Violent crime, police presence and poor sleep in two low-income urban predominantly Black American neighbourhoods

Abstract

Objective

To examine violent crime in relation to sleep and explore pathways, including psychological distress, safety perceptions and perceived police presence, that may account for associations.

Methods

In 2018, 515 predominantly Black American (94%) adults (Pittsburgh, Pennsylvania, USA) provided survey data: actigraphy-assessed sleep duration and wakefulness after sleep onset (WASO). We estimated pathways from violent crime (2016–2018) to sleep through psychological distress, perceptions of safety and perceived adequacy of police presence.

Results

WASO was most strongly associated with violent crimes that were within 1/10 mile of the participant’s home and within the month preceding the interview. Violent crimes were associated with lower perceived safety (β=−0.13 (0.03), p<0.001) and greater WASO (β=5.96 (2.80), p=0.03). We observed no indirect associations between crime and either WASO or sleep duration through any of the tested mediators. Crime was not associated with sleep duration.

Conclusions

We demonstrated that more proximal and more recent violent crimes were associated with reduced perceived safety and worse WASO. Differential exposure to violent crime among Black Americans may contribute to health disparities by reducing residents’ perceived safety and sleep health.

INTRODUCTION

Sleep is critical for optimal health, and multiple dimensions of poor sleep have been linked with adverse mental and physical health outcomes.^1,2 Racial/ethnic minorities are disproportionately burdened by both poor sleep health and chronic health disparities, such as cardiometabolic disorders.^3,4 In particular, Black Americans are more likely to have shorter objectively measured sleep duration than non-Hispanic White Americans.⁵ Multiple factors can influence these sleep dimensions, including physical and mental health, socioeconomic status and stress in the physical and social environments. Black Americans are also more likely to live in socioeconomically disadvantaged neighbourhoods with higher crime rates than White Americans.⁶ The challenges of living in disadvantaged neighbourhoods can strain mental and physical health as well as sleep through increased crime, stressful conditions and disorder.^7–9

Much of the extant work on crime and sleep has focused on perceptions of crime and perceived safety and self-reported sleep outcomes. Although these findings indicate associations between crime and sleep, they may be biased due to shared method variance when all measures are self-reported.^7,8 Underscoring this point, self-reported sleep quality has been associated with perceived safety, but not with objectively measured annual crime.⁸ Objectively measured crime has typically been measured as a total number of crimes occurring in the past year within a geographic unit.^8–11 However, the timing and proximity of violent crimes to one’sresidence may be particularly relevant to sleep: more recent crimes and those in closer proximity to individuals may have greater salience, triggering feelings of threat or vigilance.¹² Consistent with this hypothesis, using crime data from 2011 to 2013, the most recent crimes and those occurring closest to resident homes have been associated with less perceived safety.¹³

The underlying mechanisms between neighbourhood crime and sleep also remain unclear. Living in a neighbourhood with a high crime rate may foster psychological distress and feeling unsafe, both of which are associated with poor sleep health.^7,14 Another path through which violent crime may increase vigilance is perceived inadequacy of or exaggerated neighbourhood police presence^14,15 since residents of high-crime neighbourhoods may be more likely to encounter and interact with police than if they lived in lower crime areas.^16,17

Among Black Americans living in urban neighbourhoods, the presence of police may foster either feelings of safety or vigilance, depending on a numberof factors and past experiences. For example, while police presence may assuage residents’ vigilant responses to crime by increasing senses of safety and protection, a persistent history of racism and police unjustly treating Black Americans¹⁸ could have the opposite effect of increasing vigilance. Indeed, proactive policing has been linked with poor health, especially among minority populations.^19–22 Recently, the American Public Health Association issued a policy statement that the physical and psychological violence that is conducted by law enforcement systems is a public health issue that disproportionately harms the health of marginalised populations.¹⁹

We examined how proximity and timing of crime associates with objectively measured sleep in predominantly Black American low-income neighbourhoods. In addition, we explored potential mechanisms by testing mediation of this association by psychological distress, perceived safety and perceptions about the adequacy of police presence. We hypothesised that the impact of violent crime on sleep indirectly operates through psychological distress, perceived safety and perceptions about the adequacy of police presence.

METHODS

Study population and participants

PHRESH Zzz (Pittsburgh Hill/Homewood Research on Neighborhoods, Sleep, and Health) is part of a larger longitudinal study designed to examine effects of neighbourhood changes in the built and social environments on health behaviours of residents in two, low-income, predominantly Black American Pittsburgh neighbourhoods, the Hill District and Homewood. Households were randomly selected in each neighbourhood in 2011, prior to neighbourhood investments that concentrated in the Hill District, and have been followed since. The primary food shopper 18 or older in each household was recruited to the panel, to address the original focus on food shopping and availability. The present analyses are based on respondent interview data and objectively measured sleep, both collected in 2018 (May through November), as well as City of Pittsburgh crime data from 2016 to 2018. Details of the PHRESHZzz study design and procedures are described elsewhere.²³ All study protocols were approved by our institution’s Human Subjects Protection Committee.

Sleep outcomes

The Actigraph GT3x+, a wrist-worn device that has been validated to measure sleep/wake rhythms, was used to provide objective assessments of sleep duration and wakefulness after sleep onset (WASO).²⁴ Participants were asked to wear the actigraph for 7 consecutive days. We calculated sleep outcomes by averaging data across all valid nights of wear. The average number of nights of actigraphy for the analytic sample was 6.8, SD=0.6, range=4.0–7.0. Bedtimes and wake-times reported in sleep diaries defined the sleep interval and were verified by visual inspection of the actigraphy tracings. Actigraphic sleep data were scored using the GGIR R-Package which uses the raw accelerometer signal to identify sleep and wake periods. This scoring method has been validated against polysomnography and demonstrated 83% accuracy for identifying sleep epochs.²⁵ Sleep duration and WASO were selected a priori as sleep outcomes; they have been previously linked with key health outcomes²⁶ and are reliably assessed via actigraphy. Sleep latency was not considered a variable of interest, as prior work has shown that sleep latency is not reliably measured via actigraphy.²⁷

Sleep duration is the total number of minutes spent sleeping during the sleep interval (assessed via diary-reported bedtimes and wake-up times). The Cole-Kripke algorithm does not include WASO as part of the sleep duration estimate. Although prior work has demonstrated that both ‘short’ and ‘long’ sleep durations are associated with adverse health outcomes,^28,29 we modelled this variable continuously, as less than 1% of the current sample could be categorised as ‘long sleepers’ (greater than 9 hours), and to retain statistical power.

WASO is the total number of minutes awake after sleep onset based on actigraphy records. WASO was analysed as a continuous variable, with higher values indicating more wakefulness during the sleep period.

Neighbourhood-level crime

The City of Pittsburgh provided incident-level crime data that included all reported crimes for the 2 years preceding survey administration. We focused on three most serious forms of violent crimes (aggravated assault, homicide and simple assault). We calculated street network distances from each household to crime locations using ArcGIS 10.2. We geocoded 95% of the incidents using address information.

To assess how incidents of violent crime were associated with resident sleep, we summed violent crimes that occurred within 1/10-mile, 1/4-mile, 1/2-mile and 3/4-mile radial distances from each household address, and within 1, 3, 6, 12 and 24 months of interview date, based upon our prior results of crime and perceived safety. We calculated 20 continuous crime measures total, for each combination of timing and proximity (eg, crimes within 1 month and 1/10th mile), specific to each survey respondent.

Hypothesised mediators

Psychological distress was measured using the well-validated Kessler 6 (K6) scale,²⁵ (alpha =0.83). K6 is a 6-item survey that screens for clinical and subclinical psychological distress with questions about time spent in various emotional states in the past weeks on a 5-point scale from ‘all of the time’ to ‘none of the time.’ We treated K6 scores as continuous by summing the individual items (range=0–24).

Perceived neighbourhood safety was assessed using four items (eg, ‘You feel safe walking in your neighbourhood during the day’, ‘You feel safe walking in your neighbourhood during the evening’, ‘Your neighbourhood is safe from crime’ and ‘Violence is a problem in your neighbourhood’). Response options for each item ranged from 0 (strongly disagree) to 4 (strongly agree). Items were reverse-coded as necessary, and the mean was calculated. Higher scores indicate higher perceived safety (alpha=0.74).

Perceived adequacy of police presence was assessed by agreement with the question ‘A greater police presence in my neighbourhood would make me feel safer’ (rated from 1 (strongly disagree) to 5 (strongly agree) and subsequently collapsed to 0 (strongly disagree/disagree), 1 (neither agree nor disagree) or 2 (agree/strongly agree) prior to analysis).

Covariates

Variables previously associated with sleep disturbances and/or neighbourhood conditions were used as covariates, including age, gender, household annual income, marital/cohabitation status (yes/no), educational attainment (highschool, high school diploma (referent), some college and college/bachelor’s degree), presence of children in the home (any/none), neighbourhood (Hill District, Homewood or other), years in the neighbourhood and BMI which was calculated from interviewer-measured height and weight as kg/m². We did not control for race/ethnicity because 94% of the sample identified as Black American. We controlled for confounding along each of the direct and indirect paths.³⁰ We included the above-mentioned covariates because they have been associated with neighbourhood conditions, sleep outcomes, psychological distress and perceived safety.^9,31,32 We hypothesised that the covariates would also be associated with perceptions of police presence.

Month of the interview was included to account for any unobserved differences related to timing of interview.

Analytic sample

Because panel eligibility was limited to households’ primary food shopper, they are disproportionately women. In addition, older residents were more likely to be at home and available to enrol and so our sample is also disproportionately older compared with the overall age demographics of the neighbourhoods. Among 820 participants who participated in the 2018 wave of the study, we excluded participants with fewer than four nights of valid actigraphy data (n=26),²³ those who did not participate in sleep actigraphy study (n=224) and those who were missing other key study variables (n=55).

Statistical analyses

We calculated t-tests (continuous variables) and χ² tests (categorical variables) to compare outcomes and covariates in the included (n = 515) versus excluded (n = 305) participants. We calculated means and SD (continuous variables) and percentages (categorical variables) of individual-level and neighbourhood-level crimes. To identify the timing and proximity of violent crime most related to sleep measures, we used linear regression models to predict WASO and sleep duration as a function of the 20 crime measures by each time and proximity combination (eg, within 1 month and 1/10 mile), estimating separate models for counts by time and proximity. To illustrate the magnitude of the associations, we plotted a three-dimensional bar graph of the linear regression beta estimates (y-axis) by crime timing (x-axis) and distance (z-axis).

We examined indirect pathways from violent crime to sleep outcomes through hypothesised mediators (perceived safety, adequacy of police presence and psychological distress) using structural equation model (SEM). SEM is a pathway-based approach that can handle multiequation models and allows estimation among multiple effects transmitted over combinations of paths simultaneously.³³ We built a SEM employing the measure of violent crime that emerged as the best predictor of sleep outcomes in the above analyses. In addition to testing mediation, we allowed for a direct pathway from crime to sleep outcomes and adjusted for covariates in all models. Online supplemental figure 1 illustrates the relationships we hypothesise exist. We allowed sleep outcomes to covary with each other and mediators to covary with each other; these paths are not shown to facilitate readability of the figure. A statistically non-significant χ² test statistic,³⁴ root mean square error of approximation <0.06³⁵ and Comparative Fit Index (CFI) values approaching 1.0³⁶ imply that the model fits the data well. We used the robust maximum likelihood estimator which provides unbiased estimates, even when the data are not multivariate normal.^37,38 We performed descriptive analyses using Stata 15.0 (StataCorp, College Station, Texas, USA) and SEM in Mplus version 7.11.³⁹

Sensitivity analyses

We repeated our analyses twice: (1) including only Black American participants (n=30) or (2) including only men (n=98), given that the overall sample is predominantly women.

RESULTS

Participants included in our main analysis did not differ from individuals excluded, except that they had lower psychological distress scores (K6 mean 3.9 vs 5.0). In 2018, our analytic sample was on average 61 years (range: 26–94 years) of age, low income (mean per capita income US$14 700), overweight (mean BMI of 30 kg/m²) and women (81%) (table 1). Few lived in a household with children (20%) or were married or cohabiting (15.6%). While most had completed high school or some college (78%) few had graduated from college or obtained a bachelor’s degree (13%) Sleep duration averaged 5.5 hours and WASO averaged 121 min.

Table 1.

Descriptive statistics of the study population characteristics in 2018, n=515, Pittsburgh, Pennsylvania, USA

	Mean (SD) or percent
Sleep measures
Sleep duration, min	327.5 (79.7)
WASO, mins	121.2 (63.4)
Mediators
Psychological distress (K6 scale)*	3.9 (4.3)
Perceived safety	3.1 (0.8)
How much of a police presence is there in your neighbourhood?
Not enough/too much	53.2
Enough	46.8
A greater police presence in my neighbourhood would make me feel safer
Strongly disagree/disagree	28.2
Neither agree nor disagree	11.7
Strongly agree/agree	60.2
Covariates
Age (years)	60.4 (14.2)
Female sex	81.0
Household annual income (per US$1000)	14.7 (14.4)
Married or living with a partner	15.6
Education
<High school	9.3
High school	38.1
Some college	39.0
College/bachelors	13.6
Any children in household	20.8
Years in neighbourhood	31.8 (23.0)
BMI	30.8 (8.0)
Neighbourhood
Hill	62.3
Homewood	26.8
Other	10.9

^*Responses were on a 5-point scale. Scores were summed into a single score (0–24) where high values reflect psychological distress.

K6, Kessler 6; WASO, wakefulness after sleep onset.

The average number of violent crimes calculated cumulatively across time and distance ranged from about 1 for the closest and most recent (within 1/10 mile and 1 month) incidents to 278 to those including the farthest and oldest incidents (within 3/4 mile and 2 years) (table 2).

Table 2.

Number of violent crimes by distance from resident’s home and time frame relative to respondent interview in 2018^*(2016–2018), N=506, Pittsburgh, Pennsylvania, USA

	Mean (SD)	Min, Max
Within 1/10 mile & 1 month	0.6 (1.0)	0, 5
Within 1/10 mile & 3 months	1.6 (2.2)	0, 15
Within 1/10 mile & 6 months	2.8 (3.4)	0, 21
Within 1/10 mile & 1 year	5.7 (5.9)	0, 31
Within 1/10 mile & 2 years	9.1 (8.7)	0, 45
Within 1/4 mile & 1 month	2.8 (2.4)	0, 13
Within 1/4 mile & 3 months	7.4 (5.4)	0, 25
Within 1/4 mile & 6 months	13.6 (9.1)	0, 54
Within 1/4 mile & 1 year	26.5 (17.9)	0, 83
Within 1/4 mile & 2 years	42.2 (27.9)	0, 136
Within 1/2 mile & 1 month	9.0 (5.3)	0, 33
Within 1/2 mile & 3 months	24.8 (13.7)	0, 74
Within 1/2 mile & 6 months	46.0 (24.7)	0, 153
Within 1/2 mile & 1 year	88.3 (46.5)	0, 249
Within 1/2 mile & 2 years	140.1 (74.0)	0, 415
Within 3/4 mile & 1 month	18.0 (8.9)	0, 48
Within 3/4 mile & 3 months	50.5 (23.4)	0, 108
Within 3/4 mile & 6 months	93.3 (42.7)	0, 215
Within 3/4 mile & 1 year	176.7 (78.7)	0, 367
Within 3/4 mile & 2 years	277.9 (124.2)	0, 607

^*Counts of crimes obtained from Pittsburgh Police Department and aggregated by timing preceding interview and radial distance from resident household.

Figure 1A and B show plots of the magnitude of the linear regression beta estimates (y-axis) by crime timing (x-axis) and distance from participants’ residence (z-axis). The height of the bar on y-axis indicates the magnitude of the beta estimate. We did not observe a pattern of association between crime timing and proximity with sleep duration. However, WASO was most strongly associated with crimes that occurred within 1/10 mile from the participant’s home, and within 1 month preceding the interview (r=0.13 p=0.002). Therefore, we used crime counts within 1/10 of a mile and 1 month of survey in our SEM.

Figure 2.

Structural equation model results of pathways from violent crime to sleep through psychological distress, police presence and perceived safety in 2018, n=515, Pittsburgh, Pennsylvania, USA.

Figure 2 presents statistically significant pathways for the SEM; for clarity of presentation, we omit other paths, though the model tested all hypothesised pathways simultaneously. Higher counts of violent crime were directly associated with feeling unsafe (β=−0.13 (SE 0.03), p<0.001) and with WASO (β=5.96 (2.80), p=0.03). The estimates suggest that an increase by one violent crime would decrease feeling safe by 0.1 points and increase the time awake after sleep onset by about 6 min. We observed no evidence of indirect paths from violent crime to sleep duration or WASO through the tested mediators (perceived safety, adequacy of police presence and psychological distress).

Figure 1.

Beta estimates from 20 models estimating sleep outcomes as a function of violent crime measures that vary by time relative to respondent interview in 2018 and distance from participants’ home, Pittsburgh, Pennsylvania, USA. (A) Wakefulnessafter sleep onset outcome. (B) Sleep duration outcome.

Sensitivity analyses

When we restricted our sample to only Black Americans, the results were very similar to our main findings (data not shown). Likewise, the results were similar when we restricted to men (data not shown), except that only the association between counts of violent crime and feeling unsafe (β=−0.18, SE=0.06, p=0.002) was significant.

DISCUSSION

We tested multiple pathways between violent crime and sleep health for Black Americans living in low-income Pittsburgh neighbourhoods, and this analysis is the only one that examines proximity and timing of crime in relation to objectively measured sleep in adults.¹² We found that crimes occurring closest to residents’ homes and most recently were significantly associated with WASO, whereas more distal crimes (in terms of both location and timing) were not associated with WASO. None of the crime measures were associated with sleep duration; however, this may be due to the short sleep duration overall in this sample.

No single definition identifies the geographic area that best represents a neighbourhood. To our knowledge, only one study has examined how counts of crime aggregated within different distances from residents’ homes were associated with perceived safety.⁴⁰ Among 303 adults living in Winston-Salem, North Carolina, USA, the number of police service calls within 1, 1/2 and 1/8 miles, normalised by population size, had low agreement with perceived safety. However, agreement appears to have been highest when using the crime rate within the closest distance. In one previous study based in a large Midwestern city, children had later bedtimes the night following a violent crime and later bedtimes were proportionate to proximity of the crime from the child’s residential address.¹²

Our participants had high rates of poor sleep and exposure to violent crime. Average actigraphy-assessed sleep duration in this predominantly Black American sample was 5.5 hours, which is lower than other national studies using actigraphy, but consistent with prior reports showing that Black Americans have shorter sleep duration than non-Hispanic White Americans.⁴¹ In a large ethnically diverse cohort, sleep duration was on average 6.1 hours among Black Americans compared to 6.9 hours among White Americans.⁵ Residents’ sleep was also fragmented, as indicated by 2 hours of wakefulness, on average after sleep onset.

Our participants lived in neighbourhoods with comparatively higher crime rates in 2018 than the city of Pittsburgh overall. Our crime measures are person-level exposures, and so they are not directly comparable to area-level measures of crime. Pittsburgh covers an area of 58.3 mile² and in 2018, 893 violent crimes occurred, which translates to about 15 violent crimes per mile².⁴² By comparison, the 2018 rates were 34 violent crimes per mile² in the Hill District (46 crimes within 1.37 mile²) and 34 violent crimes per mile² in Homewood (107 crimes within 1.45 mile²).

Overall, we did not find significant indirect paths between violent crime and sleep, via distress, perceptions of safety or police presence, suggesting that other unmeasured factors may contribute to observed associations. Indeed, null findings could suggest that complex and competing relationships exist between crime, police and perceptions of safety and distress, particularly among Black American residents from low-income neighbourhoods. For example, proactive policing has been linked with poor health, especially among minority populations.^19–22 Recently, the American Public Health Association issued a policy statement that the physical and psychological violence that is conducted by law enforcement systems is a public health issue that disproportionately harms the health of marginalised populations.¹⁹ Police shootings of unarmed Black Americans, in particular, bring into stark light the conflict Black Americans may feel when they perceive threat from police rather than feeling protected. Increasing evidence demonstrates that these events adversely impact the mental health of Black Americans living in the community in which the killing occurred as well as nationally.⁴³

Of potential relevance to the current findings, tragically, on June 19, 2018, during our data collection period, an unarmed Black American youth was fatally shot by a police officer in Pittsburgh. Protests of hundreds of people were held throughout the summer.⁴⁴ This highly publicised and traumatic event along with other incidents that we are not aware of could have impacted our findings. Future work should seek to assess the multifaceted relationships between police and Black Americans and how they affect perceived safety.

Nevertheless, our findings are consistent with the growing literature that links neighbourhood disadvantage and disorder to poor sleep outcomes.^7–9 Yet, this study is unique in that it focuses on residents living in predominantly Black American, urban communities with high violent crime rates. Such communities are important to study as they are racially and socioeconomically segregated areas—where health (including sleep) disparities are intertwined. Our findings provide robust evidence that violent crime is an independent correlate of sleep health in a vulnerable population and therefore an important intervention target.

The study has several limitations. Participants excluded from this analysis reported less distress than those included in the analyses, which may have biased our results. Asking respondents about how safe they would feel with a greater police presence in their neighbourhood may not reflect residents’ perception of unfair treatment by police, and it is only a single-item measure. In addition, residential selection could bias our results, as unobserved characteristics (eg, health consciousness⁴⁵) may relate to an individual’s choice of residential location. Our study was set in low-income, racially isolated urban neighbourhoods; therefore, findings may not be generalisable to other neighbourhoods with residents who have different socio-demographic profiles.

Public health implications

This work adds evidence that among Black Americans living in urban low-income neighbourhoods, neighbourhood violent crime may adversely impact perceptions of safety and sleep health. Crime timing and proximity should be considered in research that linkscrime to sleep. Neighbourhood interventions that reduce crime and improve resident perception of safety and sleep remain critical for the well-being of communities and may reduce health disparities in the USA. Non-policing strategies to consider include allocating police funding to emergency and social services, restorative justice, unarmed mediation and intervention by community members.⁴⁶

What is already known on this subject.

• Black Americans are disproportionately burdened by poor sleep health.

• Black Americans are also more likely to live in neighbourhoods with higher crime rates than White Americans.

• Timing and proximity of crime have been largely overlooked in neighbourhood effects on sleep health.

• Underlying mechanisms between neighbourhood violent crime and sleep are unclear.

What this study adds.

• We modelled paths from violent crime to sleep through psychological distress, safety and police presence.

• Violent crime was associated with poor perceived safety and sleep.

• Violent crimes that occurred closest to residents’ homes and most recently were the most strongly associated with sleep.