Rev. Criminol 2 :53–71 [ Google Scholar ]
  • O’Brien DT, Kauffman RA. 2013. Broken windows and low adolescent prosociality: not cause and consequence, but co-symptoms of low collective efficacy . Am. J. Community Psychol 51 ( 3–4 ):359–69 [ PubMed ] [ Google Scholar ]
  • O’Brien DT, Sampson RJ. 2015. Public and private spheres of neighborhood disorder: assessing pathways to violence using large-scale digital records . J. Res. Crime Delinquency 52 ( 4 ):486–510 [ Google Scholar ]
  • O’Brien DT, Sampson RJ, Winship C. 2015. Ecometrics in the age of big data: measuring and assessing “broken windows” using large-scale administrative records . Sociol. Methodol 45 ( 1 ):101–47 [ Google Scholar ]
  • Rosenbaum PR, Rubin DB. 1983. Assessing sensitivity to an unobserved binary covariate in an observational study with binary outcome . J. R. Stat. Soc. B 45 ( 2 ):212–18 [ Google Scholar ]
  • Rubin DB. 1986. Comment . J. Am. Stat. Assoc 81 ( 396 ):961–62 [ Google Scholar ]
  • Rubin DB. 2006. Causal inference through potential outcomes and principal stratification: application to studies with “censoring” due to death . Stat. Sci 21 ( 3 ):299–309 [ Google Scholar ]
  • Sampson RJ. 2008. Moving to inequality: Neighborhood effects and experiments meet social structure . Am. J. Sociol 114 ( 1 ):189–231 [ PMC free article ] [ PubMed ] [ Google Scholar ]
  • Sampson RJ. 2012. Great American City: Chicago and the Enduring Neighborhood Effect . Chicago: Univ. Chicago Press [ Google Scholar ]
  • Sampson RJ, Groves WB. 1989. Community structure and crime: testing social-disorganization theory . Am. J. Sociol 94 ( 4 ):774–802 [ Google Scholar ]
  • Sampson RJ, Morenoff JD, Earls F. 1999. Beyond social capital: spatial dynamics of collective efficacy for children . Am. Sociol. Rev 64 ( 5 ):633–60 [ Google Scholar ]
  • Sampson RJ, Raudenbush SW. 1999. Systematic social observation of public spaces: a new look at disorder in urban neighborhoods . Am. J. Sociol 105 ( 3 ):603–51 [ Google Scholar ]
  • Sampson RJ, Raudenbush SW. 2004. Seeing disorder: neighborhood stigma and the social construction of “broken windows .” Soc. Psychol. Q 67 ( 4 ):319–42 [ Google Scholar ]
  • Sampson RJ, Raudenbush SW, Earls F. 1997. Neighborhoods and violent crime: a multilevel study of collective efficacy . Science 277 ( 5328 ):918–24 [ PubMed ] [ Google Scholar ]
  • Sampson RJ, Wikström P-OH. 2008. The social order of violence in Chicago and Stockholm neighborhoods: a comparative inquiry. In Order, Conflict, and Violence , ed. Kalyvas SN, Shapiro I, Masoud T, pp. 97–119. Cambridge, UK: Cambridge Univ. Press [ Google Scholar ]
  • Shaw CR, McKay HD. 1931. Report on the Causes of Crime , Vol. II . Washington, DC: Gov. Print. Off. [ Google Scholar ]
  • Shaw CR, McKay HD. 1969. Juvenile Delinquency in Urban Areas . Chicago: Univ. Chicago Press. Revis. Ed. [ Google Scholar ]
  • Skogan WG. 1990. Disorder and Decline: Crime and the Spiral of Decay in American Neighborhoods . New York: Free Press [ Google Scholar ]
  • Skogan WG. 2015. Disorder and decline: the state of research . J. Res. Crime Delinquency 52 ( 4 ):464–85 [ Google Scholar ]
  • Sobel ME. 2006. What do randomized studies of housing mobility demonstrate? Causal inference in the face of interference . J. Am. Stat. Assoc 101 ( 476 ):1398–407 [ Google Scholar ]
  • Sobel ME. 2012. Does marriage boost men’s wages?: Identification of treatment effects in fixed effects regression models for panel data . J. Am. Stat. Assoc 107 ( 498 ):521–29 [ Google Scholar ]
  • St. Jean PKB. 2007. Pockets of Crime: Broken Windows, Collective Efficacy, and the Criminal Point of View . Chicago: Univ. Chicago Press [ Google Scholar ]
  • Steenbeek W, Hipp JR. 2011. A longitudinal test of social disorganization theory: feedback effects among cohesion, social control, and disorder . Criminology 49 ( 3 ):833–71 [ Google Scholar ]
  • Taylor RB. 1996. Neighborhood responses to disorder and local attachments: the systemic model of attachment, social disorganization, and neighborhood use value . Sociol. Forum 11 ( 1 ):41–74 [ Google Scholar ]
  • Taylor RB. 2001. Breaking Away from Broken Windows: Baltimore Neighborhoods and the Nationwide Fight against Crime, Grime, Fear, and Decline . Boulder, CO: Westview Press [ Google Scholar ]
  • Taylor RB. 2015. Community Criminology: Fundamentals of Spatial and Temporal Scaling, Ecological Indicators, and Selectivity Bias . New York: NYU Press [ Google Scholar ]
  • Volker B 2017. Revisiting broken windows: the role of neighborhood and individual characteristics in reaction to disorder cues . Sociol. Sci 4 :528–51 [ Google Scholar ]
  • Weisburd D, Hinkle JC, Braga AA, Wooditch A. 2015. Understanding the mechanisms underlying broken windows policing: the need for evaluation evidence . J. Res. Crime Delinquency 52 ( 4 ):589–608 [ Google Scholar ]
  • Wheeler AP. 2017. The effect of 311 calls for service on crime in D.C. at microplaces . Crime Delinquency 64 ( 14 ):1882–903 [ Google Scholar ]
  • Wheeler AP, Kim D-Y, Phillips SW. 2018. The effect of housing demolitions on crime in Buffalo, New York . J. Res. Crime Delinquency 55 ( 3 ):390–424 [ Google Scholar ]
  • Wicherts JM, Bakker M. 2014. Broken windows, mediocre methods, and substandard statistics . Group Process. Intergroup Relat 17 ( 3 ):388–403 [ Google Scholar ]
  • Wickes R, Hipp JR. 2018. The spatial and temporal dynamics of neighborhood informal social control and crime . Soc. Forces 97 ( 1 ):277–308 [ Google Scholar ]
  • Wilson JQ, Kelling GL. 1982. Broken windows . The Atlantic . March. [ Google Scholar ]
  • Woodward J 2003. Making Things Happen: A Theory of Causal Explanation . New York: Oxford Univ. Press [ Google Scholar ]
  • Wooldridge JM. 2010. Econometric Analysis of Cross Section and Panel Data . Cambridge, MA: MIT Press. 2nd ed. [ Google Scholar ]
  • Xu Y, Fiedler ML, Flaming KH. 2005. Discovering the impact of community policing: the broken windows thesis, collective efficacy, and citizens’ judgment . J. Res. Crime Delinquency 42 ( 2 ):147–86 [ Google Scholar ]
  • Yang S-M. 2010. Assessing the spatial-temporal relationship between disorder and violence . J. Quant. Criminol 26 ( 1 ):139–63 [ Google Scholar ]

U.S. flag

An official website of the Department of Health & Human Services

  • Search All AHRQ Sites
  • Email Updates

Patient Safety Network

1. Use quotes to search for an exact match of a phrase.

2. Put a minus sign just before words you don't want.

3. Enter any important keywords in any order to find entries where all these terms appear.

  • The PSNet Collection
  • All Content
  • Perspectives
  • Current Weekly Issue
  • Past Weekly Issues
  • Curated Libraries
  • Patient Safety 101
  • The Fundamentals
  • Training and Education
  • Continuing Education
  • WebM&M: Case Studies
  • Training Catalog
  • Improvement Resources
  • Innovations
  • Submit an Innovation
  • About PSNet
  • Editorial Team
  • Technical Expert Panel

'Broken hospital windows': debating the theory of spreading disorder and its application to healthcare organizations.

Churruca K, Ellis LA, Braithwaite J. 'Broken hospital windows': debating the theory of spreading disorder and its application to healthcare organizations. BMC Health Serv Res. 2018;18(1):201. doi:10.1186/s12913-018-3012-2.

Unit-level dysfunction creates observable violations that, once normalized , can result in organizational failure . This article explains how applying the broken windows theory in health care can enable understanding of workarounds that may escalate from helpful adjustments in practice to a manifestation of conditions that contribute to patient harm.

Predictors of response rates of safety culture questionnaires in healthcare: a systematic review and analysis. October 26, 2022

"Time is of the essence": relationship between hospital staff perceptions of time, safety attitudes and staff wellbeing. December 8, 2021

Dimensions of safety culture: a systematic review of quantitative, qualitative and mixed methods for assessing safety culture in hospitals. September 1, 2021

Issues and complexities in safety culture assessment in healthcare. July 19, 2023

Changing how we think about healthcare improvement. June 13, 2018

False dawns and new horizons in patient safety research and practice. December 20, 2017

Beyond patient safety Flatland. June 30, 2010

Determination of health-care teamwork training competencies: a Delphi study. December 2, 2009

Changes in unprofessional behaviour, teamwork, and co-operation among hospital staff during the COVID-19 pandemic. September 28, 2022

Harnessing implementation science to improve care quality and patient safety: a systematic review of targeted literature. May 21, 2014

Learning from disasters to improve patient safety: applying the generic disaster pathway to health system errors. February 9, 2011

The application of the Global Trigger Tool: a systematic review. October 26, 2016

Experiences of health professionals who conducted root cause analyses after undergoing a safety improvement programme. December 20, 2006

The patient died: what about involvement in the investigation process? June 24, 2020

Implementation of a patient safety incident management system as viewed by doctors, nurses and allied health professionals. May 6, 2009

Attitudes toward the large-scale implementation of an incident reporting system. April 9, 2008

The effect of physicians' long-term use of CPOE on their test management work practices. September 27, 2006

Positive deviance: a different approach to achieving patient safety. August 20, 2014

'Between the flags': implementing a rapid response system at scale. May 14, 2014

Cultural and associated enablers of, and barriers to, adverse incident reporting. June 23, 2010

COVID-19: patient safety and quality improvement skills to deploy during the surge. June 24, 2020

Investigating patient safety culture across a health system: multilevel modelling of differences associated with service types and staff demographics. July 25, 2012

Association between organisational and workplace cultures, and patient outcomes: systematic review. January 17, 2018

Comparing rates of adverse events detected in incident reporting and the Global Trigger Tool: a systematic review. August 16, 2023

A root cause analysis of clinical error: confronting the disjunction between formal rules and situated clinical activity. May 31, 2006

Elimination of central-venous-catheter-related bloodstream infections from the intensive care unit. March 2, 2011

Transformational improvement in quality care and health systems: the next decade. November 25, 2020

Effects of an adverse-drug-event alert system on cost and quality outcomes in community hospitals. April 28, 2010

Analysis of Australian newspaper coverage of medication errors. January 25, 2012

Improving patient safety: the comparative views of patient-safety specialists, workforce staff and managers. January 30, 2005

Health professional networks as a vector for improving healthcare quality and safety: a systematic review. January 11, 2012

Putting safety on the curriculum. October 7, 2009

Towards safer, better healthcare: harnessing the natural properties of complex sociotechnical systems. March 4, 2009

Patient safety, quality care, and service utilization with PLATO (Physician Leadership for Accurate and Timely Orders): a pilot study. August 26, 2009

A qualitative content analysis of retained surgical items: learning from root cause analysis investigations. May 27, 2020

Turning the medical gaze in upon itself: root cause analysis and the investigation of clinical error. October 26, 2005

Nurses' workarounds in acute healthcare settings: a scoping review. June 19, 2013

Primary care closed claims experience of Massachusetts malpractice insurers. October 16, 2013

Disentangling quality and safety indicator data: a longitudinal, comparative study of hand hygiene compliance and accreditation outcomes in 96 Australian hospitals. October 8, 2014

Safety culture includes "good catches." September 30, 2015

Market-based control mechanisms for patient safety. April 15, 2009

Improving the accuracy of patient identification in the medication-use process. February 8, 2006

Effect of a pharmacist-led multicomponent intervention focusing on the medication monitoring phase to prevent potential adverse drug events in nursing homes. January 30, 2005

Doing right by our patients when things go wrong in the ambulatory setting. February 12, 2014

Effects of two commercial electronic prescribing systems on prescribing error rates in hospital in-patients: a before and after study. February 15, 2012

Quality improvement to decrease specimen mislabeling in transfusion medicine. September 20, 2006

Frequency and nature of communication and handoff failures in medical malpractice claims. April 6, 2022

Bad stars or guiding lights? Learning from disasters to improve patient safety. May 12, 2010

Nurses' and nursing assistants' perceptions of patient safety culture in nursing homes. August 9, 2006

Iatrogenic events resulting in intensive care admission: frequency, cause, and disclosure to patients and institutions. April 27, 2005

Dual surgeon operating to improve patient safety. September 29, 2021

Performance of a fail-safe system to follow up abnormal mammograms in primary care. September 8, 2010

Nurses' medication work: what do nurses know? November 17, 2010

Electronic prescribing: improving the efficiency and accuracy of prescribing in the ambulatory care setting. June 25, 2014

Thoughtless design of the electronic health record drives overuse, but purposeful design can nudge improved patient care. April 25, 2018

Six year audit of cardiac arrests and medical emergency team calls in an Australian outer metropolitan teaching hospital. December 19, 2007

Automated electronic reminders to prevent miscommunication among primary medical, surgical and anaesthesia providers: a root cause analysis. August 22, 2012

Observational teamwork assessment for surgery: feasibility of clinical and nonclinical assessor calibration with short-term training. May 30, 2012

The impact of nontechnical skills on technical performance in surgery: a systematic review. January 30, 2005

Appropriateness of outpatient antibiotic prescribing among privately insured US patients: ICD-10-CM based cross sectional study. February 6, 2019

Patient safety education: what was, what is, and what will be? December 18, 2013

Teamwork is associated with reduced hospital staff burnout at military treatment facilities: findings from the 2019 Department of Defense Patient Safety Culture Survey. March 22, 2023

Explaining Michigan: developing an ex post theory of a quality improvement program. August 3, 2011

Using a risk assessment approach to determine which factors influence whether partially bilingual physicians rely on their non-English language skills or call an interpreter. August 22, 2012

Accuracy of laboratory data communication on ICU daily rounds using an electronic health record. October 12, 2016

Data omission by physician trainees on ICU rounds. February 6, 2019

Implementation science: a neglected opportunity to accelerate improvements in the safety and quality of surgical care. December 21, 2016

The effect of providing staff training and enhanced support to care homes on care processes, safety climate and avoidable harms: evaluation of a care home quality improvement programme in England. August 18, 2021

Clinical handovers between prehospital and hospital staff: literature review. September 24, 2014

Identification of the barriers and enablers for receiving a speaking up message: a content analysis approach. August 16, 2023

The influence of personality on psychological safety, the presence of stress and chosen professional roles in the healthcare environment. June 21, 2023

Does incorporating medications in the surveyors' interpretive guidelines reduce the use of potentially inappropriate medications in nursing homes? July 4, 2007

A comparison of two distribution methods on response rates to a patient safety questionnaire in nursing homes. October 3, 2007

Medical costs of Alzheimer's disease misdiagnosis among US Medicare beneficiaries. August 26, 2015

Frequency of passive EHR alerts in the ICU: another form of alert fatigue? July 27, 2016

Detection of patient risk by nurses: a theoretical framework. February 17, 2010

The meaning of justice in safety incident reporting. November 7, 2007

Use of medical emergency team responses to reduce hospital cardiopulmonary arrests. March 27, 2005

Use of medical emergency team (MET) responses to detect medical errors. March 6, 2005

Improving patients' intensive care admission through multidisciplinary simulation-based crisis resource management: a qualitative study. September 20, 2023

Exploring the factors that promote or diminish a psychologically safe environment: a qualitative interview study with critical care staff. September 22, 2021

Wristbands as aids to reduce misidentification: an ethnographically guided task analysis. September 28, 2011

Errors, near misses and adverse events in the emergency department: what can patients tell us? November 5, 2008

Patients do not always complain when they are dissatisfied: implications for service quality and patient safety. December 4, 2013

Analysis of surgical errors in closed malpractice claims at 4 liability insurers. August 16, 2006

Application of surgical safety standards to robotic surgery: five principles of ethics for nonmaleficence. April 2, 2014

Case study: getting boards on board at Allen Memorial Hospital, Iowa Health System. April 2, 2008

Fidelity and the impact of patient safety huddles on teamwork and safety culture: an evaluation of the Huddle Up for Safer Healthcare (HUSH) project. October 20, 2021

Resident and RN perceptions of the impact of a medical emergency team on education and patient safety in an academic medical center. December 2, 2009

Critical laboratory value notification: a failure mode effects and criticality analysis. January 31, 2006

Vital sign abnormalities, rapid response, and adverse outcomes in hospitalized patients. January 30, 2013

Factors contributing to Registered Nurse medication administration error: a narrative review. March 18, 2015

What is the role of individual accountability in patient safety? A multi-site ethnographic study. November 25, 2015

Nurses' perceptions of patient safety climate in intensive care units: a cross-sectional study. February 27, 2013

A qualitative study comparing experiences of the surgical safety checklist in hospitals in high-income and low-income countries. September 4, 2013

Transferring responsibility and accountability in maternity care: clinicians defining their boundaries of practice in relation to clinical handover. October 10, 2012

Creating a culture of caregiver support. October 25, 2017

Safe handover. November 22, 2017

The wicked problem of patient misidentification: how could the technological revolution help address patient safety? May 1, 2019

Multiple meanings of resilience: health professionals' experiences of a dual element training intervention designed to help them prepare for coping with error. March 31, 2021

Patient Safety Innovations

Preventing Falls Through Patient and Family Engagement to Create Customized Prevention Plans

Annual Perspective


Understanding complexity in a safety critical setting: a systems approach to medication administration. April 26, 2023

Racism and electronic health records (EHRs): perspectives for research and practice. April 12, 2023

Perspectives on Safety

Managing interruptions to improve diagnostic decision-making: strategies and recommended research agenda. March 29, 2023

Speaking up as an extension of socio-cultural dynamics in hospital settings: a study of staff experiences of speaking up across seven hospitals. February 22, 2023

An evidence and consensus-based definition of second victim: a strategic topic in healthcare quality, patient safety, person-centeredness and human resource management. February 15, 2023

Weight and size descriptors for drug dosing: too many options and too many errors. January 11, 2023

Integrating principles of safety culture and just culture into nursing homes: lessons from the pandemic. January 12, 2022

DEEP SCOPE: a framework for safe healthcare design. October 13, 2021

Leadership: an effective human factor during COVID-19. September 1, 2021

Experiences and perspectives of transgender youths in accessing health care: a systematic review. August 4, 2021

From fable to reality at Parkland Hospital: the impact of evidence-based design strategies on patient safety, healing, and satisfaction in an adult inpatient environment. February 10, 2021

Organisational crisis resource management: leading an academic department of emergency medicine through the COVID-19 pandemic. October 7, 2020

Planning for a pandemic: mitigating risk to radiation therapy service delivery in the COVID-19 era. July 22, 2020

Patient Safety Primers

'Poking the skunk': ethical and medico-legal concerns in research about patients' experiences of medical injury. July 17, 2019

Failure to report poor care as a breach of moral and professional expectation. June 19, 2019

The 2018 Gosport Independent Panel report into deaths at the National Health Service's Gosport War Memorial Hospital. Does the culture of the medical profession influence health outcomes? June 12, 2019

Infection prevention in long-term care: re-evaluating the system using a human factors engineering approach. February 13, 2019

Ten principles for more conservative, care-full diagnosis. October 10, 2018

Active-shooter response at a health care facility. September 19, 2018

The application of system dynamics modelling to system safety improvement: present use and future potential. September 19, 2018

Patient safety climate: a study of Southern California healthcare organizations. September 5, 2018

Reframing and addressing horizontal violence as a workplace quality improvement concern. August 22, 2018

Patient Safety Network

Connect With Us


Sign up for Email Updates

To sign up for updates or to access your subscriber preferences, please enter your email address below.

Agency for Healthcare Research and Quality

5600 Fishers Lane Rockville, MD 20857 Telephone: (301) 427-1364

  • Accessibility
  • Disclaimers
  • Electronic Policies
  • HHS Digital Strategy
  • HHS Nondiscrimination Notice
  • Inspector General
  • Plain Writing Act
  • Privacy Policy
  • Viewers & Players
  • U.S. Department of Health & Human Services
  • The White House

PSNet Log in

  • Don't have account? Create new account

Submit Your Innovations

Please select your preferred way to submit an innovation.

Continue as a Guest

Track and save your innovation

in My Innovations

Edit your innovation as a draft

Continue Logged In

Please select your preferred way to submit an innovation. Note that even if you have an account, you can still choose to submit an innovation as a guest.

Continue logged in

New users to the psnet site.

Access to quizzes and start earning

CME, CEU, or Trainee Certification.

Get email alerts when new content

matching your topics of interest

in My Innovations.

Using Broken Windows Theory as the Backdrop for a Proactive Approach to Threat Identification in Health Care


  • 1 From the Embry-Riddle Aeronautical University.
  • 2 Halifax Health Medical Center, Daytona Beach, Florida.
  • PMID: 27617964
  • DOI: 10.1097/PTS.0000000000000328

Objectives: Historically, health care has relied on error management techniques to measure and reduce the occurrence of adverse events. This study proposes an alternative approach for identifying and analyzing hazardous events. Whereas previous research has concentrated on investigating individual flow disruptions, we maintain the industry should focus on threat windows, or the accumulation of these disruptions. This methodology, driven by the broken windows theory, allows us to identify process inefficiencies before they manifest and open the door for the occurrence of errors and adverse events.

Methods: Medical human factors researchers observed disruptions during 34 trauma cases at a Level II trauma center. Data were collected during resuscitation and imaging and were classified using a human factors taxonomy: Realizing Improved Patient Care Through Human-Centered Operating Room Design for Threat Window Analysis (RIPCHORD-TWA).

Results: Of the 576 total disruptions observed, communication issues were the most prevalent (28%), followed by interruptions and coordination issues (24% each). Issues related to layout (16%), usability (5%), and equipment (2%) comprised the remainder of the observations. Disruptions involving communication issues were more prevalent during resuscitation, whereas coordination problems were observed more frequently during imaging.

Conclusions: Rather than solely investigating errors and adverse events, we propose conceptualizing the accumulation of disruptions in terms of threat windows as a means to analyze potential threats to the integrity of the trauma care system. This approach allows for the improved identification of system weaknesses or threats, affording us the ability to address these inefficiencies and intervene before errors and adverse events may occur.

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

  • Delivery of Health Care
  • Health Personnel
  • Operating Rooms*
  • Trauma Centers*
  • All Publications
  • Priorities Magazine Spring 2018
  • The Next Plague and How Science Will Stop It
  • Priorities Magazine Winter 2018
  • Priorities Magazine Fall 2017
  • Little Black Book of Junk Science
  • Priorities Magazine Winter 2017
  • Should You Worry About Artificial Flavors Or Colors?
  • Should You Worry About Artificial Sweeteners?
  • Summer Health and Safety Tips
  • How Toxic Terrorists Scare You With Science Terms
  • Adult Immunization: The Need for Enhanced Utilization
  • Should You Worry About Salt?
  • Priorities Magazine Spring 2016
  • IARC Diesel Exhaust & Lung Cancer: An Analysis
  • Teflon and Human Health: Do the Charges Stick?
  • Helping Smokers Quit: The Science Behind Tobacco Harm Reduction
  • Irradiated Foods
  • Foods Are Not Cigarettes: Why Tobacco Lawsuits Are Not a Model for Obesity Lawsuits
  • The Prevention and Treatment of Osteoporosis: A Review
  • Are "Low Dose" Health Effects of Chemicals Real?
  • The Effects of Nicotine on Human Health
  • Traditional Holiday Dinner Replete with Natural Carcinogens - Even Organic Thanksgiving Dinners
  • A Primer On Dental Care: Quality and Quackery
  • Nuclear Energy and Health And the Benefits of Low-Dose Radiation Hormesis
  • Priorities in Caring for Your Children: A Primer for Parents
  • Endocrine Disrupters: A Scientific Perspective
  • Good Stories, Bad Science: A Guide for Journalists to the Health Claims of "Consumer Activist" Groups
  • A Comparison of the Health Effects of Alcohol Consumption and Tobacco Use in America
  • Moderate Alcohol Consumption and Health
  • Irradiated Foods Fifth Edition
  • Media/Contact
  • Write For Us

Fixing a 'Broken Window'

Related articles.

broken window theory health

The "Broken Window" theory is among the most widely-known ideas in policing. It’s been very controversial to say the least. But what if we don’t understand how to address the broken window? A new study looks at fixing the window, rather than pursuing the one who threw the rock.

broken window theory health

“ A piece of property abandoned, weeds grow up, a window is smashed.”

That line comes early in “ Broken Windows ,” one of the most widely quoted articles on policing. The theory offered up by those authors has been linked with changes in policing that ranged from zero tolerance of petty crimes, like graffiti and panhandling to significant interventions, such as New York City’s “stop and frisk.”

But as an  article  in the New Yorker asked as recently as two years ago, what if we had gotten the whole idea wrong. “What if vacant property had received the attention that, for thirty years, was instead showered on petty criminals?” Should we have then, and certainly now, focused our interest not on those small acts of disorder but on the environment where they thrive. Every day there are more and more headlines of increasing crime in our major cities. At the same time, we battle over funding or defunding the police - scientific study offers some insight and, I believe, a practical new direction. 

Before jumping into the latest research, let us briefly consider the studies on policing involving the “Broken Windows” philosophy. First, and most important, what constitutes disorder and what policing procedures should be instituted is never made explicit. We see “broken window policing” that ranges from polite reminders to ticketing and fines to stop and frisk. The outcomes, too, are somewhat unclear, is a reduction in crime, a reduction in misdemeanors, or serious violent crime. As a result, supportive studies can be found on both sides of the funding/defunding argument. As George Kelling, one of the two authors of “Broken Windows,”  writes  in 2015

“…broken windows was never intended to be a high-arrest program. Although it has been practices as such in many cities, neither Wilson nor I ever conceived of it in those terms. Broken windows policing is a  highly discretionary set of activities that seeks the least intrusive means of solving a problem  – whether that problem is street prostitution, drug dealing in a park, graffiti, abandoned buildings or actions such as public drunkenness.” [emphasis added]

What if we change the built environment?

Prior studies looking at violent crime around abandoned buildings or vacant lots in Philadelphia looked at crime rates in those areas, comparing buildings and lots that were repaired and cleaned up to those that were not. In three months, there was a 39% reduction in gun violence associated with building improvement; a more extended nine-year study demonstrated a smaller 5% reduction in gun violence with cleaning up vacant lots. Researchers estimate that every dollar spent on remediation resulted in $5 to $26 in savings to taxpayers. Those early pilot studies set the stage for this current report, again from Philadelphia, on how improving the built environment impacts crime.

The current study

Philadelphia’s Basic Systems Repair Program (BSRP) provides low-income homeowners with grants to repair damage to their homes. With many of these homeowners living at or below poverty levels, home repair is not a priority. This condition can be found in numerous urban centers; New York, Newark, Detroit all come readily to mind. These grants of up to $20,000 require an application and involve a waiting period of roughly two-and-a-half years – applicants are motivated. The researchers had access to the applications, crime statistics from the Philadelphia police creating a single variable to categorize seven crime categories. [1] Their outcomes, broken down by blocks, covered the period 2006 to 2013.

6732 blocks of 19,869 blocks within the city received BSRP interventions. The homeowners were predominantly black, about 78%, or Latino, 12% with a mean annual income of about $12,000. Blocks without BSRP remediation had smaller Black populations, less unemployment, and higher annual incomes. 

broken window theory health

The effect was modest, but one can reasonably argue that it not only reduced crime but simultaneously improved the housing of the recipients. I would offer a simple proposal to those in New York City’s mayoral race – one that is anticipated to be all about law and order.

A modest proposal

broken window theory health

Why not a grand experiment. Let’s fix the NYCHA housing, use those $! billion in funds that have already been  “defunded”  from the police. And then we can watch the changes in crime. At the very worst, there will be little movement in crime but a significant improvement in housing. If you believe the science, then we should also see a crime reduction. Let’s given broken windows another try.

[1] “homicide, assault, burglary, theft, robbery, disorderly conduct, and public drunkenness”

Source: "Association Between Structural Housing Repairs for Low-Income Homeowners and Neighborhood Crime" JAMA Network Open DOI: 10.1001/jamanetworkopen.2021.17067"

View the discussion thread.

broken window theory health

By Chuck Dinerstein, MD, MBA

Director of Medicine

Dr. Charles Dinerstein, M.D., MBA, FACS is Director of Medicine at the American Council on Science and Health. He has over 25 years of experience as a vascular surgeon.

Latest from Chuck Dinerstein, MD, MBA :

shopify analytics tool


  1. How to Apply the Broken Window Theory to Your Own Messy Life

    broken window theory health

  2. The Broken Window Theory & Why It Matters.

    broken window theory health

  3. Pin by Paulo Ramos on Projects in 2020

    broken window theory health

  4. The Broken Windows Theory

    broken window theory health

  5. The Broken Window Theory

    broken window theory health

  6. Researchers find little evidence for 'broken windows theory,' say

    broken window theory health



  2. #LUNC Broken Window Theory

  3. A Broken Windows Theory A Spiral of Disorder. Fully generated by AI

  4. Broken Window Demo

  5. Window Theory (2005) Trailer

  6. Broken windows theory


  1. Broken Windows Theory: History, Meaning, and Controversy

    This theory of broken windows was introduced in an article in 1982 by George Kelling and James Q. Wilson, but the original research dates back to the late 1960s. The theory states that ...

  2. Broken Windows Theory

    The broken windows theory states that visible signs of disorder and misbehavior in an environment encourage further disorder and misbehavior, leading to serious crimes. The principle was developed ...

  3. The Broken Windows Theory: Origins, Issues, and Uses

    The broken windows theory was proposed by James Q. Wilson and George Kelling in 1982, arguing that there was a connection between a person's physical environment and their likelihood of committing a crime. The theory has been a major influence on modern policing strategies and guided later research in urban sociology and behavioral psychology.

  4. An empirical application of "broken windows" and related theories in

    Broken windows theory (BWT) proposes that visible signs of crime, disorder and anti-social behaviour - however minor - lead to further levels of crime, disorder and anti-social behaviour. While we acknowledge divisive and controversial policy developments that were based on BWT, theories of neighbourhood disorder have recently been proposed to have utility in healthcare, emphasising the ...

  5. Broken (windows) theory: A meta-analysis of the evidence for the

    The criminological "broken windows" theory (BWT) has inspired public health researchers to test the impact of neighborhood disorder on an array of resident health behaviors and outcomes. This paper identifies and meta-analyzes the evidence for three mechanisms (pathways) by which neighborhood disorder is argued to impact health, accounting ...

  6. Broken (windows) theory: A meta-analysis of the evidence for the

    The criminological "broken windows" theory (BWT) has inspired public health researchers to test the impact of neighborhood disorder on an array of resident health behaviors and outcomes. This paper identifies and meta-analyzes the evidence for three mechanisms (pathways) by which neighborhood disorder is argued to impact health, accounting for ...

  7. An empirical application of "broken windows" and related theories in

    Broken windows: a theory of disorder in neighbourhoods. Broken windows theory (BWT), as a social-psychological theory of urban decline, was originally developed almost 40 years ago by Wilson and Kelling . Proponents of this theory argue that both physical disorder (e.g., broken windows, graffiti, litter) and social disorder (e.g., vandalism ...

  8. Broken windows theory

    broken windows theory, academic theory proposed by James Q. Wilson and George Kelling in 1982 that used broken windows as a metaphor for disorder within neighbourhoods. Their theory links disorder and incivility within a community to subsequent occurrences of serious crime.. Broken windows theory had an enormous impact on police policy throughout the 1990s and remained influential into the ...

  9. Broken (windows) theory: A meta-analysis of the evidence for the

    The criminological "broken windows" theory (BWT) has inspired public health researchers to test the impact of neighborhood disorder on an array of resident health behaviors and outcomes. This paper identifies and meta-analyzes the evidence for three mechanisms (pathways) by which neighborhood disorder is argued to impact health, accounting for methodological inconsistencies across studies.

  10. An empirical application of "broken windows" and related ...

    Background: Broken windows theory (BWT) proposes that visible signs of crime, disorder and anti-social behaviour - however minor - lead to further levels of crime, disorder and anti-social behaviour. While we acknowledge divisive and controversial policy developments that were based on BWT, theories of neighbourhood disorder have recently been proposed to have utility in healthcare ...

  11. 'Broken hospital windows': debating the theory of spreading disorder

    Broken windows: A theory of spreading disorder in neighborhoods. Almost 40 years ago, Wilson and Kelling famously used broken windows as a metaphor for disorder within neighborhoods, arguing that 'if a window in a building is broken and is left unrepaired, all of the rest of the windows will soon be broken' [].Hence, BWT proposes that visible signs of neighborhood disorder (e.g., broken ...

  12. Broken windows theory

    In criminology, the broken windows theory states that visible signs of crime, anti-social behavior and civil disorder create an urban environment that encourages further crime and disorder, including serious crimes. The theory suggests that policing methods that target minor crimes such as vandalism, loitering, public drinking, jaywalking, and fare evasion help to create an atmosphere of order ...

  13. Researchers find little evidence for 'broken windows theory,' say

    Daniel T. O'Brien et al. Broken (windows) theory: A meta-analysis of the evidence for the pathways from neighborhood disorder to resident health outcomes and behaviors, Social Science & Medicine ...

  14. How Broken Windows Do —and Do Not— Matter

    Long before broken windows theory was first formulated in 1982, urbanists emphasized neighborhood disorder, but as a symptom of social health rather than a cause of crime. C ustodial steps that community members take to prevent or eliminate disorder can be a key indicator of how well communities accomplish shared goals. Hundreds of cities have ...

  15. Broken Windows Theory of Policing (Wilson & Kelling)

    The Broken Windows theory, first studied by Philip Zimbardo and introduced by George Kelling and James Wilson, holds that visible indicators of disorder, such as vandalism, loitering, and broken windows, invite criminal activity and should be prosecuted. This form of policing has been tested in several real-world settings.

  16. Applying the "Broken Windows" Theory to Healthcare

    Consider the "Broken Windows Policing" theory, which has received a great deal of attention lately. This theory sets forth the idea that in urban areas, any buildings with broken windows, graffiti-covered walls, and trash on the ground signal that no one really cares about the area—and that it isn't as safe as other neighborhoods.

  17. Broken Windows, Informal Social Control, and Crime: Assessing Causality

    Broken Windows Theory. Wilson & Kelling's (1982) broken windows thesis posits that disorder and crime are causally linked in a developmental sequence in which unchecked disorder spreads and promotes crime. Both physical disorder (e.g., abandoned buildings, graffiti, and litter) and social disorder (e.g., panhandlers, homeless, unsupervised youths) exert causal effects on crime directly and ...

  18. (PDF) An empirical application of "broken windows ...

    For example, the "Broken Window Theory" (Wilson & Kelling, 1983) states that the visible signs of physical disorder (e.g., broken windows, graffiti, litters, etc.) and social disorder (e.g ...

  19. 'Broken hospital windows': debating the theory of spreading ...

    Unit-level dysfunction creates observable violations that, once normalized, can result in organizational failure. This article explains how applying the broken windows theory in health care can enable understanding of workarounds that may escalate from helpful adjustments in practice to a manifestation of conditions that contribute to patient harm.

  20. Using Broken Windows Theory as the Backdrop for a Proactive ...

    Using Broken Windows Theory as the Backdrop for a Proactive Approach to Threat Identification in Health Care J Patient Saf. 2021 Apr 1 ... This methodology, driven by the broken windows theory, allows us to identify process inefficiencies before they manifest and open the door for the occurrence of errors and adverse events.

  21. Fixing a 'Broken Window'

    A new study looks at fixing the window, rather than pursuing the one who threw the rock. " A piece of property abandoned, weeds grow up, a window is smashed.". That line comes early in " Broken Windows ," one of the most widely quoted articles on policing. The theory offered up by those authors has been linked with changes in policing ...

  22. Exploring The Impact Of The Broken Window Theory On Society Today

    The Broken Window Theory is a criminological concept which suggests that visible signs of crime, anti-social behavior, and civil disorder create an urban environment that encourages further crime and disorder, including serious crimes. The theory was first introduced by social scientists James Q. Wilson and George Kelling in 1982.